Vehicle Detection And Tracking Based On Roadside 3D LiDAR

The development of intelligent transportation systems has provided new effective technical method for the solution of traffic problems.Light Detection and Ranging(Li DAR)technology is applied to acquire traffic data.The spatial resolution of Li DAR is centimeter and the time resolution of Li DAR is millisecond.Li DAR changes the macroscopic characteristics of traffic data in traditional traffic data acquisition field to a certain extent.High-resolution micro traffic data collected by Li DAR is an essential part of the current autonomous vehicles and connected vehicles.Vehicle detection and tracking are necessary prerequisites for obtaining high-resolution micro traffic data.This paper proposes vehicle detection and tracking methods based on roadside Li DAR.The main contributions of this paper can be described as follows:(1)The characteristic of the 3D cloud points of Li DAR was analyzed.The background subtraction was designed to filter background points and noise points in different traffic roads.(2)For the problem of vehicle detection and tracking in the scenario of the straight road,a historical multi-state joint vehicle detection and tracking algorithm was designed.In vehicle detection,a vehicle abnormity merging algorithm based on feature description was designed to deal with the problem of vehicle abnormal merger.In vehicle tracking,a historical multi-state joint vehicle tracking algorithm based on adjacent frames was designed to deal with the problem of vehicle loss and occlusion.Experimental results show that the tracking accuracy of straight road can up to 87.71%.(3)For the problem of vehicle detection and tracking in the scenario of the intersection,a time window optimization vehicle detection and tracking algorithm was designed.In vehicle detection,the combined clustering algorithm was designed to improved vehicle detection.The vehicle movement state features were used to filter over-detection caused by combined clustering algorithm.In vehicle tracking,the global nearest neighbor algorithm was used to improve vehicle tracking.Experimental results show that the accuracies of both straight road and intersection are higher than 90%.