Abstract
We propose a method for extracting human regions from a sequence of three images using motion, head detecting and simple human countor model. First the method calculates the subtraction between successive frames to find a moving region. Next it classifies the edge segments into curves and lines, and then represents the curves as circles of candidates for a human head. Simple contour model of human is placed at the head position which the moving region and the circles decide, then converges to make a rough human region. To obtain a more accurate region, the image is segmented into regions with different color, and the regions are merged to make a human region. We conducted experiment using real scene images to evaluate the proposed method, and showed that human regions were almost extracted successfully.
