Application of Signed Distance Function Neural Network in Real-Time Feet Tracking.

Area of Science:

• Robotics
• Computer Vision
• Biomedical Engineering

Background:

• Locomotion loss is a significant challenge for the elderly.
• Gait monitoring is crucial for lower-limb rehabilitation and progress evaluation.
• Current gait measurement methods (motion capture, force plates, IMUs) have limitations like cost, portability, and sensor attachment.

Purpose of the Study:

• To propose a novel contactless method for real-time foot pose tracking.
• To develop a system compatible with over-ground rehabilitation robots.
• To improve the accuracy and practicality of gait analysis for elderly rehabilitation.

Main Methods:

• Development of a novel deep neural network, SDF-Net, modeling the signed distance function (SDF).
• Real-time tracking of foot poses using color and depth images as input.
• Evaluation of the algorithm's accuracy through subject testing with various foot movements.

Main Results:

• The proposed SDF-Net achieves accurate, angle-independent foot pose tracking.
• Dynamic errors in position and orientation were found to be less than 9 mm and 8 degrees, respectively.
• Performance surpasses existing state-of-the-art contactless gait analysis methods.

Conclusions:

• SDF-Net offers a promising solution for contactless, real-time foot pose estimation.
• This method facilitates the development of advanced, portable gait rehabilitation technologies.
• The system's accuracy and compatibility enhance its potential for clinical and robotic applications.