Indirect Estimation of Vertical Ground Reaction Force from a Body-Mounted INS/GPS Using Machine Learning.

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Traditional vertical ground reaction force (vGRF) measurement methods (force plates, instrumented treadmills) are confined to indoor settings.
• Wearable insoles offer mobility but have limited durability.
• Indirect estimation of vGRF using inertial measurement units (IMUs) and machine learning presents a promising alternative.

Purpose of the Study:

• To present a methodology for indirectly estimating vGRF and other gait analysis features using a wearable INS/GPS device.
• To evaluate the accuracy of machine learning models in predicting gait parameters and vGRF from INS/GPS data.

Main Methods:

• Extracted 27 features from INS/GPS data.
• Utilized feature analysis to identify six key features for gait parameter estimation.
• Trained bagged ensembles of regression trees for gait parameter prediction.
• Employed K-nearest neighbor (KNN) and long short-term memory (LSTM) neural networks for vGRF and ground contact time prediction.

Main Results:

• Six features from INS/GPS data accurately estimated 11 gait parameters.
• Regression trees achieved good prediction accuracy for gait parameters, even for subjects with no prior training data.
• KNN models demonstrated lower normalized root mean square error for vGRF prediction compared to LSTM networks.

Conclusions:

• Wearable INS/GPS systems combined with machine learning offer a viable method for indirect vGRF estimation.
• This technology enables gait analysis in unrestricted, outdoor environments.
• While effective, KNN models for vGRF prediction have limitations in detecting novel force patterns.