Research On Vehicle Trajectory Tracking Based On Multi-modal Sensor Fusion On Roadside

With the increasing number of vehicles,traffic congestion and traffic safety problems become increasingly prominent.It is very important to obtain reliable track information of road vehicles for the research and judgment of traffic situation and improve traffic efficiency.At present,single sensor is used to track the vehicle target,which has low tracking accuracy and great environmental impact.In view of this,this paper carries out the research of multisensor trajectory tracking fusion technology.Firstly,the time-space synchronization of the roadside camera,laser radar and millimeter wave radar sensors is studied.Secondly,aiming at the defects of point cloud sparsity of millimeter wave radar,through data analysis and DBSCAN algorithm optimization,the target tracking technology of millimeter wave radar is studied.Finally,considering the decay of lidar point cloud reflection intensity with distance,combined with the millimeter-wave radar dual-mode operating characteristics and camera detection capability,the dynamic weight fusion of three roadside sensors is studied.According to relevant research in this paper,the main innovation points and conclusions are as follows:(1)A time synchronization method based on frequency self-matching is developed to realize time synchronization between three sensors of varying acquisition frequencies with an accuracy of 99.86%.Additionally,the space synchronization between the three sensors was improved,resulting in an error of 0.19 pixels for camera and lidar reprojection,and an error of 0.08 pixels for internal parameter calibration.(2)A multi-frame adaptive clustering method based on DBSCAN is improved.Aiming at the sparsity of radar point cloud,the clustering effect of radar point cloud is effectively improved by aggregating continuous multi-frame point cloud data to increase the density of point cloud and filtering the aggregated point cloud at the same time.Through the analysis of the data preprocessing of radar,the adaptive Kalman filter algorithm is used to process the measured data of radar,which can effectively improve the data tracking accuracy,so as to determine the data processing method of radar.(3)A DWARS trajectory tracking fusion method is proposed.Based on this method,experiments are carried out in 5 different actual scenarios.Compared with track tracking by a single sensor,the track tracking range is effectively improved.The repair rates of track disconnection in 5 scenarios reached 21.32%、20.53%、19.98%、22.43%、23.15%,respectively.(4)The multi-source sensor data acquisition platform is constructed,and the topic information corresponding to the sensor can be changed according to different sensor drivers;At the same time,a tf-tree is established in ROS to unify the coordinates of multiple sensors and realize the simultaneous display of lidar and radar data in ROS system.Finally,the simultaneous acquisition and visualization of roadside camera,Lidar and radar data is realized.