Intelligent Vehicle Target Tracking System Based On Fusion Of LiDAR And Vision

Target detection and target tracking are important parts of intelligent vehicle environment perception.Multi-sensor information fusion is an important method to improve the detection ability and accuracy of target detection and tracking system.Vision sensor and Li DAR are the main sensors for target detection and tracking of intelligent vehicles.Considering the performance and cost of sensor system comprehensively,the intelligent vehicle target tracking system based on fusion of monocular vision sensor and solid-state Li DAR is studied in order to improve the development efficiency of intelligent vehicle environment sensing system in this thesis.In this thesis,the following aspects were included in the research of intelligent vehicle target tracking system based on vision and Li DAR fusion:1.An integrated lidar and vision fusion target tracking system architecture is proposed to reduce the wiring,installation and calibration required for the deployment of the environmental awareness system in the development process of intelligent vehicles,and effectively improve the development efficiency of the environmental awareness system of intelligent vehicles.2.A target tracking algorithm based on vision and Li DAR fusion is proposed for target fusion tracking,in order to improve the efficiency and accuracy of fusion tracking.In the aspect of target detection,the target detection methods of lidar and monocular vision sensor based on lane line constraint are studied.For Li DAR target detection,in order to improve the slope under the condition of ground plane segmentation effect,the grid gradient difference segmentation method is proposed to cluster the different slope plane.The ground points and non-ground points were split out by the ground plane fitting method.Finally point cloud targets are got by clustering the non-ground points and establishing obstacle model.For visual target detection,image target recognition is carried out first,and then monocular distance measurement is carried out by using homography matrix to get the image target in three-dimensional space.In the aspect of fusion tracking,the sector tracking door is established to filter measurement,the measurement and trajectory are correlated by angle information,and the associated target is modified by the target trust degree.The associated target is divided into fusion target,potential target and disposable target,which improves the reliability and accuracy of fusion results.Finally,Kalman filter is used to estimate the state of moving objects.3.Aiming at the architecture and algorithm of intelligent vehicle environment sensing system which is proposed in this thesis,the fusion tracking system is designed and developed,and the real vehicle experiment is carried out.The experimental results show that the system can meet the requirements of intelligent vehicle environment perception performance and improve the development efficiency of intelligent vehicles.