I’ve built an advanced AI security system that runs entirely on-site (no Wi-Fi required) and uses YOLOv8 for real-time people tracking. I rely on pretrained models for both object detection and facial recognition, ensuring a high level of accuracy from the start. The system tracks individuals from a distance, then only triggers facial detection once they’re close enough to be considered a potential threat, which is determined by measuring the bounding box area around the face. I store facial embeddings in JSON files, enabling quick identification of known individuals. When the system detects an unrecognized person, it automatically sends a text message alert to the property owner. This combination of object tracking, threshold-based facial recognition, and local processing helps maintain reliable and efficient security without relying on external networks.