A Robust Vehicle Tracking for Real-time Traffic Surveillance
Bo Zou, Yujie Liu, Xiao Yu Zhou · 20th ITS World CongressITS Japan · 2013
Vehicle tracking, as one of the important steps in traffic surveillance, has intensively been studied in the past decades. Although numerous approaches have been proposed, robust tracking of moving vehicles for traffic surveillance remains a huge challenge. Difficulties arise due to complex road context, various lighting and different weather conditions. Besides, multiple targets, occlusion and high speed driving are also key things which lead to tracking failure in traffic surveillance. In this paper, a recently developed Tracking-Learning-Detection (TLD) approach is used for tracking vehicles under these challenging conditions. Meanwhile, the authors improve several strategies for multi-vehicle tracking and long-term occlusion. The proposed algorithm has been tested in a traffic surveillance system that is running over a real traffic scenario. The results from 217 Red Light Cameras on 53 intersections demonstrate that the proposed approach is feasible and effective for vehicle tracking.