This study focuses on the example of the segregation of personal items in high-risk sectors such as clothing and personal protective equipment. We propose using biometric objects from Open Images V5 and DeepFashion2 datasets for pretraining mask segmentation networks for recognition and segmentation of personal protective equipment in the workplace. The preliminary results of our proposed model achieve a mean average precision with modest optimization, which results in extremely effective segmentation of welding masks, high-visibility vests, construction helmets, and ear protection in the workplace. The results of this research can be applied to improve workplace safety in high-risk industries by providing a way to ensure that personal protective equipment (PPE) is used appropriately while protecting employee privacy.