Three-dimensional Online Traffic Reconstruction And Optimization Driven By Data And Model Collaboration

It is important to realize three-dimensional(3D)online traffic reconstructions based on the video data collected by fixed safety cameras in the urban road network.It is conducive to accident analysis,traffic flow statistics,route navigation,congestion analysis,etc.Existing methods do not process unstructured video data,but extract vehicle data in videos semi-manually with the help of Multiple Object Tracking(MOT)methods.The installation position of the intersection camera is close to the edge of the road on one side,and the height is limited,which causes serious traffic occlusion in the video data,and it is difficult to obtain a complete vehicle trajectory through the existing MOT method.The data obtained by the tracking method is two-dimensional pixel coordinates and the camera parameters are unknown,which cannot be used for 3D traffic reconstruction directly.How to achieve realistic 3D traffic reconstruction based on intersection camera data has become a challenging task.This paper proposes a three-dimensional online traffic reconstruction and optimization driven by data and model collaboration method.The main research work is as follows:In order to extract vehicle trajectory coordinates from video raster data,an interactive iterative inverse perspective projection transformation method is proposed.specifically,firstly the multiple object tracking technology was used to extract the pixel coordinates of the vehicles’ trajectory in the video,and secondly transform pixel coordinates to geodesic coordinates using an inverse perspective projection transformation.Lastly model a 3D static scene based on the video.In order to get the complete trajectory of the vehicle,a vehicle trajectory completion method based on traffic simulation is proposed.Specifically,a cognition-driven traffic simulation in an open environment was built to predict vehicles’ trajectory.Then,the above obtained trajectory data was combined to get complete trajectory.Finally,optimize the direction angle of vehicle through linear regression and achieve 3D online traffic reconstructionsIn order to get the optimal trajectory of the vehicle,a vehicle trajectory optimization method based on Kalman filter is proposed.Specifically,we use the traffic simulation method to predict vehicles’ trajectories and use the Kalman filter to optimize vehicles’ trajectories according to the extracted trajectories and the predicted trajectories.We also optimize the traffic simulation parameters iteratively according to the optimized trajectories when needed.Experimental results show that the method can realize 3D online traffic reconstructions with excellent visual effects in real traffic scenes,and can deal with trajectory missing and trajectory misalignment.Movements of vehicles are smooth in the visualization,which is better than that using the trajectory data obtained by the multiple object tracking method directly.