Counting pedestrians passing through a line in crowded scenes by extracting optical flows

Abstract

Recently, pedestrian density estimation in crowded scenes based on optical flow extraction has attracted much attention. In this report, we propose a counting method of pedestrians passing through a specific line, called counting line, in crowded scenes. In our proposed method, optical flows whose segment crosses the counting line are used for estimation. In addition, we delete optical flows detected falsely or extracted from background image. Through experimental evaluations using actual video sequences, we show that the correlation coefficient between the number of optical flows and the number of pedestrians passing through the counting line is more than 0.856. We also show that the mean relative error of our proposed method is less than 0.068.