Study On Obstacle Detection And Tracking Algorithms Based On LIDAR

Intelligent vehicle has become the development focus of modern automobile industry,as well as the main direction and key competitive field of automobile technology development in the future.This trend has put forward higher requirements on the acquisition and accuracy of automobile environmental information.As the key technology of environmental information acquisition,environmental perception is very important to ensure the safety of intelligent vehicle.With high resolution,long detection range and strong anti-interference capability,3D LIDAR has been widely used in intelligent vehicle environment perception system.Environmental perception needs to provide real-time and accurate obstacle target status information,so it is necessary to construct an efficient obstacle detection and tracking algorithms based on 3D LIDAR.Aiming at the characteristics of large scale,sparsity and disorder of point cloud,this paper studied and constructed a series of algorithms for environmental modeling,obstacle detection and tracking based on 3D LIDAR,centering on the sensing requirements of intelligent vehicle environment and aiming at obtaining the status information of obstacles in real time.The main research content of this paper is as follows:(1)Aiming at the under-segmentation problem existing in the current ground segmentation algorithm when dealing with slope road surface,a ground segmentation algorithm based on adaptive regional growth plane fitting was proposed to segment the ground.In this algorithm,the ground is represented by a plane model,the terrain is represented by the angle change of the plane normal vector,and the local ground information is considered by the incremental plane fitting,so as to realize the adaptive ground segmentation.Experimental data showed that this algorithm improves the robustness of the ground segmentation algorithm for terrain when dealing with slope road surface.(2)Aiming at the interference problem of non-road boundary points to road boundary fitting,morphological filtering algorithm was used to eliminate the influence of non-road boundary points;aiming at the problem of clustering over-segmentation caused by the sparsity of point cloud,morphological closed operation was used for preprocessing,and connected component labeling algorithm was used for obstacles clustering.The comparison experiment of obstacle detection without road boundary constraint and with road boundary constraint was completed,which verified the importance of road boundary detection and the effectiveness of the algorithm.(3)Aiming at the problems of large amount of data association calculation and false association in multi-target tracking system under complex and dense traffic environment,a tracking gate based on the prior knowledge of obstacle was proposed,and the data association was carried out according to the characteristics of the two frames of obstacle in order to further simplify the calculation.The joint probabilistic data association and Extended Kalman Filter were used to estimate the state of the nonlinear system in order to obtain the state information of the obstacle target.Finally,based on real road environment simulation examples,the feasibility and effectiveness of the proposed algorithms were verified.