Voxel51/Safe_and_Unsafe_Behaviours

Dataset Card for safe_unsafe_behaviours

This is a FiftyOne dataset with 691 samples.

Installation

If you haven't already, install FiftyOne:

pip install -U fiftyone

Usage

import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/Safe_and_Unsafe_Behaviours")

# Launch the App
session = fo.launch_app(dataset)

Dataset Details

Dataset Description

A high-resolution video dataset of safe and unsafe workplace behaviors collected from security cameras at a production facility, designed for occupational accident prevention research. The dataset contains 691 video clips capturing 8 behavior classes (4 safe, 4 unsafe) that represent common safety compliance scenarios in industrial environments including walkway violations, unauthorized equipment interventions, panel cover states, and forklift load compliance.

• Curated by: Oğuzhan Önal and Emre Dandıl, Department of Computer Engineering, Faculty of Engineering, Bilecik Şeyh Edebali University, Bilecik, Turkey
• Shared by: Kafaoğlu Metal Plastik Makine San. ve Tic. A.Ş., Eskişehir, Turkey
• Language(s) (NLP): en
• License: CC BY 4.0

Dataset Sources

• Repository: https://data.mendeley.com/datasets/xjmtb22pff/1
• Paper: https://www.sciencedirect.com/science/article/pii/S235234092400756X

Uses

Direct Use

• Video classification for industrial safety monitoring systems
• Action recognition and temporal behavior detection research
• Training real-time unsafe behavior detection models
• Benchmarking video understanding models on industrial surveillance footage
• Computer vision research for occupational health and safety applications

Out-of-Scope Use

• Deployment in environments significantly different from industrial manufacturing settings
• Worker surveillance or performance monitoring without proper consent and ethical oversight
• Applications requiring detection of safety behaviors not represented in the 8 defined classes

Dataset Structure

This dataset is formatted for FiftyOne, an open-source tool for building high-quality datasets and computer vision models.

Dataset Info:

• Name: safe_unsafe_behaviours
• Media type: video
• Num samples: 691
• Splits: train (566 samples), test (125 samples) — indicated via sample tags

Sample Fields:

Field	Type	Description
id	ObjectIdField	Unique sample identifier
filepath	StringField	Path to video file
tags	ListField(StringField)	Split tags: train or test
metadata	EmbeddedDocumentField(VideoMetadata)	Video metadata (resolution, duration, fps, etc.)
ground_truth	EmbeddedDocumentField(Classification)	Video-level behavior classification label

Frame Fields:

Field	Type	Description
id	ObjectIdField	Unique frame identifier
frame_number	FrameNumberField	Frame index within video

Classes (8 total):

Class	Behavior Type	Description
Safe Walkway Violation	Unsafe	Worker goes beyond designated safe walkway boundaries
Unauthorized Intervention	Unsafe	Worker intervenes on equipment without proper safety gear/authorization
Opened Panel Cover	Unsafe	Panel cover left open after intervention
Carrying Overload with Forklift	Unsafe	Forklift carrying 3+ blocks
Safe Walkway	Safe	Worker stays within designated walkway
Authorized Intervention	Safe	Worker properly equipped for equipment intervention
Closed Panel Cover	Safe	Panel cover properly closed
Safe Carrying	Safe	Forklift carrying 2 or fewer blocks

Video Specifications:

• Resolution: 1920×1080 pixels
• Frame rate: 24 fps
• Format: MP4
• Duration: 1–20 seconds per clip

Dataset Creation

Curation Rationale

Unsafe behavior is a leading cause of workplace injuries and deaths. Despite regular safety inspections, accidents occur due to breaches of occupational health and safety protocols. This dataset was created to support the development of computer vision systems capable of real-time detection of unsafe behaviors before accidents occur, addressing the need for automated, continuous safety monitoring in industrial environments.

Source Data

Data Collection and Processing

Video footage was collected from security cameras at Kafaoğlu Metal Plastik Makine San. ve Tic. A.Ş., a production facility in an organized industrial zone in Eskişehir, Turkey. Collection occurred between November 5, 2022 and December 13, 2022 (39 days) using two different IP cameras. After collection, domain experts reviewed the footage to identify segments containing the defined safe and unsafe behaviors, extracting clips of 1–20 seconds containing the target behaviors.

Who are the source data producers?

Workers and employees at Kafaoğlu Metal Plastik Makine San. ve Tic. A.Ş. performing normal production activities. Necessary permissions were obtained from company officials and employees prior to data collection.

Annotations

Annotation process

After videos were collected, frames containing safe and unsafe behaviors were identified by domain experts including factory managers and occupational safety specialists. Video clips were then extracted from the full surveillance footage. Some videos contain a single behavior class while others may contain multiple behavior classes.

Who are the annotators?

Domain experts including factory managers and the occupational safety specialist at the facility where the videos were collected, in collaboration with the research team.

Personal and Sensitive Information

The dataset contains video footage of workers performing their duties in an industrial setting. Workers' faces and bodies are visible in the footage. Proper permissions were obtained from company officials and employees for the use of video recordings and images in academic studies. The permission documentation is maintained at the Department of Computer Engineering, Faculty of Engineering, Bilecik Şeyh Edebali University.

Bias, Risks, and Limitations

• Single environment: Data collected from one facility in Turkey, which may limit generalization to other industrial settings, equipment configurations, or geographic contexts
• Temporal scope: 39-day collection period may not capture seasonal variations in behavior or clothing
• Class definitions: Safety behaviors are specific to this facility's protocols and may not align with regulations in other jurisdictions
• Scale: 691 videos is relatively small for training deep learning models; data augmentation may be necessary
• Camera perspectives: Only two camera viewpoints represented

Recommendations

Users should consider domain adaptation techniques when applying models trained on this dataset to different industrial environments. The dataset is best suited for research and prototyping rather than direct production deployment without additional validation on target environments.

Citation

BibTeX:

@article{ONAL2024110756,
  title = {Video dataset for the detection of safe and unsafe behaviours in workplaces},
  journal = {Data in Brief},
  volume = {56},
  pages = {110756},
  year = {2024},
  issn = {2352-3409},
  doi = {https://doi.org/10.1016/j.dib.2024.110756},
  url = {https://www.sciencedirect.com/science/article/pii/S235234092400756X},
  author = {Oğuzhan Önal and Emre Dandıl}
}

APA:

Önal, O., & Dandıl, E. (2024). Video dataset for the detection of safe and unsafe behaviours in workplaces. Data in Brief, 56, 110756. https://doi.org/10.1016/j.dib.2024.110756