Research On Multi-target Detection And Trajectory Tracking Based On LiDAR In Complex Environments

At present,the multi-dimensional monitoring,precise control and collaborative service capability of the transportation system is being enhanced with the construction of intelligent vehicle-infrastructure cooperative systems(IVICS).IVICS often require cooperative sensing between roadside and onboard devices to improve traffic safety and efficiency.However,in scenarios with low penetration of onboard intelligent terminals,IVICS still rely on roadside sensing systems to obtain trajectory information of unconnected targets on the road.Unfortunately,roadside sensing devices and methods are heavily affected by complex environmental factors,including weather conditions,lighting,and road conditions,which can lead to decreased sensing performance and imprecise tracking of on-road targets.Therefore,how to improve the stability of target detection and trajectory tracking in complex environment has become one of the urgent problems to be solved in the development of IVICS.In this paper,the research of multi-target detection and trajectory tracking method under complex environment was carried out with 360° Light Detection and Ranging(LiDAR)as the main sensor,focusing on snowy days,nighttime and high-density traffic flow scenarios.First,by analyzing the data characteristics and spatial distribution pattern of snow noise points in point cloud data,the three-dimensional(3D)space was divided into reflection regions of different intensities.Based on the traditional filtering method,a multi-stage point cloud filtering superposition algorithm was proposed to filter out the noisy point clouds at different scales respectively.Secondly,based on the Point RCNN detection model,the effects of the model training dataset and different sensor coordinate systems on the detection range of the underlying model were explored.An omnidirectional point cloud information stitching and detection technique was proposed,which divided the LIDAR view into front and rear detection range by coordinate transformation.The algorithm processed the detection results within 180° of each parts in parallel.At the same time,an overlapping object estimation and removal method was designed to eliminate erroneous detection data of the same object near the front and rear stitching seams and to achieve the integration of results in the same coordinate system.Then,based on the tracked initial trajectory data,the trajectory survival cycle management strategy was optimized to adaptively adjust the trajectory survival cycle according to the target motion in different scenes,improved the correlation degree of targets between frames,iteratively update the trajectory data,and reduced the trajectory interruption caused by the masking effect.Finally,the unconstrained traffic conflict discrimination method was improved,and the application and validation of target detection and trajectory tracking methods were carried out in the actual complex environment.The research results showed that by using the point cloud overlay filtering algorithm,the data quality of snowy point clouds was significantly improved,and the false alarm rate of target detection was reduced by about 96%.The omnidirectional point cloud information stitching detection method can further control the consumption of computational resources and the increase of model training cost,and the number of output targets increased by 77.15%The base Point RCNN achieved 360° full coverage detection range in environments with snow,night-time,and high-density traffic flow,enhancing the applicability of detection algorithms on roadside LiDAR.By analyzing the detection distance,trajectory integrity,and conflict point calculation results of the proposed method in actual traffic scenarios,it was found that the related method can achieve accurate detection and stable tracking of multiple targets in complex environments,effectively improving the perception ability of roadside LiDAR in complex environments.The trajectory data can also be used for traffic conflict analysis and discrimination,providing data support and technical support for improving traffic safety and optimizing traffic decision-making.