Lower-body posture estimation with a wireless smart insole.

Area of Science:

• Biomechanical Engineering
• Machine Learning
• Wearable Technology

Background:

• Traditional optical marker and camera-based motion capture systems have limitations including indoor restrictions, difficulties in multi-person tracking, and high costs.
• Existing systems often fail to provide comprehensive lower-body movement analysis efficiently and portably.

Purpose of the Study:

• To develop a novel method for accurate lower-body posture estimation using a wireless smart insole.
• To leverage end-to-end deep neural network training for real-time human movement analysis.

Main Methods:

• A wireless smart insole was developed and integrated with a deep neural network model.
• The model was trained end-to-end to predict lower-body posture, including hip, knee, ankle, and toe movements.
• Real-time inference was achieved with a processing time of 1.62ms.

Main Results:

• The proposed method accurately predicts lower-body movements across a wide range of activities.
• The system demonstrates real-time performance suitable for dynamic applications.
• The smart insole system provides a portable and efficient alternative to conventional motion capture.

Conclusions:

• The wireless smart insole with deep neural network-based posture estimation offers an accurate, efficient, and portable solution for lower-body motion analysis.
• This technology has significant potential for applications in sports analysis, physical rehabilitation, and virtual reality environments.
• The system overcomes key limitations of existing motion capture technologies, enabling broader accessibility and use cases.