Violence Detection by Modified Descriptor Based on optical Flow in Video
The rapid growth of installation of surveillance cameras at public pace, security zone, and border area leads to demand for an automatic surveillance system at these places that monitors human activity to recognize violent and suspicious events. Nowadays, many kinds of research based on motion, trajectory, skeleton, silhouette, motion history image, and interest points, are diggingin this area to refine the existing methods. The main challenges to detect suspicious action are to estimate the velocity, orientation, texture, and densityaccurately. In this paper, we proposed a modified descriptor using magnitude and orientation of optical flow to estimate motion and orientation and support vector machine (SVM) classifier for classification of a suspicious event. In the experiment, our proposed descriptor outperforms over the histogram of Optical Flow Magnitude and Orientation (HOMO), Distribution of Magnitude and Orientation of Local Interest Frame (DiMOLIF), Violent Flow (ViF), and Oriented Violent Flow (OViF)descriptors for the parameter efficiency, standard deviation,Area undercurve on the benchmark dataset.
Keywords: Optical flow, Suspicious Event, Thresholding, Classifier, Velocity, Orientation, Feature Extraction