A comparative study of plantar pressure and inertial sensors for cross-country ski classification using deep learning

Abstract

This work presents a comparative study of low cost and low invasiveness sensors (plantar pressure and inertial measurement units) for classifying cross-country skiing techniques. A dataset was created for symmetrical comparative analysis, with data collected from skiers using instrumented insoles that measured plantar pressure, foot angles, and acceleration. A deep learning model based on CNN and LSTM was trained on various sensor combinations, ranging from two specific pressure sensors to a full multisensory array per foot incorporating 4 pressure sensors and an inertial measurement unit with accelerometer, magnetometer, and gyroscope. Results demonstrate an encouraging performance with plantar pressure sensors and classification accuracy closer to inertial sensing. The proposed approach achieves a global average accuracy of 94% to 99% with a minimal sensor setup, highlighting its potential for low-cost and precise technique classification in cross-country skiing and future applications in sports performance analysis.