SaferoadAI

SaferoadAI is a real-time car accident detection system that utilizes a YOLO model trained on a custom dataset. The system is designed to work with a video feed from CCTV or live stream, detecting accidents in real-time and sending messages to the nearest hospital using the OLA Maps API.

Features

• Real-time accident detection using YOLOv8.
• Custom-trained model on a car accident dataset.
• Integration with CCTV or live video feeds for practical applications.
• Automatic alert system using OLA Maps API to notify the nearest hospital.
• Frame capture and image conversion to URL for message attachments.
• Potential for further enhancements, such as traffic control automation.

Installation

Prerequisites

Ensure you have the following installed before proceeding:

• Python 3.8+
• PyTorch
• Ultralytics YOLOv8
• OpenCV
• Streamlit (for deployment, if needed)
• OLA Maps API access

Clone the Repository

git clone https://github.com/aayush010904/SaferoadAI.git
cd SaferoadAI

Install Dependencies

pip install -r requirements.txt

Datset

Dataset used for training : Roboflow datset URL

Usage

Running the Application

To start the accident detection system, run:

python app.py

How It Works

• app.py imports functions from SendMessage.py and NearestHospital.py to send messages and fetch the nearest hospital.
• When an accident is detected, the frame is saved.
• The saved frame is converted into a URL using Image2Url.py.
• The image URL is sent along with an alert message to the nearest hospital.

Training the YOLOv8 Model (If Needed)

python model_training.ipynb

Flask Live Stream Demo (uses app.py logic)

The Flask app reuses the same YOLO model and detection logic defined in app.py for live MJPEG streaming.

Run locally:

pip install -r requirements.txt
python flask_app.py

Then open http://localhost:5000, upload or pick a sample video, click Process Detection to start the live processed stream. Accident counts appear after the stream completes. The “See live alerts” button is a placeholder—point it to your real alerts/messages URL.

Deployment

Quick local server (Waitress)

pip install waitress
waitress-serve --port=5000 flask_app:app

Container (recommended for hosting)

Example Dockerfile:

FROM python:3.11-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY . .
ENV PORT=5000
CMD ["gunicorn", "-b", "0.0.0.0:5000", "flask_app:app"]

Build and run:

docker build -t saferoad-ai .
docker run -p 5000:5000 saferoad-ai

Host this container on services like Render, Railway, Fly.io, AWS ECS/Fargate, or Azure Web App for Containers. Ensure ffmpeg/H.264 support is available in your base image if you need MP4 output.

Project Structure

SaferoadAI/
├── app.py                # Main application script
├── requirements.txt      # Python dependencies
├── README.md             # Project documentation
├── .env                  # Environment variables
├── best.pt               # Pre-trained YOLO model
├── best_model.pt         # Additional trained model
├── model_training.ipynb  # YOLO model training notebook
├── Image2Url.py          # Converts detected accident frames to image URLs
├── NearestHospital.py    # Fetches nearest hospital using OLA Maps API
├── SendMessage.py        # Sends alert messages with accident details
├── currentLocation.py    # Determines the user's current location
└── other_files/          # Additional scripts or resources

Future Enhancements

• Improving model accuracy with more training data.
• Expanding API support for other mapping services.
• Implementing real-time traffic management integration.

Contributions

Feel free to open an issue or submit a pull request if you’d like to contribute!

License

This project is licensed under the MIT License.