🤟 ASL Detection - Sign Language Recognition System

A real-time American Sign Language (ASL) detection system using YOLOv8 for recognizing hand gestures and building sentences from sign language.

📋 Features

• Real-time ASL Detection: Recognizes 26 letters (A-Z) + backspace gesture
• Sentence Builder: Automatically builds sentences from detected gestures
• Custom Dataset Support: Merge multiple datasets for improved accuracy
• YOLOv8 Integration: State-of-the-art object detection model
• GPU Acceleration: CUDA support for faster inference and training

🚀 Quick Start

Prerequisites

• Python 3.8 or higher
• Webcam for real-time detection
• (Optional) NVIDIA GPU with CUDA for faster training

Installation

1. Clone the repository

   git clone https://github.com/PrajanManojKumarRekha/Object-Detection.git
   cd Object-Detection

2. Create a virtual environment (recommended)

   python -m venv .venv
   
   # On Windows
   .venv\Scripts\activate
   
   # On macOS/Linux
   source .venv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

📚 Usage Guide

Step 1: Prepare Your Dataset (Optional)

If you have multiple ASL datasets to merge:

python merge_datasets.py

This will:

• Combine datasets from different sources
• Create a unified ASL_Merged directory
• Generate a data.yaml configuration file

Step 2: Train the Model

Train the YOLOv8 model on your ASL dataset:

python train_model.py

Training Configuration:

• Model: YOLOv8 Nano (yolov8n.pt)
• Epochs: 100
• Batch Size: 16
• Image Size: 640x640
• Early Stopping: 10 epochs patience

What happens during training:

1. Pre-training checks (dataset validation, GPU detection)
2. Model training with automatic checkpointing
3. Validation on test set
4. Best model saved to models/best_asl_27.pt

Expected Output:

✓ GPU Available: [Your GPU Name]
✓ Merged dataset found
✓ Model: yolov8n.pt
✓ Epochs: 100
✓ Batch Size: 16

📊 Validation Results:
  mAP@0.5:    0.XXX
  mAP@0.5-95: 0.XXX
  Precision:  0.XXX
  Recall:     0.XXX

Step 3: Run Real-Time Detection

Start the ASL detection system:

python Runner.py

Controls:

• Q: Quit the application
• C: Clear the sentence builder
• S: Show statistics (if implemented)

How it works:

1. Opens your webcam
2. Detects ASL gestures in real-time
3. Displays bounding boxes with confidence scores
4. Builds sentences from detected letters
5. Shows FPS and current sentence at the bottom

📁 Project Structure

Object-Detection/
├── ASL_Merged/              # Merged dataset directory
│   ├── train/              # Training images
│   ├── valid/              # Validation images
│   ├── test/               # Test images
│   └── data.yaml           # Dataset configuration
├── models/                  # Trained model weights
│   └── best_asl_27.pt      # Best trained model
├── runs/                    # Training runs and results
├── OpenCV collector/        # Data collection scripts
├── merge_datasets.py        # Dataset merging utility
├── train_model.py          # Model training script
├── Runner.py               # Real-time inference script
├── requirements.txt        # Python dependencies
└── README.md              # This file

🎯 Recognized Gestures

The system recognizes 27 different gestures:

• A-Z: All 26 letters of the alphabet
• Backspace: Delete the last character

⚙️ Configuration

Modify Training Parameters

Edit train_model.py:

EPOCHS = 100              # Number of training epochs
BATCH_SIZE = 16          # Batch size (reduce if out of memory)
IMAGE_SIZE = 640         # Input image size
PATIENCE = 10            # Early stopping patience

Modify Inference Settings

Edit Runner.py:

CONFIDENCE_THRESHOLD = 0.6    # Minimum confidence for detection
DEBOUNCE_TIME = 1.5          # Time between same gesture detection (seconds)

🔧 Troubleshooting

Issue: "Merged dataset not found"

Solution: Run python merge_datasets.py first to create the merged dataset.

Issue: "Cannot access webcam"

Solution:

• Check if another application is using the webcam
• Verify webcam permissions in your OS settings
• Try changing camera index in Runner.py: cv2.VideoCapture(1) instead of 0

Issue: "CUDA out of memory"

Solution: Reduce batch size in train_model.py:

BATCH_SIZE = 8  # or even 4

Issue: Low detection accuracy

Solution:

• Train for more epochs
• Add more training data
• Adjust confidence threshold in Runner.py
• Ensure good lighting conditions during inference

📊 Performance Tips

1. GPU Training: Use CUDA-enabled GPU for 10-20x faster training
2. Lighting: Ensure good, consistent lighting for better detection
3. Background: Use a plain background for improved accuracy
4. Distance: Keep hand at consistent distance from camera
5. Gesture Hold: Hold each gesture steady for 1-2 seconds

🤝 Contributing

Contributions are welcome! Feel free to:

• Report bugs
• Suggest new features
• Submit pull requests
• Improve documentation

📝 License

This project is open source and available under the MIT License.

🙏 Acknowledgments

• Ultralytics YOLOv8: For the excellent object detection framework
• OpenCV: For computer vision capabilities
• ASL Community: For gesture datasets and resources

📧 Contact

For questions or support, please open an issue on GitHub.

---

Made with ❤️ for the ASL community