A Method for People Counting in Complex Scenes Based on Normalized Foreground and Corner Information

For the problem of people counting in intelligent video surveillance, a method of people counting in complex scenes based on the normalized foreground and corner information is proposed. First, based on the binary foreground, the area of normalized foreground after perspective correction is calculated. Second, the optimized corner information of foreground is extracted to compute the occlusion coefficient of crowd. Finally, the above two features are used as the inputs of the Back Propagation (BP) neural network to train and test the people counting. Experiments results show that, the proposed method exhibits good performance in complex scenes.