Combining Computer Vision with Fuzzy Logic for Real-time Evaluation of Personal Protective Equipment Compliance

Computer Vision
Fuzzy Logic
Deep Learning
Under review
Author

Amirhossein Ghadiri, Vahid Shahhosseini

Published

January 1, 2024

Under review at the International Journal of Occupational Safety and Ergonomics. Code is available on GitHub.

This was the core of my master’s research at Amirkabir University of Technology. The idea was to take the part of site safety that usually depends on someone watching a camera feed and turn it into something a model can do continuously, while still giving a safety manager a number they can act on rather than a raw detection.

Abstract

Introduction. Construction safety codes mandate the consistent use of personal protective equipment (PPE). Yet, the inadequacies of manual monitoring lead to PPE non-compliance, resulting in fatalities or severe injuries. Computer vision models are promising alternatives for effectively identifying PPE non-use.

Methods. This study leverages computer vision and fuzzy logic to detect workers and assess safety hats, vests, and gloves wearing. Firstly, the YOLOX object detector identifies workers and PPE-related objects, associating each object with its respective worker. Subsequently, the recognized workers undergo cropping and are input into three classifiers, each specialized in evaluating compliance with a specific PPE category. Multiple fuzzy inference systems (FISs) amalgamate the outcomes of the preceding stages to compute danger levels for the non-compliance of each PPE type for each worker. Lastly, another fuzzy inference system integrates the danger levels and evaluates a worker’s PPE danger generally.

Results. Empirical findings illustrate that utilizing FIS to combine outputs from Object Detection and Image Classification modules enhances precision and recall.

Conclusions. This approach enables accurate outcomes even when employing fast algorithms. With the proposed model, safety managers can confidently identify at-risk workers, enabling them to take immediate or future measures with certainty and efficacy.