3D LIDAR-based Obstacles Characteristic Analysis

In this paper,under the background of the autonomous navigation for an Autonomous Land Vehicle(ALV),we focus on the analysis of obstacles characteristic in the environment.The obstacles are divided into static obstacles and dynamic obstacles according to the characteristics of the obstacles.The static obstacles are divided into positive obstacles,negative obstacles,passable obstacles and special semantic obstacles.Based on the characteristics of different obstacles,the obstacles characteristic analysis and detection method based on LIDAR point cloud are proposed,and validates the algorithm through real vehicle experiment.The main contributions and innovations of this paper are as follows:Firstly,negative obstacles in the off-road environment have a great impact on the driving safety of ALV.Currently,two 32-lines LIDAR are used to detect the negative obstacles in front of the ALV body,which not only greatly increases the cost of ALV,but have a large blind spot.The data for the analysis of the characteristics of negative obstacles in this paper obtained from the 64-line LIDAR that installed vertically in the vehicle body.Through the ground segmentation,the reference height of the obstacles relative to the ground plane is obtained.Then,we can judge the possible negative obstacles in the environment and use the sequential information of LIDAR for multi-frame fusion to achieve a stable output.Secondly,based on the analysis of the characteristics of LIDAR depth information in the haystack,a new detection method based on LIDAR is proposed for detecting haystack in the off-road environment.We extract the depth characteristics of the single line of LIDAR in the haystack,use the Bayesian principle to extract multi-line features,and then fuse the sequential information of the LIDAR data,to achieve a stable output of haystack obstacles.Thirdly,in the city environment,there are correlated between a large number of traffic signals and bar-type obstacles(such as trunks,poles,street lights).By analyzing the characteristics of the bar-type obstacles,we can separate the bar-type obstacles as a kind of obstacle that contains semantic information,and achieve a kind of detection method of bar-type obstacles based on geometric characteristicsFourthly,a novel dynamic target detection and tracking method based on occupancy grid is proposed.We combine the occupancy grid with the dynamic target tracking based on particle filter,analyze the occupancy characteristics of the cells and then build the occupancy grid map.By using particle filter to get the dynamic characteristics of the cells and then analyzing the consistency of the dynamic characteristics of the occupied cells we can obtain accurate dynamic grid.Finally,dynamic cells clustering are used to achieve target detection and tracking.