Research On Obstacle Detection And Tracking Based On 3D LIDAR

This paper uses 3D LIDAR to detect and track obstacles in order to meet vehicles’ perception needs for obstacles in structured road environment,which can provide location and status information for active braking,active obstacle avoidance,etc.In order to establish the region of interest that changes with the location of vehicle,the paper firstly uses the camera to detect the lane line in front of the vehicle,and calculates the location of the left and right lane lines so as to acquire the lateral dynamic boundary of the region of interest.,The value of the grid map resolution is determined by the horizontal resolution of the LIDAR taking its influence on the obstacle detection into consideration.The conversion relationship between LIDAR coordinate system,vehicle body coordinate system,and image coordinate system is established so that the location parameters of LIDAR can be implemented,and the grid map is established by mapping the LIDAR data to a two-dimensional image.Secondly,the reason of noise points in LIDAR’s original point cloud data is analyzed,and the traditional median filtering algorithm is improved according to the grayscale feature of grid map.On this basis,the method based on the value of the height difference and the method combining with Bayesian theory are respectively used to divide the background point cloud of grid map so as to remove the point cloud data which has less impact on the vehicle’s safe driving and obtain an obstacle grid map which only contains the point cloud data of obstacles.Furthermore,in order to reduce the distance of point cloud data between obstacles and strengthen the correlation between them,the morphological operation is introduced to preprocess the obstacle grid map.Then a density clustering algorithm based on neighborhood search is proposed to cluster the point cloud data in the obstacle grid map.The box model is used to parameterize the clustering results,and the location as well as the state information of obstacles are extracted.Considering the number of obstacles in each frame of LIDAR,a multi-feature nearest neighbor algorithm is proposed and the correlation matrix is introduced to associate the obstacles in obstacles list from previous frame and current frame because the point cloud shape of the same obstacles from adjacent frames is similar.Taking into account the LIDAR data in each frame includes not only an obstacle,combined with the shape of adjacent frames the same obstacle point cloud approximation,and the different obstacles point cloud morphological differences,a multi-feature nearest neighbor algorithm of obstacle and obstacle in the list of frames associated,and data association the introduction of correlation matrix.Finally,a dynamic obstacle extraction algorithm based on dynamic characteristics is proposed,and an extended Kalman filter is used to track the dynamic obstacles.The results of actual vehicle experiment show that the algorithm proposed in this paper can complete the accurate detection and stable tracking of the obstacles in the region of interest.