Video Anomaly Detection and Localization in Crowded Scenes

Abstract

Nowadays, the analysis of abnormal events becomes more and more exhausting due to the divine use of surveillance cameras. This paper proposes a novel approach to predict and localize anomaly events. In this paper, a new framework for motion extraction called BQM is proposed. Then, the regions of interest are extracted and a filtering process is applied to eliminate the non-significant ones. However, for more precision, the HFG descriptor is applied for each region already divided into non-overlapping cells, Finally, we have evaluated our method using UCSD and Avenues datasets. The Sparse Auto-encoder, an instance of a deep learning strategy is presented for efficient abnormal activity detection and the Softmax for the classification.