Driver behavior analysis to reconstruct vehicle trajectory in urban area

Abstract

Recently, many applications in video surveillance, intelligent traffic system and social security management require information of vehicle trajectories in an urban area. Conventional methods tried to track moving vehicles by using either appearance matching or spatial and temporal information to estimate vehicle trajectory. However, the authors have recognized a phenomenon that vehicles have tendencies to follow some main roads. Therefore, the trajectory could be reconstructed based on a small observation-set. By using a training process, we analyze driver behavior to find out main roads of an urban area. Besides, the authors propose a new idea to predict the vehicle trajectory based on these learned main roads. Based on an observed location, these main roads that the vehicle could go through have been estimated. The full trajectory is would be fusion process of these roads. Experiments prove our hypothesis about the existence of key roads that citizens prefer to use and proposed method could improve reconstruction performance.