Lane detection in surveillance videos using vector-based hierarchy clustering and density verification

Automatic lane detection is known to facilitate the real-time traffic planning and identify traffic congestion. In this paper, we develop a visual surveillance trajectory clustering (VSTC) framework for automatic lane detection. Given a surveillance video, trajectories of vehicles are extracted at first. These trajectories contain behavior of vehicles on different lanes and are clustered by VSTC to retrieve candidate lanes. Finally, a density verification is applied to identify the correct lanes from candidate lanes. As shown in the experiments, our framework can identify the lanes by using trajectories without prior knowledge.