Pedestrian Detection Based on Active Contour Models

In this paper, we propose an improved active contour model to detect pedestrian in video sequences. Our method can detect pedestrian whose profiles are constantly changing when they are walking. First, we improve the internal energy computation by using the squares of the distance between average distance of all adjacent points in the contour curve and the control points, and construct local energy window to search for optimal solution. It could increase the continuity of the object contour and make control points distribute evenly to avoid aggregation. Second, Greedy method is proposed to obtain minimum energy and find image contours differing from the traditional variational approach. Third, let the gradient value be part of point´s external force. The experiment results prove that the algorithm has a better performance in detecting pedestrian´s contour compared with other object detecting methods or traditional active contour models.