Modular PPE Detection System for Workplace Safety Management using Pose Estimation

Improving workplace safety management is crucial to ensure the correct utilization of personal protective equipment (PPE). We introduce an effective PPE detection system with the objective of verifying PPE usage. The overall pipeline employs a pose estimation network to obtain 2D keypoints which is followed by a set of modular components where each equipment type has an individual algorithm and these modules benefit from keypoints to localize the desired RoI. Apart from IoU, we introduced and used a new metric in order to measure the crop localization quality. Finally, corresponding CNN classifiers predict if the detected people wear specific equipment. Evaluations are tested on equipment classification. Assessment of classifiers shows an overall 0.92, 0.98, 0.86 and 0.95 test accuracy for hardhat, vest, glove and eyeglass respectively. Our proposed PPE detection system demonstrates promising results in accurately verifying the usage of personal protective equipment in the workplace. These findings highlight the potential of our system to improve workplace safety management.