ADAS for Indian Road Vehicle

Advanced Driver Assistance Systems (ADAS) tailored for Indian road conditions. This project focuses on road segmentation, object detection, and instance segmentation using the YOLOv8-Seg model to enhance road safety and navigation in diverse Indian road scenarios.

---

Table of Contents

1. Project Overview
2. Features
3. Dataset
4. Project Structure
5. Installation
6. Usage
7. Results
8. Contributing
9. License
10. Acknowledgements

---

Project Overview

This project aims to develop an ADAS system specifically designed for Indian road conditions. The system leverages the YOLOv8-Seg model for:

• Road Segmentation: Identifying and segmenting road areas.
• Object Detection: Detecting vehicles, pedestrians, and other obstacles.
• Instance Segmentation: Providing detailed masks for detected objects.

The model is trained on a diverse dataset of Indian roads, including various weather conditions, lighting scenarios, and road types.

---

Features

• YOLOv8-Seg Model: State-of-the-art instance segmentation for accurate road and object detection.
• Diverse Dataset: Includes 45,000+ annotated images of Indian roads under various conditions.
• Superpixel Segmentation: Enhances segmentation accuracy using SLIC superpixel methods.
• Real-Time Performance: Optimized for mid-range resolutions (e.g., 854×480) to balance accuracy and computational efficiency.
• Data Augmentation: Techniques like brightness adjustment, noise addition, and blur effects to improve model robustness.

---

Dataset

The dataset consists of 40 videos recorded across different Indian road scenarios, converted into 45,000+ annotated images. Key features of the dataset:

• Road Types: Highways, urban roads, rural roads, and more.
• Weather Conditions: Sunny, cloudy, rainy, and nighttime.
• Annotations: Generated using the Segment Anything Model (SAM2) for efficient mask creation.

Dataset statistics:

• Training: 80%
• Validation: 10%
• Testing: 10%

---

Project Structure

This project is organized into several Python packages and directories:

• data_processing/: Scripts and modules for dataset preparation, augmentation, and format conversion, especially for YOLO. Includes YoloDatasetProcessor for creating YOLO datasets.
• training/: Contains scripts for model training, primarily train.py for YOLOv8-Seg.
• utils/: A package for common utility scripts and modules.
  • utils/dataset_helpers/: Utilities specifically for dataset manipulation (e.g., CheckMaskFile.py).
  • utils/evaluation/: Scripts and modules for model evaluation, IoU calculation, performance analysis, and system monitoring. Contains sub-packages like analysis_with_resolution, analysis_with_super_pixel, etc.
• Model/: Contains model outputs, logs, and visualizations from training runs (e.g., Model/RoadSeg/).
• ROC/: Scripts related to ROC curve generation.
• scr/: Miscellaneous source scripts.
• weights/: Intended for storing downloaded pretrained model weights (e.g., yolov8l-seg.pt).
• dataset/: Intended for storing datasets in the structure required by training scripts.

Many of these directories (like utils, data_processing, training, utils/evaluation and their subdirectories) are structured as Python packages, allowing for modular imports.

---

Installation

To set up the project locally, follow these steps:

1. Clone the Repository:

   git clone https://github.com/thippeswammy/ADAS-for-Indian-Road-Vehicle.git
   cd ADAS-for-Indian-Road-Vehicle

2. Install Dependencies:

   pip install -r requirements.txt

3. Download Pretrained Weights: You can download the YOLOv8-Seg pretrained model from the link below: 📥 Download Model Weights

   After downloading, place the file in the weights/ directory:

   weights/
   ├── yolov8l-seg.pt  # Downloaded weights
   

4. Prepare Dataset: Ensure the dataset is organized in the following structure:

   dataset/
   ├── images/
   │   ├── train/
   │   ├── val/
   │   └── test/
   └── labels/
       ├── train/
       ├── val/
       └── test/
   

---

Usage

Running the Model

To train the YOLOv8-Seg model:

python -m training/train.py 

Inference on Test Images

To run inference on test images:

python python RoadSegment/RoadSegment.py

Superpixel Segmentation

To apply superpixel segmentation:

python scr/different_type_superpixel_segmentation.py --method majority_pixel --n_segments 500

---

Results

Performance Metrics

• Mean IoU: 0.8983 (at 854×480 resolution)
• Precision: 0.9692
• Recall: 0.9501
• Accuracy: 95.21%

Visual Results

Sample Output Video

Output.mp4

📹 Watch the full demo video: Click here to view

Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository.
2. Create a new branch (git checkout -b feature/YourFeature).
3. Commit your changes (git commit -m 'Add some feature').
4. Push to the branch (git push origin feature/YourFeature).
5. Open a pull request.

---

License

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

---

Contact

For questions or feedback, please contact:

• Thippeswamy - thippeswamy636408@gmail.com
• GitHub: thippeswammy