Pressure-Based In-Bed Pose Estimation Methods Using the Skinned Multi-person Linear Model

Monitoring sleeping postures is a critical task in the healthcare field, as improper sleeping positions can result in issues such as pressure ulcers and sleep apnea. Furthermore, specific sleeping postures must be maintained by certain patients to facilitate better postoperative recovery. Currently, this task primarily relies on manual monitoring and questioning, which is labor-intensive and susceptible to subjective errors. In this study, we propose an algorithm for pose estimation based on the Skinned Multi-Person Linear (SMPL) model, which can accurately determine 3D human body postures using pressure maps and corresponding gender information. Unlike visual sensor-based approaches, our method is not affected by lighting conditions or obstructions, ensuring enhanced privacy protection for patients. Our approach involves learning three distinct sets of parameters from pressure images, which are subsequently applied to the SMPL model to generate meshes and 3D joints. Our method surpasses previous approaches in terms of performance and maintains a high inference speed. To expedite both training and inference processes, we developed a batched SMPL forward module capable of simultaneously handling inputs for both genders. This module significantly accelerates the overall speed of training and inference procedures.