Estimating ground reaction forces and moments during gait from multiple kinematic variables using a feedforward...

Area of Science:

• Biomechanics
• Wearable Technology
• Artificial Intelligence in Sports Science

Background:

• Ground reaction force (GRF) and ground reaction moment (GRM) are essential for gait analysis.
• Traditional force plates are accurate but expensive and spatially restrictive.

Purpose of the Study:

• To evaluate the accuracy and reliability of predicting GRF and GRM using a feedforward neural network (FNN).
• To integrate infrared camera positional data with wearable accelerometer (ACC) data for enhanced prediction.

Main Methods:

• Eighty participants walked over force plates while their body segment positional and ACC data were collected.
• A feedforward neural network was trained using this combined data to predict GRF and GRM.
• Prediction accuracy was quantified using metrics like RMSE and NRMSE.

Main Results:

• Combining positional and ACC data significantly improved GRF predictions across all directions (vertical, anterior-posterior, medial-lateral).
• High correlation (0.979) was achieved for medial-lateral GRF, with NRMSE at 6.07%.
• GRM predictions, particularly in the sagittal plane (R²=0.939), showed superior performance with ACC integration.

Conclusions:

• The proposed FNN model integrating camera and ACC data demonstrates superior performance compared to existing methods.
• This approach provides a cost-effective and flexible alternative to force plates for clinical and sports assessments.
• The findings suggest a significant advancement in non-invasive gait analysis techniques.