Tracking of polyhedral objects

Abstract

Here, we address the problem of tracking of polyhedral objects in images from traffic scene. The matching of the model is achieved by minimizing the distance between the projected segments of the 3D model and the segments extracted in the 2D image. The success depends on the global optimization of the cost function that quantifies the quality of alignment. The existence of numerous local minima makes it impossible to locate the global solution. In this study, we propose an effective global optimization strategy to overcome this problem. Results on simulated and real images are presented.