🧠 Face Recognition Attendance System This project is a full-stack face recognition-based attendance system built using:

InsightFace for face detection and embedding.

Pinecone vector database for face similarity search.

Flask for the web interface.

PostgreSQL for attendance record storage.

Integration with Roboflow for dataset management.

📸 Features Extract frames from videos to build datasets.

Train on face embeddings using InsightFace.

Upload embeddings to Pinecone for similarity matching.

Real-time photo capture and recognition in Google Colab.

Flask-based web portal for:

Face recognition via photo.

Attendance logging (enter/exit).

View recent attendance history.

Robust handling of repeated entries and exits.

🚀 Quick Start

1. 📦 Install Dependencies bash Copy Edit pip install roboflow insightface onnxruntime faiss-cpu opencv-python-headless matplotlib scikit-learn tqdm flask sqlalchemy python-dotenv If using Pinecone:

bash Copy Edit pip install -U pinecone-client 2. 📥 Download Training Dataset from Roboflow python Copy Edit from roboflow import Roboflow

rf = Roboflow(api_key="YOUR_API_KEY") project = rf.workspace("YOUR_WORKSPACE").project("face-tlml4") version = project.version(1) dataset = version.download("folder") 3. 🧠 Extract Face Embeddings python Copy Edit from insightface.app import FaceAnalysis

app = FaceAnalysis(name="buffalo_l", providers=['CPUExecutionProvider']) # or CUDAExecutionProvider for GPU app.prepare(ctx_id=0)

Loop through training data to extract embeddings

4. 🧠 Upload Embeddings to Pinecone python Copy Edit from pinecone import Pinecone, ServerlessSpec

pc = Pinecone(api_key="YOUR_PINECONE_API_KEY") index_name = "face-embeddings"

Create index if not exists

pc.create_index( name=index_name, dimension=512, metric="cosine", spec=ServerlessSpec(cloud="gcp", region="us-west1") )

index = pc.Index(index_name) index.upsert([...]) # Upload embeddings in batch 5. 📷 Real-Time Face Recognition in Colab Use webcam-based photo capture and run inference:

python Copy Edit from IPython.display import display, Javascript from google.colab import output Detect face → Get embedding → Query Pinecone → Return label and score.

6. 🌐 Flask Web App for Attendance ⚙️ Setup Set up environment variables in .env:

env Copy Edit SECRET_KEY=your-secret-key DATABASE_URL=postgresql://user:password@localhost/face_attendance PINECONE_API_KEY=your-pinecone-key PINECONE_INDEX_NAME=face-embeddings Initialize the database:

bash Copy Edit python initialize_db.py ▶️ Run the App bash Copy Edit python run.py 📁 Project Structure arduino Copy Edit . ├── app/ │ ├── models.py # Attendance DB model │ ├── routes.py # Flask route handlers │ ├── utils/ │ │ └── face_recognition.py # Pinecone-based recognition │ └── static/ │ └── captured.jpg # Saved recognized image ├── templates/ │ ├── home.html │ ├── recognized.html │ └── not_recognized.html ├── config.py ├── run.py ├── initialize_db.py ├── requirements.txt └── README.md 🗃️ Database Schema attendance Table: Field Type Description id Integer Primary key person_name String Detected person’s name class_id String Group/class identifier action String "enter" or "exit" timestamp DateTime Timestamp of action entry_time DateTime Entry time (nullable) exit_time DateTime Exit time (nullable) duty_duration Interval Time duration between entry & exit

🔐 Security Notes Use environment variables (.env) for all secrets and keys.

Limit the number of face queries per second if using public APIs.

Enable HTTPS for production deployments.

📚 Future Improvements Add user registration and admin panel.

Add support for mobile camera capture.

Improve recognition accuracy via face alignment.

👨‍💻 Author Arif M.

📄 License This project is licensed under the MIT License.