Real-Time Object Detection & Face Recognition

A real-time object detection and face recognition application that uses your webcam to detect objects and analyze faces. Available in both web (TensorFlow.js) and Python (YOLO/DeepFace) versions.

Features

Object Detection

• Real-time object detection from webcam feed
• Bounding boxes with labels and confidence scores
• Support for 80 object classes (COCO dataset)
• FPS monitoring and performance stats
• Modern, responsive UI (web version)
• Command-line interface (Python version)

Face Recognition & Analysis

• Real-time face detection from webcam
• Age estimation
• Gender classification
• Emotion detection (happy, sad, angry, surprise, fear, neutral, disgust)
• Race/ethnicity detection
• Bounding boxes with all detected attributes
• Optimized for 50-60 FPS with background threading
• Toggle between object detection and face detection modes (web version)
• Configurable FPS target and analysis intervals

Project Structure

ObjectDetector/
├── web/          # Web version (TensorFlow.js + COCO-SSD)
├── python/       # Python version (YOLO/COCO-SSD + OpenCV)
└── README.md     # This file

Quick Start

Web Version

1. Navigate to the web directory:

   cd web

2. Install dependencies:

   npm install

3. Start the development server:

   npm run dev

4. Open your browser and navigate to the URL shown (usually http://localhost:5173)

5. Allow camera permissions when prompted - the webcam will start automatically

Python Version

1. Navigate to the python directory:

   cd python

2. Create and activate a virtual environment (recommended):

   python3 -m venv venv
   source venv/bin/activate  # On macOS/Linux
   # or
   venv\Scripts\activate  # On Windows

3. Install dependencies:

   pip3 install -r requirements.txt

   This installs OpenCV, NumPy, and YOLO (Ultralytics). YOLO will automatically download its model on first use.

4. Run the detector:

   # For object detection (uses default webcam automatically):
   python3 detector.py
   
   # For face detection and analysis (uses default webcam automatically):
   python3 face_detector_app.py

   Note: Both applications automatically use your default webcam (camera 0). The webcam will start immediately when you run the application.

   Object Detection:

   • The detector will automatically:
     • Try YOLO first (recommended - downloads model automatically)
     • Fall back to COCO-SSD if YOLO is not available
     • Use a simple detector if no ML models are available

   Face Detection:

   • Uses OpenCV DNN face detector (with Haar Cascade fallback)
   • Analyzes faces using DeepFace (downloads models automatically on first run)
   • Optimized for 50-60 FPS with background threading for analysis
   • Analysis runs every 20 frames by default (adjustable with --analyze-interval)
   • Use --no-analysis flag for maximum FPS (face detection only)
   • Configurable target FPS with --target-fps option

5. Press q to quit, s to save a screenshot

Requirements

Web Version

• Modern web browser with camera support
• Node.js 16+ and npm
• TensorFlow.js models (load automatically)

Python Version

• Python 3.8+
• Webcam
• OpenCV, NumPy, and YOLO (Ultralytics) - see python/requirements.txt
• DeepFace and TensorFlow (for face recognition features)
• Optional: COCO-SSD model files (for fallback)

Model Information

Object Detection Models

Python Version:

• YOLO (Recommended): Uses YOLOv8n - automatically downloads on first use, detects 80 COCO classes
• COCO-SSD (Fallback): MobileNet-SSD model - requires manual download if YOLO unavailable

Web Version:

• COCO-SSD: TensorFlow.js model - automatically loads from CDN

Both models can detect 80 different object classes including:

• People, animals (cat, dog, bird, etc.)
• Vehicles (car, bus, bicycle, motorcycle, etc.)
• Furniture (chair, couch, bed, etc.)
• Electronics (laptop, mouse, keyboard, etc.)
• Food items (banana, apple, pizza, etc.)
• And many more...

Face Recognition Models

Python Version:

• Face Detection: OpenCV DNN face detector (with Haar Cascade fallback)
• Face Analysis: DeepFace library with pre-trained models for:
  • Age estimation (ApparentAge model)
  • Gender classification (Gender model)
  • Emotion detection (Ferc2013 model)
  • Race detection (Race model)
• Models download automatically on first use

Web Version:

• Face Detection: BlazeFace model via TensorFlow.js
• Face Analysis: Simplified analysis (can be extended with additional models)

Troubleshooting

Camera Not Working

For Python version on macOS:

1. Run the diagnostic script first:

   cd python
   python3 check_camera.py

2. Grant camera permissions:
   • System Preferences > Security & Privacy > Camera
   • Enable Terminal/Python in the list
3. If Terminal is not listed, try:
   • Running the detector once, then check again
   • Or use: tccutil reset Camera then restart Terminal

General troubleshooting:

• Ensure your browser/application has camera permissions
• Check that no other application is using the camera
• Try refreshing the page or restarting the Python script
• Close other apps using the camera (Zoom, FaceTime, etc.)

Model Loading Issues

Python Version:

• YOLO: Will download automatically on first run (requires internet)
• COCO-SSD: If YOLO is not available, you can download COCO-SSD manually:

  python3 models/download_models.py

• If you see "Object" labels instead of actual names, install YOLO:

  pip install ultralytics

Web Version:

• Check browser console for errors
• Ensure internet connection (model loads from CDN)

Performance Issues

For Face Detection:

• Use --no-analysis flag for maximum FPS (face detection only, no analysis)
• Increase --analyze-interval (e.g., --analyze-interval 30) for better FPS
• Lower target FPS if CPU can't keep up: --target-fps 30
• Lower resolution: --width 320 --height 240

General:

• Lower the video resolution in settings
• Reduce the frame processing rate
• Close other applications using the camera