FACIAL ATTENDANCE SYSTEM - LOGIC OVERVIEW
========================================

1. USER REGISTRATION & FACE EMBEDDING
-------------------------------------
- Users register by providing their name, email, and employee code, and by capturing/uploading 3 face images.
- Each image is processed using the insightface model (Python backend) to extract a face embedding (vector representation).
- Embeddings are stored in the database (as JSON) along with the image path and user reference.
- Only images with a valid detected face are saved; others are discarded.

2. REAL-TIME FACE DETECTION (FRONTEND)
---------------------------------------
- The browser loads face-api.js models (TinyFaceDetector) for fast, lightweight face detection.
- The user's webcam stream is analyzed in real time; when a face is detected, a frame is captured.
- The captured frame is converted to a base64-encoded JPEG and sent to the backend via AJAX (fetch POST request).
- No manual button press is needed; detection and sending are automatic.

3. ATTENDANCE/LOGOUT MARKING (BACKEND)
---------------------------------------
- The backend receives the image data and decodes it to a NumPy array (OpenCV image).
- insightface is used to extract the face embedding from the received image.
- The new embedding is compared to all stored user embeddings using cosine similarity.
- If a match (similarity > 0.5) is found, the corresponding user is identified.
- For attendance: checks if the user has already marked 'login' today; if not, creates a new 'login' event.
- For logout: checks if the user has already marked 'logout' today; if not, creates a new 'logout' event.
- The result (success, already marked, or error) is sent back to the frontend for user feedback.

4. MATCHING LOGIC
-----------------
- Cosine similarity is used to compare the new face embedding with each stored embedding.
- A threshold (0.5) is used to determine a match (can be tuned for stricter/looser matching).
- Only the first matching user is used (no multi-user detection per frame).

5. EVENT TYPE HANDLING
----------------------
- Each attendance record has an 'event_type' field: 'login' or 'logout'.
- The system prevents duplicate login/logout events for the same user on the same day.
- Logs show both event types for full traceability.

6. LOGS, ADMIN, AND EXPORT
--------------------------
- Recent attendance/logout events are shown on the main pages.
- Full logs can be filtered by name, employee code, or date.
- Logs can be exported as CSV for reporting.
- Django admin is enabled for user and attendance management.

7. ERROR HANDLING & OPTIMIZATIONS
---------------------------------
- Handles cases where no face is detected (frontend and backend).
- Handles cases where no matching user is found.
- Only sends frames to the backend when a face is detected (reduces server load).
- Uses async/await and event listeners for smooth, non-blocking UI.
- Cleans up user images from disk when a user is deleted.

This logic ensures a seamless, automatic, and robust facial attendance workflow for both users and administrators.