Real-time elliptical head contour detection under arbitrary pose and wide distance range

Abstract

In this paper, we propose a real time elliptical head contour detection method based on quadrant arcs, which is efficient and robust to arbitrary head pose and wide distance range. First, the moving object area is detected according to background model which is built on three color channels. Then, all the valid elliptical arcs are extracted out from connected edges, and classified into four kinds of quadrant arc sets according to their gradient information. Finally, the arcs lying in different sets are combined to fit out the elliptical head contour based on the least square method. Experimental results confirm the robustness and the accuracy of this method under arbitrary head pose and wider distance range, as well as the real time property, strong robustness to long or short hair and with or without hat.