Project 2: Face Recognition Attendance System

A contactless attendance system using face recognition - widely used in offices and schools.

What You’ll Learn

  1. Face Detection - Locating faces in images using Haar Cascades or DNN
  2. Face Encoding - Converting faces to numerical feature vectors
  3. Face Recognition - Comparing face encodings to identify people
  4. Database Management - Storing and retrieving face data
  5. Attendance Logging - Recording attendance with timestamps

Key Concepts

Concept Description
Face Detection Finding faces in an image
Face Encoding 128-dimensional vector representing a face
Face Matching Comparing encodings using distance metrics
Tolerance Threshold for considering faces as match

Usage 

# Register new faces
python main.py --register

# Start attendance mode
python main.py --attendance

# List registered faces
python main.py --list

# Export attendance to Excel
python main.py --export

# Run demo mode
python main.py --demo

System Flow 

Registration Mode:
1. Capture face from camera
2. Detect face location
3. Extract face encoding (128-d vector)
4. Store encoding with name in database

Attendance Mode:
1. Capture frame from camera
2. Detect all faces
3. Extract encodings
4. Compare with registered encodings
5. Identify matches and mark attendance
6. Log to CSV file

Key OpenCV/ML Functions

Function Library Purpose
face_recognition.face_locations() face_recognition Detect face locations
face_recognition.face_encodings() face_recognition Get face embeddings
face_recognition.face_distance() face_recognition Compare face embeddings
cv2.CascadeClassifier() OpenCV Haar cascade detection
pickle.dump() Python Save database to file

Real-World Applications

  • Office attendance systems
  • School/university attendance
  • Event check-in systems
  • Access control systems
  • Visitor management

Code Highlights

Face Encoding 

# Detect faces
face_locations = face_recognition.face_locations(rgb_image)

# Get 128-dimensional encoding
encoding = face_recognition.face_encodings(rgb_image, face_locations)[0]

Face Matching 

# Compare with known faces
distances = face_recognition.face_distance(known_encodings, test_encoding)

# Lower distance = better match
if distances[best_match_idx] < tolerance:
    name = known_names[best_match_idx]

Database Structure 

face_database/
├── encodings.pkl      # Pickled face encodings + names
├── Alice.jpg          # Reference photo
├── Bob.jpg
└── Charlie.jpg

attendance_logs/
├── attendance_2024-01-15.csv
├── attendance_2024-01-16.csv
└── ...

Output Format

CSV Log 

Name,Timestamp
Alice,2024-01-15 09:00:15
Bob,2024-01-15 09:02:33
Charlie,2024-01-15 09:05:12

Performance Tips

  1. Process every N frames (not every frame)
  2. Resize images before processing
  3. Use model="cnn" for GPU acceleration
  4. Pre-compute encodings during registration

References