3D Point Cloud Vehicle Detection And Tracking Based On Laser Radar

Target detection and tracking based on three-dimensional imaging lidar has become an important topic in the field of computer vision in the late years.Compared with the traditional two-dimensional imaging sensors,it can effectively obtain the three-dimensional geometric information of the target,and is less affected by external illumination changes and imaging distance.Accordingly,lidar has been widely used in unmanned driving technology,military field,service robot and other neighborhoods.But there are still some problems in the three-dimensional imaging lidar: the quantity of 3D point clouds is much more than the traditional imaging sensor in the same field of view on distance due to the high spatial resolution and three-dimensional imaging capability of Radar.The increase of measurement dimension and resolution brings the possibility of the improvement of the detection performance that requires a more intelligent and robust algorithm.In addition,the target is usually placed in a certain background environment which is integrated with the background.The target is occluded in the process of lidar detection due to the angle of view,background and other reasons and it may be difficult to detect and segment in range profile.In fact,the target and background are separated at different altitudes and can be segmented by the elevation of the scene in the real scene that can effectively classify in theory.Therefore,this paper mainly focuses on the characteristics of three-dimensional point cloud image data of ground vehicle targets and carries out the main research and analysis of target detection and tracking.The research work of this paper is mainly divided into three parts:1.Vehicle detection based on fast detection and eliminate false alarm by AdaBoost learning algorithm.Firstly,the local elevation feature be used as a strong feature to quickly pre-classify the point cloud data,combing with the moving surface fitting algorithm to separate the ground and non-ground points.The Region of Interest(ROI)is detected by clustering the non-ground points which using the geometric size of the target and the specific elevation gating.Using AdaBoost algorithm to check ROI to improve the detection accuracy of the algorithm,accurately judge the target and improve the detection accuracy and reduce false alarm.2.Proposed an improved 3-D point cloud target tracking algorithm based on Kalman filter and Mean shift.Proposed an improved target tracking mechanism to solve the problems of Mean shift tracking algorithm when the object is occluded andmoving too fast.Firstly,Mean shift is used to optimize the search direction of the target and Kalman filter is used to better predict and modify the search range of the target which can still track the target well when the target disappears for a short time.When the target tracking is lost,the whole frame is re-detected using the proposed vehicle detection method based on fast detection and false alarm rejection to retrieve the missing target which can achieve long-term target tracking.3.Finally,the feasibility and validity of the proposed method are analyzed through experimental simulation data and KITTI common data sets.