Research On Obstacle Recognition Of Intelligent Vehicle Highway Based On 3D Point Cloud

As a solution to the problems of traffic safety,energy conservation and environmental protection,the intelligent vehicle is an indispensable part of the automotive industry.As the starting point forintelligent vehicle to obtain external information,the perception system can perceive the surrounding environment information and provide information for the decision-making system of intelligent vehicle.Compared with other vehicle sensors,lidar has the advantages of high resolution and strong anti-interference ability,and is widely used in smart cars.In this paper,the VLP-16 LIDAR from Velodyne company is selected as the acquisition device to realise the obstacle recognition of 3D point clouds in highway scenarios,The specific research content is as follows:1.Building intelligent vehicle acquisition platform,understood the structure,working principle and data structure of lidar in detail,and completed the conversion of VLP-16 raw data to point cloud.Reasonable clipping of point cloud can reduce the processing burden of subsequent algorithms.2.The point cloud has a large amount of data and some noise point clouds exist,so the original point cloud is de-noised and de-sampled.The ground point cloud removal based on the fusion of RANSAC algorithm and direct filtering is realized,which improves the accuracy of obstacle clustering.The method has a good effect in the case of existence road undulation.3.The clustering method based on density and distance is discussed.In order to quickly index point cloud data,the K-D tree data structure is established by PCL,the clustering segmentation method based on Euclidean distance is improved,and the accuracy of clustering is improved by setting corresponding thresholds.Three methods of constructing rectangular bounding boxes are compared,and the minimum rectangular bounding boxes of obstacle clustering are established by convex hull method of point cloud cluster,and make a simple classification of obstacles.4.In the vehicle target recognition method,the open source Apollo model is used to project the point cloud into a 2.5D grid map for obstacle recognition,which improves the operation efficiency and ensures the real-time performance of the recognition system.Relevant function packages are compiled under ROS system to verify the feasibility and effectiveness of the method,and relevant function packages are developed through the results of experimental feedback to improve the accuracy of identification.