Abstract

In this work we address the issue of focused computation in computer vision for effectiveness and efficiency. In particular, we propose a scheme for motion segmentation and tracking that links scene-oriented contextual knowledge with the computational constraints involved. Such an approach enhances sensitivity to visual evidence and gives the selectivity we require. The approach uses Bayesian belief revision techniques to map explicit scene knowledge onto implicit causal dependent constraints in controlling computational parameters used in motion segmentation and tracking. We show experimental results from applying this method in improving existing techniques in traffic surveillance applications. 1 Introduction In the past, research in computer vision was greatly influenced by the theory of David Marr [17]. Visual processing modules in the Marr framework operate at different levels of abstraction, such as edge detection, surface reconstruction and model matching. Typically, then, a hi...

Citations