The Research Of Video-based Body Detection And Counting

Nowadays, the economy develops quickly and persons flow frequently, It is urgent to monitor and manage the persons flow on public places. Pedestrian detection and counting technology based on video is applied in several fields such as society, economy, transportation, culture tourism and so on. There is often an entrance which can controls the quantity of pedestrian in many place. And the entrance often installs a video camera to shoot the pedestrian's face. In this paper, we take the characteristic of the surveillance into account. So we mainly study the pedestrian's detection and tracking technology in video which is took by stationary inclined video camera. The main contents in this paper are as follows:First, an background subtraction algorithm based on self-adaptive background is proposed to segment the moving pedestrian through analyze the advantages and drawbacks of the existing algorithms on how to detect the pedestrian in video. It is often aimed at observe the face in video surveillance. We can improve the accuracy of target detection through face detection within the moving target scope. Because it is not in favor of extract the feature of the face in video surveillance, we detect face through skin segment. The existing algorithms about skin segment have been analyzed. Then an improved elliptic skin model has been proposed to detect face in video.Second, the GM(1,1) model combines the feature matching has been used to track the pedestrian. After analyzing the feature of the gray prediction theory, it has been used to predict the pedestrian when tracking. It is face but not body has been tracked in order to improve the accuracy of tracking. The speed of the system is improved through narrowing the detection area and matching features selectively. Third, a count area is set to count the pedestrian who has been passed the entrance in order to realize the system. The experimental results have been presented and analyzed.The proposed method is validated through testing in an actual video image sequence which is taken from an entrance. The results show that the proposed method can count the entrance pedestrians effectively.