An Integrated Vision-Based Architecture for Indoor Environment Security

Abstract

A sense of personal and community safety is crucial to create a harmonized and productive society. However, the dramatic increase in home burglary activity is affecting social cohesion. Financial, physical and psychological impacts of such crimes are eroding trust within society and therefore lead to devastating consequences on social development. An automated security system for indoor environments is the best choice to address this severe problem. Broadly classified there are two types of automated security systems for smart homes: Sensor-based and Vision-based. Comparing with sensor-based systems vision-based home security systems are easy to set up, low cost and unobtrusive to catch illegal intruders in a wide range. Most of the vision-based systems developed so far are single-layered which can be easily deceived. This study proposes a novel integrated vision-based architecture to increase the security label (robustness) of a home security system. The system incorporates two modules: a face recognition module and a human motion detection module. A primary layer face recognition at the entrance door serves as a user authentication subsystem. A secondary layer human motion detection inside a room acts as a reliable backup in case, the primary layer is failed to catch the intruder for different reasons. The face detection and recognition is implemented using the current cutting-edge technology, deep learning. Histogram of Gradient (HOG) with Support Vector Machine (SVM) is employed to implement a human motion detection module. Several comparative experiments conducted have guaranteed a promising performance assuring the reliability of the methodological approaches followed to carry out this study. In addition, a scenario-based architecture evaluation with field tests conducted has shown the robustness and the feasible application of the proposed architecture in a consumer environment.