누적 움직임벡터를 이용한 겹침현상에 강인한 보행자 추적

Abstract

In this paper, a pedestrian tracking method that meets the surveillance system using the gradient histogram and cumulative motion vector is proposed. In order to get the feature of pedestrian, a block-by-block histogram is created using the direction of the gradient based on HOG (Histogram of Oriented Gradient), and then only pedestrians are classified by learning the Adaboost in the video. Next, the motion vector is cumulated based on the area of pedestrian detection. In the case of overlap, the direction of pedestrian using the obtained cumulative motion vector is predicted. Through experiments, it is proven that the proposed method is more effective compared to the existing methods for the accurate tracking of the pedestrian.