Abnormal crowd behavior detection based on local pressure model

Abnormal crowd behavior detection is an important issue in crowd surveillance. In this paper, a novel local pressure model is proposed to detect the abnormality in large-scale crowd scene based on local crowd characteristics. These characteristics include the local density and velocity which are very significant parameters for measuring the dynamic of crowd. A grid of particles is placed over the image to reduce the computation of the local crowd parameters. Local pressure is generated by applying these local characteristics in pressure model. Histogram is utilized to extract the statistical property of the magnitude and direction of the pressure. The crowd feature vector of the whole frame is obtained through the analysis of Histogram of Oriented Pressure (HOP). SVM and median filter are then adopted to detect the anomaly. The performance of the proposed method is evaluated on publicly available datasets from UMN. The experimental results show that the proposed method can achieve a higher accuracy than that of the previous methods on detecting abnormal crowd behavior.