Research And Implementation Of Road Environment Sensing Algorithm Based On Dual Multi-line Laser Radar

Unmanned ground vehicle(UGV)has broad application prospect in both the military and civilian fields.With the development of Internet,artificial intelligence,computer science and other related technologies,the external environment of UGV is also improving.As an important part of UGV system,environment perception plays an important role in the vehicle.In this paper,a double LIDAR based environment perception and data processing architecture is designed.Based on this architecture,two difficult problems of environment perception,curb detection in the structured environment and negative obstacle detection in the unstructured environment,are researched.The major achievements and innovations are follows:As most single LIDAR like 2D or 3D LIDAR with sparse point cloud has poor perception ability for some special scenes,a new environment perception and data processing architecture based on double symmetrically installed LIDAR is designed.The quantitative analysis is used to analyze the point cloud density under the framework.The results of the experiment show that the proposed architecture can greatly improve the point cloud density of the observation area in front of UGV and reduce the blind zone around vehicle body,it makes the UGV can solve more difficult environment perception problems.According to the scanning characteristics of the LIDAR installed horizontally,the distribution characteristics of the LIDAR points in the obstacle area are analyzed.A new curb detection algorithm in the structured environment is proposed.First,the algorithm uses a gradient-consistent point cloud segmentation method to segment the LIDAR points into ground points and obstacle points quickly.Then,we select candidate curb points using ground points and grid map.Finally,the LMS and the improved RANSAC algorithms are used to extract the curbs.Experimental results show that the segmentation algorithm has a good effect and the improved RANSAC algorithm is higher real-time.A new method is proposed for the detection of negative obstacles in unstructured environments.The method does not depend on the flatness of ground,but only use local feature of LIDAR points to detect negative obstacles.Firstly,LIDAR points are projected to multi-scale gird map.Then,point density as well as relative height of each cell is computed.After that,every cell is marked according to the feature.Secondly,geometric feature of negative obstacles is extracted from point cloud,the key points in pair are searched with both statistical characteristics and geometric features.Finally,clustering algorithm is applied to divide negative obstacles.The algorithm has been successfully run in UGV,experiments show the higher real-time,reliability and good detection.The achievement of this research has been successfully used in the "Xingjian 1" UGV.This UGV achieve excellent grades in "Chinese Future Challenge" for many times.