Moving Object Segmentation Using Dynamic 3D Graph Cuts and GMM
Bo Li, Baozong Yuan, Yunda Sun · 2006
It is one of the most challenging problems in computer vision how to segment moving objects accurately. In this paper, we present a novel approach to segment moving objects with edge information and temporal information using 3D graph cuts model when cameras is fixed. Moving object segmentation is modeled as finding a minimum energy of 3D graph. Our algorithm assigns n-links in 3D graph according to spatial gradient in same frame and temporal gradient in neighboring frames. Gaussian mixture model is used to assign t-links with edge difference term and shadow elimination term. Finally, a dynamic graph cuts algorithm is used to find the minimum cut of 3D graph and segments moving objects in image sequences. Experiments show that our approach achieves nice performance