Counting pedestrians and cars with an improved virtual gate method

This paper presents a system for counting the number of the pedestrians and the cars passing through a line, with an improved virtual gate method. First, the foreground area is extracted using GMM-based background modeling. Second, the velocity field in the foreground regions are computed by an improved optical flow estimation method, which enables the estimation of the dense velocity field of the cars. Then, we separate the velocity field into different groups by using threshold constraints and perform integration on each group along a virtual gate. According to the integration results, the number of the pedestrians and the cars can be estimated. Finally, the experimental results validate our method on several real video sequences.