Seeing Under the Cover: A Physics Guided Learning Approach for In-bed Pose Estimation

摘要: Human in-bed pose estimation has huge practical values in medical and healthcare applications yet still mainly relies on expensive pressure mapping (PM) solutions. In this paper, we introduce our novel physics inspired vision-based approach that addresses the challenging issues associated with problem including monitoring a fully covered person complete darkness. We reformulated using proposed Under Cover Imaging via Thermal Diffusion (UCITD) method to capture high resolution information of body even when it is by long wavelength IR technique. physical hyperparameter concept through which achieved quality groundtruth labels different modalities. A annotated dataset called Simultaneously-collected multimodal Lying Pose (SLP) also formed/released same order magnitude as most existing large-scale human datasets support complex models’ training evaluation. network trained from scratch tested two diverse settings, one living room other hospital showed performance 98.0% 96.0% PCK0.2 standard, respectively. Moreover, multi-factor comparison state-of-the art solution based PM, significant superiority all aspects being 60 times cheaper, 300 smaller, while having higher recognition granularity accuracy.