International Journal of Geo-Information
Article Road Rutting Measurement Using Mobile LiDAR Systems Point Cloud
Luis Gézero 1,2,* and Carlos Antunes 1,3 1 Faculdade de Ciências, Universidade de Lisboa, 1749-016 Lisboa, Portugal 2 LandCOBA, Digital Cartography Consulting Lda, Av. 5 de Outubro, 323, 1649-011 Lisboa, Portugal 3 Instituto Dom Luiz, Universidade de Lisboa, 1749-016 Lisboa, Portugal * Correspondence: [email protected]
Received: 23 August 2019; Accepted: 9 September 2019; Published: 11 September 2019
Abstract: Road rutting caused by vehicle loading in the wheel path is a major form of asphalt pavement distress. Hydroplaning and loss of skid resistance are directly related to high road rutting severity. Periodical measurements of rut depth are crucial to maintenance and rehabilitation planning. In this study, we explored the feasibility of using point clouds gathered by Mobile LiDAR systems to measure the rut depth. These point clouds that are collected along roads are usually used for other purposes, namely asset inventory or topographic survey. Taking advantage of available clouds to identify rutting severity in critical pavement areas can result in considerable economic and time saving and thus, added value, when compared with specific expensive rut measuring systems. Four different strategies of cloud points aggregation are presented to create the cross-section of points. Such strategies were established to improve the precision of individual sensor measurements. Despite the 5 mm precision of the used system, it was possible to estimate rut depth values that were slightly inferior. The rut depth values obtained from each cross-section strategy were compared with the manual field measured values. The cross-sections based on averaged cloud points sensor profile aggregation was revealed to be the most suitable strategy to measure rut depth. Despite the fact that the study was specifically conducted to measure rut depth, the evaluation results show that the methodology can also be useful for other mobile LiDAR point clouds cross-sections applications.
Keywords: Mobile LiDAR systems; point clouds; LiDAR profiles; rutting; pavement distress
1. Introduction Rutting is a longitudinal, permanent pavement deformation along roads created by repetitive vehicle loading in the wheel path. Water accumulation on ruts reduces skid resistance and increases the danger of hydroplaning [1–4]. Usually, three levels of severity are used for rut classification (low, medium, and high), however, there is no consensus regarding the rut severity classification thresholds among countries and agencies [2,3,5]. Regardless of the classification values that are used, there is a wide-ranging consensus of the direct correlation between rut depth severity and increased traffic accidents. Therefore, periodical measurement of road rut depth values is essential for road maintenance intervention plans, avoiding road safety degradation and saving money [1–5]. The manual method to measure the rut depth is performed by placing a straight edge across a rut and measuring the distance between the straight edge and the rut bottom [4]. Along with being a slow procedure, this task usually requires safety measures, which implies use disruptions or even road use interruption. One of the procedures specifically used to perform the rut depth measurement includes a vehicle with a discrete number of laser beams installed on a bar perpendicular to the direction of the road. By comparing the different distances to the ground, the rut depth measurement can be performed by a
ISPRS Int. J. Geo-Inf. 2019, 8, 404; doi:10.3390/ijgi8090404 www.mdpi.com/journal/ijgi ISPRS Int. J. Geo-Inf. 2019, 8, 404 2 of 16 ISPRS Int. J. Geo-Inf. 2019, 8, x FOR PEER REVIEW 2 of 16 minimuma minimum of threeof three laser laser beams—one beams—one in each in each wheel wheel path path and anotherand another in the in car’s the center.car’s center. However, However, since thesince system the system vehicle vehicle does not does have not the have means the tomeans ensure to thatensure the that wheels the arewheels always are inalways the rut’s in the deepest rut’s point,deepest the point, number the of number laser beams of laser is directly beams correlated is directly with correlated the precision with the of the precision rut depth of measurement.the rut depth Themeasurement. lower the number The lower of laserthe number beams, theof laser lower beams, the chances the lower of the the rut’s chances deepest of the point rut’s being deepest detected point andbeing measured detected and, and consequently, measured and, the consequently, rut depth can the be underestimated.rut depth can be Initially underestimated. having three Initially laser beams,having thisthree number laser beams, has increased, this number for example, has increased, Serigos for et example, al. [6] mention Serigos a et 37-laser al. [6] mention beams system. a 37-laser beamsAnother system. rut depth measurement specific technique includes the use of continuous 3D laser systems.Another Such systemsrut depth allow measurement the measurement specific of techniqu thousandse ofincludes points inthe each use profile, of continuous increasing 3D the laser rut depthsystems. precision. Such systems However, allow these the systems measurement are developed of thousands specifically of points for pavementin each profile, distress increasing assessment the and,rut depth consequentially, precision. they However, are very these expensive systems [7]. Figureare developed1 shows schematic specifically representations for pavement of manualdistress discreteassessment laser and, profilers consequentially, and continuous they laser are systems very methodologies. expensive [7]. Figure 1 shows schematic representations of manual discrete laser profilers and continuous laser systems methodologies.
(a)
(b)
(c)
FigureFigure 1.1. SchematicSchematic transversaltransversal roadroad profileprofile andand rutrut depthdepth measurementmeasurement techniques.techniques. ((aa)) ManualManual methodmethod usingusing aa straightstraight edgeedge acrossacross thethe rut.rut. ((bb)) DiscreteDiscrete laserlaser beambeam profilerprofiler installedinstalled inin aa frontfront carcar transversaltransversal bar.bar. ((cc)) ContinuousContinuous laserlaser system system installed installed in in the the car’s car’s back.back.
SeveralSeveral studiesstudies regardingregarding discretediscrete andand continuouscontinuous laserlaser systemssystems comparisoncomparison andand precisionprecision evaluationevaluation havehave beenbeen publishedpublished [[6,8].6,8]. BesidesBesides thethe threethree methodologiesmethodologies illustratedillustrated inin FigureFigure1 1,, other other approachesapproaches toto measuremeasure thethe rutrut depthdepth usingusing geoinformationgeoinformation havehave beenbeen presented.presented. InIn [[9–11],9–11], closeclose rangerange terrestrialterrestrial photogrammetryphotogrammetry principlesprinciples areare usedused toto measuremeasure thethe rutrut depthdepthin in forestforest areas. areas. UsingUsing camerascameras suspendedsuspended byby manuallymanually carriedcarried poles,poles, thethe rutrut depthdepth measurementmeasurement isis performedperformed overover thethe three-dimensionalthree-dimensional createdcreated point point cloud. cloud. More recently, Unmanned Aerial Vehicles (UAV) have been found to be an economic method to efficiently take high resolution photos along large road areas. Marra et al. [11] show how the rut
ISPRS Int. J. Geo-Inf. 2019, 8, 404 3 of 16
More recently, Unmanned Aerial Vehicles (UAV) have been found to be an economic method to efficiently take high resolution photos along large road areas. Marra et al. [11] show how the rut depth in a forest path is measured from the automatically created point cloud [11]. Several authors have used UAV to assess pavement distress through image processing, however they did not include rut depth measurements [12–14]. Existing grass, logging residues, and water along ruts are referred to as the main factors that compromise the precision of the photogrammetric methods. Still, in forest paths, a methodology to measure the rut depth using a 2D laser is presented in [15]. The respective obtained results suggest that a LiDAR sensor mounted on a forest vehicle can be a viable method for rut depth data as part of normal forestry operations. It should be noted that all these studies were conducted on unpaved forest roads where the precision requirements are smaller due to the terrain nature and the fact that the ruts are much deeper than in paved roads. In [16–18], recent reviews of pavement distress detection and measurements, including rutting, are presented. Few studies have contributed to rut depth measurement by using LiDAR point cloud data. Li et al. [19] introduce a real-time 3D scanning specific system for pavement distortion inspections. The system uses a simple yet robust structured light method for 3D profile generation. The authors conclude that rutting can be reliably detected from 3D transverse profiles and the real-time measurements and 3D visualization of these distortions can be an output for a pavement management system. Li et al. [20] demonstrate the application of random forest classification to identify pavement distress based on the use of UAV LiDAR point cloud data. The data quality used does not allow the identification of slight pavement distresses, therefore, a classification method mainly for severe pavement distresses was established. Due to the similarity between subsidence, rutting, and cracks, they are all regarded as the same type of distress in the classification, which prevents any type of distinction between the pavement pathologies. Even so, a satisfactory classification with an accuracy of 92.3% was achieved. Mobile LiDAR Systems (MLS) have emerged in recent years as a new source of precise and very detailed georeferenced information. The gathered point clouds with a few thousand of points per square meter represent the road pavement and surrounding area with high spatial resolution. These point clouds become very popular for the collection of accurate objects geometry and their related attributes. Many country’s road agencies are using MLS point clouds to extract road assets’ geometry, attributes, and condition. These systems reveal to be much faster and cheaper than the traditional survey methods [21]. Another MLS advantage is the extraction of many asset types from the same point cloud, gathered with just one road passage and without any road traffic disruptions. Several studies support the general idea of MLS point clouds as a solution for road inventory [22–24]. Che et al. [25] present a complete and recent state of the art review regarding the object’s extraction methods from MLS point clouds. Given the MLS efficiency, it is expected that the object types and extraction methodologies from the gathered point clouds will continue to increase [26]. This will result in a large amount of existing point clouds covering the entire road pavement. Using this available data, an efficient methodology to identify road areas with high rut depth values can clearly be of added value, speeding up the process of problematic areas detection, without the need of specific laser systems for pavement distress detection. Apart from reducing the cost of the process, it can anticipate road interventions and diminish traffic accidents. At first glance, the rut depth measurement can easily be performed through cross-sections along the MLS point cloud. However, different strategies for grouping the point clouds to get the profile points can lead to very different results. Also, the lower accuracy of the commercial MLS systems when compared with the specific laser systems can compromise the rut depth measurement quality. In this work, a comparative study of different strategies to group the cloud points in order to create the cross-sections is presented. By comparing different strategies, it is intended that the precision improvement regarding the individual system sensor measurement can be evaluated. On the one hand, ISPRS Int. J. Geo-Inf. 2019, 8, 404 4 of 16 statically, by increasing the number of cloud points used for each cross-section point computation. ISPRSOn the Int. other J. Geo-Inf. hand, 2019 taking, 8, x FOR advantage PEER REVIEW of the system operating principles knowledge, grouping the 4cloud of 16 points by individual laser rotation. Thus, Section2 includes an MLS working principles description, principlesfollowed by description, each proposed followed strategy by description. each proposed Finally, strategy Sections description.3 and4 presents Finally, the discussionSection 3 ofand the 4 presentsresults, main the discussion conclusions, of the and results, future work.main conclusions, and future work.
2.2. Materials Materials and and Methods Methods SinceSince rutting rutting has has a a longitudinal longitudinal development, development, the the most most obvious obvious way way to to measure measure the the rut rut depth depth is is throughthrough the the creation creation of of road road tran transversalsversal cross-sections. cross-sections. However, However, to to evaluate evaluate the the potential potential precision precision improvementimprovement of thethe sensor’ssensor’s individualindividual measurement, measurement, four four strategies strategies to to aggregate aggregate the the cloud cloud points points for forcreating creating those those cross-sections cross-sections were were set out.set out. The The proposed proposed strategies strategies will will be described be described in this in this section section and andjustification justification of choices of choices will be will included be included in the discussionin the discussion of the results. of the Sinceresults. some Since defined some strategies defined strategiesare related are to related the way to the the data way are the gathered, data are agathered, resumed adescription resumed description of the MLS of working the MLS principles working principlesare presented. are presented.
2.1.2.1. MLS MLS Working Working Principles Principles AnAn MLS MLS consists consists of of a a set of devices and sensors, installed in a moving platform. These These systems systems allowallow gathering gathering georeferenced georeferenced points reflected reflected on the surface of the surrounding objects. TheThe data data registered registered by MLS sensors is classified classified in two types: • TrajectoryTrajectory—includes datadata collected collected by by the the system system sensors sensors to compute to compute the most the accurate most trajectory.accurate • trajectory.Namely, Global Namely, Navigation Global Navigation Satellite System Satellite (GNSS) System antennas, (GNSS) Inertial antennas, Measurement Inertial Measurement Unit (IMU), Unitand high-resolution(IMU), and high-resolution Distance Measuring Distance Instrument Measuring (DMI).Instrument (DMI). • LaserLaser—a—a set set of distances and angles to to compute the the points points coordinates coordinates in in space space regarding regarding the the • laserlaser sensor sensor position. position.
ByBy merging merging these these two two data data types, types, georeferenced georeferenced po pointint clouds clouds representing representing all all objects objects at at a a sensor sensor rangerange are are created. created. ToTo ensure a full surrounding coverage, two two type typess of of movement movement are are applied applied to to the the laser laser pulse pulse direction.direction. The The first first is is a a rotational rotational movement movement 360 360 degrees degrees around around the the sensor sensor axis, axis, creating creating a a single single profile.profile. The The second second is is the the vehicl vehiclee movement, movement, which which allows allows consecutive consecutive profiles profiles and and the the point point cloud cloud effect.effect. Figure Figure 2 illustrates thethe MLSMLS pointpoint cloudcloud gatheringgathering process.process.
Figure 2. Mobile LiDAR Systems (MLS) point cloud effect creation by consecutive sensor profiles and vehicleFigure movement. 2. Mobile LiDAR Systems (MLS) point cloud effect creation by consecutive sensor profiles and vehicle movement. Along this work, two profile types are recurrently mentioned, the original sensor cloud points profilesAlong described this work, in Figuretwo profile2, and types the road are recurrently transversal mentioned, profiles used the for original measuring sensor the cloud rut depth.points profiles described in Figure 2, and the road transversal profiles used for measuring the rut depth. The term profile will always be used when referring to the first type and the term cross-section when referring to the second type.
ISPRS Int. J. Geo-Inf. 2019, 8, 404 5 of 16 ISPRS Int. J. Geo-Inf. 2019, 8, x FOR PEER REVIEW 5 of 16
profile cross-section The termIn order towill increase always the be surround used whening objects referring area to coverage, the first type the andplatform the term laser sensor installationwhen referringis typically to the performed second type. in a way that the profiles direction performs a 45 degree angle regarding the systemIn order trajectory to increase direction, the surrounding as shown in objectsFigure 2. area coverage, the platform laser sensor installation is typically performed in a way that the profiles direction performs a 45 degree angle regarding the system2.2. Cross-Section trajectory direction, Points Creation as shown Strategies in Figure 2. 2.2. Cross-SectionAn MLS point Points cloud Creation dataset Strategies gathered along a highway was used to compare the cross-sections points creation strategies. The used system was a Trimble MX8 with the sensor RIEGL VQ-250 An MLS point cloud dataset gathered along a highway was used to compare the cross-sections configuration. In Table 1, the main sensor characteristics are presented. points creation strategies. The used system was a Trimble MX8 with the sensor RIEGL VQ-250 configuration. In Table1, the main sensor characteristics are presented. Table 1. Riegl VQ-250 sensor main characteristics.
TableCharacteristics 1. Riegl VQ-250 sensor main characteristics.Values Accuracy1 10 mm Characteristics Values Precision2 5 mm 1 MaximumAccuracy measurement rate 300,00010 points/second mm Precision 2 5 mm MaximumNumber measurement of sensor profiles rate Up 300,000 to 200 points profiles/second/second 1 Accuracy is the degreeNumber of conformi of sensorty profilesof a measured Upquantity to 200 to profiles its actual/second (true) value. One sigma at 1 Accuracya 50 m range is the under degree test of conformity conditions. of a measured quantity to its actual (true) value. One sigma at a 50 m range under test conditions. 2 Precision or repeatability is the degree to which further measurements show the same result. 2 OnePrecision sigma at aor 50 repeatability m range under is test the conditions. degree to which further measurements show the same result. One sigma at a 50 m range under test conditions. TheThe point point clouds clouds were were obtained obtained to to perform perform the the highway highway asset asset inventory. inventory. The The strategy strategy goal goal was was to evaluateto evaluate the the use use of those of those point point clouds clouds to additionallyto additionally measure measure the the rut rut depth. depth. The The data data was was already already processedprocessed using using the the IMU IMU and and GNSS GNSS data data and and subsequent subsequent integration integration of of distances distances and and angles angles were were collectedcollected by by the the laser laser sensor. sensor. The The obtained obtained georeferenced georeferenced point point cloud cloud was was exported exported to to LAS LAS format, format, versionversion 1.2. 1.2. InIn Figure Figure3, 3, a samplea sample point point cloud cloud gathered gathered along along a highwaya highway is presented.is presented. To To describe describe each each proposedproposed strategy, strategy, the the red red line line drawn drawn transversal transversal to the to roadthe road will bewill used. be used. An image An image of the of yellow the yellow area willarea be will enlarged. be enlarged. Thus, theThus, continuous the continuous cloud e ffcloudect will effect be lost,will andbe lost, the originaland the cloudoriginal points cloud profiles points willprofiles be exposed will be (Figuresexposed 4,(Figures 6, 8 and 4,6,8,10). 10). The The distance distance between between those those profiles profiles depends depends on on the the vehicle vehicle velocityvelocity and and on on the the sensor sensor rotation rotation frequency, frequency, while while each each profile profile point point density density depends depends on on the the sensor sensor measurementmeasurement rate rate frequency frequency and and on on the the distance distance between between the the reflection reflection point point and and the the sensor. sensor. AlongAlong the the gathering gathering process, process, both both maximum maximum system system frequencies frequencies were were used used and and the the vehicle vehicle velocityvelocity was was always always kept kept below below 50 km 50/h, which,km/h, consideringwhich, considering the system the rotation system frequency, rotation guarantees frequency, aguarantees maximum distancea maximum between distance consecutive between profilesconsecutive of less profiles than 7 of cm. less than 7 cm.
FigureFigure 3. 3.MLS MLS point point cloud cloud and and red red cross-section cross-section line line sample. sample.
2.2.1. Cross-Section Line Projected Points The first proposed approach to create the cross-section points is probably the most intuitive. That is, project in the cross-section line segment all cloud points below a perpendicular predefined
ISPRS Int. J. Geo-Inf. 2019, 8, 404 6 of 16
2.2.1. Cross-Section Line Projected Points The first proposed approach to create the cross-section points is probably the most intuitive. That is,ISPRS project Int. J. inGeo-Inf. the cross-section2019, 8, x FOR PEER line REVIEW segment all cloud points below a perpendicular predefined distance 6 of 16 of that line. Figure4 shows the zoomed yellow area indicated in Figure3 with the red cross-section line.distance A lower of that point line. cloud Figure density 4 shows can the be zoomed observed yellow along area the indicated sensor profiles in Figure on the 3 with left sidethe red since cross- the vehiclesection trajectoryline. A lower is closer point to cloud the right density side. can be observed along the sensor profiles on the left side since the vehicle trajectory is closer to the right side.
FigureFigure 4. Strategy 4. Strategy 1 schematic 1 schematic top view. top Each view. point Each cloud point inside cloud distance inside L is distance projected L isin projectedthe cross-section in the line. cross-section line. In Figure 4, P1 and P2 are the two end points of the cross-section line, P is a random cloud point and QIn is Figure the P4 ,projection P1 and P2 in are the the cross-section two end points line. of The the two-dimensional cross-section line, Euclidian P is a random distance cloud between point andP and Q Q is are the represented P projection by in D, the and cross-section L is the maximum line. The threshold two-dimensional distance Euclidianvalue below distance which between the cloud P andpoints Q are representedconsidered. by D, and L is the maximum threshold distance value below which the cloud pointsThe are point considered. Q XY-coordinates can then be computed by The point Q XY-coordinates can then be computed by