SelfDrivePyAuto

A Python-based autonomous driving system that uses computer vision and deep learning to control vehicle steering in simulation environments. The project implements a U-Net architecture with batch normalization for road detection and converts the detected path into steering commands.

Features

• Real-time screen capture and processing for road detection
• Deep learning-based road segmentation using U-Net architecture
• Image annotation support using LabelMe for creating training data
• Automatic steering control using PyAutoGUI
• Training pipeline for custom road detection models
• Video to image conversion utilities for training data preparation
• Support for both CPU and GPU inference

Requirements

• Python 3.6+
• PyTorch
• OpenCV
• NumPy
• PyAutoGUI
• Jupyter Notebook
• labelme

Installation

# Clone the repository
git clone https://github.com/Indirajith-jithu/SelfDrivePyAuto.git
cd SelfDrivePyAuto

# Install required packages
pip install torch torchvision
pip install opencv-python
pip install numpy
pip install pyautogui
pip install jupyter
pip install labelme

Project Structure

• video_to_images.ipynb: Utility for converting training videos to image frames
• train_model.ipynb: Training pipeline for the road detection model
• predict_steering.ipynb: Real-time prediction and steering control implementation

Usage

1. Data Preparation

   • Use video_to_images.ipynb to convert your training videos into image frames
   • Organize your training data in appropriate directories

2. Data Annotation

   • Install and launch LabelMe: labelme
   • Open your image directory in LabelMe
   • Create polygons around the road areas
   • Save annotations in JSON format
   • Annotations will be used to create binary masks for training
   • Recommended annotation guidelines:
     • Label the drivable road area
     • Be consistent with the annotation boundaries
     • Include various road conditions and lighting
     • Annotate at least 50 images for basic training

3. Model Training

   • Open train_model.ipynb in Jupyter Notebook
   • Adjust hyperparameters if needed
   • Run all cells to train the model
   • The best model will be saved as 'best_model.pth'

4. Autonomous Driving

   • Open predict_steering.ipynb in Jupyter Notebook
   • Ensure your game/simulation window is properly positioned
   • Run the notebook to start autonomous driving
   • Press Ctrl+C in the notebook to stop the program

Model Architecture

The project uses a U-Net architecture with the following features:

• Batch normalization for improved training stability
• Three encoder blocks and two decoder blocks
• Skip connections for better feature preservation
• Binary segmentation output for road detection

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• PyTorch team for the deep learning framework
• OpenCV community for computer vision tools
• PyAutoGUI developers for system control capabilities