Graph-Cut Optimization for Video Moving Objects Detection with Geodesic Active Contour

Abstract

The traditional graph-cut for video moving objects detection is a global optimization algorithm, the result may be over-smoothing. The lack of local information in graph-cut limits the ability to precisely localize object boundaries. In this paper, moving objects detection algorithm is improved by introducting geodesic active contour. By the Kalman prediction of the number of objectives pixels and objectives-background pixel-pairs, and adaptive updating of the nodes flux with geodesic active contour, the proposed algorithm is successfully applied to video moving objects detection. Though adaptive updating of the nodes flux with geodesic active contour, the proposed algorithm will have a better edge capture ability of moving objects. Experimental results show that the proposed algorithm is more effective than graph-cut for video moving objects detection in complex backgrounds.