FaceAttend — Biometric Attendance System

A desktop application that uses face recognition to automate attendance tracking. Built with Python, it uses deep learning models (SCRFD + ArcFace) running on ONNX Runtime to detect and identify faces in real-time through a webcam feed.

How It Works

1. Register — Admin captures an employee's face through the webcam. The system extracts a 512-dimensional face embedding using ArcFace and stores it in a local SQLite database alongside user details.

2. Scan — When someone stands in front of the camera, the system detects their face using SCRFD, generates their embedding, and compares it against all registered users using cosine similarity. If a match is found above the threshold (default 40%), their attendance is logged with a timestamp.

3. Dashboard — View attendance records by date, see who checked in and when, and export reports to CSV.

Tech Stack

• Face Detection: SCRFD (via InsightFace's buffalo_l model pack)
• Face Recognition: ArcFace — generates 512-D face embeddings
• Inference: ONNX Runtime (CPU, with optional GPU support)
• GUI: CustomTkinter (modern dark-themed desktop UI)
• Camera: OpenCV with threaded capture
• Database: SQLite
• Language: Python 3.10+

Setup

Prerequisites

• Python 3.10 or higher
• A working webcam
• ~500MB disk space (for model files downloaded on first run)

Installation

# clone the repo
git clone https://github.com/Avneesh11905/attendance_system-.git
cd attendance_system-

# create virtual environment
python -m venv venv

# activate it
# Windows:
venv\Scripts\activate
# Linux/Mac:
source venv/bin/activate

# install dependencies
pip install -r requirements.txt

Running

python -m app.main

On the first run, the InsightFace buffalo_l model pack (~300MB) will be downloaded automatically. This only happens once.

Usage

Registering a User

1. Click Register in the sidebar
2. Position the person's face in the camera
3. Click Capture Face — the system will detect and highlight the face
4. Fill in Name, Employee ID, and Department
5. Click Register User

Scanning Attendance

1. Click Scan in the sidebar
2. Auto-scan is on by default — just stand in front of the camera
3. The system will match your face and log attendance automatically
4. A green checkmark means attendance was recorded; orange means already checked in today

Viewing Records

1. Click Dashboard in the sidebar
2. Navigate between dates using the arrow buttons
3. Click Export CSV to save records to a file

Settings

• Similarity Threshold — controls how strict face matching is (lower = more lenient)
• Camera — switch between connected cameras

Project Structure

├── app/
│   ├── main.py              # entry point
│   ├── config.py             # settings and constants
│   ├── core/
│   │   ├── face_engine.py    # SCRFD + ArcFace inference wrapper
│   │   ├── camera.py         # threaded camera capture
│   │   └── database.py       # SQLite operations
│   ├── ui/
│   │   ├── app_window.py     # main window with sidebar
│   │   ├── scan_frame.py     # live scanning screen
│   │   ├── register_frame.py # user registration screen
│   │   ├── dashboard_frame.py# attendance records view
│   │   └── settings_frame.py # app settings
│   └── utils/
│       └── helpers.py        # image conversion, embedding utils
├── data/                     # SQLite database (auto-created)
├── face_models/              # ONNX models (auto-downloaded)
└── requirements.txt

How the Recognition Pipeline Works

Camera Frame
    ↓
SCRFD Face Detection (det_10g.onnx)
    ↓  bounding box + 5 facial landmarks
Face Alignment (affine transform using landmarks)
    ↓  normalized 112×112 face crop
ArcFace Embedding (w600k_r50.onnx)
    ↓  512-dimensional feature vector
Cosine Similarity vs. registered embeddings
    ↓  score > threshold?
Log Attendance to SQLite

Notes

• Face embeddings are stored as binary blobs in SQLite — no images are saved
• Each user can only check in once per day (duplicate prevention)
• Camera runs in a separate thread so the UI stays responsive
• Inference runs in a background thread to avoid frame drops
• The InsightFace buffalo_l models are for non-commercial/research use. See InsightFace license for details.