Research On Vehicle Tracking Model And Corresponding Algorithm Based On Data Fusion And Information Integration

Transportation management and control are always conducted based on information obtained by accurate and timely traffic state estimation.As traffic data and traffic problems in smart cities become more fine-grained,it is feasible and necessary to construct traffic models for fine-grained traffic state estimation.This paper proposed a vehicle tracking model based on a data fusion and information integration approach,which can effectively track urban vehicles to get the individual-level travel times and trajectories,via fusing mobile-sensing data and fixed-location data and integrating different kinds of traffic information.The model is low time consuming,and it could integrate more traffic information to improve the tracking performance,without much additional computation time.Therefore,it is suitable in smart cities where the heterogeneous traffic data can provide more traffic information.This paper first proposed a method to decompose the urban arterials into a series of sections and intersections,so the vehicle tracking problem was defined as the data matching problem between the upstream and downstream locations at each section and intersection.Second,data fusion was performed by constructing the first joint matching probability model,which was constructed via the one-to-one matching between the mobile-sensing data and the fixed-location data;meanwhile,the second joint matching probability model was constructed via the one-to-one matching between the upstream and downstream fixed-location data which is not involved in the data fusion,based on the probability integration of lane choice decisions/traffic merges,travel time distribution and error distribution of measured vehicle length.The two joint matching probability models constitute the vehicle tracking model in this paper.Then,a dynamic programming/Kuhn-Munkres algorithm-based two-step approach was designed to efficiently solve the model in a two-parts way.The dynamic programming and Kuhn-Munkres algorithm were used to search the optimal matching sequences which maximizes the first joint matching probability and the second joint matching probability,respectively.The matching results in the two parts are global optimal,and the two-step approach is low time consuming.Finally,the experimental data acquired from two NIGSIM(Next Generation Simulation)datasets with different data sizes was used to test the model in ideal scenario,i.e.,the fixed-location data and mobile-sensing data are perfect and well synchronized,as well as in inaccurate data scenario and detector failure scenario.The portability of proposed model was also analyzed through comparing the experimental results in the two datasets.In addition,the computation efficiency of the two-step approach was assessed quantificationally.