A robust approach for congested vehicles tracking based on Tracking-Model-Detection framework

Congested vehicles tracking is one of the most challenging problems in Intelligent Transportation System. Partial occlusions significantly undermine the performance of vehicles tracking in congested situation. Gradual occlusion often causes the drifting problem in many vehicles tracking methods. In this paper, we propose a robust algorithm for congested vehicles tracking based on Tracking-Modeling-Detection (TMD) framework system. We improve this method to track congested vehicles and apply it in traffic application. New rectangle region choosing strategy is proposed to select new tracking rectangle regions that contain best feature points when occlusion happens. Instead picking points on the rectangle grid in TMD method, we utilize points with good feature to enhance the efficiency and accuracy of tracking. The paper also presents experiment using video sequences of challenging congest traffic to verify the proposed method.