CSE 499A/B — Senior Design Project · Department of Electrical and Computer Engineering, North South University · Fall 2022
Co-authored with Zannatul Islam Proma.
The Problem: Manual Attendance Doesn't Scale
Roll-call attendance in university classrooms is slow, easy to game through proxy sign-ins, and generates no data beyond a simple present/absent mark. THIRD EYE was our senior design response: a system that uses facial recognition to automate attendance-taking, with an eye toward eventually extending that same camera feed into behavior and engagement monitoring.
System Architecture
THIRD EYE was built as a Django web application backed by SQLite, with OpenCV handling the computer-vision pipeline for face detection and recognition. The system was designed around four core workflows:
Development was done in PyCharm with Python and Git for version control, keeping the stack deliberately lightweight so it could run on ordinary classroom hardware rather than requiring specialized equipment.
Recognition Accuracy in Testing
During in-class testing, the face-recognition module was run against a small group of enrolled students, each captured under normal classroom lighting and camera angle:
The spread of results was itself informative: recognition confidence for present students varied with lighting and camera angle even within the same short test session, while an absent student correctly registered no match at all — confirming the core detection-and-matching logic worked as intended before we pushed further into edge cases.
What Actually Shipped — and What Didn't
Being accurate about scope: the facial-recognition attendance pipeline — registration, enrollment, live detection, and report generation — was fully implemented and tested. The behavior-monitoring half of the project's ambition, however, was not complete at the time of submission. The final report is explicit about this: behavior analysis was scoped as future work rather than a delivered feature, and we said so directly rather than overstating what the system could do.
That honesty about an incomplete feature was, in hindsight, as important a lesson as the technical build itself: a working, well-scoped attendance system beats an over-promised behavior-analytics platform that doesn't actually run.
Conclusion & Future Work
THIRD EYE demonstrated that a lightweight Django + OpenCV stack is sufficient to automate classroom attendance with real-world accuracy, running on standard hardware without specialized sensors. The natural next step — and the one we flagged explicitly in the final report — is extending the same camera pipeline into engagement and behavior signals: eye-gaze estimation, posture tracking, or attention scoring layered on top of the existing face-detection foundation.
Interested in Computer Vision for Education?
I'd enjoy comparing notes on classroom-facing computer vision, facial recognition pipelines, or Django-based systems in general. Reach out.
Contact Me