Gear Classification in Skating Cross-Country Skiing Using Inertial Sensors and Deep Learning.

Area of Science:

• Sports Science
• Biomechanical Analysis
• Machine Learning in Sports

Background:

• Cross-country skiing, particularly the skating style, involves complex techniques.
• Accurate identification of different skiing gears is crucial for performance optimization and injury prevention.
• Existing methods for technique analysis can be cumbersome or lack precision.

Purpose of the Study:

• To develop and validate a system for identifying three distinct gears in cross-country ski skating.
• To utilize embedded inertial measurement units (IMUs) and deep learning for automated technique classification.
• To provide a tool for enhanced ski training and performance analysis.

Main Methods:

• Two experienced skiers performed uphill skating techniques, utilizing symmetric (G3) and asymmetric (G2L, G2R) pole-pushing gears.
• Inertial measurement units (IMUs) were attached to skis to record acceleration and Euler angle data.
• Convolutional Neural Networks (CNNs) combined with Long Short-Term Memory (LSTM) networks were employed for spatio-temporal feature extraction and classification.

Main Results:

• The developed deep learning model achieved 90% accuracy in cross-user evaluations.
• The system demonstrated a high accuracy of 98% in cross-scene evaluations for individual users.
• The model effectively distinguished between the symmetric (G3) and asymmetric (G2L, G2R) ski gears.

Conclusions:

• The system shows significant promise for accurately identifying different ski gears in the skating technique.
• The integration of IMUs and deep learning offers a valuable, non-invasive tool for ski training and biomechanical analysis.
• This technology can aid coaches and athletes in refining technique and improving performance.