Walking Speed and Uncertainty Estimation Using Mixture Density Networks for Dynamic Ambulatory Environments.

Area of Science:

• Gerontology
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Aging leads to reduced muscle strength and cardiovascular function, decreasing walking speed and activity levels.
• Wearable walking rehabilitation devices aim to assist elderly individuals by reducing effort or enhancing muscle function.
• Accurate measurement of walking speed is crucial for objectively assessing rehabilitation progress.

Purpose of the Study:

• To develop an advanced method for accurately estimating walking speed in elderly individuals using wearable devices.
• To address the limitations of existing walking speed estimation methods in diverse environmental conditions.
• To incorporate uncertainty estimation into walking speed measurements for a more comprehensive assessment.

Main Methods:

• Introduction of a novel Convolutional Neural Network (CNN)-based Mixture Density Network (CMDN) structure.
• Validation of the CMDN model through experiments with 20 elderly participants.
• Testing the system's performance in varied scenarios, including flat surfaces and stair descent.

Main Results:

• The CMDN structure demonstrated enhanced accuracy in estimating walking speed compared to existing methods.
• The network provided valuable uncertainty information, indirectly reflecting the complexity of the walking environment.
• Successful performance was observed across different terrains, indicating robustness.

Conclusions:

• The CMDN offers a promising foundation for reliable walking speed estimation in diverse real-world scenarios.
• The uncertainty estimation capability of CMDN can provide deeper insights into the wearer's walking condition.
• This technology has the potential for widespread application in personalized elderly rehabilitation programs.