Traffic Data Extraction Based On Roadside 3D LiDAR

Intelligent traffic monitoring is an effective means to alleviate traffic problems.Advanced traffic information sensing technology can promote the rapid development of intelligent traffic monitoring.As an active sensor,Light Detection and Ranging(LiDAR)technology has the advantages of strong anti-interference ability,high spatial resolution and strong environmental adaptability.In this paper,LiDAR is deployed at the roadside for traffic data collection and road monitoring.By analyzing and processing 3D point cloud data,high-resolution micro traffic information can be extracted,which can be used to reveal the internal formation mechanism and evolution process of traffic congestion,traffic safety and other traffic problems.The main contributions are described as follows:(1)For the problem of background scattered point filtering and object point extraction in sparse point cloud,an object point extraction method based on background reconstruction was designed.Firstly,the background data set is constructed using the average and maximum distances of each azimuth.Then,an automatic frame selection algorithm for background construction was designed based on the characteristics of the point height value when vehicle passed.Finally,points are divided into continuous points,accidental points and split points.Background point filtering and object point extraction methods are designed for different types of points.Experimental results show that the effective range can reach 100 meters,and the accuracy rate can reach up to 93%.(2)For the problem of object detection and tracking in sparse point clouds,an object detection and tracking method based on multi-frame data fusion was designed.Firstly,an object detection algorithm based on DBSCAN was designed based on the inherent correlation and density characteristics of traffic object points.Then,for the problem of abnormal clustering,combined with the characteristics of point clouds of moving traffic objects,an object tracking algorithm based on historical frame data fusion was designed.The experimental results show that the accuracy of object detection can reach up to 98%,and the accuracy of object tracking can reach up to 87%.(3)For high-resolution micro traffic data extraction,a traffic object bounding box model construction algorithm was designed.Then,according to the characteristics of moving object points,trajectory information extraction,speed and head distance calculation methods were designed.In the experimental part,the acceleration and deceleration behavior,lane changing behavior,and vehicle following behavior were analyzed with case studies,which reveals the characteristics and change rule of vehicle behavior at a micro scale.