Human Identification using Histogram of Oriented Gradients (HOG) and Non-Maximum Suppression (NMS) for ATM Video Surveillance

Abstract

Today video surveillance is the scorching  topic in the research field of Computer Vision. Around 400  million surveillance cameras are used in various sectors  which merely act as blind and record videos for post incident analysis. With the increase in crimes around the  world, video surveillance plays a key role in identifying  anomaly activities in a day to day application. One such  application is a robbery in Automated Teller Machine  (ATM). Identifying and tracking unlawful human activities  is a challenging task in the ATM video surveillance system.  Thus, an effective Human detection method using computer  vision and image processing is proposed to create  phenomenal results. The proposed approach, incorporate the  most acclaimed Histogram of Oriented Gradient (HOG)- Support Vector Machine (SVM) with a combination of  Adaptive Background Generation Model and Non Maximum Suppression (NMS) algorithm to detect human  with appropriate results in a video sequence. The  experimentation result of the proposed method is applied on  ATM Video Surveillance, real-time dataset. The  performance analysis is evaluated by considering the  average value of metrics such as the number of an input  frames, ground truth elements and frames with human  identified which acquired 97% of accuracy.