3D Foot Kinetics Estimation From Distributed VGRF From Smart Insoles via 1D Domain Transformation.

Area of Science:

• Biomechanics
• Human Locomotion Analysis
• Wearable Sensor Technology

Background:

• Understanding foot kinetics is crucial for analyzing human locomotion and mechanical loads.
• While vertical ground reaction force (vGRF) is common, comprehensive 3D kinetic data (3D-GRF&M, CoP) offer deeper insights.
• Smart insoles offer portable vGRF measurement but often yield lower data quality compared to lab equipment.

Purpose of the Study:

• To develop a deep learning model for generating instrumented treadmill-level 3D-GRF&M-CoP from smart insole vGRF data.
• To enhance the quality of vGRF data captured by smart insoles.
• To identify optimal plantar sensor placement for accurate 3D kinetics estimation.

Main Methods:

• Leveraging deep learning-based domain transformation using 1D sequence-to-sequence (1D-s2s) models.
• Applying multi-segment analysis to determine key plantar regions for kinetic parameter estimation.
• Developing and evaluating Ke2KeNet, a novel deep learning architecture.

Main Results:

• Successfully transformed distributed vGRF signals into comprehensive 3D-GRF&M-CoP data.
• Demonstrated enhanced insole vGRF data quality comparable to instrumented treadmills.
• The novel Ke2KeNet model outperformed existing 1D-s2s benchmarks in accuracy.

Conclusions:

• Deep learning enables high-fidelity 3D foot kinetics estimation from portable smart insoles.
• This approach significantly advances the potential of wearable sensors for detailed biomechanical analysis.
• Optimized sensor layouts and advanced deep learning models are key to unlocking the full potential of smart insoles.