Research And Implementation Of Road Environment Perception Algorithm Based On Lidar

In recent years,environment awareness for unmanned driving has become an important research area in smart transportation and the Lidar-based road environment awareness technology has attracted much attention.Four-line Lidar is widely used due to its high measurement accuracy,large amount of data,and robustness to environment.Considering these advantages,this paper uses the four-line Lidar in the study.This paper studies multiple Lidar-based awareness algorithms on the unmanned system aiming at the three important tasks of road environment awareness during unmanned driving.The three tasks are obstacle detection,road boundary detection,and simultaneous localization and mapping.The main contents consists of the following parts:(1)A multi-density DBSCAN obstacle detection algorithm based on the distribution of concavity and convexity called CM-DBSCAN is proposed and implemented.The properties of the Lidar sensor data varies sharply at the boundary of the obstacles.In order to overcome the disadvantages of the traditional DBSCAN algorithm which has long execution time and a single parameter,the obstacle candidate points are obtained according to the distance features of data distribution;the adaptive radius function is adopted based on the distribution characteristic that the density of Lidar data decreases with the increasing of the scanning distance.The experimental results show that comparing with the traditional DBSCAN algorithm,the performance of the proposed CM-DBSCAN algorithm is improved significantly.(2)A moving obstacle detection algorithm based on GPS/INS information and multi-feature matching is proposed and implemented.Basing on the result of the obstacle detection,analyzes vehicle offset,distance and degree between obstacle and vehicle in neighbor frames,moving obstacle is detected by filtering and matching obstacles considering all these analysis,point amount and area features.(3)A road boundary detection algorithm based on ROI and constraints is proposed and implemented.Analyzes the distribution characteristic of the road boundary by using the straight line model.Considering the characteristics of regionalization and continuity of the road boundary,points in ROI are extracted,the Hough transform is used to obtain multiple candidate road boundaries,the point amount,distance and length features of the candidate road boundaries are constrained to obtain the best fit straight line.This line is the road boundary.(4)A simultaneous localization and mapping algorithm based on bilinear interpolation and LM algorithm is proposed and implemented.The algorithms of simultaneous localization and mapping based on parrticle filter and scan matching are analyzed.Inspired by the Hector SLAM algorithm,the grid occupancy probability is calculated by using bilinear interpolation and without odometer,so that the map can be regarded as a continuous probability distribution model approximately.According to the grid occupancy probability of each Lidar point in the world coordinate system,the LM algorithm is used to find the optimal mapping between the Lidar data and the existing map by using the spatial gradient information of this map.Experimental results show that the proposed method has good performance and satisfies the real-time requirement in the regular outdoor scenarios.