Journal of Geo-information Science ›› 2020, Vol. 22 ›› Issue (11): 2109-2117.doi: 10.12082/dqxxkx.2020.190738

    Next Articles

Vehicle Trajectory-map Matching based on Particle Filter

ZHENG Shichen1,2(), SHENG Yehua1,2,*(), LV Haiyang3   

  1. 1. Key Laboratory of Virtual Geographic Environment, Nanjing Normal University, Ministry of Education, Nanjing 210023, China
    2. Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application, Nanjing 210023, China
    3. Smart Health Big Data Analysis and Location Services Engineering Lab of Jiangsu Province, Nanjing University of Posts and Telecommunications, Nanjing 210023, China
  • Received:2019-11-27 Revised:2020-02-13 Online:2020-11-25 Published:2021-01-25
  • Contact: SHENG Yehua E-mail:shichen_zheng@163.com;shengyehua@njnu.edu.cn
  • Supported by:
    National Natural Science Foundation of China(41631175)

Abstract:

Vehicle trajectory is a time series geospatial location sampling data. The traditional vehicle trajectory-map matching methods are mainly computed by ways of global or local incremental optimization, which limited the relative independence in matching process of the trajectory data in spatial temporal situation. To address this problem, this paper proposes the method of computing matching relationships between vehicle trajectory and road map based on the Particle Filter (PF) method. First, construct the road network from the road dataset, and search the neighboring nodes from the road network based on the vehicle sampling locations along the moving direction that are detected from the vehicle trajectory. Then, construct the motion model based on the vehicle trajectories, randomly generate particles on the road arcs that are related to the searched nodes, and move the particles along the sampled road segments according to the trajectory motion model. Second, compute the motion states of the particles according to the motion model in each time state, get the distance errors between the particles and the vehicle position sampling locations, obtain the particle weights based on the Gaussian probability density function, resample particles based on the random resampling method, and then update the motion states of particles iteratively. Finally, compute the accumulated weights of the particles in each of the topologically related road arcs, which are searched by the neighboring nodes, and calculate the matching relations between the vehicle trajectories and the map based on the accumulated weights of the particles. With this method, the experiments were conducted based on the vehicles' trajectories, which were two long sequenced trajectories with the total length > 102 km. The results showed that 85.51% and 93.01% correctness rates of vehicle trajectory-map matching experiments had been achieved for each of the vehicle trajectories. Besides, the motion of the vehicle sampling locations could be reflected by the spatial-temporal movements of the particles, where particles started to follow the motion of the vehicle sampling locations after a few time states. The results showed that it could achieve the accurate matching relations between the vehicle trajectories and the road map.

Key words: vehicle trajectory, time series, particle filter, road network, map matching, random particles, related road arcs, motion model