Detecting Robbery and Violent Scenarios

Abstract

In this paper, we devise a method to detect robbery and violent scenarios for the goal of improving the security of self-service banks. The method first extracts the motion region from the video and denotes this region with a rectangle. Then, the method calculates the optical flow and energy of the rectangular region. The method takes the length and width of the rectangle, the energy, and the orientation variance of the motion region which is denoted by the same rectangle as features to distinguish the video where robbery and violent segments occur from other videos. The experimental results from a number of surveillance videos show that our devised method is feasible and can achieve a very good performance.