Haven AIWatchdog

Abstract

Popular Home Security Systems are reactive in nature and mostly do not have any kind of intelligent component able to assist with timely distress signalling. Haven AI Watchdog is a home security system that provides a discreet method for triggering emergency alerts thorugh configurable gesture sequences mapped to protocols. Instead of relying on traditional/static/physical means of calling for help that may be inaccessible, time consuming or infeasible, the system allows users to signal for help using pre-configured hand gestures. The system combines real-time gesture recognition with facial verification to reduce false alarms. The implementation consists of a Flutter mobile application for all types of users that wll be involved in the system, a Django main server backend, and a locally installed processing unit with a camera with local script server. Gesture detection is performed using a YOLO v11m model trained on a hand gesture dataset from Kaggle. Facial verification uses the FaceNet-512 model via the DeepFace library. The system supports multi-user households with role-based access and includes a training mode for users to practice gestures safely.