摘要: A method for segmenting monocular images of people in motion from a cinematic sequence frames is described. This based on image intensities, motion, and an object model-i.e., model the person motion. Though each part may move different directions at any instant, time averaged all parts must converge to global average value over few seconds. People be occluded by other people, usually it not easy detect their boundaries. These boundaries can detected with information if they directions, even there are almost no apparent differences among intensities or colors. Each scene divided into several parts, distinct The merged single group iterative merging algorithm because coherently. analogous property perceptual grouping human visual perception Experiments complex real scenes produced results that supportive authors approach segmentation >