What you build
A system that detects and recognises faces from a classroom camera and marks attendance automatically. You measure how accurate it is and where it fails.
Modules
- Student face enrolment
- Face detection in camera frames
- Face recognition using face embeddings
- Attendance records and reports
- A confidence threshold to reduce wrong matches
What you will understand
- The difference between face detection and face recognition
- What a face embedding is and how matching works
- False accepts versus false rejects, and how the threshold trades one for the other
How it is tested
- Accuracy on a test set of enrolled and unknown faces
- Tests under poor lighting, side angles and spectacles
- Confusion matrix and error analysis
What goes in the report
- Literature survey of face recognition methods
- System architecture and model choice
- Results, limitations and privacy considerations
Viva questions to prepare
- Why an embedding instead of comparing pixels?
- How did you choose the threshold?
- What are the privacy risks of this system?