Trajectory Topic Modelling to Characterize Driving Behaviors with GPS-Based Trajectory Data

Abstract

The rapid accumulation of large-scale driving data represents an opportunity to improve our understanding of driving behavior patterns and driver traveling intentions. However, limited efforts have been devoted to understanding these patterns and the travel intentions behind them. This study proposes a new trajectory topic model (TTM) to explore latent driving patterns from driving trajectory data and to qualitatively analyze drivers’ main traveling intentions. These trajectory data were collected from more than 150,000 commercial vehicles in Fujian Province, China. After data preprocessing, the TTM was then established to decompose trajectory data into various topics with corresponding probabilities, which were correlated to drivers’ preferences. Several experiments conducted in Fuzhou City were performed to evaluate the feasibility and efficiency of the TTM using a real trajectory dataset. The results show that the TTM could effectively mine users’ driving behavior patterns with topic probability. The model would enable us to understand the context in which drivers travel and learn their individual preferences. It is also beneficial in that it can predict drivers’ behaviors, analyze traffic patterns in an entire city, and even help autonomous vehicles to learn from drivers.