Knowledge Distillation Method Improves Wearable Gait Analysis

A new deep learning method, Selective Correlation Based Knowledge Distillation, has been proposed to enhance ground reaction force (GRF) estimation from wearable insole sensors. GRF is crucial for gait analysis in healthcare, rehabilitation, and sports, but traditional measurement requires expensive force plates in labs. Wearable sensors offer portability but suffer from noise and interference. Deep learning models can improve accuracy but demand high computing resources, limiting real-time use on portable devices. The proposed technique addresses these issues by selectively transferring knowledge from a complex teacher model to a lightweight student model, reducing computational load while maintaining accuracy. The method was validated on a dataset, showing improved performance over existing approaches. This advancement could enable practical, real-time gait monitoring outside laboratory settings.