Research On Vehicle Circumferential Multi-source Sensing Target Motion Information Acquisition For Autonomous Driving

Autonomous driving decision planning and vehicle driving behavior recognition both depend on accurate and reliable target information.At present,most of the researches on target motion information process the sensor information to improve the quality of the target information,and then directly as the input of intelligent driving path planning module.However,the target information detected by the sensor has a relative relationship with the vehicle.Under different road conditions or when the vehicle changes lanes,the target motion behavior cannot be accurately described by relying only on the target sensor information.In the current relevant studies,the movement of the analysis target along the lane direction is less,and improving the quality of the target movement information is of great significance for the decision planning requirements and the recognition of vehicle driving behavior.Therefore,sensor fusion and motion analysis were carried out based on the experimental vehicle equipped with multi-source sensors.It is of research value and application significance to use real vehicle data to test and verify This paper mainly carries out the following research:(1)In order to solve the problem of multi-source sensor target information fusion,distributed information fusion architecture is used for multi-source sensor decision level fusion.By using Kalman filter to track the single source sensor target,the detection error of the sensor is reduced and the quality of the detection target information is improved.Based on the nearest neighbor data association method,Euclidean distance is used to match and associate the target,and the sensor covariance matrix is used to fusion the target,which synthesizes the advantages of different sensors and improves the quality of fusion target information.Through the correlation and matching of fusion targets at adjacent moments,the target is managed while the vehicle circumferential target is tracked,so as to ensure that the target always has a unique identifier when it is in the vehicle circumferential sensing field of view.(2)By analyzing the target in the lane,the target information decoupled from the vehicle is obtained,and the quality of the target information is improved by fitting the target trajectory.The forward lane line information was combined with the vehicle motion information to generate the backward lane line.The backward lane lines were stored in the form of discrete points in the way of rolling update to avoid the fitting distortion problem of lane line fitting in the transition stage of straight line and curve.Considering the quality of lane line detection,the inner lane line of the target was selected as the reference lane line,so as to reduce the influence of lane line on the calculation of target information.When the vehicle is driving on a curved road or changing lanes,the sensor detects the target information has a relative relationship with the vehicle,which is not enough to accurately and intuitively describe the target motion behavior.Therefore,it is necessary to analyze the motion of the target relative to the reference lane line.Based on the historical position information of the target,the horizontal and longitudinal trajectories of the target are generated.The cubic curves are used to describe the target trajectories by referring to the Mobileye lane model,and the least square method is used to perform recursive fitting of the trajectories,so as to complete the correction of the target position information by using the historical track information.In the simulation environment,the accuracy of the above algorithm is verified under the straight line and curve road.(3)The driving behavior of the target vehicle is identified based on the lateral information of the target vehicle.By analyzing the driving characteristics of vehicles,the characteristic variables of vehicles are extracted.A hidden Markov model based on Gaussian mixture distribution was established to identify the vehicle driving behavior.The vehicle lateral information calculated above was used as the observation variable of the model,and the vehicle driving behavior under straight and curved road conditions was identified.SCANe RMATLAB/Simulink simulation platform was used to collect the data set,so as to complete the training and verification of the model,which verified the robustness of the model to identify the driving behavior of vehicles on different roads.(4)Based on real vehicle data,the relevant algorithms are verified.Through the establishment of real vehicle test platform and the establishment of real vehicle data acquisition model,the data acquisition work under the real vehicle environment is carried out.The multi-source sensor target fusion algorithm,target motion information acquisition algorithm and vehicle driving behavior recognition model were verified by data recharge technology.The results prove that,driven by real vehicle data,the target fusion algorithm can still accurately complete the complete tracking of the vehicle circumferential target.The object information correction method based on trajectory fitting established in this paper effectively improves the quality of the object information,and the driving behavior recognition model can still accurately identify the vehicle lane changing behavior under the real vehicle data.