Dynamic Target Detection And Motion Estimation Of Unmanned Vehicle Based On 3D Lidar

Unmanned ground vehicle,as an important category of intelligent robots,has become a strategic step in the field of high technology research all over the world,and has caused extensive research in both civil and military fields.In real traffic environment,there are dynamic objects,such as pedestrians and vehicles,and the recent traffic accidents of unmanned ground vehicles are due to the lack of accurate detection of dynamic targets.The unmanned ground vehicle needs to detect the surrounding dynamic objects and get accurate their motion parameters in order to make safe driving planning and decisions.Compared with other environment perception sensors,such as camera and radar,3D lidar can provide accurate environment depth distance information,which is lessly interfered by environmental factors such as ambient light,and is important sensor for unmanned ground vehicle.In this paper,the detect-tracking algorithm framework is used to detect and track the dynamic targets through the point cloud data of 3D lidar,including the detection and tracking of dynamic objects,and the estimation of their motion state.In this paper,in the detection process,for the problem of inaccurate parameters of the object contour,obtained through the connected component analysis,the edge contour of the clustered target is fitted to get the accurate rectangle contour of the target.For the problemthat the position of the center point of the target rectangle,observed through the lidar,is instability,a dynamic target feature point observation model is established by analyzing the observation characteristics of the 3d lidar,and the dynamic objects are tracked through combiming the tracking gate method and the joint probability data correlation method.In order to accurately estimate the motion state of the dynamic target,Gaussian mixture model method is used to identify the motion pattern of the dynamic target,then a linear motion estimation model and a turn motion estimation model are established,thereby selecting the motion model corresponding to estimate the current motion state of dynamic targets.The dynamic target detection algorithm proposed in this pape is tested through KITTI dataset and in real traffic with unmanned ground vehicle.Using the three-dimensional target detection dataset in the KITTI dataset and the road test verified the effect of the algorithm for dynamic target detection and motion pattern recognition,beside,real vehicle test verified the effect of the dynamic target motion state estimation model for straight line motion and turning motion.The test results show that the algorithm can detect the dynamic targets around the unmanned vehicle stably and provide reliable dynamic target information for the dynamic obstacle avoidance of unmanned ground vehicles.