Tracking Multiple Pedestrians in Crowd

Abstract

In the domain of city design, the demand for the development of the visual tracking method is increasing, which provides the pedestrian paths. This paper proposes the tracking method which reduces the ambiguity of matching. When features of a pedestrian are ambiguous in the original image, the proposed method generates plural candidates for the pedestrian path. In the following frames the proposed method veri es them to eliminate false ones. We experiment on real image sequences.