Vehicle Trajectory Reconstruction And Traffic State Perception In Connected Environment

Traffic monitoring is one of the most essential roles of the intelligent transport system.The overall perception of traffic state is the basis of traffic operation,management,and control.Over the past two decades,a breakthrough has been made in intelligent connected vehicles.Many advanced driver assistance systems,such as adaptive cruise control and lane-keeping assist system,have been deployed in various types of production vehicles.For a long time in the future,intelligent connected vehicles and ordinary vehicles will coexist in the transportation network.In this mixed environment,how to realize accurate traffic state perception is a problem worth researching.In fact,with the massive data collected by intelligently connected vehicles,various objects in the traffic network can be detected and tracked.Compared with stationary sensors data,the data collected by the intelligently connected vehicle is more comprehensive and detailed,which can meet the research needs of both macro and micro levels,and is a better data source.However,due to data transmission,vehicle occlusion,sensor detection range,and other reasons,the trajectory dots collected by intelligently connected vehicles usually contain missing segments,which need to be estimated and reconstructed.In order to form a complete traffic state perception framework of vehicle detection,trajectory completion and state estimation,the lane-changing decision model,vehicle trajectory reconstruction model and traffic state estimation model are developed successively with the detection information of intelligently connected vehicles.In this framework,the incomplete trajectory information is used to make decisions through the vehicle lane-changing behavior decision model.Correspondingly,the trajectory reconstruction model of following and lanechanging is entered to complete the missing trajectory.Then the traffic state estimation model is used to obtain the accurate estimation of traffic parameters in the whole space and time.The main research work is summarized as follows:Firstly,in order to identify lane-changing decisions,based on the UAV aerial data,the characteristics of traffic parameters under different vehicle operating states are analyzed,and the characteristic parameters affecting lane-changing decisions are obtained.The Gaussian mixture hidden Markov model is used to model lane-changing decisions,and the relationship between vehicle interaction and lane-changing decisions is established.The results of the test have shown that the accuracy rate of lane-changing decisions is 92.37%.Secondly,in order to the resolve trajectory missing problem,vehicle trajectory changes under different vehicle operating states are analyzed,the contributing factors to trajectory change of following and lane-changing behavior and the time step of historical data are determined.Based on the lane-changing decision model,a vehicle trajectory prediction model based on the bidirectional long short-term memory network was constructed.Combined with the smooth connection algorithm,the complete vehicle trajectory was finally obtained.The reconstructed trajectory is compared with the measured trajectory,and the validity of the model is verified.Finally,the problem of traffic state estimation is transformed into a matrix estimation problem by spatio-temporal region discretization,the matrix estimation problem is transformed into a matrix completion problem by dividing the probing region and the unprobing region.Using the complete vehicle reconstruction trajectory,the matrix factorization method was used to establish the relationship between the missing value and the measured value,and the traffic parameters in the whole space-time area were calculated and estimated,the complete spacetime evolution map of the traffic state was obtained.The test results of the model show that the proposed matrix factorization method can achieve accurate estimation of velocity and density at the penetration rate of 15% and 20%.