Loitering Detection in Home Surveillance System

Abstract

An automated system which recognizes human action from continuous videos, though seem like a work of fiction, is possible now due to the magnificent increase in processing power and improvements in the field of automated analysis. Human Action Recognition (HAR) has a wide range of applications in fields of Gaming, Animation, Assisted Living and Surveillance etc. The home surveillance systems are easily available for anyone to use, the surveillance camera records video 24/7, reviewing the same to find out malicious activity will be a time critical work, the proposed method focuses on creating a system which looks for the loitering action in home surveillance videos and records the loiterer information in a file with the photo identity of the loiter. Loitering is an action that is defined as walking around for no obvious reason. In a home environment, if a random person seems to be walking around the house, then it is a suspicious activity, which needs to be addressed. The proposed system uses YOLOv3 (You Only Look Once) for identifying the person of interest and loitering is tracked using DeepSORT (Simple Online Real Time Tracking), Once the loiterer is identified and tracked, his face is captured by using the Haar Cascade algorithm and stored in for future reference. People who have installed home surveillance CCTV system can use this to system to identify loiterers.