Video Abnormal Event Detection Based on ELM
Guangli Wu, LIU Liping, Zhang Chen, Dengtai Tan · 2019
To the problem of poor real-time detection of abnormal behaviors of the crowd, low recognition rate of classification algorithm and less characteristic quantity, the characteristics of crowd movement based on optical flow field and Harris corner are presented. Firstly, the optical flow vector of the crowd in video is extracted by the Lucas-Kanade optical flow method, the average kinetic energy and the direction entropy are calculated, and the distance potential energy of the crowd is extracted using the Harris corner detection algorithm. Secondly, the Extreme Learning Machine (ELM) and Support Vector Machine (SVM) are used to establish the classification model for the crowd. The experimental results show that by comparing the two classification algorithms, the time reduction of the ELM is 1.6ms (nearly a quarter) when the recognition rate is close to the SVM (the recognition rate of ELM is 96.92%, and that of SVM is 96.15%) and the characteristics of crowd movement based on optical flow field and Harris corner have higher robustness.