Real-Time Estimation Method of the Number of Pedestrians in Video Sequences

In a ubiquitous network society, information on mobile nodes such as pedestrians with cell phones in a target region is useful for network configuration, communication control, and so on. In this paper, we propose a novel method to estimate in real time the number of pedestrians in video sequences using a stationary camera. The proposed method extracts moving regions from video frame grabs using background difference and frame difference, and estimates the number of pedestrians based on the size of the moving region using a conversion function constructed based on prior learning. We evaluate the effectiveness of the proposed method by comparing the estimated values with the actual values, and show that the estimation error of the number of pedestrians is smaller than 0.1 in approximately 50% of frames, and that the mean absolute error is 0.37.