🚗 Vehicle Detection using YOLOv8

A machine learning project to detect and classify vehicles in video streams using YOLOv8, including dataset preprocessing, model training, ONNX/OpenVINO exports, and real-time inference.

🧠 Key Features

✅ Full YOLOv8 training and inference pipeline
✅ Dataset preparation, augmentation, and class correction
✅ Real-time object detection in videos with annotation
✅ Frame-level confidence scoring and CSV logging
✅ Exports to ONNX and OpenVINO for efficient deployment

📁 Project Structure

.
├── Code/
│   ├── Augmentation.py              # Image augmentation functions
│   ├── Download_youtube_video.py    # Download video data
│   ├── Inference_for_class.py       # Run inference by class
│   ├── Splitting.py                 # Split data into train/val
│   ├── Training.py                  # YOLOv8 model training script
│   ├── changing_class_nos.py        # Remap class numbers
│   ├── combining_Files.py           # Merge files for dataset
│   └── run_the_model.py             # Main runner for YOLOv8 pipeline
├── CNN Object Detection.ipynb       # Colab-compatible training + inference notebook
├── data.yaml                        # YOLOv8 data config (paths + class info)
├── onnx_export.py                   # Export model to ONNX
├── openvino_export.py               # Export model to OpenVINO format
├── run1.py                          # Run inference and save annotated video
└── README.md                        # Project documentation

🗂 Dataset Structure

dataset/
├── images/
│   ├── train/
│   └── val/
├── labels/
│   ├── train/
│   └── val/
└── data.yaml

Example data.yaml:

train: C:/path/to/train/images
val: C:/path/to/val/images
nc: 4
names: ['person', 'motorcycle', 'car', 'truck']

🚀 How to Run

1️⃣ Install Requirements

pip install ultralytics opencv-python tqdm pillow numpy

Or install via requirements.txt:

pip install -r requirements.txt

2️⃣ Train the Model

from ultralytics import YOLO

model = YOLO('yolov8n.pt')
model.train(data='path/to/data.yaml', epochs=50, imgsz=640)

3️⃣ Validate the Model

model.val(data='path/to/data.yaml', model='path/to/best.pt')

4️⃣ Export the Model

# Export to ONNX
model.export(format="onnx")

# Export to OpenVINO
model.export(format="openvino")

5️⃣ Run Inference on Video

python run1.py

• Annotated video will be saved (e.g., output_video.mp4)
• Frame-wise accuracy (confidence) will be logged into results.csv

🎯 Model Evaluation

• mAP50-95: Evaluated using YOLOv8 .val() method
• Precision / Recall: Tracked during training
• Accuracy Logging: CSV output via OpenVINO inference

🧪 Sample Outputs

⚠️ Note: Outputs are not included in this repo.

Running the pipeline will generate:

• Annotated video (e.g., output_video.mp4)
• Frame-wise accuracy report (results.csv)

📊 Results Example

Model Format	Speed (FPS)	Accuracy (avg)
YOLOv8n	35 FPS	~82%
OpenVINO	50 FPS	~80–85%

Note: Results may vary depending on system specs and dataset.

📦 Dependencies

• ultralytics
• opencv-python
• tqdm
• pillow
• numpy

Install all with:

pip install -r requirements.txt

📌 Notes

• Works in both local and Google Colab environments
• Ensure correct paths in your data.yaml file
• You can export and deploy the best model in multiple formats

🤝 Contributing

Feel free to fork this project and submit pull requests for improvements!

📄 License

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

🙏 Acknowledgments

• Ultralytics YOLOv8 for the amazing YOLO implementation
• OpenVINO toolkit for model optimization
• The open-source community for various tools and libraries used