2019/08/17 by Trong-Nguyen Nguyen, Nguyen, Trong Nguyen, Jean Meunier +1 · 5 citations
Computer Science · #Anomaly Detection Techniques and Applications #Video Surveillance and Tracking Methods #Human Pose and Action Recognition
paper · pdf · doi:10.48550/arxiv.1908.06351
Anomaly detection in surveillance videos is currently a challenge because of\nthe diversity of possible events. We propose a deep convolutional neural\nnetwork (CNN) that addresses this problem by learning a correspondence between\ncommon object appearances (e.g. pedestrian, background, tree, etc.) and their\nassociated motions. Our model is designed as a combination of a reconstruction\nnetwork and an image translation model that share the same encoder. The former\nsub-network determines the most significant structures that appear in video\nframes and the latter one attempts to associate motion templates to such\nstructures. The training stage is performed using only videos of normal events\nand the model is then capable to estimate frame-level scores for an unknown\ninput. The experiments on 6 benchmark datasets demonstrate the competitive\nperformance of the proposed approach with respect to state-of-the-art methods.\n