All projects

Computer vision · Edge authentication

Smart Face Door Lock.

A webcam-based authentication prototype combining face embeddings, blink checks, and a local control interface.

Status
Documented prototype
Context
Facial authentication · February–April 2026
Technology
Python · ArcFace · MediaPipe · PyTorch · ONNX · Flask
Concept diagram connecting webcam capture, face matching, blink verification, and an access decision
CONCEPTUAL SYSTEM DIAGRAM · Smart Face Door Lock

Problem

Recognizing an enrolled face and deciding whether a live person is present are different tasks. An access-control prototype needs to consider both.

Solution

The documented pipeline uses MediaPipe facial landmarks, ArcFace embeddings for identity comparison, and blink-based multi-frame verification. A Flask interface brings enrollment, camera authentication, and mobile monitoring together.

Architecture

  1. 01Webcam + face landmarks
  2. 02ArcFace identity match
  3. 03Blink / multi-frame check
  4. 04Local access decision
Simplified data flow through Smart Face Door Lock.

My contribution

  • Connected MediaPipe face landmarks to alignment and ArcFace embedding comparison.
  • Added multi-frame blink checks and a Flask interface for enrollment and monitoring.

Engineering challenge

Identity matching must be combined with a liveness signal before an access decision. The project explores this separation using standard camera input and local inference.

Results & demo

The prototype combines ArcFace identity matching and a separate blink check in one local workflow.

Project walkthrough (opens in a new tab)

What I’d improve

Measure false accepts and rejects with consented participants under different lighting and angles.

A prototype documented by the author, not a certified security product. No security guarantee, anti-spoofing benchmark, timing claim, or independently measured recognition accuracy is asserted. The diagram illustrates the reported pipeline; it is not an application screenshot.

Project references

Next projectLeafLens

Find a page