An adaptive graph cut algorithm for video moving objects detection

The algorithms based on graph cut have the advantage to detect the moving objects effectively and robustly. The main trouble of the algorithm based on graph cut is that its model parameters will be determined empirically. In this paper, a novel algorithm of adaptive graph cut is proposed to detect video moving objects. Based on Markov random field model, the proposed algorithm uses the numbers of moving objects pixels and objectives-background pixel-pairs to describe the geometric features of the moving objects. And the relationship between the geometric features of the moving objects and the model parameters are set up. In this paper, the model parameters are adaptively optimized through the extraction and prediction of the geometric features of moving objects. Then the detection based on the graph cut is preformed on ROI, which well achieves the balance between the computation and accuracy. Finally, the experimental results show the proposed algorithm can hold the details of moving objects more effectively compared with other algorithms, and improve the detection performance of moving object in the video surveillance.