Skeleton-based detection of anomalous personal protective equipment doffing behaviors among healthcare workers

Abstract

Identification of doffing behaviors of personal protective equipment (PPE) plays a crucial role in ensuring the safety of healthcare workers. With the continuous emergence of new infectious diseases, accurate detection of anomalous behaviors during PPE doffing procedures has become increasingly critical. In complex medical environments, conventional visual methods have demonstrated limited capability in accurately capturing the subtle movements involved in the multistep PPE doffing process. To address the challenges of low motion heterogeneity and minimal amplitude variations in PPE doffing procedures, this study presents a skeleton keypoint-based anomaly detection model. The proposed model innovatively integrates spatiotemporal embedding modules and adaptive attention mechanisms, allowing the precise detection of subtle changes in localized hand movements. In contrast to the limitations of conventional methods in characterizing fine-grained feature differences, this model demonstrates significantly enhanced capability in identifying anomalous PPE doffing behaviors. Extensive experimental results indicate that the model outperforms existing methods in key metrics, including precision and recall, providing novel technical support for the management of standardized PPE in medical settings.