Abnormal-Trajectory Detection Method Based on Variable Grid Partitioning

Abstract

Abnormal-trajectory detection can be used to detect fraudulent behavior by taxi drivers when carrying passengers. Existing methods usually detect abnormal trajectories based on the characteristics of “few and different”, which require large data sets and, therefore, may identify “few and near” trajectories chosen by drivers according to their driving experience as abnormal situations. This study proposed an abnormal-trajectory detection method based on a variable grid to address this problem. First, the urban road network was divided into three regions: high-, medium-, and low-density road network regions using a kernel density analysis method. Second, grids with different sizes were set for different types of road network regions; trajectory tuples were obtained based on the grid division results, and the abnormality rate of the trajectory was calculated. Finally, a trajectory-abnormality probability function was developed to calculate the deviation of each trajectory from the benchmark trajectory to detect abnormal trajectories. Experimental results on a real taxi trajectory dataset demonstrated that the proposed method achieved a higher accuracy in detecting abnormal trajectories than similar methods.