Unusual crowd event detection: An approach using probabilistic neural network

Abstract

Unusual event detection is a disputing problem in the field of security assets for public places. In this paper, we proposed detection of unusual crowded events in video established on interest point information. The distribution of magnitude and orientation holistic feature descriptor extracted from the histogram of optical flow for detection of usual and unusual events. In the proposed method, the probabilistic neural network approach is employed for unusual event detection in video. The reason for using PNN is more accurate, fast in training process and much faster than support vector machine. The proposed method used three benchmark datasets, UMN, PETS2009, and violent flows datasets, for the experiment. The obtained results compared with state-of-the-art methods, and improved outcomes are depicted to evaluate the outperformance of the proposed method.