Video dataset for the detection of safe and unsafe behaviours in workplaces

Abstract

Real-time detection of safe and unsafe behaviours in production facilities is very important to prevent these behaviours before they occur. In this context, this study presents a highresolution video-based dataset of safe and unsafe behaviours obtained from a closed production facility for use in occupational accident prevention. The dataset was collected from the security cameras of a production facility operating in an organised industrial zone in Eski ¸sehir, Turkey, in November and December 2022, after obtaining the necessary permissions from company officials and employees. A total of 8 behaviour classes, 4 classes of safe and 4 classes of unsafe behaviours, were identified for the dataset and 691 video clips containing these behaviours were obtained. The video clips created for the dataset are in MP4 format at 1920×1080 pixels and 24 frames per second. In the dataset, the safe behaviour classes are Safe Walkway, Authorized Intervention, Closed Panel Cover and Safe Carrying, while the unsafe behaviour classes are Safe Walkway Violation, Unauthorized Intervention, Opened Panel Cover and Carrying Overload with Forklift.