An energy model approach to people counting for abnormal crowd behavior detection

摘要: Abnormal crowd behavior detection plays an important role in surveillance applications. We propose a camera parameter independent and perspective distortion invariant approach to detect two types of abnormal behavior. The typical activities are people gathering running. Since counting is necessary for detecting the behavior, we present potential energy-based model estimate number public scenes. Building histograms on X- Y-axes, respectively, can obtain probability distribution foreground object then define entropy. Crowd Distribution Index by combining results with entropy represent spatial crowd. set threshold gathering. To running, kinetic energy determined computation optical flow Index. With threshold, be used test performance our algorithm, videos different scenes densities experiments. Without calibration training data, method robustly behaviors low load.