NeuroFace Recognition System

Highlights

• dlib 128-dimensional face embeddings and 68-point landmarks, with centroid tracking across frames that skips re-embedding while the face count is stable
• Multi-face detection with real-time FPS monitoring
• Tkinter registration GUI plus a Flask dashboard for records
• SQLite attendance persistence and CSV feature storage

How it was built

For my final-year project I built a face-recognition attendance system. A Tkinter tool captures face images, and a batch script averages each person's embeddings into one 128-dimensional vector using dlib's pretrained models. A live OpenCV loop matches faces by Euclidean distance with a strict 0.4 threshold and writes attendance to SQLite, with a small Flask dashboard to view records. To save compute, it skips re-embedding while the face count is unchanged and tracks people by centroid between frames. Limits: no tests or measured accuracy are recorded.