Evaluation of computer vision techniques for automated hardhat detection in indoor construction safety applications

Abstract

Construction is considered among the most dangerous industries and is responsible for a large portion of total worker fatalities. A construction worker has a probability of 1-in-200 of dying on the job during a 45-year career, mainly due to fires, falls, and being struck by or caught between objects. Hence, employers must ensure their workers wear personal protective equipment (PPE), in particular hardhats, if they are at risk of falling, being struck by falling objects, hitting their heads on static objects, or coming in proximity to electrical hazards. However, monitoring the presence and proper use of hardhats becomes inefficient when safety officers must survey large areas and a considerable number of workers. Using images captured from indoor jobsites, this paper evaluates existing computer vision techniques, namely object detection and color-based segmentation tools, used to rapidly detect if workers are wearing hardhats. Experiments are conducted and the results highlight the potential of cascade classifiers, in particular, to accurately, precisely, and rapidly detect hardhats under different scenarios and for repetitive runs, and the potential of color-based segmentation to eliminate false detections.