Evaluation of the Effects of Pavement Marking Characteristics on Detectability by Adas Machine Vision

Project No. 20-102 (06)

EVALUATION OF THE EFFECTS OF PAVEMENT MARKING CHARACTERISTICS ON DETECTABILITY BY ADAS MACHINE VISION

FINAL REPORT

Prepared for the National Cooperative Highway Research Program Transportation Research Board of The National Academies of Sciences, Engineering, and Medicine

Disclaimer

The information contained in this report was prepared as part of National Cooperative Highway Research Program Project 20-102(06).

SPECIAL NOTE: This report IS NOT an official publication of the National Cooperative Highway Research Program, the Transportation Research Board, or the National Academies of Sciences, Engineering, and Medicine.

Adam M. Pike, Timothy P. Barrette Texas A&M Transportation Institute The Texas A&M University System College Station, Texas

Paul J. Carlson Road Infrastructure Inc. College Station, TX

May 2018

ACKNOWLEDGMENT OF SPONSORSHIP This study was conducted for the National Cooperative Highway Research Program (NCHRP) Project 20-102(06) “Road Markings for Machine Vision” by the Texas A&M Transportation Institute, a member of The Texas A&M University System. The NCHRP is supported by annual voluntary contributions from the state Departments of Transportation.

DISCLAIMER

The opinions and conclusions expressed or implied in this report are those of the research agency that performed the research and are not necessarily those of the Transportation Research Board or its sponsoring agencies. This report has not been reviewed or accepted by the Transportation Research Board Executive Committee or the National Academies of Sciences, Engineering, and Medicine or edited by the Transportation Research Board.

AUTHOR ACKNOWLEDGMENTS

The research team would like to thank the following for their contributions of materials and expertise to research: the 3M Company, Ennis-Flint, and Gamma Scientific.

CONTENTS

List of Figures and Tables ...... v Acknowledgments ...... Error! Bookmark not defined. Abstract ...... viii Executive Summary ...... 1 Chapter 1. Background ...... 5 Algorithms and Information Complexity ...... 5 System Requirements ...... 8 Testing9 Performance Metrics ...... 10 Other Systems ...... 10 Effect on Users ...... 11 Summary ...... 11 Chapter 2. Research Approach ...... 13 Facilities ...... 13 Roadway and Ambient Lighting Conditions ...... 14 Equipment ...... 15 Vehicles...... 15 Machine Vision System ...... 15 Data Acquisition System...... 16 Pavement Marking Color and Retroreflectivity Characterization ...... 17 CCD Luminance Camera ...... 17 Pavement Marking Samples ...... 18 Pavement Marking Properties ...... 20 Daytime Conditions ...... 20 Nighttime Conditions ...... 28 Nighttime Glare Conditions ...... 37 Chapter 3. Data Collection and Analysis ...... 43 MV Data Collection ...... 43 Analysis ...... 48 Consideration of Factors Affecting MV Marking Evaluations...... 48 Evaluation of Marking Performance Relative to Material Properties ...... 76

iii Chapter 4. Findings, Recommendations, and Suggested Research ...... 87 Findings ...... 87 MV System ...... 87 Influencing Factors ...... 88 Marking Characteristics ...... 89 Recommendations ...... 92 Study Limitations ...... 93 Suggested Research ...... 93 References ...... 95 Appendix A: Tabular Material Properties ...... A-1 Appendix B: Marking Images ...... B-1 Phase I Data Collection ...... B-1 Phase I Lighted Section Samples ...... B-5 Phase II Data Collection ...... B-7 Appendix C: Box-and-Whisker Plots Using Additional Material Properties ...... C-1

iv LIST OF FIGURES AND TABLES

Figure 1. Lane Tracking Algorithm Showing the Estimated Lane Boundary against Shadows, Obstacles, and Misleading Markings [9] ...... 5 Figure 2. False Alarm Caused by Lack of Lane-Curvature Data ...... 7 Figure 3. Challenges in Detecting Faded Lane Markings ...... 8 Figure 4. Misdetection Caused by Strong Shadows ...... 8 Figure 5. RELLIS Campus Test Areas ...... 13 Figure 6. Isolux Plot of Roadway Lighting Illuminance ...... 14 Figure 7. Water Distribution System on Semi-Trailer ...... 14 Figure 8. Ford Explorer and F-150 Used for Data Collection ...... 15 Figure 9. Mobileye Camera from Exterior and Interior of Ford Explorer ...... 16 Figure 10. PolySync Screen Used for Data Reduction ...... 17 Figure 11. CCD Camera Output ...... 18 Figure 12. Spectrophotometer CIE Y Measurements ...... 22 Figure 13. Retroreflectometer Coefficient of Diffuse Illumination (Qd) Measurements ...... 23 Figure 14. CCD Luminance during Daytime Dry Conditions, Phase I ...... 24 Figure 15. CCD Luminance during Daytime Dry Conditions, Phase II ...... 25 Figure 16. CCD Luminance Measurements during Daytime Wet Conditions, Phase I ...... 26 Figure 17. CCD Luminance Measurements during Daytime Wet Conditions, Phase II ...... 27 Figure 18. Daytime Wet CCD Image ...... 28 Figure 19. Marking Dry Retroreflectivity, Phase I ...... 29 Figure 20. Marking Dry Retroreflectivity, Phase II ...... 30 Figure 21. Marking Recovery Retroreflectivity, Both Observation Periods ...... 31 Figure 22. CCD Luminance Measurements during Nighttime Dry Conditions, Phase I ...... 33 Figure 23. CCD Luminance Measurements during Nighttime Dry Conditions, Phase II ...... 34 Figure 24. CCD Camera Display during Nighttime Dry Data Collection ...... 35 Figure 25. Phase II CCD Luminance during Nighttime Wet Conditions ...... 36 Figure 26. CCD Interface during Nighttime Wet Data Collection ...... 37 Figure 27. CCD Luminance during Nighttime Dry Conditions with Glare ...... 38 Figure 28. CCD Interface during Nighttime Dry Glare Data Collection ...... 39 Figure 29. CCD Luminance during Nighttime Wet Conditions with Glare ...... 40 Figure 30. CCD Interface during Nighttime Wet Glare Data Collection ...... 41 Figure 31. Daytime Dry Phase II Speed Assessment Using Explorer ...... 49 Figure 32. Daytime Dry Phase II Speed Assessment Using F-150 ...... 50 Figure 33. Nighttime Dry Phase II Speed Assessment Using Explorer ...... 52 Figure 34. Nighttime Dry Phase II Speed Assessment Using F-150 ...... 53 Figure 35. Daytime, Dry, Edge Line Marking Phase I Speed Assessment Using Explorer ...... 55 Figure 36. Nighttime, Dry, Edge Line Marking Phase I Speed Assessment Using Explorer ...... 56 Figure 37. Daytime, Dry, Broken Marking Phase I Speed Assessment Using Explorer ...... 58 Figure 38. Nighttime, Dry, Broken Marking Phase I Speed Assessment Using Explorer ...... 59 Figure 39. Daytime Dry Phase II Cloud Assessment Using Explorer ...... 61 Figure 40. Daytime Dry Phase II Cloud Assessment Using F-150 ...... 62 Figure 41. Daytime Wet Phase II Cloud Assessment ...... 63 Figure 42. Daytime Wet Phase I Cloud Cover Assessment...... 64

v Figure 43. Daytime Wet Glare Phase I, Winter 2016, Facing South (left), North (right)...... 65 Figure 44. Solid vs. Broken Markings during Daytime Dry Conditions ...... 66 Figure 45. Solid vs. Broken Markings during Nighttime Dry Conditions...... 67 Figure 46. Solid vs. Broken Markings during Daytime Wet Conditions...... 69 Figure 47. Solid vs. Broken Markings during Nighttime Wet Conditions ...... 70 Figure 48. Standard vs. Contrast Marking under All Weather Conditions ...... 72 Figure 49. Phase I Overheard Lighting Section ...... 73 Figure 50. Luminance Contrast Ratio under Streetlight Presence ...... 75 Figure 51. Average Rating vs. Y Contrast Ratio by Sample during Daytime Dry Conditions, Explorer Only...... 77 Figure 52. Average Rating vs. Y Contrast Ratio by Sample during Daytime Dry Conditions, F- 150 Only...... 78 Figure 53. Average Rating vs. Y Contrast Ratio by Sample during Daytime Wet Conditions, Explorer Only...... 80 Figure 54. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Dry Conditions, Explorer Only...... 82 Figure 55. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Dry Conditions, F- 150 Only...... 83 Figure 56. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Wet Conditions, Explorer Only...... 84 Figure 57. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Dry Glare Conditions, Explorer Only ...... 85 Figure 58. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Wet Glare Conditions, Explorer Only ...... 86

Table 1. Summary of Marking Samples ...... 19 Table 2. Marking Length and Spacing ...... 20 Table 3. Disaggregate Counts of Explorer Daytime Dry Observations ...... 43 Table 4. Disaggregate Counts of Explorer Daytime Wet Observations ...... 44 Table 5. Disaggregate Counts of Explorer Nighttime Dry Observations ...... 45 Table 6. Disaggregate Counts of Explorer Nighttime Wet Observations ...... 46 Table 7. Disaggregate Counts of Explorer Nighttime Glare Observations ...... 47 Table 8. Disaggregate Counts of F-150 Observations ...... 47 Table 9. Average confidence rating percent change from 50 mph ratings, Phase II markings under dry, daytime conditions ...... 51 Table 10. Percent change from 50 to 65 mph, Phase II Dry, Nighttime conditions ...... 54 Table 11. Percent rating change from 50 mph ratings, Phase I edge line markings under dry conditions ...... 57 Table 12. Percent rating change from 50 mph ratings, Phase I broken markings under dry conditions ...... 60 Table 13. Percent Change from Solid to Broken Markings during Dry Conditions...... 68 Table 14. Percent Change from Solid to Broken Markings, Wet Conditions...... 71 Table 15. Percent Change from Standard to Contrast Marking ...... 72 Table 16. Dry MV System Geometry Luminance (Lv) Comparison ...... 73 Table 17. Count of Observations in Lighted Section ...... 74 Table 18. Average Sample Detection Confidence Rating in Lighted Section ...... 74

vi Table 19. Lighted vs. Unlighted Performance ...... 75

vii ABSTRACT Advanced driver assistance systems (ADAS) such as lane-departure warning (LDW) and lane-keeping assistance (LKA) are becoming more common on new vehicles. LDW and LKA typically use machine vision (MV) technology in the form of cameras to detect longitudinal pavement markings. Standards and policies on pavement marking design and maintenance have been developed with the human driver in mind (more specifically, the human vision system). Vehicles equipped with LDW and LKA are already on the road and their numbers are projected to continue to increase. The most commonly heard highway infrastructure comment regarding the effectiveness of LDW and LKA is that the pavement markings need to be maintained to a good state of repair. This project was developed to help understand and define how pavement markings could be designed and maintained to provide reliable MV detection. This research explored the effect of longitudinal pavement marking quality on the detectability of pavement markings by MV systems. White and yellow pavement markings with a variety of performance levels (specifically, color and retroreflectivity) meant to represent a range of in-service markings were used to evaluate the reliability of MV detection. The detection confidence ratings from aftermarket Mobileye devices were recorded and used as the key measure of effectiveness. The testing occurred under eight scenarios representing various lighting and roadway conditions. The key results of this study suggest that in order to achieve consistently high MV detection confidence ratings, the contrast ratio of the longitudinal pavement markings relative to the adjacent pavement needs to be of an adequate level. The daytime dry testing showed that all markings with a CIE Y value of 23 or higher (1.6 contrast ratio) resulted in adequate MV detection confidence ratings. The nighttime dry testing showed that all markings with a retroreflectivity level of 34 mcd/m2/lux or higher (2.5 contrast ratio) provided adequate MV detection confidence ratings. During daytime wet conditions, the glare from the sun severely impacted the MV detection confidence ratings and no specific findings were developed. For nighttime wet conditions, the results showed that markings with a wet recovery retroreflectivity level of 4 mcd/m2/lux or higher (2.1 contrast ratio) provided adequate MV detection confidence ratings. Other factors were also examined that provide some additional understanding of how various factors impact MV detection rates.

viii

EXECUTIVE SUMMARY Advanced driver assistance systems (ADAS) such as lane-departure warning (LDW) and lane-keeping assistance (LKA) are becoming more common on new vehicles. LDW and LKA typically use machine vision (MV) technology in the form of cameras to detect longitudinal pavement markings. Standards and policies on pavement marking design and maintenance have been developed with the human driver in mind (more specifically, the human vision system). Vehicles equipped with LDW and LKA are already on the road and their numbers are projected to continue to increase. The most commonly heard highway infrastructure comment regarding the effectiveness LDW and LKA is that the pavement markings need to be maintained to a good state of repair. This project was developed to help understand and define how pavement markings could be designed and maintained to provide reliable MV detection. This research explored the effect of longitudinal pavement marking quality on the detectability of pavement markings by MV systems. White and yellow pavement markings with a variety of performance levels (color and retroreflectivity) meant to represent a range of in- service markings were used to evaluate the reliability of MV detection. Two aftermarket Mobileye ADAS devices were installed in separate test vehicles and adapted such that the detection confidence ratings that the LDW algorithm assigned to the pavement markings were recorded and used as the key measure of effectiveness. The testing occurred under eight scenarios representing various lighting and roadway moisture conditions: daytime dry, nighttime dry, nighttime dry with glare, daytime wet, nighttime wet, nighttime wet with glare, nighttime dry with overhead lighting, and nighttime wet with overhead lighting. The wet conditions were wet recovery conditions, i.e. the evaluation took place after the markings and pavement were wetted, but not while being wetted. Data were collected in two phases, providing some insight into seasonal characteristics (sun position) that can affect MV performance. Additional factors were also explored but were not the focus of the research. These factors include the speed at which the data were collected, ambient lighting conditions, marking pattern, and direction of travel. Data were collected during the daytime and nighttime, but daytime data collection had varying lighting conditions. These varying conditions were due to the presence (or lack thereof) of clouds or no clouds, which was documented during the different days of data collection. To understand the interactions, the MV detection confidence ratings were analyzed using several pavement marking performance contrast ratios (marking performance characteristic directly compared to adjacent pavement performance characteristic) that are generally considered to be good indicators of visibility. For daytime markings, the contrast ratios were the luminance (CIE Y), luminance coefficient under diffuse illumination (Qd), and MV geometry daytime luminance (Lv). For nighttime observations, the contrast ratios were coefficient of retroreflected luminance (RL, retroreflectivity) and MV geometry nighttime luminance (Lv). During Phase I of the collection period, 14 pavement markings were used, nine of which were white preformed tape markings (including one contrast marking), and five of which were yellow preformed tape markings. The markings were installed in pairs such that two markings were observed simultaneously, one on each side of the vehicle. All markings were evaluated as 4-inch wide markings. The contrast marking consisted of a 4-inch white marking with 2-inches of black marking on both sides of the white. A second testing area included a subset of the markings that were evaluated under continuous overhead lighting. The runways where the testing occurred are oriented in a north-south direction. The markings were applied longitudinally along the runway in the north-south direction. The Phase II data collection included 11 different

pavement markings consisting of five preformed tapes similar to those evaluated in the Phase I data collection as well as six water-borne paint pavement markings. The daytime visibility of the markings was characterized with measures of luminance (CIE Y), luminance coefficient under diffuse illumination (Qd), and MV system geometry luminance (Lv). The testing indicated that all markings that had a CIE Y value of 23 or higher provided adequate MV detection confidence ratings for features such as LDW. For the testing conducted herein, this resulted in a contrast value of 1.6. In wet daytime conditions, the results were influenced by the presence of the sun causing glare on the markings and surrounding pavement, greatly reducing the MV detection confidence ratings. There were no specific findings that could be distilled from the test results. It is worth noting that low-angle sun glare was not part of this testing. It is expected that low-angle sun glare would have similar MV detection confidence rating reduction effects in dry or wet daytime conditions. The nighttime visibility of the markings was evaluated with measures of coefficient of retroreflected luminance (RL) and the MV system geometry luminance (Lv). All markings (except one sample in one direction by one test vehicle) with a retroreflectivity level of 34 mcd/m2/lux or higher provided adequate MV detection confidence ratings. For the testing conducted herein, this resulted in a contrast value of 2.5. In wet night conditions, markings with a wet recovery retroreflectivity level of 4 mcd/m2/lux or higher provided adequate MV detection confidence ratings. For the testing conducted herein, this resulted in a contrast value of 2.1. The results of the testing showed that the continuous markings were more easily detected compared to the lane line markings (broken markings). The largest discrepancies between the continuous and broken marking detection performance were observed during the daytime data collection. Although not a primary factor of the research, several other factors were also examined to better understand their possible interactions. For instance, the testing was conducted at vehicle speeds of 40, 50, and 65 mph. During the daytime testing, the MV system detection confidence ratings generally decreased with increased travel speeds. However, the nighttime testing generally showed no impact of speed, and in some cases, even a slight improvement in the MV detection confidence rating as speed increased. Regarding testing under different levels of cloud coverage, no clear trend could be identified. Initially, it was expected that cloud cover would be associated with improved performance due to the mitigation of glare caused by the sun. This study examined only a limited number of observations under cloudy/partially cloudy conditions, which likely contributed to the lack of clear trends in this area. Also, by design, much of the testing occurred with the sun in a high position directly overhead, limiting the impact of glare. Detrimental glare was observed in some instances, especially during the winter data collection period for southbound wet testing. When present, the glare resulted in lower detection confidence ratings for the markings. Testing that included markings with and without continuous overhead lighting showed that the lighting had an adverse effect on MV detection confidence ratings. The combination of the vehicle testing speed, the relatively short MV detection range (compared to human vision), and the overhead lighting patterns and color may have resulted in conditions that made the markings more difficult to detect compared to markings only illuminated with the vehicle headlamps. This study examined one particular marking as both a standard marking and a contrast marking, where the white 4-inch wide marking was paralleled by 2-inch wide black striping on

each side. The addition of the contrast marking striping resulted in mixed findings. Improved performance of the contrast marking was observed for the continuous and lane line markings, as well as under daytime, nighttime, wet, and dry conditions. The improved performance was typically for the northbound travel direction, with the largest benefit during the dry day conditions. However, the southbound data collection showed negative effects of the contrast marking in dry day data collection. This may indicate the black portion of the contrast may create glare problems instead of mitigating them when the sun is causing oncoming glare. Overall, the results of this study suggest that in order to achieve consistently high MV detection confidence ratings, the contrast ratio of the pavement markings relative to the adjacent pavement needs to be of an adequate level to facilitate detection. While the recommendations will generally yield high detection confidence, there are situations where outside influences can degrade the MV detection confidence ratings. Most notably is the influence of glare from the sun. Glare at night from oncoming vehicles also resulted in the need for a higher contrast ratio. More in-depth evaluation of glare, and ways to reduce or account for its impact are needed. The recommendations from this study are different than similar research conducted on roads open to public travel with in-service pavement markings. In general, the results from the research with in-service markings indicate that higher levels of pavement marking maintenance are needed to generate adequate MV detection confidence ratings. This difference indicates that there may be in-situ factors such as marking wear patterns that impact MV detection of pavement markings. Other factors that may have also impacted the results herein are that testing area had a relatively uniform roadway surface; roadways with conflicting messages from previously removed markings, blackout markings, crack seal, varying road surfaces, cracking, or rutting may require higher-quality pavement markings. Where insufficient contrast can be achieved by the combination of marking and pavement alone, a contrast pavement marking may be able to improve marking detection to improve MV performance. The research on the MV system indicates that the camera sees similarly to a human. Good pavement marking practices for a human driver will provide good conditions for an MV system. Both the human driver and the MV system detection of markings decreases if mixed signals are present. Pavement marking practices should provide good markings in a good state of repair without other signals that could be mistaken for longitudinal delineation. In addition, both the human driver and the MV system detection of markings decreases if glare signals are present. This glare can be from the sun, oncoming headlights, or other light sources at night. Methods to mitigate the impacts of glare need to be developed to benefit both the human driver and MV systems. These methods could be related to the pavement marking characteristics or to the MV system hardware or software.

CHAPTER 1. BACKGROUND

ALGORITHMS AND INFORMATION COMPLEXITY In the relatively recent past, machine vision has been shown to be a versatile tool in tackling a wide variety of problems facing the transportation industry. Studies have illustrated the usefulness of MV in monitoring railcar roll angle [1] and inspecting railroad tracks [2]. The trucking industry is beginning to use MV on in-cab cameras to assess driver behavior, as well as for insurance purposes [3]. Specific to automobiles, algorithms have been developed to identify a variety of infrastructure components including roadway signs [4], to discriminate between various types of lane delineators [5], and to supplement global positioning system (GPS) information in complex environments [6]. By providing warnings prior to lane-departure events, these systems have been shown to reduce lane-departure crashes, specifically head-on and single-vehicle crashes [7]. Given the affordability of cameras and considering that ADASs are more easily implemented than fully autonomous (or automated) control functions, MV algorithms have been proven popular in target detection and tracking [8]. To successfully improve driver lane-keeping behavior, algorithms must be able to interpret nuances within the region of interest. This includes dealing with strong shadows, misleading lines, obstacles, and more, as demonstrated in Figure 1.

Figure 1. Lane Tracking Algorithm Showing the Estimated Lane Boundary against Shadows, Obstacles, and Misleading Markings [9]

In the case of LDW, these algorithms must not only detect the lane itself but also extract other important data from the detected lanes. Double- or continuous-line boundaries that separate the directions of traffic, discontinuous boundaries that separate lane markings in the same direction, and merge-type markings (dense, discontinuous markings) that separate the road from the roadside parking area are only a few examples of such additional information that these algorithms need to provide for proper functionality of an LDW system [10]. In addition to identifying lane markings, previous literature has examined robust detection algorithms with the capability to operate on both structured and unstructured roadways (with and without markings, respectively) [11]. Urban environments with various lighting patterns and obstacles add yet another level of intricacy to the information that lane-detection algorithms must be able to account for [12], as well as necessitate the development of programs that are capable of not just detecting lane markings but determining which lane a vehicle is in [13]. Pavement markings such as arrows and crosswalks will also need to be interpretable as vehicles hedge toward autonomy [14]. Furthermore, high-congestion areas have been demonstrated to be problematic for some detection software since vehicles can obstruct or be mistaken for markings [15]. Various environmental conditions have been documented as causing problems for LDW systems. The presence of adverse weather conditions, such as fog or rain, are well documented in the extant literature as scenarios that are problematic for vision-based detection systems [16, 17, 18, 19]. This is due to (a) wet markings having markedly lower retroreflectivity levels in comparison to dry markings [20], and (b) increased detection ratings as retroreflectivity increases [21, 22]. Retroreflectivity has also been documented as having a significant effect on detectability in cases of medium and high rain intensities [16, 17, 23]. A report from Europe suggests that a good road marking should have a minimum maintained dry retroreflectivity of 150 mcd/m2/lux, wet retroreflectivity of 35 mcd/m2/lux, and a minimum line width of 150 mm [24]. Another European report indicates that the retroreflectivity required for LDW to work is 70 mcd/m2/lx in dry night conditions, 20 mcd/m2/lx in wet night conditions, and a marking luminance coefficient 5 mcd/m2/lx higher than the pavement surface in daylight conditions [25]. Early efforts in lane detection focused on examining individual images and classifying pixels as markings or marking-adjacent pavement [26]. Lane-detection algorithms are frequently developed using an image binarization approach, whereby pixel intensity is used to assess and identify longitudinal markings using a Hough transformation [27, 28, 29]. The algorithms that identify lane markings are statistical classifier algorithms, which rely on training on video and images to identify lane delineators from the surrounding pavement; however, some of these algorithms struggle to analyze scenarios that they are unfamiliar with. To counteract this problem, some algorithms make use of machine learning techniques such as deep learning to improve performance beyond the conditions for which they were designed [30]. Advancing technology has allowed development of algorithms that assess images more precisely than in the past (i.e., at a pixel level as opposed to a bounding box) [31]. Several studies illustrate that research is moving into multicolor processing, rather than relying on grayscale or binary images, to improve the detection algorithms [32, 33]. Following a comprehensive literature review, it was proposed that a public benchmarking dataset be created to assess the performance of algorithms [34]. Yet, as industry introduces more sophisticated ADAS applications with additional or increasing degrees of autonomous (or automated) control, such as lane keeping/centering, collision avoidance, overtaking, and turning, it needs additional information about each lane marking [35]. Although lane position and type are sufficient for some applications, such as

LDW, other applications require lane-curvature information. For example, collision warning systems can generate false alarms when the lane curvature is not known [36], as shown in Figure 2, and are generally associated with lower efficacy rates [37]. False alarms have also been shown to be more frequent at night, when the road surface is wet, and during rain events [19]. Curve- detection algorithms sometimes rely on the estimation of a series of possible curve trajectories, which can be complicated by faded lane markings or obstruction by other vehicles [37]. A field comparison of two aftermarket systems found that LDWs were more repeatable on straight segments as opposed to curves [38].

Figure 2. False Alarm Caused by Lack of Lane-Curvature Data

The amount and complexity of data and information that need to be detected and interpreted by LDW, LKA, AVs, and other similar applications demonstrates the importance of processes and algorithms that process and derive this information. Investigations of the lowest layer of an LDW system, the algorithms, have primarily focused on performance of these algorithms, with not much attention being paid to the properties of the detected lane markings. Reliability of feature detection as a function of intrinsic marking properties, ambient lighting and weather conditions, and viewing geometry is an equally important aspect of algorithm performance that must be explored if progress is to continue in this area of research. While detection methods have been shown to be effective in current LDW systems, many have only worked well with particular types of road markings or conditions [39]. To this end, unsupervised learning algorithms are enabling the development of software capable of recognizing not just lane markings but other pavement markings that are present at various points on roadways [40]. There are a wide variety of factors that a vehicle may encounter in a roadway environment which could adversely impact the performance of ADAS systems. Glare caused by oncoming headlamps [41], bright sunlight (as noted by Elon Musk in Figure 3), or reflections can also cause the system to fail [17, 39]. In some cases, lighting conditions can be overcome through the use of specialized hardware that uses high modulation speeds [42]. In addition, bright reflections caused by surface water, very faint lane markings, and zigzag lane markings have also been found to cause inaccurate readings [39, 43], and some misdetection has been caused by strong shadows created by overpasses, as shown in Figure 4 [36]. Along the same lines of problematic lane line detection, Caltrans has discontinued the use of non-reflective pavement markers (Botts Dots) as lane line guidance opting instead to install 6-inch wide highly reflective pavement markings in place of the Botts Dots, based on industry input [44].

Figure 3. Challenges in Detecting Faded Lane Markings

Figure 4. Misdetection Caused by Strong Shadows

In addition to the examples listed above influencing the performance of an image- processing algorithm to detect road lane markings, the performance is also subject to various factors, such as [26]: • Viewing geometry that defines distance to the target area of the examined scene. • Viewing angle with respect to horizontal positioning to the target area (sun location). • Lighting conditions (directness of illumination [clear vs. overcast vs. foggy]). • Physical properties of the feature in the target area (intrinsic visual properties of the white/yellow stripe, such as width, contrast, etc., as described before). • Environmental conditions (amount, rate, and type of precipitation). The above list is not comprehensive, but it covers the majority of the factors and demonstrates the importance to develop and evaluate the performance of algorithms while considering other criteria. Nevertheless, algorithms, regardless of their robustness, rely on the data they receive from cameras.

SYSTEM REQUIREMENTS Automotive cameras are built to conform to a stringent set of requirements. These requirements, especially for safety-related applications such as LDW, have made many of the challenging scenarios mentioned before an easier obstacle to overcome. Following are some of these requirements:

• Wide dynamic range: a property required to guarantee performance under major disparities in ambient lighting. This requirement translates into the ability to accurately capture visual information in conditions such as approaching headlights, glare from other vehicles, tunnel entrances and exits, and rising or setting sun. To accomplish this, it is necessary to reduce the exposure response time. • Signal-to-noise ratio: a requirement to efficiently convert light to signal and produce minimum noise. This sensitivity helps the camera to yield a good signal-to-noise ratio and thus usable images in environments with less light. • Sensitivity to near-infrared wavelengths: required in order to provide better performance at night. This becomes especially relevant because near-infrared can also be projected by special headlights as an additional lighting source since it is invisible to drivers’ eyes. • Reliability: an undeniable characteristic of any automotive part or component in order to withstand the harsh operating environment. Automotive parts, including cameras, should be manufactured and delivered under specific standards and guidelines to ensure a high level of product quality and reliability. • Competitive prices: desired for automotive systems. This is accomplished by controlling the cost of the camera component itself, as well as designing the camera to enable lower costs in the overall system while balancing other requirements mentioned before. However, in spite of these stringent requirements, the many advances that have been introduced to the automotive camera systems (e.g., high dynamic range CMOS cameras compared to CCD cameras), and improved algorithms for detection and recognition, lane marking detection is still a challenge.

TESTING In the testing regiment proposed for verification and validation of LDW, ISO 17361:2007 [45] does not explicitly state the types of road markings that the system has to detect. However, the system has to be able to pass a series of performance tests. These are performed in a test location where the “lane markings are in good condition in accordance with the nationally defined visible lane markings.” There are no requirements on the environmental conditions that the system must be capable of operating under. However, the performance testing must be carried out where the visibility range is greater than 1 km [46]. A similar approach can also be found in the National Highway Traffic Safety Administration LDW test procedure document, requiring high-contrast and uniform pavement; lane marking specifications adhering to the Manual on Uniform Traffic Control Devices (e.g., standard marking widths of 4–6 inches [47]) and considered in good condition; and avoidance of tests in inclement weather, including rain, fog, snow, hail, smoke, or ash [48]. Additionally, the National Institute of Standards and Technology has developed guidance on the proper documentation of factors when collecting and analyzing data to evaluate intelligent vehicle systems [49]. A report for the Federal Motor Carrier Safety Administration outlined several proposed requirements for LDW systems in commercial vehicles [50], such as performance of a self-test; detection of vehicle position relative to various lane marker types and levels of wear; tracking at speeds in excess of 37 mph and in various lighting conditions; use of a variety of warning thresholds; accuracy within 4 inches; and ability to track 95 percent of the time in ideal conditions, issue warnings under various curve scenarios, not issue warnings when the driver uses the turn signal, and function properly when windshield wipers are used. Beyond scenarios where LDWs must issue warnings,

LDWs may do the following: issue warnings related to direction of drift, widen the warning thresholds on curves, provide warnings based on time to lane crossing, use differential warnings based on type of lane marking, warn about the road edge in absence of lane markings, report system faults where conflicting boundaries exist, and issue warnings when turn signals are left on. Real-world road infrastructure may not necessarily follow the minimum performance requirements set by the abovementioned guidelines, especially when the tests, developments, and requirements are performed independent of intrinsic properties of lane markings. For example, worn yellow markings often have similar grayscale intensity to the road pixels [36], making detection a more challenging task [51]. In order to test the effectiveness of LDW/LKA systems under real-world scenarios, some researchers have used functional performance testing whereby a series of trips are made on roadways which are selected based on having similar geometric and operational conditions associated with roads that are over represented in lane departure crashes [52]. Although this methodology can help identify differences in the performance of ADAS systems, it lacks the repeatability needed for a standardized test. Crack sealing and discontinuous markings have also been noted as problematic for detection algorithms [53].

PERFORMANCE METRICS Several performance measures are frequently utilized in the existing literature to assess the quality of LDW systems. The rate/frequency of false alarms and efficacy rate (the ratio of alarms to actual encroachments) are used particularly often [16, 17, 23, 54], with false alarms being generally associated with less-effective systems [55]. In a U.S. Department of Transportation report outlining objective test procedures, false alarms were termed false positives and used as performance metrics alongside false negative rates (encroachment without alarm) [56], which were metrics previously outlined in the field [57]. Availability, which is the percentage of time that the system is actually able to track the lane delineators, has also been utilized in various studies [25, 56]. Performance of the algorithm can also be evaluated at a more microscopic level by considering the error distribution of the rate of change of lane position (i.e., the accuracy of the lane model) [58]. In addition to identifying proper performance metrics, another task confronting researchers and policy makers is establishing a threshold for satisfactory performance [52].

OTHER SYSTEMS As vehicles head further toward large-scale connectivity and autonomous operation, MV systems for lane detection will likely be supplemented with other systems on both the hardware side and the software side [59]. Trade magazines have recently documented technology that has been designed to allow connected vehicles to avoid high-severity, multiple-vehicle collisions [60, 61]. In the same vein, it has been noted that driver monitoring systems in vehicles are necessary to achieve Level 2 (partially automated) driving [62]. Systems such as RADAR and LiDAR will allow vehicles to sense both moving and static objects [63]. Passive license plates and LED brake lights flashing in specific patterns have also been identified as potential strategies for connected vehicles, specifically in regard to multiple-vehicle collision reduction [64, 65, 66]. In order to simplify detection algorithms, the integration of various vehicle technologies, such as adaptive cruise control, have been shown to have potential [67]. Vehicle assistance systems may involve cybernetic drivers that take control of the steering wheel to avoid crashes [68]. From the perspective of vehicle owners that are considering retrofitting a vehicle with a commercially

available system, LDW systems are likely to have higher return on investment compared to lane change assist devices (capable of partially or fully performing a lane change maneuver) [69]. A Matlab simulation recently demonstrated the ability of an algorithm to adapt to particular situations, thereby reducing the driver workload on long trips or function as a lane-keeping or LDW system under other conditions [70].

EFFECT ON USERS A paper published by SAE International indicated that 40 percent of single vehicle lane- departure crashes in the crash sample evaluated would be the crashes most likely prevented by a LDW system [71]. The 40 percent of single vehicle crashes had a pre-crash scenario of no maneuver, occurred on a paved road, and did not occur near an intersection or interchange. LDW primarily serves to prevent lane-departure crashes; however, its use has been demonstrated to reduce lane position variation and cause drivers to decrease the frequency of travel near lane edges [57]. Rumble strips are often utilized as an infrastructure-based solution to lane-departure crashes; however, simulator research has shown that LDW systems are equally effective at preventing such crashes [72]. Another simulator study indicated that LDW systems were generally less effective when higher false-alarm rates were present, but there was no change in driver behavior immediately after an individual false alarm [55]. Recent research has outlined a learning algorithm that accounts for an individual driver’s behavior to reduce false-alarm rates [73]. A field operations test demonstrated that alerts issued for vehicles that were about to cross edge line lane markings resulted in improved lane-keeping behavior, while systems that provided warnings for crossing broken lines without signaling resulted in improved signal usage [74]. A study of teenage drivers also found that lane-keeping and turn-signal behavior improved among drivers of vehicles with lane-departure and collision warning systems, but unsafe following behavior tended to increase [75]. Based on the results of a telephone survey, 71 percent of drivers whose current vehicles were equipped with LDW devices wanted the devices on their next vehicle, while 14 percent of people found the system annoying [76]. One potential reason for a relatively low level of irritation among users is that warnings, at least from aftermarket systems, are typically issued approximately 0.5 m after the marking has been crossed [38]. The timing of the alarm relative to the lane-departure event is not an ironclad policy, however, and a simulator study found a 25 percent crash reduction when the waring was triggered 1 second prior to the lane crossing, while only a 7 percent reduction was observed when the alarm was triggered 1 second after the event [77]. The sentiment that LDW is useful has been echoed by a study using a combination field operations test and follow-up survey, which also found that the system had no adverse effects on other in-vehicle activities [74].

SUMMARY The discussion presented in this section provides a summary of how LDW and other MV- based ADAS systems operate, possible benefits of the systems, and what challenges are present in detection of lane markings. However, there is more research that needs to be done to empirically establish the needs of these systems from an infrastructure standpoint, and more specifically from a pavement marking standpoint across a range of conditions. Findings from the literature review and interactions with the industry revealed that there is an overall lack of understanding of the reliability of vision-based lane detection as a function of pavement marking and environmental conditions. Given the performance of cameras and the lack of reliability when detecting lane markings in adverse lighting and weather conditions, the

importance of improving the quality and availability of lane markings has increased. The importance of this proposition is realized by the European Union. They reviewed existing national practices and available research, and also conducted industry discussions that included representatives from consumer associations, safety organizations, vehicle manufacturers, and sign and marking industries. What they found was that the combination of inadequate maintenance of roads and differences in national regulations for road markings and traffic signs across Europe were a major obstacle to the effective implementation of ADAS technologies, specifically LDW/LKA and traffic sign recognition [24]. The European Road Assessment Programme (EuroRAP) concluded that the road markings on Europe’s roads should adopt a simple 150 × 150 standard. This indicates that the lane and edge markings should be a consistent 150 mm (~5.9 inches) wide, and these markings should reflect light at 150 millicandela (formally 150 mcd/lux/m²) under dry conditions [24]. Although the recommendations provided in the EuroRAP report are specifically targeted for EU policy makers, there was no specific justification for the values provided. This plan would provide more uniform markings which has been requested by automakers and MV suppliers.

CHAPTER 2. RESEARCH APPROACH The overarching goal of this research was to identify pavement marking characteristics and their associated performance levels that impact the ability of MV systems to confidently detect the markings. Various pavement marking samples of varying performance levels were evaluated in a variety of conditions using an aftermarket MV system that was installed on two different vehicles. The following section describes the equipment and approach used to collect and reduce data for this project.

FACILITIES Data collection activities for this study were conducted at The Texas A&M University System’s RELLIS campus. The facility, which was previously an Air Force base, has a network of runways and taxiways that served as the testbed for the evaluations as the research team has substantially more control over the characteristics of the markings than would be available in a field test setting and researchers are unencumbered by the presence of other road users. Utilizing a closed course test facility does limit the study in several ways. First, the markings have not been degraded by weather (due to the duration of the study) or traffic. Second, the concrete pavement is relatively consistent throughout the facility. Concrete is generally lighter in color than asphalt, which means that marking materials will have different contrast ratios on other pavement surfaces. The test areas had a one percent cross slope to facilitate drainage. Data were collected in two phases, with Phase I data collection activity occurring during winter 2016 and Phase II data collection activity occurring during summer 2017. The specific timeframe of the data collection is an important factor due to the changes in the position of the sun and its impact on daytime data collection. Figure 5 provides an aerial image of the RELLIS campus test areas. Testing area one was used in both phases and for most test conditions. Testing area two was only used during the Phase I testing that took place under the overhead illumination. The overhead illumination consisted of 4 high-pressure sodium street lights. The lights were 30 feet tall and spaced at 160 foot intervals. Figure 6 provides an isolux plot of the illuminance from the overhead lighting.

Figure 5. RELLIS Campus Test Areas

Figure 6. Isolux Plot of Roadway Lighting Illuminance

ROADWAY AND AMBIENT LIGHTING CONDITIONS Eight scenarios representing various lighting and roadway moisture conditions were considered: daytime dry, nighttime dry, nighttime dry with glare, daytime wet, nighttime wet, nighttime wet with glare, nighttime dry with overhead lighting, and nighttime wet with overhead lighting. The wet conditions were wet recovery conditions, i.e. the evaluation took place after the markings and pavement were wetted, but not while being wetted. When appropriate, condition- specific data were matched to each condition. For example, luminance was measured separately for daytime and nighttime observations during each data collection period. To simulate wet road conditions, a truck was used to tow a modified tanker trailer that applied water to the roadway surface. This water temporarily flooded the markings and road surface. Figure 7 shows the truck.

Figure 7. Water Distribution System on Semi-Trailer

EQUIPMENT Vehicles Two vehicles, a 2015 Ford Explorer and a 2015 Ford F-150 were used to collect data for this project. Figure 8 provides images of the two test vehicles. The Explorer was used in both phases of the study to collect data at 40, 50, and 65 miles per hour during the various evaluation conditions. The F-150 was only used during Phase 2 data collection. The F-150 collected data at 50 or 65 mph in dry conditions during both day and night evaluations. During night observations, both vehicles only used low beam illumination. The headlights on both vehicles were the standard OEM headlights with halogen bulbs. During night glare testing the F-150 served as the glare vehicle with low beam illumination. The truck was stationary near the end of the test markings being evaluated, in a position representing an opposing vehicle in a two-lane two-way alignment.

Figure 8. Ford Explorer and F-150 Used for Data Collection

Machine Vision System Both vehicles used in this study were outfitted with a Mobileye 5 series advanced driver assistance system. Figure 9 shows the Mobileye system and additional forward-facing camera to capture the forward scene during data collection. The Mobileye system uses a relatively low resolution monochrome camera (<1 megapixel) that focuses on pavement markings located 30-50 feet in front of the vehicle. Initial testing to determine when the start or end of a marking test section was detected indicated the 30 to 50 foot range. A paper describing a test with the same ADAS system indicated a detection “sweet spot” between 30 and 40 feet, though detection went out further [22]. The camera has a horizontal field of view of approximately 40 degrees, and a vertical field of view of approximately 30 degrees. The system process images at 15 frames per second. The system algorithm assigns a detection confidence rating to the pavement markings on either side of the vehicle. The detection confidence rating is an integer value between 0 and 3, with 3 being the highest confidence. The systems requires a confidence value of 2 or greater in order to provide LDW assistance. Pavement markings that resulted in detection confidence ratings of 2 or higher were considered adequate for this study. Mobileye literature indicates the system cannot see better than the driver. The device setup in each vehicle required slightly different approaches to extracting the detection confidence rating assigned to each longitudinal pavement marking. The data extraction process for reach system is described in the following section.

Figure 9. Mobileye Camera from Exterior and Interior of Ford Explorer

Researchers attempted to acquire other MV systems for testing. The attempts were unsuccessful in terms of obtaining equipment with the necessary functionality to allow the researchers to determine the confidence level that the MV system had in detecting the markings. The Mobileye system tested was by far the most common system on the market when the project started. At the time of writing this report newer versions of the Mobileye system have been released that use newer hardware and software. Data Acquisition System In the Explorer, the MV system output was integrated into PolySync, a data logging system that simultaneously presents a graphical representation of the lane model developed by the MV system, the detection confidence rating overlaid on a forward viewing camera image, and other streaming data output from the MV system. The goal of the data reduction process was to develop an analysis database that would consider each time one of the vehicles passed by a longitudinal pavement marking as a unique observation. For observations made using the Explorer, screen capture video (see Figure 10) of the data collection software was manually reviewed. The researcher reviewing the video identified the most prevalent detection confidence rating for the first half and second half of each observation of the longitudinal pavement marking. These two values were then averaged to determine an overall average rating for the observation. In the F-150, automated software was used to extract detection confidence ratings at a frequency of 10 hertz over the duration of the data collection. The automated data logging software associated with the MV of the F-150 created a spreadsheet output and uploaded via cellular connection to a data cloud for remote download, which eliminated the human aspect present in the data collected using the Explorer. Using GPS points, the beginning and end of each marking section were identified in the output. The data between the beginning and end of a particular marking were then averaged over the first half and second half of the marking, and then the two halves were averaged to create an overall detection confidence rating for the marking. After the detection confidence ratings were extracted from each of the MV systems, they were then matched with field measurements of the pavement marking performance

characteristics obtained using the CCD luminance camera and the other pavement marking performance characterization equipment as described in the following subsections.

Figure 10. PolySync Screen Used for Data Reduction

Pavement Marking Color and Retroreflectivity Characterization Delta LTL-XL Mark II and Delta LTL-XL handheld retroreflectometers were used to obtain measurements of coefficient of retroreflected luminance (RL), which is indicative of visibility at night, and the luminance coefficient under diffuse illumination (Qd), which is indicative of visibility during the day. The retroreflectometers were used to evaluate each marking and the adjacent pavement at 20 foot intervals along the length of the markings in both directions. The recovery retroreflectivity, which measures the coefficient of retroreflected luminance of a pavement marking after it has been wetted, was also captured using the handheld retroreflectometer following ASTM E2177. Recovery retroreflectivity readings were conducted at 3 locations along each marking. A HunterLab MiniScan XE Plus portable spectrophotometer was used to obtain color (x, y chromaticity coordinates) and luminance (CIE Y) of the markings and pavements. This device measures data in the CIE color space. Measurements were collected using a two degree standard observer and illuminant D65. Color measurements were conducted at 5 locations along each marking. For all measurements the adjacent road surface was also evaluated. CCD Luminance Camera

A CCD luminance camera (imaging colorimeter) was used to measure the luminance (Lv) of the markings under various lighting and wetting conditions. The camera, a Prometric I29, was mounted inside the Explorer near the MV system and captured information at four distances in front of the vehicle. The geometry of the evaluation was not a standard geometry but rather a field geometry representing the geometry at which the MV system was viewing the markings. To

provide a frame of reference for each of the three nearest ranges, a ceramic Spectralon tile was placed adjacent to the location of the measurement. Figure 11 provides a screenshot of the CCD output. The output provides luminance and color information for the pavement markings and the surrounding pavement.

Figure 11. CCD Camera Output

Figure 11 shows a series of boxes in green on the left side. These boxes identify the locations where the CCD image was being analyzed. The measurement locations were at 45, 85, 125, and 165 ft away from the measurement device, with each box being 10 ft long and centered at the aforementioned distances. Ultimately, the measurements taken at the 45-ft distance were used in the analysis since they were generally reflective of the area of interest being used by the MV system. A single location was used for CCD luminance measurements in each direction of observation for each measurement condition.

PAVEMENT MARKING SAMPLES During Phase I, 14 pavement markings were used, nine of which were white preformed tape markings (including one contrast marking), and five of which were yellow preformed tape markings. The markings were installed in pairs such that two markings were observed simultaneously, one on each side of the vehicle. Table 1 provides a summary of the pavement marking materials. All markings were evaluated as 4-inch wide markings. The contrast marking consisted of a 4-inch white marking with 2-inches of black marking on both sides of the white. The side indicated in Table 1, refers to the side of the vehicle the marking was on when traveling northbound. The marking would be on the opposite side when traveling southbound. The location of each of the seven Phase I test sections is shown as Testing Area 1 in Figure 5. Testing Area 2 consisted of a subset of the markings that were evaluated under continuous overhead lighting. The runways where the testing occurred are oriented in a north-south direction. The markings were applied longitudinally along the runway in the north-south direction.

Phase II data collection evaluated seven test sections containing 11 different pavement markings. These test sections consisted of five preformed tapes similar to those evaluated in the Phase I data collection as well as six water-borne paint pavement markings. The location of each of the pavement markings evaluated in Phase II was in Testing Area 1 shown in Figure 5. During the Phase I data collection, all markings were specially manufactured pavement marking tape. The tape was produced with specific color and retroreflectivity properties to cover a wide range of pavement marking quality to simulate varying levels of wear. The quantity of optics and quality of the pigments were modified by the manufacturer to produce markings that have performance similar to that of aged markings. The majority of the tape had the standard profiled tape pattern, but some of the tape was flat due to the modifications to the color of the product. The Phase I markings were evaluated as broken (lane line) and single solid (edge line) markings. The Phase II data collection focused on solid edge line markings and included a subset of similar performance preformed tape from Phase I. The Phase II marking set was expanded to include a variety of water-borne paint markings. The paint markings utilized two water-borne based binders that were faded white or yellow. All the paint markings were thinly applied to the pavement so that the pavement surface color still showed through the material. One pair of the paint markings were just applied as binder material with no glass beads added. The other two pairs of paint markings were applied with varying rates of AASHTO Type I beads and black beauty abrasive material to provide some discoloration. Each of the marking samples had various characteristics that affect its visibility to the human eye and MV systems. The research aimed to control marking quality to help determine minimum necessary marking properties for acceptable detection by the MV system.

Table 1. Summary of Marking Samples Phase I Data Phase II Data Collection 2016 Collection 2017 Marking Label Material Color Structure Section Side Section Side WT1 Tape White Profiled 6 Right 4 Right WT2 Tape White Profiled 3 Right - - WT3 Tape White Flat 7 Left - - WT4 Tape White Flat 5 Right - - WT5 Tape White Profiled 2 Right - - WT6 Tape White Profiled 1 Right - - WT6C Tape White Profiled 7 Right - - WT7 Tape White Profiled 4 Right 2 Right YT1 Tape Yellow Flat 1 Left - - YT2 Tape Yellow Profiled 6 Left 4 Left YT3 Tape Yellow Profiled 3 Left 2 Left YT4 Tape Yellow Profiled 5 Left 3 Left YT5 Tape Yellow Profiled 4 Left - - YT6 Tape Yellow Profiled 2 Left - - WP1 Paint White Flat - - 5 Right WP2 Paint White Flat - - 6 Right WP3 Paint White Flat - - 7 Right YP1 Paint Yellow Flat - - 5 Left YP2 Paint Yellow Flat - - 6 Left YP3 Paint Yellow Flat - - 7 Left The dash symbol (-) indicates sample not studied during a given Phase

The length of each of the markings and the spacing between the markings during both phases of data collection are documented in Table 2.

Table 2. Marking Length and Spacing Phase I Data Collection Phase II Data Collection Distance Distance Distance To Distance To From Marking From Marking Sample Next Next Previous Length (ft) Previous Length (ft) Marking (ft) Marking (ft) Marking (ft) Marking (ft) WT1 130 410 130 785 420 NA WT2 170 410 130 - - - WT3 130 250 NA - - - WT4 150 410 130 - - - WT5 190 410 170 - - - WT6 NA 490 190 - - - WT6C 130 250 NA - - - WT7 130 450 150 660 395 685 YT1 NA 490 190 - - - YT2 130 410 130 725 380 NA YT3 170 410 130 785 410 670 YT4 150 410 130 670 415 725 YT5 130 450 150 - - - YT6 190 410 170 - - - WP1 - - - NA 500 200 WP2 - - - 200 500 200 WP3 - - - 200 500 280 YP1 - - - NA 500 200 YP2 - - - 200 500 200 YP3 - - - 200 500 280 -Indicates material not studied during a particular data collection period, NA indicates material was the first observed going northbound

PAVEMENT MARKING PROPERTIES This study used data obtained from two data collection periods, Phase I and Phase II. The markings characteristics for each of the marking samples were observed in each of the data collection periods. The following sections document the properties of each sample and the adjacent pavement surface in each of the two data collection periods. Several of the measurements were taken for each direction of travel to account for conditions when the directionality of the marking or observation would influence the performance. The following sections provides a graphical summary of the properties collected during this study. Appendix A provides tabular summaries of the various marking properties. Appendix B provides images of the pavement markings. Daytime Conditions Figure 12 documents the luminance (CIE Y) measurements taken using the MiniScan during Phases I (top) and II (bottom) of data collection. The value for Y is a scaled value between 0 and 100, with 0 representing a perfect black and 100 representing a perfect white. The brighter (whiter) a marking is the higher the luminance it will have. Appendix A provides plots

of the chromaticity coordinates of each of the samples in relation to the ASTM color boxes for yellow and white markings. Figure 13 contains the Qd measurements obtained through a Delta LTL-XL retroreflectometer. In addition to the spectrophotometer and retroreflectometer measurements, a CCD imaging camera was used to measure the luminance of the pavement markings and the adjacent pavement. Figure 14 documents those measurements that were taken during daytime dry conditions during the Phase I data collection period. Figure 15 presents the data from the Phase II data collection period. The visibility characteristics of both the pavement and the marking vary substantially depending on whether the samples are wet or dry. To account for this, CCD measurements were also taken after wetting the pavement using the water truck described in Chapter 2. These CCD measurements are documented in the following figures. Figure 16 presents the CCD luminance measurements during daytime wet conditions. Figure 17 presents comparable measurements taken during the Phase II data collection period. The figures illustrate that the values obtained for these measurements were extremely dependent on the lighting conditions present at the time since the values were typically higher during the Phase II data collection period.

Figure 12. Spectrophotometer CIE Y Measurements

Figure 13. Retroreflectometer Coefficient of Diffuse Illumination (Qd) Measurements

) 2 Luminance (cd/m Luminance

Figure 14. CCD Luminance during Daytime Dry Conditions, Phase I

) 2 Luminance (cd/m Luminance

Figure 15. CCD Luminance during Daytime Dry Conditions, Phase II

) 2 Luminance (cd/m Luminance

Figure 16. CCD Luminance Measurements during Daytime Wet Conditions, Phase I

) 2 Luminance (cd/m Luminance

Figure 17. CCD Luminance Measurements during Daytime Wet Conditions, Phase II

Figure 18 illustrates the CCD camera display during daytime wet data collection. Dry pavement can be seen in the upper-right corner of the image, providing a visual comparison of how the pavement markings contrast with the pavement under wet and dry conditions.

Figure 18. Daytime Wet CCD Image

Nighttime Conditions Figure 19 documents the retroreflectivity measurements during the Phase I data collection period, while Figure 20 displays the measurements from the Phase II data collection period. The pavement retroreflectivity level is also provided in both figures. Figure 21 presents the measurements for the wet recovery retroreflectivity, which captures the retroreflectivity of the pavement marking while recovering after being wetted. The wet recovery retroreflectivity of the pavement was 2 mcd/m2/lux for all test areas.

Figure 19. Marking Dry Retroreflectivity, Phase I

Figure 20. Marking Dry Retroreflectivity, Phase II

Figure 21. Marking Recovery Retroreflectivity, Both Observation Periods

The visibility characteristics observed by the CCD camera are dependent on the amount of light falling on the sample (illuminance), the geometry of the light source, and the geometry of the camera in relation to the target. Figure 22 documents the nighttime CCD luminance observations where the explorer test vehicle with low beam headlights served as the illumination source. There are large discrepancies between the luminance values from the daytime and nighttime CCD observations. This is due to the aforementioned fact that the luminance values are dependent on the amount of light (both in terms of intensity and angle) hitting the sample. Subsequently, the observations that occurred at night were substantially lower because only the vehicle headlights were providing the illumination. Figure 23 documents the luminance characteristics during the Phase II data collection period.

) 2 Luminance (cd/m Luminance

Figure 22. CCD Luminance Measurements during Nighttime Dry Conditions, Phase I

) 2 Luminance (cd/m Luminance

Figure 23. CCD Luminance Measurements during Nighttime Dry Conditions, Phase II

Figure 24 illustrates the CCD display during nighttime dry data collection.

Figure 24. CCD Camera Display during Nighttime Dry Data Collection

Figure 25 illustrates the observations of the CCD luminance captured during nighttime wet conditions. Relative to the observations taken during the daytime, these values are substantially lower since the lighting is due to the vehicle headlights as opposed to the sun. Wet CCD measurements were not taken during Phase I because the CCD camera was under repair.

) 2 Luminance (cd/m Luminance

Figure 25. Phase II CCD Luminance during Nighttime Wet Conditions

Figure 26 illustrates the CCD interface during nighttime wet data collection. In the upper- right corner of the camera view shown in Figure 26, the pavement is dry, while the remainder of the pavement shown is wet. This figure helps demonstrate the effect of moisture on pavement

color and, subsequently, on the contrast between the pavement marking and the pavement. In the upper-left portion, a vehicle can be observed. During this portion of the data collection activities, the vehicle had no lights on and was merely in position for the glare portion of the study.

Figure 26. CCD Interface during Nighttime Wet Data Collection

Nighttime Glare Conditions During the Phase II data collection period, measurements were taken at night in the presence of glare applied by the headlights of a stationary vehicle located to simulate oncoming traffic under both dry and wet pavement conditions. This data was collected as a mini-study to try and gain additional knowledge of the impact of oncoming glare, so that it could be explored in more depth in the future. Figure 27 illustrates the CCD luminance during nighttime dry conditions with glare applied from an opposing vehicle’s headlights. Figure 28 illustrates the user interface of the CCD camera during nighttime dry data collection with glare applied by oncoming headlights. Figure 28 demonstrates what is generally observed by any human driver of an automobile at night, the vision-inhibiting glare of the headlights of an oncoming vehicle.

) 2 Luminance (cd/m Luminance

Figure 27. CCD Luminance during Nighttime Dry Conditions with Glare

Figure 28. CCD Interface during Nighttime Dry Glare Data Collection

CCD measurements were also taken under nighttime wet conditions with glare applied, in a manner similar to that shown in Figure 27. Figure 29 presents the measurements of luminance captured by the CCD camera under nighttime wet conditions with glare applied from the headlights of an opposing vehicle. Of the samples that were observed under dry glare conditions, the values obtained for wet glare conditions were generally lower.

) 2 Luminance (cd/m Luminance

Figure 29. CCD Luminance during Nighttime Wet Conditions with Glare

Figure 30 illustrates the interface of the CCD camera during nighttime wet data collection with glare applied by oncoming headlights. The effect of headlights illustrated in Figure 28 is also noticeable in Figure 30. The effect appears to be potentially enhanced by the additional reflection due to water on the roadway.

Figure 30. CCD Interface during Nighttime Wet Glare Data Collection

The glare condition testing was not included in the scope of the original project. Researchers added it to see potential impacts and to generate some initial data to help influence future research. Observations of CCD MV geometry luminance (Lv) were not collected for samples WT1, YT2, YT4, and WT3 during nighttime dry glare conditions due to time constraints. Of the samples that were observed under dry glare conditions, the values obtained for wet glare conditions were fairly comparable, with samples YT3 and YT1 having slightly lower observations. As previously indicated the luminance values used were at the 45 foot distance. This distance did not fall directly in the glare path of the static evaluation.

CHAPTER 3. DATA COLLECTION AND ANALYSIS

MV DATA COLLECTION For each of the scenarios previously described, each of the marking samples were observed multiple times using the MV systems in the Explorer and F-150. Table 3 through Table 8 document the number of observations made for each of the various conditions, beginning with the daytime dry observations using the Explorer. Rather than simply showing the counts of observations by each sample, the counts are disaggregated based on cloud cover, speed, direction of travel, and marking pattern (broken or solid).

Table 3. Disaggregate Counts of Explorer Daytime Dry Observations Marking Pattern Broken Solid Cloud Condition No Clouds No Clouds Clouds Speed (mph) 40 50 65 50 65 50 65 Direction NB SB NB SB NB SB NB SB NB SB NB SB NB SB Total WT1—Phase I 3 3 4 4 2 2 5 5 1 1 0 0 0 0 30 WT1—Phase II 0 0 0 0 0 0 2 1 3 2 1 2 0 1 12 WT2—Phase I 3 3 4 4 2 2 5 5 1 1 0 0 0 0 30 WT3—Phase I 3 3 4 4 2 2 2 2 1 1 0 0 0 0 24 WT4—Phase I 3 3 4 4 2 2 3 3 1 1 0 0 0 0 26 WT5—Phase I 3 3 4 4 2 2 5 5 1 1 0 0 0 0 30 WT6C—Phase I 3 3 4 4 2 2 5 5 1 1 0 0 0 0 30 WT6—Phase I 3 3 4 4 2 2 5 5 1 1 0 0 0 0 30 WT7—Phase I 3 3 4 4 2 2 5 5 1 1 0 0 0 0 30 WT7—Phase II 0 0 0 0 0 0 2 2 3 2 1 1 0 1 12 YT1—Phase I 3 3 4 4 2 2 2 2 1 1 0 0 0 0 24 YT2—Phase I 3 3 4 4 2 2 4 4 1 1 0 0 0 0 28 YT2—Phase II 0 0 0 0 0 0 2 1 3 2 1 2 0 1 12 YT3—Phase I 3 3 4 4 2 2 4 4 1 1 0 0 0 0 28 YT3—Phase II 0 0 0 0 0 0 2 2 3 2 1 1 0 1 12 YT4—Phase I 3 3 4 4 2 2 4 4 1 1 0 0 0 0 28 YT4—Phase II 0 0 0 0 0 0 2 2 3 2 1 1 0 1 12 YT5—Phase I 3 3 4 4 2 2 4 4 1 1 0 0 0 0 28 YT6—Phase I 3 3 4 4 2 2 4 4 1 1 0 0 0 0 28 WP1—Phase II 0 0 0 0 0 0 2 3 2 2 1 0 0 0 10 WP2—Phase II 0 0 0 0 0 0 2 2 1 2 1 1 1 0 10 WP3—Phase II 0 0 0 0 0 0 2 2 2 1 1 1 0 1 10 YP1—Phase II 0 0 0 0 0 0 2 3 2 2 1 0 0 0 10 YP2—Phase II 0 0 0 0 0 0 2 2 1 2 1 1 1 0 10 YP3—Phase II 0 0 0 0 0 0 2 2 2 1 1 1 0 1 10 Total 42 42 56 56 28 28 79 79 39 34 11 11 2 7 514

Table 4. Disaggregate Counts of Explorer Daytime Wet Observations Marking Pattern Broken Solid Cloud Cover No Clouds No Clouds Some Clouds Clouds Direction NB SB NB SB NB SB NB SB Total WT1—Phase I 3 6 3 2 1 0 3 1 19 WT1—Phase II 0 0 2 3 0 0 0 0 5 WT2—Phase I 6 4 7 2 1 0 4 1 25 WT3—Phase I 3 6 0 0 0 0 0 0 9 WT4—Phase I 4 6 0 0 0 0 0 0 10 WT5—Phase I 7 4 8 2 1 0 4 1 27 WT6C—Phase I 3 6 3 2 0 0 3 1 18 WT6—Phase I 8 3 9 2 1 0 4 1 28 WT7—Phase I 5 5 6 2 1 0 3 1 23 WT7—Phase II 0 0 0 1 2 1 0 0 4 YT1—Phase I 8 3 0 0 0 0 0 0 11 YT2—Phase I 3 6 3 2 1 0 3 1 19 YT2—Phase II 0 0 2 3 0 0 0 0 5 YT3—Phase I 6 4 7 2 1 0 4 1 25 YT3—Phase II 0 0 0 1 2 1 0 0 4 YT4—Phase I 4 6 5 2 1 0 3 1 22 YT4—Phase II 0 0 2 3 0 0 1 0 6 YT5—Phase I 5 5 6 2 1 0 3 1 23 YT6—Phase I 7 4 8 2 1 0 4 1 27 WP1—Phase II 0 0 0 0 0 0 3 2 5 WP2—Phase II 0 0 0 0 1 1 2 1 5 WP3—Phase II 0 0 0 0 0 0 2 2 4 YP1—Phase II 0 0 0 0 0 0 3 2 5 YP2—Phase II 0 0 0 0 1 1 2 1 5 YP3—Phase II 0 0 0 0 0 0 2 2 4 Total 72 68 71 33 16 4 53 21 338

Table 5. Disaggregate Counts of Explorer Nighttime Dry Observations Marking Pattern Broken Solid Speed (mph) 40 50 65 50 65 Row Labels NB SB NB SB NB SB NB SB NB SB Total WT1—Phase I 0 0 0 0 0 0 2 2 2 2 8 WT1—Phase II 2 2 5 5 2 2 2 2 1 1 24 WT2—Phase I 2 2 6 6 2 2 2 2 1 1 26 WT3—Phase I 2 2 6 6 2 2 2 2 1 1 26 WT4—Phase I 2 2 5 5 2 2 2 2 1 1 24 WT5—Phase I 2 2 6 6 2 2 2 2 1 1 26 WT6C—Phase I 0 0 0 0 0 0 2 2 2 1 7 WT6—Phase I 2 2 6 6 2 2 2 1 1 1 25 WT7—Phase I 2 2 6 6 2 2 2 2 1 1 26 WT7—Phase II 0 0 0 0 0 0 2 2 2 2 8 YT1—Phase I 2 2 6 6 2 2 2 2 1 1 26 YT2—Phase I 0 0 0 0 0 0 2 2 2 2 8 YT2—Phase II 2 2 5 5 2 2 2 2 1 1 24 YT3—Phase I 0 0 0 0 0 0 2 2 2 2 8 YT3—Phase II 2 2 6 6 2 2 2 2 1 1 26 YT4—Phase I 2 2 5 5 2 2 2 2 1 1 24 YT4—Phase II 0 0 0 0 0 0 2 2 2 2 8 YT5—Phase I 2 2 5 5 2 2 2 2 1 1 24 YT6—Phase I 2 2 5 5 2 2 2 2 1 1 24 YP1—Phase II 0 0 0 0 0 0 2 2 2 1 7 YP2—Phase II 0 0 0 0 0 0 2 2 2 1 7 YP3—Phase II 0 0 0 0 0 0 2 2 2 1 7 WP1—Phase II 0 0 0 0 0 0 2 2 2 1 7 WP2—Phase II 0 0 0 0 0 0 2 2 2 1 7 WP3—Phase II 2 2 5 5 2 2 2 2 1 1 24 Total 28 28 77 77 28 28 50 49 36 30 431

Table 6. Disaggregate Counts of Explorer Nighttime Wet Observations Marking Pattern Broken Solid Speed (mph) 50 50 Direction NB SB NB SB Total WT1—Phase I 4 2 4 2 12 WT1—Phase II 0 0 3 2 5 WT2—Phase I 6 2 6 2 16 WT3—Phase I 3 2 0 0 5 WT4—Phase I 4 2 0 0 6 WT5—Phase I 7 2 7 2 18 WT6C—Phase I 3 2 3 2 10 WT6—Phase I 8 2 8 2 20 WT7—Phase I 5 2 5 2 14 WT7—Phase II 0 0 3 2 5 YT1—Phase I 8 2 0 0 10 YT2—Phase I 4 2 4 2 12 YT2—Phase II 0 0 3 2 5 YT3—Phase I 6 2 6 2 16 YT3—Phase II 0 0 3 2 5 YT4—Phase I 4 2 4 2 12 YT4—Phase II 0 0 3 2 5 YT5—Phase I 5 2 5 2 14 YT6—Phase I 7 2 7 2 18 YP1—Phase II 0 0 3 2 5 YP2—Phase II 0 0 3 2 5 YP3—Phase II 0 0 1 3 4 WP1—Phase II 0 0 3 2 5 WP2—Phase II 0 0 3 2 5 WP3—Phase II 0 0 1 1 2 Total 74 28 88 44 234

Table 7 presents the observations taken during nighttime glare conditions using the Explorer. This includes nighttime dry glare (NDG) observations and nighttime wet glare (NWG) observations. During nighttime observations, cloud cover was not considered since the moon generally does not provide enough light to substantially illuminate the pavement markings. During wet glare conditions, observations were only made at 50 mph. The Phase II glare assessment only took place on solid edge line markings. Finally, the glare observations were only made in the northbound direction of travel. Table 8 presents the observations from the F-150. The F-150 was only used during the Phase II data collection. The F-150 travelled at 50 or 65 mph, and only collected data in dry day and dry night conditions.

Table 7. Disaggregate Counts of Explorer Nighttime Glare Observations NDG NWG Sample NB NB WT1—Phase II 2 2 WT7—Phase II 2 2 YT2—Phase II 2 2 YT3—Phase II 2 2 YT4—Phase II 2 2 WP1—Phase II 2 2 WP2—Phase II 2 2 WP3—Phase II 2 2 YP1—Phase II 2 2 YP2—Phase II 2 2 YP3—Phase II 2 2 Total 22 22

Table 8. Disaggregate Counts of F-150 Observations Cloud Cover No Clouds Some Clouds Speed (mph) 50 65 50 65 Direction NB SB NB SB NB SB NB SB Total Dry, Daytime WT1—Phase II 2 1 2 1 0 1 0 1 8 WT7—Phase II 2 2 2 2 0 0 0 0 8 YT2—Phase II 2 1 2 1 0 1 0 1 8 YT3—Phase II 2 2 2 2 0 0 0 0 8 YT4—Phase II 2 2 2 2 0 0 0 0 8 WP1—Phase II 2 2 1 2 0 0 1 0 8 WP2—Phase II 2 2 2 2 0 0 0 0 8 WP3—Phase II 2 2 2 2 0 0 0 0 8 YP1—Phase II 2 2 1 2 0 0 1 0 8 YP2—Phase II 2 2 2 2 0 0 0 0 8 YP3—Phase II 2 2 2 2 0 0 0 0 8 DD Total 22 20 20 20 0 2 2 2 88 Dry, Nighttime WT1—Phase II 2 2 2 2 0 0 0 0 8 WT7—Phase II 2 2 2 2 0 0 0 0 8 YT2—Phase II 2 2 2 2 0 0 0 0 8 YT3—Phase II 2 2 2 2 0 0 0 0 8 YT4—Phase II 2 2 2 2 0 0 0 0 8 WP1—Phase II 2 2 2 2 0 0 0 0 8 WP2—Phase II 2 2 2 2 0 0 0 0 8 WP3—Phase II 2 2 2 2 0 0 0 0 8 YP1—Phase II 2 2 2 2 0 0 0 0 8 YP2—Phase II 2 2 2 2 0 0 0 0 8 YP3—Phase II 2 2 2 2 0 0 0 0 8 ND Total 22 22 22 22 0 0 0 0 88

ANALYSIS Consideration of Factors Affecting MV Marking Evaluations Prior to conducting the investigation, researchers anticipated that factors unrelated to the properties of the pavement marking samples themselves could influence the detection confidence rating of the markings by the MV system. These factors included the speed at which the observation was collected, ambient lighting conditions, marking pattern, and direction of travel. Data were collected during the daytime and nighttime, but daytime data collection had varying lighting conditions. These varying conditions were due to the presence of or lack of clouds. The cloud conditions were subjectively documented during the different days of data collection. Nearly all of the figures included in this section are divided into two directional sections to account for direction-specific sample and lighting characteristics, with the top half of the figure documenting northbound observations and the bottom half reserved for southbound observations (with the exceptions being scenarios where only northbound observations were considered). The northbound and southbound sections have each been divided into two smaller subsections, for a total of four vertically stacked subsections. The top half of each section illustrates the spread of the individual observations for each of the samples via box-and-whisker plots, while the lower half illustrates the average value of the observations for each of the scenarios. To identify the impact of the factors, the MV results were plotted against several performance contrast ratios (marking performance characteristic directly compared to adjacent pavement performance characteristic) that are generally considered to be good indicators of visibility. For daytime markings, the contrast ratios were the luminance (CIE Y), luminance coefficient under diffuse illumination (Qd), and MV geometry daytime luminance (Lv). For nighttime observations, the contrast ratios were coefficient of retroreflected luminance (RL, retroreflectivity) and MV geometry nighttime luminance (Lv). In the figures, the samples are ordered from left to right in order of ascending value of the contrast ratio. The calculated contrast ratios and the pavement marking performance characteristic are provided for each marking. The figures contained in the following section are based on the luminance (CIE Y) contrast ratio for daytime observations and retroreflectivity contrast ratio for nighttime observations. These performance metrics were used for several reasons. First, these material properties are among the most readily available options for road agencies to quantify the performance of their road markings. Second, Y and RL remain consistent within a sample from observation to observation because they are a standard measurement, whereas the luminance measurements using the CCD camera were taken under viewing conditions at the time of the observation. The CCD luminance will vary as the ambient lighting varies because it is not a standard measurement since it relies on an uncontrolled light source. Qd is only observable using a specific retroreflectometer that is yet to be widely adopted in the United States. Plots using Qd and MV geometry daytime or nighttime luminance are provided in Appendix C. Influence of Speed The speed at which the observations were recorded could potentially affect the rating assigned to a sample. During data collection, three speeds (40, 50, and 65 mph) were used to collect data. Researchers examined the Explorer and F-150 data by the observation speed for each of the combinations of moisture and lighting for which they were used. Figure 31 through Figure 34 contain data for observations taken during the Phase II data collection period. These rankings are explicitly based on solid edge line pavement markings. Figure 31 presents a box-

and-whisker plot illustrating the ratings observed in the Explorer at various speeds during the Phase II data collection period.

Figure 31. Daytime Dry Phase II Speed Assessment Using Explorer

Figure 31 indicates the effect of speed variations from sample to sample. Considering the observations taken while traveling northbound, samples WT1, YT4, WT7, WP2, YP3, and WP3 had lower detection confidence ratings at higher speeds, while YP1 had higher ratings. Other samples performed similarly between the various speeds. Considering the southbound

observations, samples YP1, WP2, WP3, and YT2 had higher detection confidence ratings at the higher speed, whereas samples YP3 and WT1 had lower detection confidence ratings at the higher speed. The other samples performed similarly at the two speeds. Figure 32 presents similar data to Figure 31, except using observations that were obtained using the F-150.

Figure 32. Daytime Dry Phase II Speed Assessment Using F-150

The northbound observations taken during the daytime with the F-150 suggest that increased speed reduces the confidence rating of the MV detection algorithm to varying degrees depending on the sample. Southbound observations using the F-150 do not present a clear trend. Considering only the northbound observations, all samples had lower or the same detection confidence ratings at 65 mph as opposed to 50, although the discrepancy was more pronounced in some instances than others. Regarding the southbound observations, samples WT1, YT4, WP1, and WP3 had higher detection confidence ratings at higher speeds. Sample YT2, YT3, WT7, YP2, and WP2 had lower detection confidence ratings at higher speeds; and the other samples performed comparably. Table 9 presents the average detection confidence rating change (in percent) between the observations taken at 65 mph relative to the observations taken at 50 mph for each marking sample that was presented in Figure 31 and Figure 32. Negative values indicate that the MV system had lower detection confidence rating during the 65 mph observations, while positive values indicate that the MV system had higher detection confidence rating during the 65 mph observations. The F-150 shows reductions in detection confidence ratings for most markings, whereas the Explorer showed some increases and reductions.

Table 9. Average confidence rating percent change from 50 mph ratings, Phase II markings under dry, daytime conditions Explorer F-150 NB SB NB SB Sample 65 mph % Change 65 mph % Change 65 mph % Change 65 mph % Change WT1 -13.33% -16.67% -9.09% 8.08% WT7 -14.28% 0.00% -3.78% -3.49% WP1 0.00% 0.00% -19.28% 3.16% WP2 -20.59% 10.00% -8.55% -6.71% WP3 -11.76% 20.00% -15.81% 2.85% YT2 0.00% 25.00% -7.46% -16.55% YT3 0.00% 0.00% -1.50% -8.30% YT4 -6.68% 0.00% -5.36% 1.64% YP1 10.00% 7.16% 0.00% -1.59% YP2 0.00% 0.00% -6.15% -6.91% YP3 -15.63% -9.09% -11.50% -1.14%

Figure 33 presents the observations taken at night during the Phase II data collection period under dry conditions using the Explorer.

Figure 33. Nighttime Dry Phase II Speed Assessment Using Explorer

During the Phase II observation periods, samples YT2 and YT3 had lower detection confidence ratings at higher speeds during northbound observations, while WT1, YP1, and WP1 had lower detection confidence ratings during southbound observations. Samples YP1 and WP1 had higher detection confidence ratings at higher speeds during northbound observations. Figure 34 contains the nighttime dry data by speed for the Phase II F-150 observations.

Figure 34. Nighttime Dry Phase II Speed Assessment Using F-150

With the exception of samples WT7 and WT1, all samples had higher or the same detection confidence ratings at higher speeds for the northbound observations at night. Southbound observations indicated samples WT7, WP3, YP1, and WP1 had higher detection confidence ratings at higher speeds, while the other samples had the same or lower detection confidence ratings at the higher speeds.

Table 10 summarizes the percent change in detection confidence rating of the samples as speed was changed from 50 mph to 65 mph for the Phase II dry night conditions. For both vehicles the change in speed had mixed results on the confidence rating impact.

Table 10. Percent change from 50 to 65 mph, Phase II Dry, Nighttime conditions Explorer F-150 NB SB NB SB Sample 65 mph % Change 65 mph % Change 65 mph % Change 65 mph % Change WT1 0.0% -8.3% -2.5% -11.3% WT7 0.0% 9.1% -21.2% 7.2% YT2 -8.3% 0.0% 2.0% -2.7% YT3 -8.3% 0.0% 11.4% -6.2% YT4 0.0% 9.1% 14.4% -0.9% WP1 16.7% -25.0% 24.7% 0.7% WP2 9.1% 9.1% 7.8% -19.0% WP3 33.3% 0.0% 7.4% 80.1% YP1 33.3% -25.0% 0.0% 2.0% YP2 0.0% 9.1% 6.4% -6.0% YP3 20.0% 0.0% 2.0% -2.3%

Figure 35 through Figure 38 contain observations from the Phase I data collection period. These figures present observations taken on both solid edge line and broken lane line markings during dry conditions. Figure 35 specifically contains daytime dry edge line markings by speed during Phase I for the explorer.

Figure 35. Daytime, Dry, Edge Line Marking Phase I Speed Assessment Using Explorer

Samples WT6 and YT6 northbound, as well as samples YT4, YT5, YT3, WT5, and WT6C southbound, had lower detection confidence ratings at higher speeds, while the other samples had the same or higher detection confidence ratings at higher speeds in comparison to lower speeds.

Figure 36 presents the nighttime observations on edge line markings during dry conditions for the Phase I observation period using the explorer.

Figure 36. Nighttime, Dry, Edge Line Marking Phase I Speed Assessment Using Explorer

Figure 36 illustrates that the samples performed relatively consistently at 50 and 65 mph, with some exceptions. The percent change of the average ratings for the Phase I day and night, dry, edge line assessments with the explorer are summarized in Table 11.

Table 11. Percent rating change from 50 mph ratings, Phase I edge line markings under dry conditions Daytime Nighttime NB SB NB SB Sample 65 mph % Change 65 mph % Change 65 mph % Change 65 mph % Change WT1 11.1% 11.1% 0.0% 0.0% WT2 3.4% 15.4% 0.0% 0.0% WT3 0.0% 0.0% 20.0% 11.1% WT4 5.9% 12.5% 0.0% 0.0% WT5 7.1% -13.0% 33.3% 0.0% WT6 -10.7% 13.6% -27.3% 0.0% WT6C 7.1% -9.1% 0.0% 0.0% WT7 7.1% 8.7% 0.0% 0.0% YT1 0.0% 0.0% -100.0% 0.0% YT2 9.1% 26.3% 0.0% 0.0% YT3 9.1% -15.8% 0.0% 0.0% YT4 20.0% -11.1% 0.0% 0.0% YT5 14.3% -20.0% 9.1% -16.7% YT6 -9.1% 26.3% 9.1% 0.0%

Figure 37 presents observations taken on broken lane line markings under dry, daytime conditions. The figure presents observations taken at 40 mph, in addition to 50 and 65 mph.

Figure 37. Daytime, Dry, Broken Marking Phase I Speed Assessment Using Explorer

The most substantial observation that can be made from Figure 37 is that sample YT1 had markedly lower detection confidence ratings than the other markings during the northbound observations. Sample WT4 southbound presented consistently improving detection confidence

ratings with respect to increasing speed, while the other samples performed somewhat similarly across the three speeds. Figure 38 presents the data collected from observations of on broken markings under dry, nighttime conditions.

Figure 38. Nighttime, Dry, Broken Marking Phase I Speed Assessment Using Explorer

Figure 38 illustrates that the pavement markings had consistently high detection confidence rating across the variety of speeds examined (40, 50, and 65 mph). Sample YT1 had notably worse detection confidence ratings than the other samples. Table 12 is a synthesis of the change between average detection confidence rating values presented in Figure 37 and Figure 38. These differences represent a percent change of the MV performance during 40 mph and 65 mph observations relative to the 50 mph observations. Positive values indicate that the MV assigned a higher detection confidence rating at a given speed for a specific sample. For the various conditions and markings the results are mixed. Some show increases in detection confidence ratings with increasing speeds, whereas other show decreases.

Table 12. Percent rating change from 50 mph ratings, Phase I broken markings under dry conditions Daytime Nighttime NB SB NB SB 40 mph 65 mph 40 mph 65 mph 40 mph 65 mph 40 mph 65 mph Sample % Chg % Chg % Chg % Chg % Chg % Chg % Chg % Chg WT1 -4.8% -4.8% -1.5% -4.3% 2.8% 2.8% 2.8% 2.85% WT2 3.0% -9.1% 3.7% 0.0% 0.0% 0.0% 0.0% 0.00% WT3 -7.2% -30.4% 8.3% 0.0% 7.1% 7.1% -2.2% -2.17% WT4 20.0% 0.0% -21.2% 9.1% 0.0% 0.0% 0.0% 0.00% WT5 4.3% -4.3% 5.3% 5.3% 0.0% -8.3% 0.0% 0.00% WT6 15.6% 6.7% -15.8% -15.8% 11.1% 11.1% 0.0% 0.00% WT6C 7.9% -4.8% 0.0% 0.0% 0.0% 0.0% 0.0% 9.09% WT7 17.6% 17.6% -13.3% 0.0% 2.8% 2.8% 0.0% 0.00% YT1 -100.0% -60.0% 14.3% 28.6% -100.0% -100.0% 0.0% 0.00% YT2 4.3% -4.3% 2.0% -5.9% 0.0% 0.0% 0.0% 0.00% YT3 0.0% 10.0% -11.1% -11.1% 7.1% 7.1% 0.0% 0.00% YT4 11.1% 11.1% 0.0% 0.0% 0.0% 0.0% 0.0% 0.00% YT5 12.3% 15.8% -5.9% -5.9% 0.0% 0.0% -1.8% 7.14% YT6 -4.8% -14.3% -6.7% -20.0% 11.1% 1.9% -1.8% 7.14%

Influence of Daytime Clouds The effect of cloud cover during the daytime was examined on data collected in Phase II during dry and wet conditions, as well as in Phase I during wet conditions. The impact of glare due to the sun was not part of the original data collection plan, but environmental conditions during the data collection efforts were noted. Data collection during both Phase I and Phase II took place during midday. The Phase II data collection occurred during the summer with the sun in a high position overhead, whereas Phase I data collection occurred during the winter with the sun lower in the sky to the south of the test area. Figure 39 presents the data with respect to cloud cover for daytime dry Phase II observations using the Explorer.

Figure 39. Daytime Dry Phase II Cloud Assessment Using Explorer

Figure 39 demonstrates that during northbound observations, cloud cover either did not have a noticeable effect or resulted in modest improvement in detection confidence rating. The southbound observations present mixed results, with some samples having lower detection confidence ratings in the presence of clouds. Figure 40 presents similar data observed using the F-150. The southbound data are based on observations taken at 50 mph, while the northbound observations include 50 and 65 mph data to present more scenarios that actually had clouds present.

Figure 40. Daytime Dry Phase II Cloud Assessment Using F-150

It is difficult to ascertain whether cloud cover impacted the MV system’s detection confidence ratings for this scenario since cloudy conditions were not observed for all lane markings. The problem is further compounded by the fact that even when clouds were present, the conditions were only partly cloudy. Figure 41 examines the impact of cloud cover during Phase II under daytime wet conditions.

Figure 41. Daytime Wet Phase II Cloud Assessment

Once again, it is difficult to ascertain any effect of cloud cover since it was not present for all samples. Some samples also lacked a clear sky condition. Among the samples that had multiple cloud cover conditions, results were somewhat mixed in that some samples had higher detection confidence ratings with cloud cover while others had lower detection confidence ratings. Figure 42 presents similar observations taken during the Phase I data collection period.

Figure 42. Daytime Wet Phase I Cloud Cover Assessment

Results for the northbound observations indicate mixed results. The southbound observations tend to suggest that samples had higher detection confidence ratings in the presence of cloud cover. These results make sense since the largest glare effects due to the sun were during the Phase I southbound data collection efforts. Some of the lowest detection confidence ratings collected during the testing were during the wet southbound sunny data collection conditions in Phase I. Figure 43 provides comparison images of a southbound (left image) and

northbound (right image) data collection area. The only thing that changed between the images was the direction the vehicle was facing.

Figure 43. Daytime Wet Glare Phase I, Winter 2016, Facing South (left), North (right).

Influence of Solid Markings vs. Broken Markings Figure 44 presents a comparison of solid edge line markings versus broken lane line markings during daytime dry conditions. Solid and broken comparison data is only available for the Phase I data collection.

Figure 44. Solid vs. Broken Markings during Daytime Dry Conditions

Figure 44 demonstrates that the MV system is generally able to detect solid markings with a higher detection confidence rating than broken markings; however, the difference is usually not very pronounced. There are a few exceptions where broken markings had higher detection confidence ratings. Figure 45 presents the observations taken during nighttime dry conditions.

Figure 45. Solid vs. Broken Markings during Nighttime Dry Conditions

Figure 45 indicates that sample YT1 had a higher detection confidence rating as a solid marking northbound, while sample WT3 had a higher detection confidence rating as a broken marking northbound. Sample YT2 had a higher detection confidence rating as a broken marking during the southbound observations, while sample WT5 had a higher detection confidence rating as a broken marking northbound. The remaining samples had similar detection confidence ratings as solid and broken markings. The percent change in the average detection confidence rating indicated in the previous two figures are shown in terms of percent change in Table 13. A

negative value indicates the detection confidence rating decreased when the marking went from a solid marking to a broken marking.

Table 13. Percent Change from Solid to Broken Markings during Dry Conditions Dry Daytime Dry Nighttime NB SB NB SB Sample % Change % Change % Change % Change WT1 -2.8% 6.5% -2.8% 16.7% WT2 -5.2% -13.5% 0.0% 0.0% WT3 -4.2% 0.0% 12.0% 2.2% WT4 -11.8% 3.1% 0.0% 0.0% WT5 2.7% 3.3% 33.3% 0.0% WT6 -33.0% 8.0% -18.2% 0.0% WT6C -6.2% -9.1% 0.0% -8.3% WT7 -24.1% 8.7% -2.8% 0.0% YT1 NA -12.5% -60.0% 0.0% YT2 4.5% -10.5% 0.0% 20.0% YT3 -9.1% -15.0% -6.7% 0.0% YT4 -10.0% -11.1% 0.0% 0.0% YT5 -9.5% -10.5% 9.1% -6.7% YT6 -4.5% -5.3% -1.8% -6.7%

Figure 46 presents the solid vs broken marking comparison data during daytime wet conditions.

Figure 46. Solid vs. Broken Markings during Daytime Wet Conditions

The information presented in Figure 46 again suggests that the solid markings had higher detection confidence ratings than the broken markings overall, but there are cases where the broken markings had higher detection confidence ratings.

Figure 47 presents the solid vs broken observations during nighttime wet conditions.

Figure 47. Solid vs. Broken Markings during Nighttime Wet Conditions

No clear pattern is visible from these data since high detection confidence ratings were found for both conditions for most samples. Sample YT5 had a higher detection confidence rating as a broken line during southbound observations. The percent change in the average detection confidence rating from the previous two figures have been summarized in Table 14.

Table 14. Percent Change from Solid to Broken Markings, Wet Conditions Wet Daytime Wet Nighttime NB SB NB SB Sample % Change % Change % Change % Change WT1 6.7% -67.6% -8.3% 0.0% WT2 11.5% 0.0% 0.0% 0.0% WT5 -1.3% -40.0% 2.4% 0.0% WT6 -2.8% -19.9% -2.4% 0.0% WT6C 6.7% -66.7% -5.6% 9.1% WT7 26.1% 79.9% -3.3% 0.0% YT2 -14.0% 0.0% 9.5% 0.0% YT3 -8.8% -5.0% 5.9% 0.0% YT4 -13.8% 20.9% 9.1% 9.1% YT5 -20.0% -10.0% 7.1% 33.3% YT6 -20.4% 3.8% 8.1% 0.0%

Influence of Contrast Markings During the Phase I data collection period, sample WT6 was examined as both a standard marking (WT6) and as a contrast marking (WT6C) for both solid and broken marking patterns. Researchers collected these data as a pilot to benefit future data collection efforts focused on evaluating contrast markings. Figure 48 illustrates that the contrast marking improved the detection confidence rating compared to the non-contrast version of the marking during northbound observations. Southbound observations saw some negative impact on detection confidence rating with the use of the contrast compared to the non-contrast version of the marking. The percent change between the detection of the standard and contrast version of sample WT6 are shown in Table 15. A small additional evaluation was conducted on the black material that was used to make the contrast marking. A 4-inch wide piece of the material was installed as a typical edge line with no other marking material. This solid black marking was evaluated with the MV system and returned a zero detection confidence rating, indicating it was not detected. This has implications for contrast marking patterns, in that the lead lag pattern (10 feet of white marking followed by 10 feet of black marking) may not be beneficial to the MV system evaluated.

Figure 48. Standard vs. Contrast Marking under All Weather Conditions

Table 15. Percent Change from Standard to Contrast Marking Northbound Southbound Broken Solid Broken Solid % Change % Change % Change % Change DD 40.0% 0.0% -15.8% 0.0% DW 21.9% 11.1% -25.0% 80.1% ND 33.3% 9.1% -8.3% 0.0% NW 10.6% 14.3% 0.0% -8.3%

Influence of Overhead Lighting Researchers evaluated the effect of continuous overhead lighting on the MV systems ability to detect the same markings that were evaluated in the other portions of the study. One of the runways at the Texas A&M RELLIS Campus has been outfitted with four high-pressure sodium street lights, as shown in Figure 49.

Figure 49. Phase I Overheard Lighting Section

Eight marking samples were examined under the lights; the sample properties of the markings were consistent with those of the markings previously documented. The one exception was sample YT7, which was a profiled version of sample YT1. Sample YT7 was only evaluated at the lighted test area during Phase I. All markings evaluated under the lights were solid edge line markings. During a portion of the Phase I data collection activities, the CCD camera was unavailable due to servicing. Subsequently, MV system geometry luminance (Lv) observations were not taken for all of the samples under the overhead lighted section, nor were observations taken during wet conditions. For those sections where dry measurements were observed, the data are summarized in Table 16.

Table 16. Dry MV System Geometry Luminance (Lv) Comparison Sample WT5 WT6 YT7 YT6 Marking Lighting Off 0.854 0.609 0.609 0.879 2 (cd/m ) Lighting On 2.449 0.687 1.461 4.083 Pavement Lighting Off 0.415 0.397 0.359 0.526 2 (cd/m ) Lighting On 1.296 0.517 0.619 2.037 Contrast Lighting Off 2.056 1.534 1.694 1.671 Ratio Lighting On 1.889 1.330 2.359 2.004

A count of the number of MV observations for the combinations of lighting, speed, pavement condition (dry or wet), and travel direction are documented in Table 17.

Table 17. Count of Observations in Lighted Section Lights On Off Speed (mph) 40 50 40 50 Pavement Dry Wet Dry Wet Dry Wet Dry Wet Direction NB SB NB SB NB SB NB SB NB SB NB SB NB SB NB SB YT7 0 0 0 0 2 2 0 1 0 0 0 0 2 2 1 1 YT6 0 0 0 0 2 2 0 1 0 0 0 0 2 2 1 1 WT6 0 0 0 0 2 2 0 1 0 0 0 0 2 2 1 1 WT5 0 0 0 0 2 2 0 1 0 0 0 0 2 2 1 1 YT3 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 YT5 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 WT2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 WT7 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Table 18 documents the average detection confidence rating for each of the marking samples under the various conditions. Cells with dashes indicate that no observations were made under that specific combination of conditions.

Table 18. Average Sample Detection Confidence Rating in Lighted Section

Lights On Off Speed (mph) 40 50 40 50 Pavement Dry Wet Dry Wet Dry Wet Dry Wet Direction NB SB NB SB NB SB NB SB NB SB NB SB NB SB NB SB YT7 - - - - 2 2 - 3 - - - - 2.5 2.25 2.5 3 YT6 - - - - 2.5 2.5 - 2.5 - - - - 3 3 3 2.5 WT6 - - - - 3 3 - 3 - - - - 3 3 3 3 WT5 - - - - 3 3 - 3 - - - - 3 3 3 3 YT3 3 3 2.5 3 3 2.5 2.5 2.5 3 3 3 3 3 3 3 3 YT5 3 2.5 3 2.5 3 2 3 2 3 2.5 3 3 3 2.5 2.5 3 WT2 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 WT7 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3

The lights-on and lights-off results from Table 18 are compared by lighted detection confidence rating minus unlighted detection confidence rating in Table 19. Positive values indicate that the presence of overhead lighting had a positive effect, while negative numbers indicate that the MV system performed better in the absence of overhead lighting. Cells with dashes again indicate no observations for a given combination of light, speed and pavement moisture. No observations were recorded under northbound wet conditions with lights on. Although a small number of observations were made under the overhead lights, the results suggest that the presence of overhead street lighting resulted in lower MV performance for the given test conditions. The potential benefits of the overhead lighting for pavement marking detection may not be realized by the MV system evaluated due to its observation distance. Overhead lighting provides illumination beyond where the vehicle headlamps can illuminate the road and markings. The MV system evaluated views the markings where the headlamp illumination is high. The effect of overhead lighting on pavement marking luminance is demonstrated in Figure 50. The

figure illustrates the luminance contrast ratio of the marking to the pavement surface at various locations relative to the vehicle. In the absence of overhead lighting, the luminance contrast ratio tended to decrease as the distance from the vehicle increased (sample WT6 did not follow the trend). In comparison, when the lights were activated, the contrast ratio at 165 ft was greater than or equal to the contrast ratio at 125 ft. Assuming even illumination from the overhead lighting, the contrast ratio should remain similar as observation distance changes.

Table 19. Lighted vs. Unlighted Performance Lights On – Off Speed (mph) 40 50 Pavement Dry Wet Dry Wet Sample NB SB NB SB NB SB NB SB YT7 - - 0 0 −0.5 −0.25 −2.5 0 YT6 - - 0 0 −0.5 −0.5 −3 0 WT6 - - 0 0 0 0 −3 0 WT5 - - 0 0 0 0 −3 0 YT3 0 0 −0.5 0 0 −0.5 −0.5 −0.5 YT5 0 0 0 −0.5 0 −0.5 0.5 −1 WT2 0 0 0 0 0 0 0 0 WT7 0 0 0 0 0 0 0 0

Figure 50. Luminance Contrast Ratio under Streetlight Presence

Evaluation of Marking Performance Relative to Material Properties The detection confidence rating of the MV system lane marking determination algorithm was examined by developing box-and-whisker plots with separate columns for each sample. In order to provide a direct comparison, the potentially confounding factors examined previously (cloud cover, speed, marking pattern) were removed so that only the observations of solid edge line markings during clear sky conditions at 50 mph were considered. Given that a goal of this research was to identify minimum marking performance levels necessary for satisfactory MV performance, the columns were ordered based on the contrast ratios for each sample. The contrast ratios (marking performance characteristic compared to adjacent pavement performance characteristic) are generally considered to be good indicators of visibility. The marking performance metric value is also included in the figures. For daytime markings, the contrast ratio utilized was the luminance (CIE Y); similar plots for the luminance coefficient under diffuse illumination (Qd) and MV system geometry daytime luminance are available in Appendix C. All three of the ratios were relatively consistent for a given sample (i.e., if a marking had a high Y ratio, it also had a high luminance or Qd ratio). For nighttime observations, the contrast ratio utilized was coefficient of retroreflected luminance (RL); MV system geometry nighttime luminance plots are available in Appendix C. Similar to the daytime performance, if a marking had a high RL, it generally had a high luminance value. Daytime Dry Conditions Figure 51 represents the average (by sample) observations taken during daytime dry conditions. In general, the samples exhibited relatively high detectability, with only a couple of markings having an average detection confidence rating value of 2 or lower. Only sample YT1 had an average confidence value lower than 2 (only in one direction), and it had a very low contrast value. No sample with a contrast value of 2.8 or higher had an average rating of 2 or less. There were several markings that had a lower Y contrast level but had higher detection confidence rating by the MV system. There were also a few markings with higher Y values, but resulted in lower contrast ratios, and resulted in lower confidence ratings. In order to develop a clearer picture of the effect of pavement marking sample performance on MV detection confidence rating during daytime dry conditions, future research will need to focus on test conditions that are more challenging to the MV system. These conditions could include test areas with conflicting signals that may make it more difficult to detect the markings and glare from the sun. Figure 52 presents the summarization of data collected using the F-150 during the daytime dry Phase II observations with respect to the Y contrast ratio. Similar to the data from the Explorer, the ratings are generally high, with all averages at 2 or higher.

Figure 51. Average Rating vs. Y Contrast Ratio by Sample during Daytime Dry Conditions, Explorer Only

Figure 52. Average Rating vs. Y Contrast Ratio by Sample during Daytime Dry Conditions, F-150 Only

Daytime Wet Conditions Figure 53 presents the detection confidence ratings of the pavement markings under daytime wet conditions. When the data for the Phase I and Phase II data collection periods are examined for various contrast ratios, the performance of the MV system varies depending on direction of travel. Figure 53 indicates that many of the markings had much lower performance

in Phase I in the southbound direction. Samples WP1, YP1, WP3, and YP3 are omitted from the figure as they were only observed under full cloud conditions. Five markings from Phase I southbound data had average detection confidence ratings of less than 2. All other markings had average confidence ratings of 2 or higher. The MV system tended to perform better during the Phase II data collection period for daytime wet conditions. The performance difference is likely due to the angle of the sun and its impact on the wet road conditions. The actual performance measurements of the markings did not correlate with the MV detection confidence ratings due to the influence of the sun during the MV data collection, see Figure 43.

Figure 53. Average Rating vs. Y Contrast Ratio by Sample during Daytime Wet Conditions, Explorer Only

Nighttime Dry Conditions Figure 54 presents the ratings observed under nighttime dry conditions. While not a perfectly increasing relationship between the retroreflectivity contrast ratio and MV detection confidence rating, Figure 54 does demonstrate that all samples that performed poorly (average detection confidence rating less than 2) had relatively low contrast ratios. Figure 55 presents data

collected using the F-150 during nighttime dry conditions with respect to retroreflectivity contrast ratio. Contrast ratios below 2.4 tended to yield the lowest detection confidence ratings. With the exception of one sample observed by the F-150, no sample with a contrast value of 2.4 or higher had an average detection confidence rating of 2 or less. The sample with the 2.4 contrast values had a retroreflectivity level of 30 mcd/m2/lux. With the exception of the one sample observed by the F-150, no sample with a retroreflectivity value of 34 mcd/m2/lux or higher had an average detection confidence rating of 2 or less. Nighttime Wet Conditions Figure 56 presents the ratings of the observations taken during nighttime wet conditions for both the Phase II and Phase I data collection periods. Figure 56 indicates a general increasing trend in MV detection confidence rating versus contrast ratio. Once again, the lowest performers were those markings with the lowest contrast ratios. No sample with a contrast value of 2.1 or higher had an average detection confidence rating less than 2. This sample had a wet recover retroreflectivity level of 4 mcd/m2/lux. All marking with wet recovery retroreflectivity level of 9 mcd/m2/lux or higher (contrast ratio of 4.7) had average detection confidence ratings of greater than 2.

Figure 54. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Dry Conditions, Explorer Only

Figure 55. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Dry Conditions, F-150 Only

Figure 56. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Wet Conditions, Explorer Only

Nighttime Dry Glare Conditions Figure 57 illustrates the performance of the pavement markings at night under dry glare conditions. These observations were only collected during Phase II and only in the northbound direction of travel. All markings with a retroreflectivity contrast ratio of 3 or greater had high detection confidence ratings by the MV system. All markings with a retroreflectivity level of 53 mcd/m2/lux or higher exceeded an average detection rating of 2 or higher.

Figure 57. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Dry Glare Conditions, Explorer Only

Nighttime Wet Glare Conditions Similar to Figure 57, Figure 58 presents the results of data collection activities during nighttime glare conditions; however, these observations occurred when the pavement was wet. Figure 58 illustrates a similar trend to those that have been presented in several other figures. The markings with higher contrast tend to have higher detection confidence ratings. All markings with a contrast ratio of 2.1 or higher had detection confidence rating of at least 2. This sample had a wet recover retroreflectivity level of 4 mcd/m2/lux. All marking with wet recovery retroreflectivity level of 9 mcd/m2/lux or higher (contrast ratio of 4.7) had average detection confidence ratings of greater than 2.

Figure 58. Average Rating vs. RL Contrast Ratio by Sample during Nighttime Wet Glare Conditions, Explorer Only

CHAPTER 4. FINDINGS, RECOMMENDATIONS, AND SUGGESTED RESEARCH

FINDINGS

This study assessed the detection capabilities of a MV system when evaluating longitudinal pavement markings that had various levels of performance that were intended to represent markings at various stages of wear. To evaluate the performance of the markings relative to the MV system, various 4-inch wide markings were examined on dry pavement during the day and night, as well as on wet pavement during the day and night with and without glare applied from an oncoming vehicle. While the relationship between material properties of longitudinal markings and the detectability by the MV systems was the primary focus of this study, several possible confounding factors were assessed in order to provide as much information as possible. The speed of observation, cloud cover during daytime observations, overhead lighting, and glare conditions (sun and oncoming vehicle) were all looked at as factors that could influence the detectability of the markings by the MV system. Marking characteristics such as the marking pattern (edge line or lane line) and added contrast were also evaluated. The performance characteristics of the markings, either day or night, were observed and compared to the MV detection confidence rating in the various test conditions. The following sections describe findings concerning the MV system, factors influencing the detectability of markings by MV, and pavement marking characteristics that influence detectability by MV systems. The study recommendations and limitations follow. The final section includes suggested research that would provide critical information needed to develop a better understanding of how pavement markings can be designed and maintained to improve MV detectability. MV System The MV system used for this study was the Mobileye 5 series ADAS system. It was installed in two different data collection vehicles that were used throughout the study. Summary information about the MV system is provided below.

• The system utilizes a monochrome camera with a resolution of about 1 MP. • The camera has a horizontal field of view of approximately 40 degrees, and a vertical field of view of approximately 30 degrees. • The system process images at 15 frames per second. • The system focuses on pavement markings located 30-50 feet in front of the vehicle. o The system may not be realizing the full benefit of high-retroreflectivity markings at night. The vehicle headlamp illumination pattern puts a large amount of light close to the vehicle, lighting up the area more so than further down the road. The high illumination level and viewing geometry (which is different than the industry standard geometry for retroreflectivity: 30-meters) lessen the need for high retroreflectivity from the markings. Future MV systems with higher resolution that look further down the road may benefit from increased retroreflectivity or overhead lighting. • The system algorithm assigns a detection confidence rating of the pavement markings on either side of the vehicle. The confidence rating is an integer value between 0 and

3, with 3 being the highest confidence. The systems requires a confidence value of 2 or greater in order to provide LDW assistance. • The 0 to 3 integer value detection confidence rating output leaves little room for identifying the impact of slight changes to the pavement markings in terms of color or retroreflectivity. • The MV system views markings like a human driver views them. Things that degrade a human driver’s ability to detect a marking will also degrade the MV systems ability to detect a marking.

The Mobileye system tested was by far the most common system on the market when the project started. At the time of writing this report newer versions of the Mobileye system have been released that use newer hardware and software. Influencing Factors Speed Several vehicle speeds (40, 50, 65 mph) were used during this study under dry and wet conditions to assess how the MV system performs. Generally speaking, the performance of the MV detection system tended to be adversely affected by increased travel speeds during the daytime, regardless of the vehicle used to collect the observation. This was true of both broken and solid markings. During the nighttime, more often than during the day, speed tended to improve or have no effect on MV performance. Cloud Cover No clear trend could be identified in this study regarding the impact of cloud cover on MV performance. From a purely intuitive standpoint, one would expect that cloud cover would be associated with improved performance due to the mitigation of glare caused by the sun. This study examined only a limited number of observations under cloudy/partially cloudy conditions, which likely contributes to the lack of clear trends in this area. Also, much of the testing occurred with the sun in a high position directly overhead, limiting the impact of glare. Detrimental glare was observed in some instances, especially during the winter data collection period for southbound wet testing. When present the glare resulted in lower detection confidence ratings for the markings. Overhead Lighting Based on a limited sample of observations taken under continuous overhead lighting, it appears that overhead lighting may have an adverse effect on MV performance. The potential benefits of the overhead lighting for pavement marking detection may not be realized by the MV system evaluated due to its short observation distance. It is important to consider that only one type of lighting was used in this study. Therefore, future research should examine the impact of a variety of lighting intensities and color temperatures. Glare Conditions The impact of sun glare was discussed in the cloud cover section. A pilot test for the impact of on-coming vehicle glare did not indicate a large impact on the detection confidence rating of the markings, but there was some negative impact. The limited impact may be the result of the study design that used a single stationary glare vehicle, coupled with the short range

viewing distance of the MV system. CCD images of the on-coming vehicle headlight glare conditions indicated the oncoming headlights cause glare on the center line markings that reduces the marking visibility. Marking Characteristics Daytime Marking Visibility Performance The daytime visibility of the markings was evaluated with measures of luminance (CIE Y), luminance coefficient under diffuse illumination (Qd), and MV system geometry luminance (Lv). When comparing the measured values to the pavement surface, researchers found that the markings had similar contrast ratios across the various pavement marking performance measures. The results focused on the CIE Y value, as it is more commonly evaluated compared to the other two measures. Testing showed that all markings that had a CIE Y contrast ratio of 2.8 or higher had MV detection confidence ratings of greater than 2 during dry daytime conditions. The markings with the 2.8 contrast ratio had a Y value of 32. All markings with a Y value of 23 or higher had detection confidence ratings of 2 or greater. The markings with the Y value of 23 had a 1.6 contrast ratio. In wet daytime conditions, the results were influenced by the presence of the sun causing glare on the markings and surrounding pavement, reducing the detection confidence ratings. There was no correlation between detection confidence rating and the marking Y value due to the presence of sun glare during some of the tests. Low-angle sun glare was not part of this testing. It is anticipated that low-angle sun glare would have similar effects on the detection confidence rating as the wet daytime testing did in the presence of sun glare. Nighttime Marking Visibility Performance The nighttime visibility of the markings was evaluated with measures of coefficient of retroreflected luminance (RL) and MV system geometry luminance (Lv). When comparing the measured values to the pavement surface, researchers found that the markings had similar contrast ratios across the various pavement marking performance measures. The results focused on the RL value, as it is more commonly evaluated compared to the luminance value. Contrast ratios below 2.0 tended to yield the lowest detection confidence ratings. With the exception of one sample observed by the F-150, no sample with a contrast value of 2.4 or higher had an average detection confidence rating of 2 or less. The sample with the 2.4 contrast values had a retroreflectivity level of 30 mcd/m2/lux. With the exception of the one sample observed by the F- 150, no sample with a retroreflectivity value of 34 mcd/m2/lux or higher had an average detection confidence rating of 2 or less. The limited nighttime dry glare testing indicated markings with a retroreflectivity contrast ratio of 3 or greater had adequate detection confidence by the MV system. All markings with a retroreflectivity level of 53 mcd/m2/lux or higher exceeded an average detection confidence rating of 2 or higher. In wet night conditions, every sample with a contrast value of 2.1 or higher had an average detection confidence rating of 2 or higher. The sample with the 2.1 contrast had a wet recover retroreflectivity level of 4 mcd/m2/lux. All marking with a wet recovery retroreflectivity level of 9 mcd/m2/lux or higher (contrast ratio of 4.7) had average detection confidence ratings of greater than 2. The results were the same for the limited opposing vehicle glare testing.

Marking Pattern During the Phase I data collection period, both solid and broken markings of the same samples were examined. The results of this study suggest that solid markings are more easily detected by the MV system, although in most instances the difference is minor. The largest discrepancies between solid and broken marking detection performance were observed during the daytime data collection. Contrast Markings This study examined one particular sample as both a standard and a contrast marking, where the white 4-inch wide marking was paralleled by 2-inch wide black striping on each side. The addition of the contrast striping resulted in mixed results compared to same marking without the contrast striping. Improved performance of the contrast marking was observable for both solid and broken markings, as well as under daytime, nighttime, wet, and dry conditions. The improved performance was typically for the northbound travel direction, with the largest benefit during the dry day conditions. Southbound data collection showed negative effects of the contrast marking in dry day data collection. This may indicate the black portion of the contrast may create glare problems instead of mitigating them when the sun is causing oncoming glare. It is important to point out that these results are only representative of one sample (a profiled tape) and contrast marking pattern. It should also be noted that a solid 4-inch wide black marking with no other marking was given a detection confidence rating of zero by the MV system evaluated. This has implications for contrast marking patterns, in that the lead lag pattern may not be beneficial to the MV system evaluated. Marking Characteristics for LDW and LKA This study evaluated several marking characteristics and how they impacted the ability of the MV system to detect the marking. Marking characteristics evaluated were the marking pattern (solid or broken), coefficient of retroreflected luminance (RL), luminance coefficient under diffuse illumination (Qd), color (x, y chromaticity coordinates) and luminance (CIE Y), and MV geometry luminance(Lv) (day and night) using an imaging colorimeter. Each of these characteristics may differ in wet and dry conditions. Each of the measures were used to generate contrast ratios between the marking and the adjacent pavement surface. It is the level of contrast between the marking and the road surface that is a key factor in the detectability of the marking by a MV system, when other influencing factors (i.e. glare, other signals, etc.) are not present. Marking characteristics not evaluated, but thought to impact detectability of the markings by a MV system are, width, non-standard lane line patterns, marking presence, marking edge quality (how well defined is the edge), and marking texture. There are other factors that will influence the detectability of the markings that do not deal with the marking characteristics. These factors are speed of observation, geometry of observation, glare from the sun, glare from other vehicle lights, vehicle head light quality, street lighting, road surface characteristics, rain, fog, snow, supplemental raised pavement markers, shadows, and other signals on the road surface that could confuse marking detection (removed markings, pavement joints, crack seal, etc.). It is difficult to develop a prioritized list of marking characteristics for LDW and LKA without studying all available characteristics that may impact the performance of MV systems. Based on the research described in this report, the literature review, and experience in the field,

the following pavement marking characteristics are thought to be the most important for current MV systems: 1. Pavement marking presence o There must be a sufficient amount of the marking present to be detected. o If there is sufficient presence the other characteristics will determine how easily the marking is detected. 2. Contrast ratio between the marking and road surface o Daytime – CIE Y luminance . Higher CIE Y value will yield a more visible marking. . Evaluate with a handheld spectrophotometer. o Nighttime - coefficient of retroreflected luminance (RL) . Brighter markings will have greater benefit the further the MV system looks ahead. . Evaluate with a handheld retroreflectometer. . If the MV system is looking close to the vehicle like the one evaluated in this study, CIE Y luminance may be as important as retroreflectivity. o Both day and night characteristics depend on the viewing geometry of the MV system. o Contrast ratio can be improved by installing a marking with higher CIE Y or RL characteristics, or by incorporating black marking material in conjunction with the marking. o Other performance metrics may be used (such as Qd) but CIE Y and RL are the most commonly used. 2. Pavement marking width o Wider markings will have greater benefit the further the MV system looks ahead. o Wider markings may extend the useful life of the marking if the edges deteriorate or presence starts to reduce. o Wider markings may reduce negative impacts of conflicting signals, due to their increased signal of the wider marking. 3. Wet-weather characteristics o The marking characteristics in bullet point 2 also pertain to wet-weather situations. o Marking texture/structure may be beneficial. o Glare situations may be worse in wet situations. 4. Lane line pattern o Is 10 feet of marking followed by a 30 foot gap optimal? o Testing indicated broken markings were not detected as well as solid markings. 5. Marking texture/structure o May provide benefits in wet conditions. o May provide higher rates of detection when glare is present.

RECOMMENDATIONS This research was conducted to identify performance characteristics of pavement markings that affect the ability of MV systems to detect the markings. The scope of the study allowed for several areas of pavement marking properties and other influencing factors to be evaluated. Further studies are needed to replicate the findings, address areas not covered by the research, and diversify the methods for data collection (e.g., using other MV systems, or other sensors). The premise of this study is that in order to achieve consistently high detection confidence by the MV system evaluated, the contrast ratio of the longitudinal pavement markings relative to the adjacent pavement needs to be of an adequate level to facilitate detection. The pavement marking performance characteristics and associated contrast ratios needed for adequate detection for the study conditions are described below. • Daytime dry testing results indicated that all marking samples with a Y value of 23 or higher had an average confidence ratings of 2 or greater. This resulted in a 1.6 contrast ratio. • Daytime wet testing was inconclusive. There was no correlation between detection confidence rating and the marking Y value—likely due to the presence of sun glare during some of the tests. • The dry night testing results indicated that marking samples with a retroreflectivity value of 34 mcd/m2/lux or higher (with the exception of one sample observed by the F-150), had an average detection confidence ratings of 2 or greater. This resulted in a 2.5 contrast ratio. • In wet night conditions, every marking sample with a wet recover retroreflectivity level of 4 mcd/m2/lux or higher had an average detection confidence rating of 2 or greater. This resulted in a 2.1 contrast ratio.

These recommendations are based on testing conducted in a closed-course environment. There are also other factors that can influence the pavement marking detection. Most notably is the influence of glare from the sun. Glare at night from oncoming vehicles also resulted in the need for a higher performance levels. More in-depth evaluation of glare, and ways to reduce or account for its impact are needed. The recommendations from this study are different than similar research conducted on roads open to public travel with in-service pavement markings [25]. In general, the results from the research with in-service markings indicate that higher levels of pavement marking maintenance are needed to generate reliable MV detection. This difference indicates that there may be in-situ factors such as marking wear patterns that impact MV detection of pavement markings. The results of this study are directly applicable for the testing conditions described in this report and may not be directly transferable to policy or specifications. The testing area had a relatively uniform roadway surface; roadways with conflicting messages from previously removed markings, blackout markings, crack seal, varying road surfaces, cracking, or rutting may require higher-quality pavement markings. The research on the MV system evaluated indicates that the camera sees similarly to a human. Good pavement marking practices for a human driver will provide good conditions for the MV system. Both the human driver and the MV system detection of markings decreases if conflicting signals are present. Pavement marking practices should provide markings in a good

state of repair without other signals that could be mistaken for longitudinal delineation. In addition, both the human driver and the MV system detection of markings decreases if glare signals are present. This glare can be from the sun, oncoming headlights, or other light sources at night. Methods to mitigate the impacts of glare need to be developed to benefit both the human driver and MV systems. These methods could be related to the pavement marking characteristics or to the MV system hardware or software. At this stage of the research specific specification language or recommendations for changes to the MUTCD are premature. There are still many questions left to be answered, and rapid advances in ADAS technology are continually altering what performance requirements are needed by new systems. Common requests from industry wish lists are increased uniformity and higher maintenance standards for markings. Improving marking uniformity can be achieved without setting specific maintenance requirements.

STUDY LIMITATIONS The scope of the study and the selected research approach resulted in several limitations that impact the results. These limitations need to be considered when evaluating the results and when developing future research that evaluates MV and pavement marking detection.

• The scope of the study did not allow for several factors to be evaluated at all. Raised pavement markers, reflective or non-reflective, were not evaluated. Markings in conjunction with markers were not evaluated. Markings wider than 4 inches were not evaluated. Markings in continuous wetting conditions were not evaluated. The impact of shadows or other conflicting signals (removed markings, joints, crack seal, shadows etc.) in the area of the evaluated marking were not evaluated. • The scope of the work did not allow for several factors to be explored in-depth. Factors such as glare and contrast markings were studied, but not fully. The length and spacing of broken lane line markings were only evaluated at the standard layout (10 feet of marking followed by a 30 foot gap). Profiled tape samples were part of the markings studied but the impact of the profile pattern was not specifically evaluated. • Despite efforts to engage with several MV providers, only one MV system was used in the study. • The closed course study approach limited the ability to evaluate naturally aged markings. The markings evaluated had various levels of retroreflectivity and color representing different levels of worn markings, but they still had 100 percent presence and well-defined edges. • All of the testing was performance on a concrete road surface. • The visual appearance of the road surface was consistent allowing the system to possibly detect the markings easier than if the road surface was more variable in appearance or if conflicting signals were present near the markings. • The testing was only conducted in straight tangent sections. • Only a single overhead lighting style/pattern was evaluated.

SUGGESTED RESEARCH Additional research should build upon this work by addressing the limitations described above. Future research should investigate other MV based ADAS systems. This study evaluated

only 4-inch-wide markings. Many states are implementing 6-inch-wide markings to improve detection by human drivers, and MV systems may also benefit from wider markings. This study was also somewhat limited due to the consistency of the pavement that was used for the assessment (i.e., all pavement was of a concrete runway surface). Although this study assessed multiple marking samples, the visibility of these markings is affected by the pavement behind them. Therefore, future studies should assess a variety of pavement types to develop a more robust understanding of the effect of varying marking properties and the resulting contrast between the marking and darker or lighter pavements. This study was conducted on a closed- course facility. To gain a better understanding of how various marking characteristics affect the performance of MV products, real-world investigations on open roadways should be utilized. Test conditions could include areas with conflicting signals that may make it more difficult to detect the markings, such as roadways with conflicting messages from previously removed markings, blackout markings, crack seal, varying road surfaces, rutting, and shadows. Testing should include markings with varying levels of wear that may impact how well defined the edge of the marking is. The impact of different overhead lighting systems on nighttime detection of markings on different pavement surfaces in different weather conditions needs to be investigated further. Future testing should include glare from the sun, increased testing on glare from opposing vehicles, and testing in rain conditions. Further testing of MV systems using various contrast marking patterns/designs will provide better insight into the effectiveness of contrast markings in conditions where insufficient contrast is achieved between a standard marking and the pavement. Evaluations of raised pavement markers (reflective or non-reflective) should be studied to see if they provide benefit when used to supplement markings, or if they can provide enough signal when substituting for markings. Lane line patterns, length of the marking and gap spacing should be evaluated to determine if alternative patterns may provide additional benefit. Although the MV systems are generally designed to be utilized at speeds in excess of 35 or 40 mph (since it serves as an LDW system), there is a need to perform tests at a wider variety of speeds, particularly lower speeds. As AVs become popular, it is likely that their use will proliferate from cities and urban centers where road users will need to travel at lower speeds and under start/stop conditions. Along the same lines, AV systems that use MV systems at higher speeds may need to look further down the road than the system evaluated in this study. The distance the markings are viewed at can have implications on the required marking properties in varying viewing conditions.

REFERENCES

[1] D. Liu, Z. Lu, T. Cao and T. Li, "A real-time posture monitoring method for rail vehicle bodies based on machine vision," Vehicle System Dynamics, vol. 55, no. 6, pp. 853-874, 2017.

[2] X. Gilbert, V. M. Patel and R. Chellappa, "Deep Multitask learning for Railway Track inspection," IEEE Transactions on Intelligent Transportation Systems, vol. 18, no. 1, pp. 153-164, 2017.

[3] B. Lily, "Cab Cameras Gain 'Machine Vision'," Transport Topics Online, 2017.

[4] X. Lu, Y. Wang, X. Zhou, Z. Zhang and Z. Ling, "Traffic Sign Recognition via Multi- Modal Tree-Structure Embedded Multi-Task Learning," IEEE Transactions on Intelligent Transportation Systems, vol. 18, no. 4, pp. 960-972, 2017.

[5] M. B. de Paula and C. R. Jung, "Real-time detection and classification of road lane markings," in XXVI SIBGRAPI Conference on Gaphics, Patterns and Images, Arequipa, Peru, 2013.

[6] K. Yoo, H. W. S. I. Hwayoung Kim and H. S. Lee, "A Sensor Fusion Digital-Map System for Driver Assistance," SAE International, 2013.

[7] S. Sternlund, j. Strandroth, M. Rizzi, A. Lie and C. Tingvall, "The effectiveness of lane departure warning systems-A reduction in real-world passenger car injury crashes," Traffic Injury Prevention, vol. 18, no. 2, pp. 225-229, 2017.

[8] J. D. Rupp and A. G. King, "Autonomous Driving - A Practical Roadmap," SAE International, 2010.

[9] L. Fletcher, N. Apostoloff, L. Petersson and A. Zelinsky, "Vision in and Out of Vehicles," IEEE Intelligent Systems, vol. 18, no. 3, pp. 12-17, 2003.

[10] V. Popescu and S. N. Rada Danescu, "On-Road Position Estimation by Probabilistic Integration of Visual Cues," in Proceedings of the 2012 Intelligent Vehicles Symposium, Alcala de Henares, Spain, 2012.

[11] H.-Y. Cheng, C.-C. Yu, K.-C. F. C.-C. Tseng, J.-N. Hwang and B.-S. Jeng, "Environment classification and hierarchical lane detection for structured and unstructured roads," IET Computer Vision, vol. 4, no. 1, pp. 37-49, 2009.

[12] C.-L. Huo, Y.-H. Yu and T.-Y. Sun, "Lane Depearture Warning System based on Dynamic Vanishing Point Adjustment," in Proceedings of the 1st IEEE Global Conference on Consumer Electronics, Tokyo, Japan, 2012.

[13] F. Homm, N. Kaempchen and D. Burschka, "Fusion of Laserscanner and Video Based Lanemarking Detection for Robust Lateral Vehicle Control and Lane Change Maneuvers," in Proceedings of the 2011 IEEE Intelligent Vehicles Symposium (IV), Baden-Baden, Germany, 2011.

[14] P. Foucher, Y. Sebsadji, P. C. Jean-Philippe Tarel and P. Nicolle, "Detection and Recognition of Urban Road Markings Using Images," in Proceedings of the 14th International IEEE Conference on Intelligent Transportation Systems, Washington, DC, 2011.

[15] Q. Li, L. Chen, M. Li, S.-L. Shaw and A. Nuchter, "A Sensor-Fusion Drivable-Region and Lane-Detection System for Autonomous Vehicle Navigation in Challenging Road Scenarios," IEEE Transactions on Vehicular Technology, vol. 63, no. 2, pp. 540-555, 2014.

[16] M. Hadi, P. Sinha and J. R. Easterling IV, "Effect of Environmental Conditions on Performance of Image Recognition-Based Lane Departure Warning Systems," Transportation Research Record, no. 2000, pp. 114-120, 2007.

[17] M. Hadi, J. Schuerger and P. Sinha, "Lane Marking/Striping ot Improve Image Processing Lane Departure Warning Systems," California Department of Transportation, Sacramento, CA, 2007.

[18] R. Goplan, T. Hong, M. Shneier and R. Chellappa, "A Learning Approach Towards detection and Tracking of Lane Markings," IEEE Transactions on Intelligent Transportation Systems, vol. 13, no. 3, pp. 1088-1098, 2012.

[19] B. H. Wilson, M. D. Stearns, J. Koopmann and C. D. Yang, "Evaluation of Road-Departure Crash Warning Systems," United States Department of Transportation, Washington, DC, 2007.

[20] A. M. Pike, J. Gene Hawkins and P. J. Carlson, "Evaluating the Retroreflectivity of Pavement Marking Materials Under Continuous Wetting Conditions," Transportation Research Record, no. 2015, pp. 81-90, 2007.

[21] R. B. Gibbons and B. M. Williams, "Assessment of the Durability of Wet Night Visible Pavement Markings: Wet Visibility Project Phase IV," Virginia Department of Transportation, Richmond, VA, 2012.

[22] C. Davies, "Effects on Pavement Marking Characteristics on Machine Vision Technology," in TRB 96th Annual Meeting Compendium of Papers, Washington, DC, 2017.

[23] M. Hadi and P. Sinha, "Effect of Pavement Marking Retroreflectivity on the Performance ov Vision-Based Lane Departure Warning Systems," Journal of Intelligent Transportation Systems, vol. 15, no. 1, pp. 42-51, 2011.

[24] European Road Assessment Programme (EuroRAP), Roads that Cars can Read: A Quality Standard for Road Markings and Traffic Signs on Major Rural Roads, Hampshire, United Kingdom: EuroRAP, 2011.

[25] S.-O. Lundkvist and C. Fors, "Lane Departure Warning Sstem-LDW: Correlation between LDW's and road markings," Swedish National Road and Transport Research Institute (VTI), Linkoping, Sweden, 2010.

[26] K. Kluge and G. Johnson, "Statistical Characterization of the Visual Characteristics of Painted Lane Markings," in Proceedings of the Intelligent Vehicles '95 Symposium, Detroit, MI, 1995.

[27] H. Ding, B. Zou, K. Guo and C. Chen, "Comparison of Several Lane Marking Line Recognition Methods," in Proceedings of the Fourth International Conference on Intelligent Control and Information Processing (ICICIP), Beijing, China, 2013.

[28] G. Liu, S. Li and W. Liu, "Lane Detection Algorithm based on Local Feature Extraction," in Proceedings of the Chinese Automation Congress, Changsha, Hunan, China, 2013.

[29] J. Huang, H. Liang, Z. Wang, Y. Song and Y. Deng, "Lane Marking Detection based on Adaptive Threshold Segmentation and Road Classification," in Proceedings of the 2014 IEEE International Conference on Robotics and Biomimetics, Bali, Indonesia, 2014.

[30] V. D. Nguyen, H. V. Nguyen and J. W. Jeon, "Robust Stereo Data cost With a Learning Strategy," IEEE Transactions on Intelligent Transportation Systems, vol. 18, no. 2, pp. 248-258, 2017.

[31] M. Mueller, "Vision heroes," Vision Zero International, pp. 4-10, January 2017.

[32] H.-Y. Cheng, B.-S. Jeng, P.-T. Tseng and K.-C. Fan, "Lane Detection With Moving Vehicles in the Traffic Scenes," IEEE Transactions on Intelligent Transportation Systems, vol. 7, no. 4, p. 571, 2006.

[33] H. Kuang, Y.-J. L. Xianshi Zhang and L. L. H. Chan, "Nighttime Vehicle Detection Based on Bio-Inspired Image Enhancement and Weighted Score-Level Feature Fusion," IEEE Transactions on Intelligent Vehicle Systems, vol. 18, no. 4, pp. 927-936, 2017.

[34] A. Bar Hillel, R. Lerner, D. Levi and G. Raz, "Recent progress in road and lane detection: a survey," Machine Vision and Applications, vol. 25, pp. 727-745, 2014.

[35] D. Ding, J. Yoo, J. Jung, S. Jin and S. Kwon, "Various Lane marking Detection and Classification for Vision-based Navigation System," in Proceedings of IEEE International Conference on Consumer Electronics (ICCE), Las Vegas, NV, 2015.

[36] Z. Kim, "Robust Lane Detection and Tracking in Challenging Scenarios," IEEE Transactions on Intelligent Transportation Systems, vol. 9, no. 1, pp. 16-26, 2008.

[37] H.-J. Jang, S.-H. Baek and S.-Y. Park, "Curved lane detection using robust feature extraction," in Proceedings of 2014 2nd International Conference on Systems and Infomatics, Shanghai, China, 2014.

[38] R. L. Hoover, S. J. Rao, G. Howe and F. S. Barickman, "Heavy-Vehicle Lane Departure Warning Test Development," National Highway Traffic Safety Administration (NHTSA), Washington, DC, 2014.

[39] D. Cualain, C. Hughes, M. Glavin and E. Jones, "Automotive standards-grade lane departure warning system," IET Intelligent Transport Systems, vol. 6, no. 1, pp. 44-57, 2012.

[40] H. Sun, C. Wang and N. El-Sheimy, "Automatic Traffic Lane Detection for Mobile Mapping Systems," in Proceedings of 2011 International Workshop on Multi- Platform/Multi-Sensor Remote Sensing and Mapping, Xiamen, China, 2011.

[41] W.-w. Zhang, X.-l. Song and G.-x. Zhang, "Real-time departure warning system based on principal component analysis of grayscale distribution and risk evaluation model," Journal of Central South University, vol. 21, pp. 1633-1642, 2014.

[42] K. Throngnumchai, H. Nishiuchi, Y. Matsuno and H. Satoh, "Application of Background Light Elimination Technique for Lane Marker Detection," SAE International, 2013.

[43] S. Suchitra, R. Satzoda and T. Srikanthan, "Identifying Lane Types," in Proceedings of the 16th International IEEE Annual Conference on Intelligent Transportation Systems, The Hague, Netherlands, 2013.

[44] California Department of Transportation (Caltrans), "Mile Marker: A Caltrans Performance Report," Caltrans, 2017.

[45] International Organization for Standardization (ISO), ISO 17361:2007(E): Intelligent transport systems - Lane departure warning systems - Performance requirements and test procedures, Geneva Switzerland: ISO, 2007.

[46] D. Cualain, M. Glavin and E. Jones, "Multiple-camera lane departure warning system for the automotive environment," IET Intelligent Transport Systems, vol. 6, no. 3, pp. 223-234, 2012.

[47] Federal Highway Administration (FHWA), Manual on Uniform Traffic Control Devices (MUTCD), Washington, DC: FHWA, 2009.

[48] National Highway Traffic Safety Administration (NHTSA), Lane Departure Warning System Confirmation Test and Lane Keeping Support Performance Documentation, Washington, DC: NHTSA, 2013.

[49] S. Szabo and R. Norcross, "An Independent Measurement System for Performance Evaluation of Road Departure Crash Warning Systems," National Institute of Standards and Technology (NIST), Gaithersburg, MD, 2006.

[50] A. Houser, J. Pierowicz and D. Fuglewicz, "Concept of Operations and Voluntary Operational Requirements for Lane Departure Warning System (LDWS) On-board Commercial Vehicles," Federal Motor Carrier Safety Administration (FMCSA), Washington, DC, 2005.

[51] M. Hadi, P. Sinha and H. Ozen, "Effect of Pavement Marking Contrast on the Performance of Vision-Based Lane Departure Warning Systems," in Proceedings of 15th World Congress on Intelligent Transport Systems and ITS America's 2008 Annual Meeting, New York, NY, 2008.

[52] F. S. Barickman, L. Smith and R. Jones, "Lane Departure Warning System Research and Test Development," in Paper No. 07-0495, 2007.

[53] European Road Assessment Programme (EuroRAP), Roads that Cars Can Read: A Consultation paper, Hampshire, United Kingdom: EuroRAP, 2011.

[54] S. Szabo and R. Norcross, Recommended Objective Test Procedures for Road Departure Crash Warning Systems, Gaithersburg, MD: National Institute of Standards and Testing (NIST), 2006.

[55] J. Navarro, E. Yousfi, J. Deniel, C. Jallais, M. Bueno and A. Fort, "The impact of false warnings on partial and full lane departure warnings effectiveness and acceptance in car driving," Ergonomics, vol. 59, no. 12, pp. 1553-1564, 2016.

[56] S. Szabo and R. Norcross, "Final Report: Objective Test Procedures for Road Departure Crash Warning Systems," National Highway Traffic Safety Administration (NHTSA), Washington, DC, 2007.

[57] F. S. Barickman, "Lane Departure Warning Systems," 9 May 2005. [Online]. Available: https://www.nhtsa.gov/document/lane-departure-warning-systems. [Accessed 4 August 2017].

[58] J. C. McCall and M. M. Trivedi, "Performance Evaluation of a Vision Based Lane System Designed for Driver Assistance Systems," in Proceedings of IEEE Intelligent Vehicles Symposium, Las Vegas, NV, 2005.

[59] I. Kington, "Twin win," Vision Zero International, pp. 48-52, January 2017.

[60] M. During and K. Lemmer, "Cooperative Maneuver Planning for Cooperative Driving," IEEE Intelligent Transportation Systems Magazine, pp. 8-22, Fall 2016.

[61] B. Peters, C. Kuhrt and K. Rink, "I spy: V2X technology can provide information about objects and conditions although they may be obscured-and Continental has already demonstrated the first use cases," Vision Zero International, pp. 60-61, January 2017.

[62] K.-H. Glander, "Human factors: The safe development of partially automated driving functions requires a strong focus on the driver," Vision Zero International, pp. 52-53, June 2016.

[63] K. Kim, K. Lee, B. Ko, B. Kim and K. Yi, "Design of Integrated Risk Management-Based Dynamic Driving Control of Automated Vehicles," IEEE Intelligent Transportation Systems Magazine, pp. 57-73, Spring 2017.

[64] J. A. Misener, C. Thorpe, R. Hearne, L. Johnson and A. C. Segal, Enhancing Driver-Assist Sensors: Background and Concepts for Sensor-Friendly Vehicles and Roadways, Misener, 1999.

[65] J. A. Misener, P. Griffiths, L. Johnson and A. Segal, "Sensor-Friendly Freeways: Investigation of Progressive Roadway Changes to Facilitate Deployment of AHS," California Partners for Advanced Transportation Technology (PATH), Berkeley, CA, 2001.

[66] J. A. Misener, C. Thorpe, R. Ferlis, R. Hearne, M. Siegal and J. Perkowski, "Sensor- friendly Vehicle and Roadway Cooperative Safety Systems: Benefits Estimation," Transportation Research Record, no. 1746, pp. 22-29, 2001.

[67] H. G. Jun, Y. H. Lee, P. J. Yoon and J. Kim, "Forward Sensing System for LKS + ACC," in Proceedings of SAE International 2008 World Congress, Detroit, MI, 2008.

[68] L. Saleh, P. Chevrel, F. Claveau, J.-F. Lafay and F. Mars, "Shared Steering Control Between a Driver and an Automation: Stability in the Presence of Driver Behavior Uncertainty," IEEE Transactions on Intelligent Transportation Systems, vol. 14, no. 2, pp. 974-983, 2013.

[69] C. Visvikis, T. Smith, M. Pitcher and R. Smith, "Study on lane departure warning and lane change assistant systems," Transport Research Laboratory, Berkshire, United Kingdom, 2008.

[70] A. Merah, K. Hartani and A. Draou, "A new shared control for lane keeping and road departure prevention," Vehicle System Dynamics, vol. 54, no. 1, pp. 86-101, 2016.

100

[71] K. D. Kusano and H. C. Gabler, "Field Relevance of the New Car Assessment Program Lane Departure Warning Confirmation Test," SAE International Journal of Passenger Cars-Mechanical Systems, vol. 5, no. 1, pp. 253-264, 2012.

[72] L. Eriksson, A. Bolling, T. Alm, A. Andersson, C. Ahlstrom, B. Blissing and G. Nilsson, "Driver acceptance and performance with LDW and rumble strips assistance in unintentional lane departures," Virtual Prototyping and Assessment by Simulation (ViP), Linkping, Sweden, 2013.

[73] W. Wang, D. Zhao, J. Xi and W. Han, "A Learning-Based Approach for Lane Departure Warning Systems with a Personalized Driver Model," arXiv:1702.01228[cs.LG], 2017.

[74] D. LeBlanc, J. Sayer, C. Winkler, R. Ervin, S. Bogard, J. Devonshire, M. Mefford, M. Hagan, Z. Bareket, R. Goodsell and T. Gordon, "Road Departure Crash Warning System Field Operational Test: Methodology and Results," National Highway Traffic Safety Administration (NHTSA), Washington, DC, 2006.

[75] J. S. Jermakian, S. Bao, M. L. Buonarosa, J. R. Sayer and C. M. Farmer, "Effects of an integrated collision warning system on teenage driving behavior," Journal of Safety Research, vol. 61, pp. 65-75, 2017.

[76] A. H. Eichelberger and A. T. McCartt, "Toyota drivers' experiences with Dynamic Radar Cruise Control, Pre-Collision System, and Lane-Keeping Assist," Journal of Safety Research, vol. 56, pp. 67-73, 2016.

[77] S. Tanaka, T. Mochida, M. Aga and J. Tajima, "Benefit Estimation of a Lane Departure Warning System using ASSTREET," SAE International Journal of Passenger Cars - Electronic and Electrical Systems, vol. 5, no. 1, pp. 133-145, 2012.

101

102

APPENDIX A: TABULAR MATERIAL PROPERTIES MiniScan Observations, Phase I Sample Color Marking Y Pavement Y Ratio x y WT1 White 64.23 13.33 4.8 0.3178 0.3368 WT2 White 59.78 12.95 4.6 0.3164 0.335 WT3 White 27.61 16.20 1.7 0.3229 0.341 WT4 White 23.35 14.70 1.6 0.3101 0.3279 WT5 White 52.03 15.74 3.3 0.3169 0.3357 WT6 White 46.66 15.92 2.9 0.3177 0.3365 WT6C White 48.54 4.64 10.5 0.3177 0.3369 WT7 White 48.00 12.60 3.8 0.3183 0.337 YT1 Yellow 13.00 14.75 0.9 0.3545 0.3709 YT2 Yellow 36.53 14.71 2.5 0.4799 0.4312 YT3 Yellow 31.95 11.50 2.8 0.4612 0.4272 YT4 Yellow 31.09 14.24 2.2 0.4602 0.4274 YT5 Yellow 29.28 11.40 2.6 0.4497 0.4205 YT6 Yellow 42.36 14.65 2.9 0.4874 0.4374

MiniScan Observations, Phase II Sample Color Marking Y Pavement Y Ratio x y WT1 White 62.95 14.47 4.4 0.318 0.3382 WT7 White 49.63 16.09 3.1 0.3191 0.3391 WP1 White 30.23 15.91 1.9 0.3256 0.3415 WP2 White 26.76 14.92 1.8 0.3275 0.3458 WP3 White 30.05 12.93 2.3 0.3258 0.3422 YT2 Yellow 38.33 14.55 2.6 0.4872 0.4322 YT3 Yellow 35.42 16.12 2.2 0.4705 0.4311 YT4 Yellow 30.31 12.90 2.4 0.4665 0.4285 YP1 Yellow 31.88 16.57 1.9 0.3407 0.3597 YP2 Yellow 29.60 14.66 2.0 0.3526 0.3701 YP3 Yellow 29.46 12.41 2.4 0.3501 0.3674

A-1

Daytime Chromaticity Coordinates White Samples Phase I

0.6 White Color Box Yellow Color Box 0.5 WT1 WT2 WT3 0.4 YT4 WT5

y 0.3 WT6 WT6C 0.2 WT7

0.1

0 0 0.1 0.2 0.3 0.4 0.5 0.6

Phase I White Marking Samples Chromaticity Coordinates

Daytime Chromaticity Coordinates Yellow Samples Phase I

0.6 White Color Box

Yellow Color Box 0.5 YT1

YT2 0.4 YT3

YT4

y 0.3 YT5

0.2 YT6

0.1

0 0 0.1 0.2 0.3 0.4 0.5 0.6

Phase I Yellow Marking Samples Chromaticity Coordinates

A-2

Daytime Chromaticity Coordinates White Samples Phase II

0.6 White Color Box

0.5 Yellow Color Box WT1

0.4 WT7

WP1

y 0.3 WP2

WP3 0.2

0.1

0 0 0.1 0.2 0.3 0.4 0.5 0.6

Phase II White Marking Samples Chromaticity Coordinates

Daytime Chromaticity Coordinates Yellow Samples Phase II

0.6 White Color Box

0.5 Yellow Color Box

YT2 0.4 YT3

y 0.3 YT4

YP1 0.2 YP2

0.1 YP3

0 0 0.1 0.2 0.3 0.4 0.5 0.6

Phase II Yellow Marking Samples Chromaticity Coordinates

A-3

Delta LTL-XL Measurements, Phase I Qd Marking Pavement Sample Direction (mcd/m2/lux) (mcd/m2/lux) Ratio WT1 NB 227 58 3.9 WT2 NB 200 62 3.2 WT3 NB 137 55 2.5 WT4 NB 101 56 1.8 WT5 NB 153 69 2.2 WT6 NB 125 70 1.8 WT6C NB 117 59 2.0 WT7 NB 128 59 2.2 YT1 NB 85 66 1.3 YT2 NB 125 55 2.3 YT3 NB 99 59 1.7 YT4 NB 90 50 1.8 YT5 NB 95 52 1.8 YT6 NB 176 69 2.6 WT1 SB 243 53 4.6 WT2 SB 198 60 3.3 WT3 SB 135 52 2.6 WT4 SB 102 54 1.9 WT5 SB 154 65 2.4 WT6 SB 128 66 1.9 WT6C SB 117 55 2.1 WT7 SB 126 55 2.3 YT1 SB 85 63 1.4 YT2 SB 128 52 2.5 YT3 SB 102 59 1.7 YT4 SB 91 49 1.9 YT5 SB 94 51 1.8 YT6 SB 175 67 2.6

A-4

Delta LTL-XL Measurements, Phase II Qd Marking Pavement Sample Direction (mcd/m2/lux) (mcd/m2/lux) Ratio WT1 NB 249 65 3.9 WT7 NB 127 71 1.8 WP1 NB 88 62 1.4 WP2 NB 77 60 1.3 WP3 NB 57 53 1.1 YT2 NB 136 63 2.2 YT3 NB 120 69 1.7 YT4 NB 95 55 1.7 YP1 NB 97 65 1.5 YP2 NB 81 60 1.3 YP3 NB 52 54 1.0 WT1 SB 248 65 3.8 WT7 SB 127 68 1.9 WP1 SB 86 62 1.4 WP2 SB 79 60 1.3 WP3 SB 65 53 1.2 YT2 SB 136 63 2.2 YT3 SB 119 68 1.8 YT4 SB 93 55 1.7 YP1 SB 99 64 1.5 YP2 SB 80 60 1.3 YP3 SB 51 54 1.0

A-5

Delta LTL-X Mark II Measurements, Phase I

RL Recovery RL Marking Pavement Marking Sample Direction (mcd/m2/lux) (mcd/m2/lux) Ratio (mcd/m2/lux) Ratio WT1 NB 1021 13 75.9 100 49.8 WT2 NB 119 15 7.8 34 17.1 WT3 NB 31 13 2.3 20 9.9 WT4 NB 105 13 7.9 180 89.8 WT5 NB 46 16 2.9 28 14.1 WT6 NB 24 17 1.4 16 7.9 WT6C NB 34 14 2.5 21 10.4 WT7 NB 101 14 7.3 34 17.1 YT1 NB 33 14 2.3 1 0.3 YT2 NB 404 13 31 52 26.1 YT3 NB 100 14 7.3 17 8.5 YT4 NB 193 13 14.9 21 10.3 YT5 NB 66 13 5.2 15 7.5 YT6 NB 31 17 1.8 11 5.3 WT1 SB 931 12 76.5 100 49.8 WT2 SB 134 14 9.7 34 17.1 WT3 SB 30 12 2.4 20 9.9 WT4 SB 108 12 8.9 180 89.8 WT5 SB 49 15 3.4 28 14.1 WT6 SB 23 15 1.6 16 7.9 WT6C SB 36 12 3.0 21 10.4 WT7 SB 102 13 8.0 34 17.1 YT1 SB 33 14 2.3 1 0.3 YT2 SB 374 13 29.8 53 26.1 YT3 SB 86 13 6.6 17 8.5 YT4 SB 183 12 15.2 21 10.3 YT5 SB 61 12 5.1 15 7.5 YT6 SB 28 16 1.8 11 5.3 2 The pavement recovery RL was 2 mcd/m /lux for each section.

A-6

LTL-X Mark II Recovery RL Measurements, Phase II

RL Recovery RL Marking Pavement Marking Sample Direction (mcd/m2/lux) (mcd/m2/lux) Ratio (mcd/m2/lux) Ratio WT1 NB 928 15 61.5 172 86.1 WT7 NB 88 17 5.2 44 22.2 WP1 NB 24 17 1.4 4 2.0 WP2 NB 27 16 1.7 4 2.1 WP3 NB 14 16 0.9 3 1.7 YT2 NB 466 17 28.1 49 24.4 YT3 NB 79 18 4.5 23 11.7 YT4 NB 197 15 12.8 17 8.5 YP1 NB 31 18 1.7 5 2.3 YP2 NB 53 18 3.0 9 4.7 YP3 NB 17 16 1.0 5 2.6 WT1 SB 861 15 57.1 172 86.1 WT7 SB 92 17 5.4 44 22.2 WP1 SB 22 17 1.3 4 2.0 WP2 SB 33 16 2.0 4 2.1 WP3 SB 15 16 1.0 3 1.7 YT2 SB 474 17 28.6 49 24.4 YT3 SB 86 18 4.9 23 11.7 YT4 SB 201 15 13.1 17 8.5 YP1 SB 31 18 1.8 5 2.3 YP2 SB 41 18 2.3 9 4.7 YP3 SB 16 16 1.0 5 2.6 2 The pavement recovery RL was 2 mcd/m /lux for each section.

A-7

Phase I Daytime Dry CCD Luminance Measurements

Northbound Southbound

Marking Pavement Marking Pavement Sample Color (cd/m2) (cd/m2) Ratio (cd/m2) (cd/m2) Ratio WT1 White 12286 2217 5.5 7481 2246 3.3 WT2 White 10404 2121 4.9 5515 1897 2.9 WT3 White 4683 1887 2.5 7453 2880 2.6 WT4 White 5250 3121 1.7 4190 2552 1.6 WT5 White 7268 2829 2.6 4679 2474 1.9 WT6 White 5397 3215 1.7 3547 2655 1.3 WT6C White ------WT7 White 6358 2659 2.4 4475 2479 1.8 YT1 Yellow 2253 2803 0.8 5385 2909 1.9 YT2 Yellow 6400 2033 3.2 3953 2403 1.6 YT3 Yellow 5065 1872 2.7 5159 2409 2.1 YT4 Yellow 4544 2419 1.9 4445 2215 2.0 YT5 Yellow 3927 1993 2.0 5556 2444 2.3 YT6 Yellow 7413 2761 2.7 6028 3099 1.9 The dash symbol (-) indicates no data available for a particular sample

Phase II Daytime Dry CCD Measurements

Northbound Southbound

Marking Pavement Ratio Marking Pavement Sample Color (cd/m2) (cd/m2) (cd/m2) (cd/m2) Ratio WT1 White 16794 3226 5.2 13810 3141 4.4 WT7 White 7745 3513 2.2 7192 3853 1.9 WP1 White 6089 3151 1.9 6583 3584 1.8 WP2 White 5502 3555 1.6 5375 3068 1.8 WP3 White 4170 3013 1.4 5072 2592 2.0 YT2 Yellow 8717 3107 2.8 7390 3738 2.0 YT3 Yellow 6767 3654 1.9 6765 4347 1.6 YT4 Yellow 5701 3233 1.8 5514 3328 1.7 YP1 Yellow 6568 3494 1.9 6597 3654 1.8 YP2 Yellow 5614 3532 1.6 6170 3824 1.6 YP3 Yellow 3918 2834 1.4 4085 2618 1.6

A-8

Phase I Daytime Wet CCD Luminance Measurements Northbound Southbound Marking Pavement Marking Pavement Sample Color (cd/m2) (cd/m2) Ratio (cd/m2) (cd/m2) Ratio WT1 White 2888 555 5.2 2102 744 2.8 WT2 White 2698 681 4.0 2698 681 4.0 WT3 Yellow 479 1202 0.4 479 1202 0.4 WT4 White 1743 789 2.2 1743 789 2.2 WT5 White 3344 1025 3.3 3344 1025 3.3 WT6 White 3745 1564 2.4 3745 1564 2.4 WT6C White ------WT7 White 3555 1080 3.3 3555 1080 3.3 YT1 Yellow 2415 2010 1.2 2415 2010 1.2 YT2 Yellow 1584 1399 1.1 1818 1821 1.0 YT3 Yellow 1716 1640 1.1 1716 1640 1.1 YT4 Yellow 1884 1464 1.3 1884 1464 1.3 YT5 Yellow 2268 1670 1.4 2268 1670 1.4 YT6 Yellow 3137 1790 1.8 3137 1790 1.8 The dash symbol (-) indicates no data available for a particular sample

Phase II Daytime Wet CCD Measurements Northbound Southbound Marking Pavement Marking Pavement Sample Color (cd/m2) (cd/m2) Ratio (cd/m2) (cd/m2) Ratio WT1 White 13707 2810 4.9 12979 2455 5.3 WT7 White 4311 2194 2.0 4596 2204 2.1 WP1 White 4943 3465 1.4 4900 3044 1.6 WP2 White 5213 3606 1.5 4501 2605 1.7 WP3 White 4278 3147 1.4 3826 2096 1.8 YT2 Yellow 7873 3810 2.1 7960 3341 2.4 YT3 Yellow 4466 3276 1.4 4688 3001 1.6 YT4 Yellow 4838 3536 1.4 5409 2952 1.8 YP1 White 5467 3237 1.7 5429 2774 2.0 YP2 White 5484 3268 1.7 5111 2317 2.2 YP3 White 3682 2661 1.4 2881 1578 1.8

A-9

Phase I Nighttime Dry CCD Luminance Measurements Northbound Southbound Marking Pavement Marking Pavement Sample Color (cd/m2) (cd/m2) Ratio (cd/m2) (cd/m2) Ratio WT1 White 11.71 0.34 34.22 12.82 0.37 34.89 WT2 White 3.32 0.46 7.27 4.52 0.40 11.43 WT3 Yellow 0.67 0.36 1.85 0.59 0.32 1.85 WT4 White 1.99 0.29 6.93 1.86 0.36 5.14 WT5 White 1.33 0.44 3.03 1.22 0.44 2.76 WT6 White 0.58 0.35 1.66 0.86 0.56 1.53 WT6C White ------WT7 White 1.55 0.33 4.64 2.01 0.44 4.52 YT1 Yellow 0.54 0.52 1.05 0.41 0.44 0.94 YT2 Yellow 7.24 0.39 18.70 6.02 0.35 17.24 YT3 Yellow 2.15 0.51 4.19 1.58 0.35 4.47 YT4 Yellow 3.57 0.37 9.59 2.48 0.33 7.55 YT5 Yellow 1.24 0.35 3.55 1.01 0.37 2.75 YT6 Yellow 1.18 0.57 2.07 0.91 0.46 1.95 The dash symbol (-) indicates no data available for a particular sample

Phase II Nighttime Dry CCD Measurements Northbound Southbound Marking Pavement Marking Pavement Sample Color (cd/m2) (cd/m2) Ratio (cd/m2) (cd/m2) Ratio WT1 White 11.98 0.29 41.28 13.78 0.38 35.82 WT7 White 1.07 0.28 3.81 1.50 0.40 3.78 WP1 White 0.46 0.32 1.43 0.57 0.43 1.32 WP2 White 0.79 0.33 2.38 1.25 0.45 2.75 WP3 White 0.36 0.34 1.07 0.53 0.38 1.37 YT2 Yellow 8.89 0.31 28.22 6.28 0.38 16.38 YT3 Yellow 1.32 0.36 3.72 1.66 0.39 4.25 YT4 Yellow 3.36 0.29 11.51 3.43 0.32 10.57 YP1 Yellow 0.67 0.37 1.78 0.66 0.40 1.66 YP2 Yellow 1.35 0.36 3.76 1.40 0.45 3.13 YP3 Yellow 0.55 0.36 1.53 0.49 0.32 1.52

A-10

Phase II Nighttime Wet CCD Measurements

Northbound Southbound Marking Pavement Marking Pavement Sample Color (cd/m2) (cd/m2) Ratio (cd/m2) (cd/m2) Ratio WT1 White 3.25 0.10 31.73 3.25 0.10 31.73 WT7 White 0.85 0.10 8.58 0.85 0.10 8.58 WP1 White 0.20 0.08 2.62 0.20 0.08 2.62 WP2 White 0.16 0.05 3.22 0.16 0.05 3.22 WP3 White 0.14 0.05 2.49 0.14 0.05 2.49 YT2 Yellow 1.70 0.06 26.17 1.70 0.06 26.17 YT3 Yellow 0.62 0.05 11.86 0.62 0.05 11.86 YT4 Yellow 1.12 0.07 17.12 1.12 0.07 17.12 YP1 Yellow 0.31 0.13 2.43 0.31 0.13 2.43 YP2 Yellow 0.43 0.11 4.02 0.43 0.11 4.02 YP3 Yellow 0.22 0.08 2.71 0.22 0.08 2.71

Phase II Nighttime Dry Glare CCD Measurements Northbound Southbound Marking Pavement Marking Pavement Sample Color (cd/m2) (cd/m2) Ratio (cd/m2) (cd/m2) Ratio WT1 White ------WT7 White 1.05 0.30 3.46 - - - WP1 White ------WP2 White 0.81 0.36 2.26 - - - WP3 White ------YT2 Yellow ------YT3 Yellow 1.41 0.46 3.03 - - - YT4 Yellow ------YP1 Yellow ------YP2 Yellow 1.43 0.46 3.13 - - - YP3 Yellow ------The dash symbol (-) indicates no data available for a particular sample

A-11

Phase II Nighttime Wet Glare CCD Measurements

Northbound Southbound Marking Pavement Marking Pavement Sample Color (cd/m2) (cd/m2) Ratio (cd/m2) (cd/m2) Ratio WT1 White 7.81 0.16 47.39 - - - WT7 White 1.00 0.14 7.23 - - - WP1 White 0.22 0.12 1.77 - - - WP2 White 0.24 0.11 2.20 - - - WP3 White 0.17 0.10 1.81 - - - YT2 Yellow 2.57 0.23 11.01 - - - YT3 Yellow 0.87 0.24 3.69 - - - YT4 Yellow 1.62 0.28 5.82 - - - YP1 Yellow 0.52 0.32 1.61 - - - YP2 Yellow 0.80 0.23 3.50 - - - YP3 Yellow 0.36 0.20 1.77 - - - The dash symbol (-) indicates no data available for a particular sample

A-12

APPENDIX B: MARKING IMAGES

PHASE I DATA COLLECTION

Phase I Marking Section 1—Samples YT1 [L] and WT6 [R]

B-1

Phase I Marking Section 2—Samples YT6 [L] and WT5 [R]

Phase I Marking Section 3—Samples YT3 [L] and WT2 [R]

B-2

Phase I Marking Section 4—Samples YT5 [L] and WT7 [R]

Phase I Marking Section 5—Samples YT4 [L] and WT4 [R]

B-3

Phase I Marking Section 6—Samples YT2 [L] and WT1 [R]

Phase I Marking Section 7—Samples WT3 [L] and WT6C [R]

B-4

PHASE I LIGHTED SECTION SAMPLES

Phase I Overheard Lighting Section 1—Samples WT6 [L] and YT7 [R]

Phase I Overheard Lighting Section 2—Samples WT5 [L] and YT6 [R]

B-5

Phase I Overheard Lighting Section 3—Samples YT3 [L] and WT2 [R]

Phase I Overheard Lighting Section 4—Samples YT5 [L] and WT7 [R]

B-6

PHASE II DATA COLLECTION

Phase II Marking Section 2—Samples YT3 [L] and WT7 [R]

Sample YT3

B-7

Sample WT7

Phase II Marking Section 3—Samples YT4 [L]

B-8

Sample YT4

Phase II Marking Section 4—Samples YT2 [L] and WT1 [R]

B-9

Sample YT2

Sample WT1

B-10

Phase II Marking Section 5—Samples YP1 [L] and WP1 [R]

Sample YP1

B-11

Sample WP1

Phase II Marking Section 6—Samples YP2 [L] and WP2 [R]

B-12

Sample YP2

Sample WP2

B-13

Phase II Marking Section 7—Samples YP3 [L] and WP3 [R]

Sample YP3

B-14

Sample WP3

B-15

APPENDIX C: BOX-AND-WHISKER PLOTS USING ADDITIONAL MATERIAL PROPERTIES

Average Rating vs. Qd Contrast Ratio by Sample during Daytime Dry Conditions, Explorer Only

C-1

Average Rating vs. Qd Contrast Ratio by Sample during Daytime Dry Conditions, F-150 Only

C-2

Average Rating vs. Qd Contrast Ratio by Sample during Daytime Wet Conditions, Explorer Only

C-3

Average Rating vs. Luminance Contrast Ratio by Sample during Daytime Dry Conditions, Explorer Only

C-4

Average Rating vs. Luminance Contrast Ratio by Sample during Daytime Dry Conditions, F-150 Only

C-5

Average Rating vs. Luminance Contrast Ratio by Sample during Daytime Wet Conditions, Explorer Only

C-6

Average Rating vs. Luminance Contrast Ratio by Sample during Nighttime Dry Conditions, Explorer Only

C-7

Average Rating vs. Luminance Contrast Ratio by Sample during Nighttime Dry Conditions, F-150 Only

C-8

Average Rating vs. Luminance Contrast Ratio by Sample during Nighttime Wet Conditions, Explorer Only

C-9

Average Rating vs. Luminance Contrast Ratio by Sample during Nighttime Dry Glare Conditions, Explorer Only

C-10

Average Rating vs. Luminance Contrast Ratio by Sample during Nighttime Wet Glare Conditions, Explorer Only

C-11