AI Based Home Anomaly & Intruder Detection System

The rapid expansion of smart home technologies has increased the need for intelligent and autonomous security systems. Conventional home security solutions that rely on motion sensors and manual CCTV surveillance are reactive in nature and frequently generate false alarms due to pets, lighting changes, or routine household activities. This paper presents an AI based home anomaly and intruder detection system that provides real-time security monitoring using computer vision and deep learning techniques. The proposed system utilizes MediaPipe for efficient face detection and a FaceNet-based deep learning model to generate facial embeddings for identity recognition. Cosine similarity is used to compare detected faces with registered users stored in a local database. When an unknown individual is detected, the system automatically activates a continuous audible alarm, records video evidence, and sends instant alerts with the recorded clip to the homeowner via Telegram. Experimental results show that the system operates reliably in real time, significantly reduces false alarms, and enhances overall smart home security.