Vehicle anomalous trajectory detection algorithm based on road network partition

• Published in: Applied intelligence (Dordrecht, Netherlands) Vol. 52; no. 8; pp. 8820 - 8838
• Main Authors: Zhao, Xujun, Su, Jianhua, Cai, Jianghui, Yang, Haifeng, Xi, Tingting
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.06.2022; Springer Nature B.V
• Subjects: Algorithms; Anomalies; Artificial Intelligence; Computer Science; Consumption; Dijkstra's algorithm; Fraud; Machines; Manufacturing; Matching; Mechanical Engineering; Processes; Roads; Roads & highways; Series (mathematics); Taxicabs; Training; Trajectories; Map matching; Consumption threshold matrix; Anomaly trajectories detection; Road network division
• ISSN: 0924-669X; 1573-7497
• DOI: 10.1007/s10489-021-02867-5

In the process of carrying passengers, taxi drivers may have fraud problems. To solve the problem, we propose an abnormal trajectory detection algorithm( RNPAT ) via road network partition in the paper which is divided into four stages: map matching, road network partition based on insert points, off-line training, and anomaly detection. A trajectory is converted into a series of ordered combinations of points sequence to make it follow the actual direction of the road network at the stage of map matching, and the problem of low data quality obtained by location devices is solved. In the road network partition phase, the missing point at the intersection of adjacent roads of trajectory is calculated, and then according to insert points, trajectories are divided to train the road consumption. At the stage of off-line training, the consumption of the road is modeled, and the Dijkstra algorithm is used to train the minimum consumption between each S-D pair, in which S is the starting point of vehicle operation and D is the destination of vehicle operation. In the anomaly detection phase, we calculate the consumption threshold matrix supporting anomaly detection and the consumption of each trajectory, and compare the trajectory’s consumption with corresponding threshold to judge whether the trajectory is abnormal. Finally, the effectiveness of RNPAT is verified by Shanghai Taxi data. In addition, RNPAT is compared with TADSS and TRAOD validates that RNPAT has higher efficiency and higher accuracy.