Vehicle tracking at intersection in image sequences with multilayer concept

摘要

Multi-object tracking is an important subject in computer vision. Vehicle tracking can be used in surveillance and traffic management systems. Tracking at intersection has gained attention in recent years. Because of different movements, motion flows and congestion, tracking at intersection becomes a complex task. In this paper, we develop a multilayer concept for tracking at intersections. We construct multiple layers for covering the field of view of intersection and assign each vehicle to a specific layer after entering the intersection according to its motion. After defining neighbors in each layer based on motion flows, tracking is done in that layer. So we divide different motions into specific layers. Our method is independent of detection phase and can be used with different detection methods. Experiments show that accurate tracks can be acquired even when vehicles pass by each other. According to experiments, it is possible to apply this method in real time vehicles tracking applications at intersections.