A Detection Method of Abnormal Event in Crowds Based on Image Entropy
Haijiang Hao, Xin Yan Li, Mengting Li · 2019
The occurrence of group abnormal events will pose a harm to social public safety. In order to improve the detection efficiency of abnormal events in dense populations. This paper proposes an algorithm for crowd abnormal event detection based on image entropy. The algorithm firstly extracts the amplitude of the optical flow of each frame in the video by Farneback optical flow method, and then constructs the image representation of the amplitude of the optical flow. The difference of the amplitude of the optical flow of two consecutive frames will provide us with the characteristic map; A population anomaly event is identified by comparing the entropy difference between the feature maps to a particular threshold. The experimental results show that the algorithm has high detection efficiency and good real-time performance.