Computer vision techniques for background modelling in urban traffic monitoring

Abstract

In this chapter, several background modelling techniques have been described, analyzed and tested. In particular, different algorithms based on sigma-delta filter have been considered due to their suitability for embedded systems, where computational limitations affect a real-time implementation. A qualitative and a quantitative comparison have been performed among the different algorithms. Obtained results show that the sigma-delta algorithm with confidence measurement exhibits the best performance in terms of adaptation to particular specificities of urban traffic scenes and in terms of computational requirements. A prototype based on an ARM processor has been implemented to test the different versions of the sigma-delta algorithm and to illustrate several applications related to vehicle traffic monitoring and implementation details.