Facial Recognition Attendance System

The problem

Operational challenge

Roll-call wastes teaching time and proxy attendance is trivial. But a camera in a real classroom deals with bad light, partial occlusion and students who simply sit at the back — a recognition-only system fails exactly when it matters.

The approach

Implementation strategy

I trained an MTCNN + CNN pipeline to detect and recognise several faces per frame, and deliberately paired it with a manual entry path and per-role dashboards. The AI handles the common case; the fallback means an instructor is never blocked by a bad frame.

Engineering trade-offs

I accepted that full automation was the wrong goal. Designing the manual path as a first-class feature rather than an error state is what made the system usable day to day — and is why it was judged Best Industrial Project of the Year.

What I built

Delivery highlights

• Trained and evaluated a multi-face detection and recognition pipeline on a self-collected dataset.
• Built four role-specific portals covering enrolment, attendance, leave requests and reporting.
• Designed a hybrid AI/manual model so degraded camera conditions never block attendance capture.
• Awarded Best Industrial Project of the Year.

How it fits together

System architecture