Surface Target Detection And Tracking Based On Lidar

With the development of sensor,communication,artificial intelligence and other technologies,the demand for unmanned platforms is increasing all over the world.As an important part of unmanned platforms,unmanned surface vehicles have broad application prospects in both military and civil fields.The premise of improving the autonomous navigation performance of unmanned craft is the accurate perception of the surrounding water environment,so as to avoid obstacles quickly and optimize the route in real time.As an important part of the sensing system,lidar has the advantages of omnidirectional sensing of the environment of the unmanned vehicle and obtaining the depth information of the obstacle target.Therefore,it is very important to carry out the research on the detection and tracking of targets on the water based on Lidar for improving the perception ability of USVs and ensuring the safe navigation of USVs.The main research work of this paper includes:(1)Aiming at the problems of large amount of surface Lidar point cloud data,redundancy and noise,a set of pretreatment methods suitable for surface Lidar point cloud is designed.Firstly,an improved statistical filtering method is used to effectively remove outliers and noise points from the point cloud.Then the 3D point cloud is transformed into grayscale image,and the region of interest is extracted by using the method of image edge detection.Finally,the required data of the obstacle point cloud is obtained.Through experiments,the designed preprocessing method is used to process the surface Lidar data set collected by unmanned craft,and a good processing effect is obtained,which verifies the effectiveness of the designed preprocessing method.(2)To solve the problem that traditional DBSCAN algorithm is difficult to select two parameters of neighborhood and density threshold,and it is difficult to achieve good clustering of all data by using fixed neighborhood and density threshold,an improved DBSCAN algorithm with adaptive parameters is proposed,which takes independent ε for each point cloud to cluster.The improved method reduces the difficulty of parameter selection,reduces the under-segmentation of the short-range target and prevents the missed detection of the long-range target.Through experiments,the clustering effect of the traditional DBSCAN algorithm and the improved DBSCAN algorithm is compared,and the effectiveness of the improved DBSCAN algorithm is verified.(3)For single target tracking using a single model extended Kalman filter effect is not ideal and multi-target tracking clutter environment in each measured target source problem,a target tracking method suitable for surface lidar point cloud was designed.For single target tracking,different motion models are selected as the state transfer equations of the filtering algorithm according to the different motion states of the target.Then for the nonlinear model,the high order term is preserved in Taylor expansion,and the high order polynomial Kalman filter is obtained,which improves the accuracy of target tracking.In order to solve the problem of target source in clutter environment is uncertain in multi-target tracking,a joint probabilistic data association algorithm is adopted to consider the competition of multiple track pairs,so that the correct matching between track and measurement is realized.Through the tracking experiments of single target straight line,turning motion and multi-target cross motion,the effectiveness of the proposed target tracking algorithm is verified.