COMPARISON AND EVALUATION OF ROADSIDE ANIMAL SENSING AND DRIVER WARNING SYSTEMS In Response to 2020 NDOT Problem Statement 20Q2E101 Hao Xu (PI), Ph.D., P.E., Associate Professor (775) 784-6909 Department of Civil and Environmental Engineering University of Nevada, Reno [email protected] Fraser Shilling, PhD., Co-Director Road Ecology Center Department of Environmental Science and Policy University of California, Davis 530-752-7859 [email protected] Jeff Gagnon, Regional Supervisor Wildlife Contracts Branch Connectivity Group Arizona Game and Fish Department Phoenix, AZ 86056 928-814-8925 [email protected] Tim Hazlehurst, President CrossTek EB, LLC Seattle, Washington 330-414-1995 [email protected] 20Q2E101: Comparison and Evaluation of Roadside Animal Sensing and Driver Warning Systems 1. PROBLEM DESCRIPTION Animals regularly enter and attempt to traverse roads, putting themselves and vehicle occupants in danger. This can cause drivers to collide with animals or change direction abruptly to try and avoid a collision, resulting in property damage and injury or death. In areas where fencing or crossing structures are not present or feasible, roadside animal detection systems (RADS) can alert drivers and sometimes the animals, to the presence of the other and reduce the risk of collision. There is an increasing array of technologies that detect animals in the roadside, classify animal type, map movement trajectory, and warn drivers. A comprehensive evaluation and comparison of currently available sensing-technologies for RADS is needed for the Nevada Department of Transportation (NDOT) and other states to select appropriate technologies and reduce animal-vehicle-collisions for driver safety and wildlife conservation. 2. BACKGROUND SUMMARY NDOT and partner agencies have been successful in building wildlife crossing structures and highway fences to reduce wildlife-vehicle collisions. However, for areas with limited NDOT right-of-way or requiring high accessibility, crossing structures and fences are not feasible. The USA Parkway (SR 439) through the Industrial Center is one of these situations. RADS are considered to detect animals nearby or on roads and warn motorists to reduce driver speed and increase attentiveness. Animal detection technologies are normally classified into two groups – break-the-beam sensors and area coverage sensors. Area coverage sensors detect animals within the detection zone of the sensor, but break-the-beam only “see” animals crossing a beam. A major interest in animal sensing and warning systems is area coverage sensing technologies, including video/still camera, Radar, thermal camera, and advanced Light Detection and Range (LiDAR) sensors. Several studies have compared options for animal detection sensors in controlled environments, but these are dated and there is still a lack of widely-agreed-upon sensing solution. Video and still cameras with infrared illumination and real-time data transmission provide visual evidence of animal presence. They have promising detection and tracking capabilities when combined with cutting-edge AI image processing algorithms and hardware. Radar sensors provide longer detection distance than other options, but some radar systems have resulted in high percentages of false positives (sensing systems report animal crossing when there is actually no animal). New radar sensors and AI data processing have been developed for higher roadside detection accuracy, so a further evaluation of the latest radar sensing for animal crossings is needed. The advancements in LiDAR sensors, such as 360-degree 3D LiDAR and image-grade LiDAR sensors, are also promising solutions for wildlife detection, providing rapid detection of large wildlife, vehicles, and pedestrians. However, accurate object classification decreases with distance, which could impact detection and driver warning. Thermal sensors have been used successfully in Arizona at an at-grade crossing where fencing guides animals to a discrete location, but the use of this technology has not been evaluated over longer stretches of road without fencing. While these sensors are also used for driver-assistance programs and autonomous vehicles, the requirements for roadside deployment and data processing are quite different from onboard sensing systems. 1 20Q2E101: Comparison and Evaluation of Roadside Animal Sensing and Driver Warning Systems There is no systematic comparison of these latest area coverage sensing technologies in a real- world environment with various traffic and weather conditions. A comparison of these newer and advanced sensing technologies will provide reference information to traffic and animal agencies for a cost-effective solution and reduction in animal-vehicle collisions. Besides the sensing and classification functions, RADS also need to warn drivers of animal-vehicle collision risks when animals approach or step on roads. Two major warning methodologies are real-time messages to connected and autonomous vehicles (CAVs) and automatic wildlife- crossing warning systems. When different sensing technologies provide various data output, testing appropriate communication for the sensors is critical to RADS efficiency. 3. PROPOSED RESEARCH AND FIELD DEPLOYMENT Our project team includes some of the foremost investigators in the field of animal detection and movement. The team proposes to meet NDOT’s needs by deploying mature instrument technologies side-by-side, the majority of which will be available without cost to the project. Our technologies (instruments and processing) are mature and deployable on the roadside. At the same time, technologies are changing fast enough, including price points, that we also propose to evaluate emerging technologies selectively. This project will be a comparative study in real-world field conditions to compare the ability of multiple sensor types (LiDAR, Thermal, Camera/Video, Radar) for rapid detection and correct classification of animals in varied roadside environments. This project will also develop a deployable combination of one or more sensor types alongside roadways with varying conditions, to determine the most cost-effective stand-alone or combination of sensors. The project team will use an Intelligent Transportation System (ITS) trailer, shown in Figure 1, as the platform for deployment of the integrated animal sensing and warning system. The trailer has been borrowed from NDOT District II for current UNR research on roadside sensing. The trailer with solar panel and wind turbine power devices can provide 24/7 power supply, house edge processing units, remotely control roadside warning signals through radio communication, and support remote system monitoring and operation through LTE connection. A CCTV camera on the top of the trailer can be remotely accessed and operated for equipment security, or to record animal presence. The test system will be placed for one year each at 1) USA Parkway, where wild horses cross the roadway in several places, 2) US 395 between Stead Blvd and White Lake Parkway, the #2 priority site statewide for NDOT for deer-vehicle collisions. NDOT can modify the choice of the second site during the project to meet other needs. To evaluate performance of roadside animal sensing and driver-warning technologies, the project team will integrate the available area-coverage wildlife sensing technologies on the ITS trailer. Data output from each sensor will be processed, validated, and compared in terms of cost, power requirements, installation requirements, communication requirements, range, sensitivity, accuracy, and reliability in different weather conditions. The project teams’ expertise, ready-to- use hardware and software, and long-term field deployment experience in wildlife and traffic detection systems are the key to this project’s success. LiDAR solution - Xu with UNR has been performing research on multi-modal traffic monitoring and tracking with roadside sensing systems for more than five years, especially advanced LiDAR sensing technologies. Software for logging and processing roadside high- resolution LiDAR data have been developed and used at seven sites with permanent LiDAR 2 20Q2E101: Comparison and Evaluation of Roadside Animal Sensing and Driver Warning Systems installation, and multiple portable platforms in northern and southern Nevada. The software converts LiDAR point-clouds to geolocated trajectories of all traffic (including vehicles, pedestrians, animals, and other road users). Figure 1 includes an example of half-hour geolocated vehicles (green dots) and horse trajectories (red dots) from the LiDAR ITS trailer at USA Pkwy. Presence and crossing of horses were confirmed using motion-triggered cameras (Shilling, UC Davis). Accuracy of presence, volume, speed, and location-detection by LiDAR has been validated in previous research and published in journals and conference papers (Xu et al. 2018, Wu et al. 2018). Besides detection of animals nearby and on roads, vehicle trajectories from LiDAR sensors can also report whether and how much a roadside warning reduces traffic speeds. Figure 1. NDOT ITS trailer equipped with equipment of power supply, sensing, edge computation, and wireless communication by UNR; Sample all-traffic trajectories from roadside LiDAR at USA Pkwy (half- hour data of 11/18/2019 2:00-2:30 PM). Green dots represent vehicles and red dots horses. Velodyne VLP-32c sensors scan the surrounding environment with 32 laser beams, and 200- meter detection radius, and they are used at most UNR’s LiDAR deployment