Face Recognition Attendance System
Team-built attendance system using face recognition, with stranger rejection, reporting, and a dashboard.
- Computer Vision
- Applied AI
- MLOps / Deployment
TL;DR
An end-to-end applied-AI system that recognizes enrolled people, rejects unknown faces, logs attendance to a database, and presents results in a dashboard — built with a small team using Agile practices. (Placeholder narrative.)
Problem
Manual attendance is slow and error-prone. The system needed to be accurate, reject strangers to avoid false check-ins, and produce trustworthy reports.
Note: Placeholder case-study content — replace with your real write-up. Avoid publishing any real face images or personal data.
Why it matters
This project shows the full arc of an applied-AI product: capturing input, making a confident decision, storing results reliably, and communicating them to non-technical stakeholders.
Dataset / inputs
Enrolled reference images plus live camera frames. (Placeholder — do not commit real personal images or identifiable data. Use synthetic or consented sample data for any public demo.)
Technical decisions
The key decision was the recognition threshold. Setting it too loose lets strangers check in; too tight and legitimate users are rejected. We treated this as a tunable product trade-off and documented the chosen operating point.
Challenges
Lighting variation and camera angles drove most recognition errors. Coordinating a team around clear component boundaries (detection, decision, storage, reporting) kept integration manageable.
Methods
- Face detection + embedding-based recognition
- Distance-threshold stranger rejection to control false accepts
- Attendance logging to a SQL database
- Reporting/dashboard layer for stakeholders
Results
- Working end-to-end demo with reporting (placeholder)
- Tuned threshold balancing recall vs false accepts (placeholder)
Lessons learned
- False-positive control is a product decision, not just a metric
- Clear interfaces between components made team integration smoother
Limitations
- Sensitive to lighting and camera angle
- Not evaluated for demographic bias — would require careful auditing before real use
Next steps
- Add liveness/anti-spoofing checks
- Bias and robustness auditing before any real deployment
Related projects
Gap Junction Segmentation in Electron Microscopy
Deep-learning segmentation pipeline for detecting thin gap-junction structures in electron-microscopy volumes.
Outcome: Example result to update — improved recall on thin structures vs. a baseline U-Net.
- Biomedical AI
- Computer Vision
- Research
YOLO Object Detection & Tracking
Real-time object detection and multi-object tracking experiments comparing YOLO with several tracking algorithms.
Outcome: Example result to update — identified the best speed/accuracy trade-off for the target hardware.
- Computer Vision
- Robotics