Suspicious and violent activity detection of humans using HOG features and SVM classifier in surveillance videos

Abstract

Crimes such as theft, violence against people, damage to property, etc., have become quite common in a society, which a serious concern. The traditional surveillance systems act like post mortem tools in the sense that they can be used for the investigation to detect the person behind the theft, but it is only after the crime has already occurred. In this chapter, we propose a method for automatically detecting the suspicious or violent activities of a person from the surveillance video. We train the SVM classifier with the HOG features extracted from the video frames of two types: frames showing no violent activities and those showing violent activities like kicking, pushing, punching, etc. In the testing phase, the frames from the surveillance video are read and processed in order to classify them as violent or normal frames. If the frames classified as violent frames are detected, an alarm is raised to alert the controller. It can be used to keep track of the time duration for which a person is found loitering at a place being monitored. If the time exceeds a predefined threshold, the alarm is raised to alert about any potential suspicious activity so that it can be checked on time.