Fast Anomaly Detection Based on 3D Integral Images

摘要

In this paper, we propose a method to detect abnormal events from videos based on the integral image under Bayesian framework. In our implementation, we consider a regular cube in the videos as one event. Each event is represented as a motion histogram which can be calculated fast from our proposed 3D integral images. Furthermore, we estimate the anomaly probability under the Bayesian framework, where we estimate the prior knowledge from the motion magnitudes and calculate the likelihood based on our maximum histogram templates. Experiments on the public datasets show that our method can effectively and efficiently detect abnormal events in complex scenes.