Wearable Motion Capture: Reconstructing and Predicting 3D Human Poses From Wearable Sensors

Area of Science:

• Biomedical Engineering
• Computer Science
• Rehabilitation Technology

Background:

• 3D human pose reconstruction is crucial for health monitoring, particularly for individuals with movement disabilities.
• Current methods often rely on motion capture systems with third-person cameras, limiting outpatient use.
• Wearable sensors offer a potential solution for remote and unconstrained pose estimation.

Purpose of the Study:

• To develop a wearable motion capture system for reconstructing and predicting 3D human poses using IMU sensors and wearable cameras.
• To enable remote diagnosis and monitoring of patients outside clinical settings.
• To address the limitations of third-person camera-based pose estimation for solo outpatients.

Main Methods:

• Introduction of a novel Attention-Oriented Recurrent Neural Network (AttRNet).
• AttRNet features a sensor-wise attention-oriented recurrent encoder, a reconstruction module, and a dynamic temporal attention-oriented recurrent decoder.
• Development of a new WearableMotionCapture dataset using wearable IMUs and cameras with ground truth joint angles.

Main Results:

• The AttRNet model demonstrated high accuracy in reconstructing and predicting 3D human poses on the new WearableMotionCapture dataset.
• The proposed method outperformed state-of-the-art techniques on two public datasets: DIP-IMU and TotalCapture.
• Successful validation of wearable motion capture for lower-limb and full-body pose estimation.

Conclusions:

• The AttRNet model provides an effective solution for wearable motion capture, enabling accurate 3D human pose reconstruction and prediction.
• This technology has significant potential for remote health monitoring, rehabilitation assessment, and assistive device control.
• The developed system overcomes the limitations of traditional motion capture, facilitating patient care outside clinical environments.