Vehicle Lane-level Location And Lane-changing Behavior Prediction Based On Low-channel Roadside LiDAR

In the traffic system,lane changing behavior of vehicles has been a fundamental characteristic of road traffic flow.In recent years,the number of traffic accidents caused by drivers’ lane-changing behaviors accounted for about 27% of the total number of accidents,among which,75% of the traffic accidents caused by lane-changing were due to drivers’ misjudgment of other road users and the environment,which directly led to drivers making wrong behavioral decisions and thus traffic accidents.Therefore,the prediction of vehicle lane level lane change behavior has been an important topic of research at home and abroad.Due to the advantages of high accuracy of LiDAR(Light Detection and Ranging,LiDAR),LiDAR has become a new technology and means to study vehicle lane level localization and lane change behavior prediction at present.Compared with the widely used on-board high-channel LiDAR,lane level lane change behavior prediction by roadside low-channel LiDAR is less costly,so in this paper,a16-channel LiDAR was chosen as roadside equipment to study lane level lane change behavior prediction of vehicles.However,the current research on lane level lane change behavior prediction of vehicles using low-channel roadside LIDAR was still insufficient.Based on this status,this paper firstly detected the road lane lines within the LIDAR scanning range,and carried out multi-target tracking of vehicles and pedestrians,matched LIDAR and GPS information of the same object to obtain lane level positioning of vehicles,and finally used high resolution data to achieve lane change behavior prediction of vehicles.The research in this paper is organized as follows.(1)Research on roadway lane marking recognition method based on 3D laser point cloud.Firstly,the ground recognition algorithm was improved to ensure that the lowchannel roadside LIDAR can acquire more ground point cloud data.Secondly,the point cloud data was divided into grid cells according to the distribution density of the point cloud,and the laser point cloud data of the ground lane line was identified based on the laser point cloud intensity.Finally,this paper proposed a clustering method based on linear density,with the target of maximizing linear density,and achieved lane line identification and detection by stripe segmentation and stripe search of the scanned area.(2)Vehicle lane level localization method based on LiDAR and GPS fusion.Based on the instance segmentation and target information extraction of laser point cloud data,the Io U(intersection over Union)and laser intensity histogram were used as efficiency values to track the target based on the improved Hungarian algorithm to obtain the vehicle’s driving path.Based on Hidden Markov Model(HMM),LiDAR vehicle trajectory data was matched and position information was fused with GPS vehicle trajectory,so that GPS information of the vehicle can be corrected to achieve vehicle lane level positioning and tracking.(3)Vehicle lane change behavior prediction based on high-resolution vehicle trajectory data.Firstly,the vehicle lane change behavior was studied and analyzed to obtain the vehicle characteristics and the influence of the surrounding vehicles on the lane change behavior when the vehicle makes the lane change behavior.The highresolution vehicle and surrounding vehicle information,such as vehicle coordinates,speed,acceleration,angle,and relative distance of surrounding vehicles,were used as input to train the LSTM neural network to predict the vehicle trajectory at future moments.It also used a nonlinear SVM-based method to identify whether a vehicle has lane-changing behavior based on the relative distance between the vehicle and the lane,position coordinates,so as to achieve a complete process of predicting vehicle lanechanging behavior.The method proposed in this paper used data collected by a low-channel roadside LIDAR to efficiently and accurately obtain lane line information,lane level localization of vehicles,and prediction results of vehicle lane change behavior.The research in this paper can provide the accurate location of the vehicle and the predicted lane change behavior for vehicles around the device with less equipment and map building cost,and provide data support for V2I-based driving information transfer,thus providing drivers with sufficient driving information and reducing traffic accidents caused by lane change.