Machine Learning face recognition

The system uses no pretrained models: it implements Eigenfaces (PCA) and Local Binary Patterns, plus a fusion that combines both. In testing it reached 85 to 92% accuracy depending on lighting.

The FastAPI service exposes 25+ endpoints to register people, recognize in real time, retrain the model on registration and raise alerts. It has 95% test coverage.

The React Native app built with Expo captures with the camera, calls the API and stores data offline with AsyncStorage.