Research On Parallel Streaming Vehicle Trajectory Reconstruction Method For Urban Road Network

Vehicle trajectory reconstruction refers to the process of inferring the continuous movement trajectory of the vehicle in the actual road network based on its discrete GPS trajectory data.Vehicle trajectory reconstruction mainly includes two processing stages,namely map matching and path inference.In the first stage,the original discrete GPS trajectory data of a certain vehicle is mapped to the corresponding road section of the actual road network through the map matching method.At this time,for the GPS trajectory data at two consecutive sampling moments,the mapped road network segment is not necessarily topologically continuous,so the continuous movement trajectory of the vehicle cannot be obtained in the map matching stage;In the second stage,combining the road network topology and the global GPS trajectory history statistics of the road network,the path inference method infers the continuous movement trajectory of the vehicle between the matching road segment at the current sampling time and the previous sampling time.Vehicle trajectory reconstruction has important application value in various location-based traffic information service fields,such as urban planning,traffic flow prediction,route inference,etc.However,due to the sparsity of GPS trajectory sampling points caused by low-frequency sampling and its measurement noise,the accuracy of the map matching method is difficult to be guaranteed in the process of reconstructing vehicle trajectory in the actual urban road network environment.Meanwhile,the long sampling time interval in the low-frequency trajectory data makes the set of candidate trajectories existing between the matching road sections corresponding to the trajectory data at two adjacent sampling moments larger.It is challenging to infer a reasonable continuous trajectory for the vehicle in the above candidate trajectory set.First of all,in view of the accuracy of map matching of the low-frequency sampling of GPS trajectory data in the urban road network is not high,this paper proposes a map matching method based on hidden Markov filter.Compared with the existing algorithms,the proposed method can further effectively improve the accuracy of the map matching.Furthermore,a Bayesian path inference model under temporal and spatial constraints is proposed,which can effectively infer the movement trajectory of the vehicle between two adjacent matching road sections.On the basis of the above work,a parallelized streaming vehicle trajectory reconstruction model is proposed,which can handle large-scale vehicle trajectory reconstruction problems in real time while ensuring accuracy.The specific contents of this paper are as follows:(1)A preprocessing method for large-scale vehicle GPS trajectory data is realized.Firstly,we analyze the characteristics of the original GPS trajectory data,and propose a cleaning rule for the vehicle GPS trajectory data.Secondly,in order to verify the performance of the algorithm at different sampling intervals,non-overlapping sampling was performed on the GPS trajectory data that is cleaned using proposed rule.Finally,the Geo Hash grid method is used to divide the urban road network map,which not only reduces the number of candidate road sections,but also can more efficiently extract the candidate road sections of GPS track points,thereby effectively improving the performance of the map matching method.(2)Aiming at the low-frequency sampling GPS trajectory data in the urban road network,a map matching method based on hidden Markov filter is proposed.This method can further improve the robustness of the map matching method on the basis of ensuring accuracy.In order to further improve the accuracy of map matching,the traditional observation probability calculation model in map matching is revised,and space constraints is introduced in the calculation of transition probability.(3)Considering the randomness of the driver’s path selection,a Bayesian path inference model under space-time constraints is proposed.The model can effectively infer the vehicle trajectory between two consecutive matching road sections.This method transforms the path inference problem into a problem of searching for K most likely candidate paths according to the joint posterior selection probability of candidate paths.When estimating the model parameters,the frequency of each road segment in the historical GPS trajectory,rather than the number of road segment transfers,is considered,which can reduce the impact of sparse GPS sampling and improve the accuracy of path inference.In addition,this paper introduces space-time constraints and probability threshold constraints into the model to reduce the search space,which can significantly improve the performance of the model.(4)Aiming at the problem of large-scale vehicle trajectory reconstruction in urban road networks,a parallelized streaming vehicle trajectory reconstruction model is proposed.At the same time,the model is implemented on the Spark Structured Streaming framework,then the performance of the proposed model is tested and evaluated in a distributed cluster.