Research On Traffic Information Detection Method Based On Three-Dimensional LiDAR In Urban Scene

With the development of social economy,the number of automobiles in China is increasing,as well as the traffic pressure.Traffic information is the basis of solving traffic problems,and effective means and methods of obtaining traffic information are the key.In this paper,three-dimensional light detection and ranging(LiDAR)i5 used as a technical means to obtain traffic information.Firstly,LiDAR detection system is built to collect road traffic information through roadside deployment.Secondly,the collected 3D point cloud data are used to achieve the purpose of urban traffic information acquisition through background filtering,road recognition,vehicle target detection and vehicle target classification.The contents of this paper are described as follows:1.A background filtering algorithm based on background differenee method is designed to solve the interference of irrelevant background in point cloud scene to target detection and classification.This algorithm combines the characteristics of point cloud data to improve the background difference method.Firstly,the multi-layer grid background point cloud model is established,and the multi-layer grid background point cloud model and the collected point cloud data frame are used to do the differential operation.The background subtraction in traffic scenes is realized,and the vehicle target point cloud data is extracted.The validity of the algorithm is validated by using experimental data of multiple traffic scenarios.2.Aiming at the problem of incomplete and unclear road boundary caused by low vertical resolution of low-channel LiDAR,an automatic road space recognition algorithm,which is based on the C4.5 Decision Tree,is designed.What to be studied in this paper are the dynamic changes of point clouds caused by vehicles in road space and three characteristics are extracted by analyzing vehicle trajectories,which are used to train C4.5 Decision Tree classifier to realize automatic recognition of road space.In view of the discontinuity of road space identified,mean filtering and Minimum Bounding Rectangle Algorithm are proposed.Finally,the accurate extraction of road space is realized.3.Vehicle detection algorithm and vehicle target classification algorithm are designed.Aiming at the missing of effective vehicle points and misdetection of vehicle in vehicle detection algorithm based on DBSCAN clustering,this paper designs a vehicle detection algorithm based on Hierarchical Maximum Density Clustering of Application with Noise algorithm.This algorithm improves the accuracy of vehicle target recognition by improving clustering center and adding hierarchical clustering.Through analyzing the vehicle point cloud data,the shape,proportion and contour features are extracted and then two classifier algorithms are trained,Random Forest and Support Vector Machine.Experimental comparison proves that the vehicle target classification algorithm based on Random Forest is more accurate,with an accuracy rate of 95.2%.