A fast algorithm based on human visual system for abnormal event detection

Abstract

Fast abnormal event detection algorithm has high application value. But it is difficult to select appropriate feature representation to realize fast abnormal event detection. In view of HVS's dual pulse propagation theory and computational complexity, LBP and OF are used as temporal and spatial feature representation of video in this paper. Since human understanding involves the abstraction of the high-level features from low-level features, a streamlined depth learning network, PCANet, is used to extract high-level fusion features of LBP and OF. And three fusion methods are proposed in this paper. Finally, these high-level features are used to detect abnormal events. Experimental results show that the proposed algorithm performs better compared with other algorithms.