A mask-wearing detection method in elevators based on improved deformable DETR

Wearing masks is an effective way to mitigate the spread of the virus, and it becomes even more crucial in confined spaces like elevators, where the virus can spread more easily. However, traditional manual detection methods are inefficient and not real-time. In this paper, a mask-wearing detection method in elevators based on improved deformable DETR (Mask deformable DETR(Detection Transformer)) is proposed. This method builds upon the foundation of deformable DETR, incorporating an efficient channel attention mechanism to assist residual networks in extracting mask features comprehensively. This mechanism enhances information on small-scale feature maps. Additionally, a conditional attention mechanism is introduced to expedite model convergence while improving the detection capability for small targets. Furthermore, a two-stage data augmentation technique tailored to the unique characteristics of mask-wearing data in elevators is designed. This method can increase sample quantity, balance sample categories, and enhance model performance. Experimental results demonstrate that this method can effectively recognize the mask-wearing status of elevator passengers, exhibiting a high level of recognition accuracy. It holds significant implications for practical elevator monitoring scenarios.