A Low Cost on-Device Intruder Detection System for Smart Home Environment

Robust security is tritical for smart homes to protect families and goods from potential intruder threats. Vision-based intruder detection systems are commonly used for this task. However, only a few households have adopted such security technology because of its high cost, which includes both expensive devices and service costs. To address this issue, a low-cost intruder detection system is investigated in this study, leveraging a Raspberry Pi 4 with a camera module. Smart resource utilization strategies are employed to enable on-device local processing for privacy preservation of sensitive video data. To achieve real-time processing on resource constrain hardware, efficient motion detection is accomplished using the frame subtraction technique, while rapid face detection is accomplished through the haar cascade utilizing Kalman filter. Face identification using OpenFace model is speed up by subsequent template matching technique. The proposed algorithm achieved an average frame rate of 22 FPS making it capable for real-time intruder detection with environmental constrain.