Home Surveillance System using Computer Vision and Convolutional Neural Network

In this paper, we introduce a home surveillance system that utilizes computer vision techniques to recognize intrusions and detailed threat information including classifying types of trespassers and specific weapons used. Process of identifying intrusions is achieved by our Smart Intruder Detection and Surveillance System (SIDSS) which involves 3 stages of computer vision algorithms. 3-stage SIDSS includes an optimized convolutional neural network (CNN) for threat and intrusion detections, cascading classifiers for locating any potential intruders with correcting mechanism to overcome undetected threats from the previous stage, and principal component analysis (PCA) to efficiently train the facial recognizer to accurately differentiate passersby from potential intruders. Our system is scalable for various surveillance events and can be expanded with additional pre-processed datasets added to the SIDSS model to manage greater surveillance areas. Through this enhanced configuration, the system can achieve enhanced accuracy of recognizing broad range of weapons used and intruders.