🪖 Helmet Detection System

A robust helmet detection system using YOLOv8 and YOLOv8-nano models that specifically detect helmets in videos.

Features

• ✅ Helmet-specific detection in videos using specialized models
• ✅ Uses YOLOv8 models (standard and helmet-trained)
• ✅ Pre-trained models (no training required)
• ✅ High accuracy for helmet identification
• ✅ Easy to use with simple commands

Installation

1. Clone or download this repository

2. Install Python dependencies:

   pip install -r requirements.txt

3. The YOLO models will be automatically downloaded when you first run the script

Usage

Basic Usage

Using helmet-specific model (Recommended):

python helmet_detector.py --video path/to/your/video.mp4 --model sharathhhhh/safetyHelmet-detection-yolov8

Using standard YOLO models:

python helmet_detector.py --video path/to/your/video.mp4

Advanced Usage

Use a larger model for better accuracy:

# YOLOv8 Small (faster)
python helmet_detector.py --video input.mp4 --model yolov8s.pt

# YOLOv8 Medium (balanced)
python helmet_detector.py --video input.mp4 --model yolov8m.pt

# YOLOv8 Large (best accuracy)
python helmet_detector.py --video input.mp4 --model yolov8l.pt

# YOLOv8 Extra Large (maximum accuracy)
python helmet_detector.py --video input.mp4 --model yolov8x.pt

Adjust confidence threshold:

# Lower threshold (detect more objects)
python helmet_detector.py --video input.mp4 --conf 0.20

# Higher threshold (detect only high confidence)
python helmet_detector.py --video input.mp4 --conf 0.40

Specify output file:

python helmet_detector.py --video input.mp4 --output detected_output.mp4

All options:

python helmet_detector.py \
  --video input.mp4 \
  --model yolov8m.pt \
  --conf 0.25 \
  --output output.mp4

Available Models

Model	Size	Speed	Accuracy	Use Case
YOLOv8n	6MB	Fastest	Good	Quick processing
YOLOv8s	22MB	Fast	Better	Balanced
YOLOv8m	52MB	Medium	Best Balance	Recommended
YOLOv8l	87MB	Slower	Very Good	High quality
YOLOv8x	136MB	Slowest	Excellent	Maximum quality

Requirements

• Python 3.8 or higher
• CUDA-capable GPU (optional, for faster processing)
• 4GB+ RAM

How It Works

1. The script loads a pre-trained YOLO model (YOLOv8)
2. Each frame of the input video is analyzed
3. The model detects objects including helmets
4. Detections are drawn on the video with bounding boxes and confidence scores
5. The output video is saved with all detections

File Structure

yolo/
├── helmet_detector.py    # Main detection script
├── requirements.txt       # Python dependencies
├── README.md             # This file
└── input.mp4             # Your input video (user provided)

Troubleshooting

Issue: Slow processing

• Use a smaller model (yolov8n.pt or yolov8s.pt)
• Lower the input video resolution
• Ensure you have a GPU for faster processing

Issue: Not detecting helmets

• Try different YOLO models (larger models have better accuracy)
• Lower the confidence threshold with --conf 0.20
• Ensure helmets are clearly visible in the video

Issue: Too many false positives

• Increase the confidence threshold with --conf 0.40
• Use a larger, more accurate model

Notes

• Supported video formats: MP4, AVI, MOV, MKV
• The first run will download the YOLO model automatically
• Processing time depends on video length and model size
• GPU acceleration is used automatically if available

Credits

• YOLOv8 by Ultralytics
• OpenCV for video processing