Parallel Personal Protective Equipment Recognition Network Based On Multimodal Fusion

Le résumé fourni par la source

The emergence of the Omicron variant in February 2022 resulted in widespread challenges in Shanghai, necessitating proper monitoring of Personal Protective Equipment (PPE) in mobile cabin hospitals for staff and patient safety. Camera-based PPE approaches offer effective monitoring for compliance and safety at work. However, existing camera-based approaches proved inefficient in accurately detecting diverse PPE components. To address this issue, we propose the Parallel Multimodal PPE Recognition Network (PMPRN), an end-to-end solution that combines target and keypoint detection in parallel. PMPRN integrates the finetuned MobileNetv2-YOLOX object detector and MobileNetv2-CID model to enhance detection performance. We introduce the novel PPELoss function to improve geometric position judgments and detection efficiency in various scenarios. Additionally, an adaptive Hybrid fusion strategy handles the exponential growth of equipment categories, ensuring precise detection and identification of PPE items like gloves. Compared to existing detectors, PMPRN demonstrated a significant 27.6-point accuracy improvement and a 12.76 Fps speed increase. Notably, it accurately detects gloves held in hand and identifies improper usage of glasses, masks, and other equipment. The proposed PMPRN was validated on a dataset of prevention personnel, and the test results demonstrated the effectiveness and real-time performance of the proposed model and framework.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.