Facial Recognition Attendance

Faster attendance

up to 80%

The problem

Manual sign-in at chapter meetings was slow and error-prone, and the records still had to be cleaned up afterward.

How it works

1. Enrollment. Each member's reference photo is encoded into a face embedding once.
2. Detection. OpenCV reads frames from the camera, and faces are located in each frame.
3. Verification. Each detected face is embedded and compared against the enrolled set. The closest match under a distance threshold is accepted.
4. Logging. Recognized members are written to an attendance record with Pandas, ready to export.

Try the first stage

The demo below rebuilds just the detection stage for the browser with a tiny BlazeFace model. By design, it doesn't identify anyone. Recognition requires enrolled face data, and that doesn't belong on a public website.

Live demo

Real-time face detection

The detection stage of my attendance system, rebuilt for the browser. It finds faces and landmarks in your webcam feed. It does not identify anyone.

model

BlazeFace (short range)

size

~230 KB

runtime

MediaPipe · GPU

Uses your camera locally. Frames are processed on your device and never uploaded or stored.

Start camera

Faces

—

Inference

—

FPS

—

Trade-offs

• Privacy. Face data is sensitive. Embeddings rather than raw photos, local processing, and opt-in enrollment matter as much as accuracy.
• Threshold tuning. A strict threshold rejects real members in bad lighting; a loose one accepts look-alikes. The right setting depends on the room and the camera.