Abnormal event detection by variation matching

Sungmin Cho, Junseok Kwon

Abstract: Recently, surveillance systems have been widely used to analyze video recordings captured by surveillance cameras and to detect abnormal or irregular events in real-world scenes. In this study, we present a novel system that detects abnormal events. Unlike conventional methods, we consider abnormal event detection as variation matching problems. In approaching this problem, we transform from a single video to multiple ones by imposing variations on the video. Using a fully connected cross-entropy Monte Carlo method, we match multiple videos in a fully connected manner and detect abnormal events in all the videos concurrently. The experimental results show that our method can accurately detect abnormal events in multiple videos. Our proposed method can be used to automatically recognize abnormal events included in multi-view CCTV videos, which are available at airport terminals and underground stations.