Roadblock Detection Project

Overview

This project implements a roadblock detection system using the YOLOv8 model to identify roadblocks (e.g., cones, barriers) in images. The model is trained on a custom dataset and can be used for real-time detection in traffic scenarios.

Requirements

• Hardware: CPU or NVIDIA GPU (e.g., RTX 3060 with CUDA 11.8)
• OS: Windows/Linux
• Software:
  • Python 3.8+
  • PyTorch 2.4.1+cu118
  • torchvision 0.19.1+cu118
  • ultralytics 8.3.x
  • opencv-python, numpy

Installation

1. Set up Python environment:

   python -m venv yolo_env
   source yolo_env/bin/activate  # Linux/Mac
   yolo_env\Scripts\activate     # Windows

2. Install dependencies:

   pip install torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 --index-url https://download.pytorch.org/whl/cu118
   pip install ultralytics opencv-python numpy

Dataset

• Location: block_data/
• Structure:

  block_data/
  ├── images/
  │   ├── train/
  │   ├── val/
  │   ├── test/
  ├── labels/
  │   ├── train/
  │   ├── val/
  │   ├── test/
  ├── data.yaml
  

• Format:
  • Images: .jpg
  • Labels: .txt (YOLO format: class_id x_center y_center width height, normalized to 0-1)
  • data.yaml:

    train: ./images/train
    val: ./images/val
    test: ./images/test
    nc: 1
    names: ['block']

Training

1. Train the model (in Jupyter Notebook or Python script):

   from ultralytics import YOLO
   
   model = YOLO("yolov8n.pt")
   model.train(
       data="block_data/data.yaml",
       epochs=50,
       imgsz=416,
       batch=8,
       device=0,  # Use GPU (or 'cpu' if no GPU)
       amp=False,
       name='block_train',
       project='runs/train',
       exist_ok=True
   )

2. Output:
   • Trained weights: runs/train/block_train/weights/best.pt
   • Training logs: runs/train/block_train/results.png

Testing

1. Test on a single image:

   from ultralytics import YOLO
   
   model = YOLO("runs/train/block_train/weights/best.pt")
   results = model("block_data/images/test/test_image.jpg", device=0)
   results.show()
   results.save(save_dir='runs/detect/test')

2. Evaluate on validation set:

   model.val(data="block_data/data.yaml", device=0)

Deployment

• Local:
  • Run training/testing in Jupyter Notebook or Python scripts.
• Remote (Company Server):
  • Upload block_data and scripts to ~/yolo_project via FileZilla (IP: 10.31.15.53, user: zhusongyu, password: zhusongyu123, port: 22).
  • Run on server:

    ssh zhusongyu@10.31.15.53 -p 22
    cd ~/yolo_project
    source yolo_env/bin/activate
    python train_yolov8.py

Notes

• GPU Issues: If NotImplementedError occurs, ensure torchvision is CUDA-enabled (0.19.1+cu118). Use device='cpu' as fallback.
• Dataset: Ensure each image has a corresponding label file. Split dataset into train/val/test (80:10:10) using split_dataset.py.
• Performance: Adjust epochs, imgsz, or batch based on dataset size and hardware.