README.md

metadata

annotations_creators:
  - expert-generated
language_creators:
  - crowdsourced
license: cc-by-4.0
multilinguality: monolingual
language:
  - en
size_categories:
  - 1K<n<10K
tasks:
  - object-detection
  - classification
dataset_type: image-folder
tags:
  - object-detection
  - classification
  - safety-helmet
  - hardhat-detection
  - construction-safety
  - drone-imagery
  - night-vision
  - computer-vision
  - yolo-format
  - color-classification

IITU Safety-Helmet Dataset v1.0

Overview:

This dataset contains annotated images of safety helmets captured both by drone and at ground level, designed for helmet detection and color classification tasks in computer vision.

---

📖 Dataset Summary

This dataset contains 1,664 images annotated for safety-helmet detection and color classification.

• 6,473 helmet instances
• Captured by drone (3–5 m, 10–15 m; angles 0°, 45°, 90°) and at ground level
• Lighting conditions: daytime (clear) and night (lamp light)

📂 Dataset Structure

iitu-safety-helmet/
├── data
│   ├── frames_daytime
│   │   ├── horizontal_0deg
│   │   │   ├── 3-5_meters
│   │   │   │   ├── images/
│   │   │   │   └── annotations/
│   │   │   └── 10-15_meters
│   │   │       ├── images/
│   │   │       └── annotations/
│   │   ├── tilted_45deg
│   │   └── vertical_90deg
│   └── frames_night_time
│       ├── horizontal_0deg
│       │   ├── images/
│       │   └── annotations/
│       ├── tilted_45deg
│       └── vertical_90deg
├── classes.txt
└── README.md

Annotation Format:

• Each .txt annotation file corresponds to one image and contains one line per helmet.
• Line format (YOLO v5-style with high-precision floats): "class_id", "x_center", "y_center", "width", "height".
• All five values are space-separated; the four geometry values are normalized to [0, 1] and may have up to 16 decimal places.

Classes:

0 blue helmet
1 green helmet
2 red helmet
3 white helmet
4 yellow helmet

Paper / Citation
Bektemyssova, G.; Bykov, A.; Keresh, A.; Yergazy, Y.; Shaikemelev, G.;

IITU Safety-Helmet Dataset: Drone-Captured and Ground-Level Images with Color Annotations. 2025.
DOI: soon

License:
CC BY 4.0