A Two-Layer AI-Integrated Multiple-Camera System for Detecting Human Intrusions

Human intrusion is crucial to residential security. To detect intruders, conventional residential security systems mainly rely on a complex network of sensors, responsive to sound, light, and door/window activities, in conjunction with surveillance cameras equipped with motion detection. However, not only are these systems costly through service providers, but they also bring much inconvenience to homeowners’ regular life (e.g., desired window openings need disabling sensors first). The continuous camera recording incurs excessive data that overwhelms homeowners, and streaming data online is also a privacy concern for homeowners. This paper proposes to solve these multifaceted issues via a two-layer AI-integrated multiple-camera system, automating the detection of strangers and weapons (knives in this paper for illustration), while enabling the cross-camera tracking to pinpoint and monitor suspicious behavior. We demonstrate its effectiveness through implementation and evaluation.