BodyPressure - Inferring Body Pose and Contact Pressure From a Depth Image.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Human-Computer Interaction

Background:

• Contact pressure is crucial for human comfort, safety, posture, and health.
• Preventing pressure injuries in healthcare settings is a significant challenge, particularly for individuals at rest in bed.
• Accurate measurement of body-mattress contact pressure is difficult, especially with occlusions like bedding.

Purpose of the Study:

• To develop a novel method for inferring human body contact pressure on a mattress using depth imaging.
• To address the challenge of occlusions, such as bedding, in pressure inference.
• To improve the prevention of pressure injuries through enhanced monitoring capabilities.

Main Methods:

• Augmenting a real dataset with synthetic data generated from soft-body physics simulations.
• Developing and training a novel deep network incorporating an embedded human body mesh model.
• Utilizing a white-box model for depth and pressure image generation within the network.
• Evaluating the network's performance on real-world data, including occluded scenarios.

Main Results:

• The deep network successfully infers human body pose, outperforming previous methods.
• The network achieves the novel capability of inferring contact pressure across a 3D human body mesh.
• Accurate pressure inference is demonstrated even in the presence of occlusions from blankets.

Conclusions:

• The proposed deep network offers a promising solution for non-invasive contact pressure inference from depth images.
• This technology has direct applications in healthcare for the prevention of pressure injuries.
• The method's ability to handle occlusions represents a significant advancement in the field.