Tarko, A. P., Ariyur, K. B., Romero, M. A., Bandaru, V. K., & Lizarazo, C. G. (2016). TScan: Stationary LiDAR for traffic and safety studies—Object detection and tracking (Joint Transportation Research Program Publication No. FHWA/IN/JTRP-2016/24). West Lafayette, IN: Purdue University. https://doi.org/10.5703/1288284316347
The ability to accurately measure and cost-effectively collect traffic data at road intersections is needed to improve their safety and operations. This study investigates the feasibility of using laser ranging technology (LiDAR) for this purpose. The proposed technology does not experience some of the problems of the current video-based technology but less expensive low-end sensors have limited density of points where measurements are collected that may bring new challenges. A novel LiDAR-based portable traffic scanner (TScan) is introduced in this report to detect and track various types of road users (e.g., trucks, cars, pedestrians, and bicycles). The scope of this study included the development of a signal processing algorithm and a user interface, their implementation on a TScan research unit, and evaluation of the unit performance to confirm its practicality for safety and traffic engineering applications.
The TScan research unit was developed by integrating a Velodyne HDL-64E laser scanner within the existing Purdue University Mobile Traffic Laboratory which has a telescoping mast, video cameras, a computer, and an internal communications network. The low-end LiDAR sensor’s limited resolution of data points was further reduced by the distance, the light beam absorption on dark objects, and the reflection away from the sensor on oblique surfaces. The motion of the LiDAR sensor located at the top of the mast caused by wind and passing vehicles was accounted for with the readings from an inertial sensor atop the LiDAR. These challenges increased the need for an effective signal processing method to extract the maximum useful information.
The developed TScan method identifies and extracts the background with a method applied in both the spherical and orthogonal coordinates. The moving objects are detected by clustering them; then the data points are tracked, first as clusters and then as rectangles fit to these clusters. After tracking, the individual moving objects are classified in categories, such as heavy and non-heavy vehicles, bicycles, and pedestrians. The resulting trajectories of the moving objects are stored for future processing with engineering applications. The developed signal-processing algorithm is supplemented with a convenient user interface for setting and running and inspecting the results during and after the data collection.
In addition, one engineering application was developed in this study for counting moving objects at intersections. Another existing application, the Surrogate Safety Analysis Model (SSAM), was interfaced with the TScan method to allow extracting traffic conflicts and collisions from the TScan results. A user manual also was developed to explain the operation of the system and the application of the two engineering applications.
Experimentation with the computational load and execution speed of the algorithm implemented on the MATLAB platform indicated that the use of a standard GPU for processing would permit real-time running of the algorithms during data collection. Thus, the post-processing phase of this method is less time consuming and more practical.
Evaluation of the TScan performance was evaluated by comparing to the best available method: video frame-by-frame analysis with human observers. The results comparison included counting moving objects; estimating the positions of the objects, their speed, and direction of travel; and counting interactions between moving objects. The evaluation indicated that the benchmark method measured the vehicle positions and speeds at the accuracy comparable to the TScan performance. It was concluded that the TScan performance is sufficient for measuring traffic volumes, speeds, classifications, and traffic conflicts. The traffic interactions extracted by SSAM required automatic post-processing to eliminate vehicle interactions at too low speed and between pedestrians – events that could not be recognized by SSAM. It should be stressed that this post processing does not require human involvement. Nighttime conditions, light rain, and fog did not reduce the quality of the results.
Several improvements of this new method are recommended and discussed in this report. The recommendations include implementing two TScan units at large intersections and adding the ability to collect traffic signal indications during data collection.
LiDAR, tracking vehicles, tracking pedestrians, traffic conflicts, traffic measurements
Joint Transportation Research Program
Indiana Department of Transportation
West Lafayette, Indiana
Date of this Version