Smart Security System

A real-time AI security platform for face recognition, multi-pose enrollment, and attendance logging.

The system combines computer vision and vector search to register people, recognize them from a live camera stream, and maintain attendance records. It also includes an optional security chatbot for querying enrolled people and attendance data.

Key Features

• Real-time face detection and recognition from webcam video.
• Multi-pose enrollment workflow (10 guided capture steps) with face-quality checks.
• Pose validation using head-pose estimation (pitch, yaw, roll) and facial cues.
• Attendance logging with deduplication for same-day repeated recognitions.
• Admin management APIs and UI for listing/deleting registered persons.
• Optional chatbot endpoint for natural-language queries over enrollment and attendance data.

Tech Stack

• Backend: FastAPI, Uvicorn
• Vision: OpenCV, MediaPipe, keras-facenet (FaceNet), TensorFlow
• Vector DB: ChromaDB
• Data storage: CSV files (persons.csv, attendance.csv)
• Frontend: HTML/CSS/JavaScript served as static files

Project Structure

Smart_Security_System/
  main.py
  requirements.txt
  backend/
    admin.py
    attendance.py
    camera.py
    chatbot.py
    config.py
    detection.py
    embedding.py
    enrollment.py
    pose_validation.py
    recognition.py
    db/
      create_chroma_db.py
  frontend/
    home.html
    index.html
    chatbot.html
    manage_persons.html
    script.js
    style.css

Setup

1. Clone and enter project

git clone https://github.com/mak4x13/Smart_Security_System.git
cd Smart_Security_System

.git at the end of the URL is optional for GitHub HTTPS cloning; both with and without .git work.

2. Create virtual environment and install dependencies

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt

3. Configure environment variables

Create a .env file in the project root.

GROQ_API_KEY=your_key_here

If you do not use the chatbot, the rest of the system still runs without this key.

Run the Application

uvicorn main:app --host 0.0.0.0 --port 8000

Open:

• Home: http://localhost:8000/
• Dashboard: http://localhost:8000/dashboard
• Chatbot: http://localhost:8000/chatbot

Enrollment and Recognition Flow

1. Switch dashboard mode to Enrollment.
2. Capture guided multi-pose samples (10 stages).
3. Confirm enrollment with person metadata.
4. Switch to Recognition for live identification.
5. Attendance gets logged automatically for recognized identities.

API Endpoints (Core)

• GET /video_feed - MJPEG live stream
• GET /recognition/live - latest recognition + attendance snapshot
• POST /system/mode/{mode} - switch recognition or enrollment
• POST /enroll/start - start enrollment session
• POST /enroll/capture - capture one validated sample
• POST /enroll/confirm - finalize enrollment
• GET /attendance/today - today attendance records
• GET /admin/persons - list registered persons
• DELETE /admin/person/{person_id} - delete person
• POST /chat - chatbot query

Research Context

This project is being used for a research workflow focused on improving recognition robustness through multi-pose enrollment:

• Phase 1: real-participant enrollment and logging experiments.
• Phase 2: BIWI dataset-based pose evaluation and benchmarking.

Notes

• Camera index defaults to 0 in backend/camera.py.
• Runtime data is stored under backend/db/.
• Tune thresholds and detection parameters in backend/config.py and backend/pose_validation.py based on your camera and environment.

License

This project is released under the MIT License.