YOLOv8 Safety Equipment Detection

This project implements a YOLOv8 model to detect three classes of safety equipment:

• Toolbox
• Oxygen Tank
• Fire Extinguisher

The model is trained to handle challenging conditions including:

1. Varying lighting conditions
2. Occlusions and object overlap
3. Diverse camera angles and distances

Optimized for i5 1335U CPU

The model parameters have been adjusted to run efficiently on an i5 1335U CPU:

• Using YOLOv8s (medium) model instead large variants
• Reduced image size (416px vs. 640px)
• Smaller batch size (8 vs. 16)
• Fewer training epochs (20 vs. 50)
• Reduced data augmentation complexity
• Early stopping with patience=10

Project Structure

.
├── data/   (from the falcon dataset move to data folder here)
│   ├── test/
│   ├── train/
│   └── val/
├── HackByte_Dataset/
│   ├── ENV_SETUP/
│   ├── classes.txt
│   ├── predict.py
│   ├── train.py
│   └── visualize.py
├── predictions/
│   ├── images/
│   └── labels/
├── runs/
│   └── detect/
├── test_case/
├── analyze_model.py
├── predict_custom.py
├── README.md
├── sample.py
├── yolo_params.yaml
├── yolo_safety_detector.bat
└── yolov8m.pt

Required Libraries

• PyTorch
• Torchvision
• CUDA 11.8
• Ultralytics
• OpenCV

   pip install torch torchvision ultralytics opencv-python

Running the Model

1. Use the yolo_safety_detector.bat file to run the model with different settings:

   • Default setting (20 epochs)
   • Quick test (5 epochs)
   • Evaluation only (skip training)
   • Custom number of epochs

2. Alternatively, run the Python script directly:

   python sample.py
   

Expected Training Time

On an i5 1335U CPU, expect:

• Full training (20 epochs): ~4-6 hours
• Quick test (5 epochs): ~1-1.5 hours
• Evaluation only: ~10-15 minutes

Results

After training, the model will generate:

• mAP50 score (primary metric for object detection)
• Visualization of predictions on test images
• Detailed metrics report (precision, recall, mAP)

Results are saved in the results/ directory.

Analysis

For deeper analysis of the model's performance, run:

python analyze_model.py

This will generate:

• Per-class performance metrics
• Confusion matrix
• Training curves
• Side-by-side prediction comparisons