KlugOculus: A Vision-Based Intelligent Architecture for Security System

Abstract

Vision-based system has gained significant attention in detecting the abnormal activities of intruders and alerting security with the amalgamation of adaptive video analytics techniques. The implementation of this kind of system works on face recognition, where the dedicated hardware with better computation power is limited in the previous studies. In this study, vision-based intelligent architecture and systems are proposed to detect intruders through facial recognition and sensors with customized hardware. As a part of the training, each subject was trained with 6 different pictures for a total of 120 images. Facial recognition implemented with machine learning (ML) inspired support vector machine (SVM) along with a histogram of oriented gradients (HOG). During the real-time implementation, the SVM model loaded in Raspberry Pi 3 has attained 99.9% accuracy for 20 different subjects. The proposed system can provide an accuracy of 99.9% even with tilted images of the subject, so it can be adopted by the different security personnel to boost the security system for the identification of intruders.