3-D Data Processing to Extract Vehicle Trajectories from Roadside LiDAR Data
Civil and Environmental Engineering
StatisticsView Usage Statistics
Accurate, high-resolution vehicle data including location, speed, and direction is essential to connected-vehicle applications, micro-level traffic performance evaluation, and adaptive traffic control. The connected-vehicle system provides longer detection distance for drivers (or autonomous vehicles) and pedestrians to “see” around corners or “through” other vehicles so that threats can be perceived earlier. The current connected-vehicle system highly relies on information broadcasted by each vehicle. The maximum safety benefits of current connected-vehicle deployment would need all vehicles to have connected-vehicle devices and broadcast their information in real time. However, the mixed traffic with connected-vehicles and unconnected-vehicles will exist for the next decades or even longer. Therefore, supplemental data to help the connected-vehicle deployment needs to be considered. The traditional traffic sensors only generate macro information like occupancy and estimated average speed. Hence the existed sensors would not be able to aid the research and application of connected vehicles with high accuracy vehicle data. Hence another innovative method which could yield high-resolution traffic data is desired. This research developed a data processing procedure for detection and tracking of multi-lane multi-vehicle trajectories with a roadside Light Detection and Ranging (LiDAR) sensor. Different from existing perception methods for the autonomous vehicle system, this procedure was explicitly developed to extract trajectories from a roadside LiDAR sensor. This procedure includes six steps. They are preprocessing of the raw data, statistical outlier removal, least median of squares-based ground estimation method to accurately remove the ground points, vehicle data clouds clustering, principal component-based oriented bounding box method to estimate the location of the vehicles and geometrically based tracking algorithm. The developed procedure has been tested against the intersection of Evans Street and Enterprise Road; a two way stops sign intersection; and Kietzke lane, an arterial road with 40 mph speed limit in Reno, Nevada. Then, the data extraction procedure has been validated by comparing tracking results and speeds logged from a testing vehicle through the onboard diagnostics interface (OBD-I), at a parking lot of University of Nevada, Reno. The validation results suggest that the tracking speed matches real driving speed accurately.A case study was conducted to examine the accuracy of tracking multiple objects on the roads. 1000 data frames from the intersection of 15th Street and Virginia Street in Reno, Nevada, were used as source data frames. The proposed data processing framework successfully tracked 37 objects out of 38 objects on the road, which gives an accuracy of 97.4%. Then a support vector machine-based algorithm was developed to differentiate pedestrians/bicyclists and cars/buses. With the Radial Basis Function (RBF) kernel, this algorithm correctly classifies 35 objects among 38 objects, which gives an accuracy of 92.1%. The result of this case study indicates that the proposed data processing framework has a satisfactory tracking and clustering accuracy which could be used for traffic micro information extraction.This data processing procedure not only could be applied to extract high-resolution trajectories for connected-vehicle applications, but it could also be valuable to practices in traffic safety, traffic mobility, and fuel efficiency estimation. The ordinary Rectangular Rapid Flash Beacon (RRFB) could be upgraded to detect pedestrians automatically; this is especially important during night time. Adaptive traffic signal control which could adapt to special events or economic changes also becomes feasible from this research. Driving cycle development, which mainly relies on sampling vehicles, could become much more accurate because this research enables the possibilities to extract every vehicle’s speed profile. In sum, this research provides a reliable way to extract high-resolution traffic data of all vehicles in the detection range of a roadside LiDAR, and it would benefit research in connected vehicles, traffic safety, traffic mobility and fuel consumption estimation.