Crowd Estimation Using Image Processing Based on Gray Scale Edge Histogram in Subway Station of Beijing

Traffic surveillance systems that use cameras are an important tool for controlling pedestrian crowds in subway stations. Large stations need many cameras which cause difficulty for traffic-control operators who need to react quickly when an accident takes place at a specific location. This paper reports on a new technology based on computer vision for use in a pedestrian crowd estimation and safety detection system. The system first collects data from video cameras and instantly converts the data to gray-brightness images frame by frame. Then the outline of each pedestrian in each image is extracted using Sobel filter convolution. Parameters of the processed images are obtained for each frame. The functional relationship between pedestrian density and image parameters (such as bin30) is set up. An alert system is developed based on this proposed method. The paper presents a test-case of the Lama Temple interchange station of the Beijing subway. The density and fundamental diagram of the flow of pedestrians is tested and compared with the same data collected manually. Results show that this methodology is effective, fast, and practical for use.