Research On Obstacle Detection Method Based On LiDAR

Intelligent vehicles are of great significance for improving road safety,alleviating traffic congestion,reducing driving labor costs,and promoting the upgrading of the automobile industry.Environmental perception is a key step to realize autonomous driving of intelligent vehicles.It provides real-time environmental information,which is an important prerequisite for the subsequent intelligent vehicle planning and decision-making and execution control.Lidar has become the most important sensor in the perception of intelligent driving environment because of its high resolution,high accuracy,strong anti-active interference ability,and the ability to obtain a wealth of information.The driving environment of intelligent vehicles is complex and diverse.In order to ensure the accuracy and stability of the environment perception system of intelligent vehicles,intelligent driving vehicles need to perceive the movement state of moving obstacles in the passable area in a dynamic environment.Based on the above reasons,this article adopts a method of obstacle detection using lidar.The research is mainly carried out from three aspects: ground segmentation method,passable area detection method,and dynamic obstacle detection and tracking method.First of all,in order to accurately segment the ground point cloud and obstacle point cloud in different scenes,this paper uses the velodyne VLP-16 lidar sensor.And analyzes its working principle and laser point cloud data format,then uses the scanning feature of laser transmitter to construct the filter based on the distance between the rings,to separate most of the obstacle points,and then divide the point cloud space into different regions,the multi-region RANSAC plane fitting algorithm is used to assist the filtering to filter out the remaining ground points.It overcomes the influence of irregular roads such as ramps and curves on the ground point cloud segmentation,and can effectively segment the ground point cloud of different road conditions.Secondly,in order to detect the driving area of intelligent vehicles,this paper firstly uses the improved Euclidean clustering algorithm to cluster the obstacle point cloud,and uses the L-shape fitting combined with Hough line detection method to extract the 3D bounding box of the obstacles after clustering,and obtains the central point position and rotation Angle of each obstacle.Then according to the road geometric characteristics and point cloud space feature information,the method based on dynamic sliding window is used to extract the road boundary points,and the quadratic curve is used to fit the road boundary points to obtain the road boundary line.Finally,based on the extraction result of the road boundary,the obstacles within the road boundary are clustered and marked,so that the passable area of the road can be determined.And provide necessary information about obstacles in the passable area for the subsequent modules.Then,in order to deal with the external disturbance and motion uncertainty in the process of dynamic obstacle detection and tracking,this paper tracks the dynamic obstacles within the road boundary constraints,and uses the data association algorithm optimized based on the Hungarian algorithm to analyze all obstacles in the front and rear frames.At the same time,linear Kalman filtering algorithm is used to filter the predicted value at the previous moment and the measured value at the current moment to obtain the optimal estimation value of the obstacle motion state.This method overcomes the problem of unpredictable point cloud data distribution in a dynamic environment.Finally,real vehicle data tests are carried out in different scenarios,and the results verify the feasibility and stability of the method we adopted.