Research On Obstacle Detection Algorithm Based On Laser Point Cloud Data

Since the birth and development of automobiles,it has made tremendous t echnological progress.With the development of automobile technology,the revo lution of computer technology and the widespread rise of artificial intelligence,research on unmanned vehicle technology has also emerged,and has gradually become the focus of attention of universities,enterprises and governments.Th e current key technologies related to unmanned vehicles mainly include environ mental perception,decision planning,motion control,etc.while perception is c onsidered to be the basis of unmanned vehicles and has a huge role in the ent ire system.On the other hand,the increase in vehicle ownership has caused so me potential risks in certain vehicle usage scenarios.This paper combines unm anned vehicle environmental perception technology with research and analysis o f typical scenes,using point cloud data collected by lidar sensors to detect obs tacles during the driving of unmanned vehicles,and on this basis,the potential in typical scenes involved Risk analysis to improve the safety redundancy of unmanned vehicle systems.The main work and contributions of this paper are as follows:1.In response to the under-segmentation of ground point cloud data and o bstacle point cloud data during the process of detection and filtering of lidar p oint cloud data,a method combining the geometric characteristics of point clou d data with grid maps has been developed to carry out ground point detection and filtering.First,divide the collected one frame of lidar point cloud data in to the overall ROI area,and divide the ROI area into quadrants according to t he coordinates of the unmanned vehicle.Secondly,the lowest point is extracted from the single-layer data in each quadrant,which is used as the preliminary estimate value of the ground point,and the road surface reflection intensity fea ture data is introduced to estimate and filter the ground point according to the quadrant.Finally,the point cloud data filtered based on the geometric features is subjected to grid map,the "overhanging" point clouds are filtered out in th e grid,and the second ground detection filtering optimization is performed to s olve the under-segmentation caused by the grid-based ground detection filtering process.At the same time,ROI and ground detection filtering are used to re duce the amount of point cloud data and improve data processing efficiency.2.In view of the phenomenon that the obstacles are broken in the grid map and the over-segmentation of the obstacles existing in the existing rapid a rea marking clustering process,an eight-neighborhood area labeling algorithm b ased on European clustering is proposed for the grid map obstacles clustering.First,morphological processing is performed on the occupied grid map to solve the regional connectivity of long-distance and large lateral penetration obstacle s.Secondly,on the basis of regional connectivity,an eight-neighborhood cluster ing algorithm based on European clustering is used to obtain a complete obsta cle.Finally,the original occupancy grid data is revisited,and European clusteri ng is used to solve the segmentation of dense obstacles.It provides a research basis for obstacle detection.3.Aiming at the detection of the potential risks of the typical scenes in t he research of this paper,a potential risk detection method combining vehicle detection information is proposed.First,on the basis of clustering segmentation of obstacles perform rough bounding box coding for cluster analysis of obstac les,static obstacle detection and subsequent scene risk detection.Extract the he ight-area ratio H/S feature of obstacles for static target detection.For the remai ning obstacles,using mature feature descriptors combined with H/S and maxim um height Zmax features,the SVM classifier with radial basis kernel function is used to detect obstacle vehicles.Finally,based on the nearest neighbor assoc iation algorithm,using the bounding box information of obstacle vehicles,the potential risks in typical scenarios are analyzed to provide safety information w arning for the safe driving of unmanned vehicles.The experiments have been s uccessful in prompting the risk information of the scene with the static obstacl e vehicle as the benchmark for a long distance.