In this paper a simple and effective crowd behavior normality method is proposed. We use the histogram of oriented social force (HOSF) as the feature vector to encode the observed events of a surveillance video. A dictionary of codewords is trained to include typical HOSFs. To detect whether an event is normal is accomplished by comparing how similar to the closest codeword via z-value. The proposed method includes the following characteristic: (1) the training is automatic without human labeling; (2) instead of object tracking, the method integrates particles and social force as feature descriptors; (3) z-score is used in measuring the normality of events. The method is testified by the UMN dataset with promising results.
Proceedings of the International Conference on IT Convergence and Security (ICITCS 2013), 4p.