Privacy-Preserving In-Bed Pose and Posture Tracking on Edge

Area of Science:

• Biomedical Engineering
• Computer Vision
• Human-Computer Interaction

Background:

• In-bed behavior monitoring is crucial for bed-bound patients but limited by current technologies like wearables or pressure mapping systems.
• Vision-based pose tracking faces privacy and computational challenges for in-bed applications.
• Existing pose estimation models require significant resources, hindering edge device deployment.

Purpose of the Study:

• To develop a privacy-preserving, edge-deployable system for in-bed human pose and posture tracking.
• To enable detection of stable motion and user-specific pose alerts.
• To overcome privacy concerns associated with vision-based in-bed monitoring.

Main Methods:

• Developed a novel privacy-preserving system for in-bed pose and posture tracking.
• Implemented the system to run entirely on an edge device.
• Utilized infrared (LWIR) imaging for data acquisition.
• Evaluated system accuracy using retrospective and real-world data.

Main Results:

• Achieved over 93.6% estimation accuracy for in-bed poses.
• Reached over 95.9% accuracy in classifying three distinct in-bed posture categories.
• Demonstrated feasibility of edge device implementation for privacy-preserving pose tracking.

Conclusions:

• The developed system offers an accurate and privacy-preserving solution for in-bed patient monitoring.
• Edge computing enables efficient and secure pose and posture tracking.
• The system has potential applications in healthcare for patient behavior analysis and alerts.