Point Cloud Data Registration And Traffic Target Detection Based On Multiple Roadside LiDARs

Traffic congestion and traffic safety are the current worldwide problems,which seriously restrict social and economic development.Intelligent transportation characterized by vehicle-road collaboration has become an effective way to solve traffic problems.Various advanced sensors are deployed on the roadside to sense and acquire traffic information,which has become an important mode to solve traffic problems.LiDAR sensors have the characteristics of high temporal and spatial resolution and active ranging,and are an effective means to perceive and acquire high-resolution micro traffic data in traffic scenes.Since one single roadside LiDAR has some shortcomings such as limited detection distance and mutual occlusion between targets.To obtain large-scale and complete traffic information,it is necessary to deploy multiple LiDARs on the roadside,constructing a multi-LiDAR system.This paper conducts research on point cloud registration and traffic target detection with multiple roadside LiDARs.The main contributions are described as follows:(1)Aiming at the problems of large scanning parallax between multiple LiDARs,obstacle occlusion,and missing features in point cloud registration,a method for LiDAR point cloud data registration based on retroreflective reference objects is proposed.This method uses the reference object point cloud registration to represent the global point cloud registration.A LiDAR point cloud registration calibration plate is designed,and the surface is coated with retroreflective material to enhance the high reflectivity features.At the same time,an automatic extraction algorithm for reference point cloud is designed.The Iterative Closest Point algorithm is used to register the reference object point cloud data,which solves the problems that the registration algorithm is time-consuming and easy to fall into local optimal.Using the actual traffic scene data for experimental verification,it is found that this LiDAR point cloud data registration method works well.(2)Considering that the application of self-made reference objects is not convenient in complex traffic scenes,a point cloud registration method based on traffic signs is proposed.The method takes the traffic signs widely used in the traffic scene as the registration reference,and designs the registration algorithm based on the high reflectivity characteristics of the reflective film on the surface of the traffic sign.The experimental results show that the method can not only achieve high registration accuracy,but also has the characteristics of fast and accurate in complex traffic scenes.(3)Focusing on the background filtering problem of fused point cloud data,the background difference method,3DDSF algorithm and background reconstruction method are compared and analyzed from the background filtering process,background construction and the filtering accuracy.The experimental results show that the 3DDSF algorithm has high filtering accuracy,can realize automatic background update,and is suitable for roadside multi-LiDAR used in the actual engineering.(4)Aiming at the problem of traffic target detection,a traffic target detection method based on multilateral adaptive threshold is proposed.The method first divides each group of LiDAR point clouds into three sub-regions,and then calculates the clustering threshold of each sub-region according to the LiDAR sensor parameters.In the process of density clustering,a multilateral adaptive threshold strategy is used to achieve accurate detection of traffic targets.The method is experimentally verified by the actual traffic scene data,and the experimental results show that the method can achieve a good detection effect,and the accuracy of traffic target detection can reach more than 95%.