Alpine Skiing Activity Recognition Using Smartphone's IMUs.

Area of Science:

• Sports Science
• Biomechanical Engineering
• Data Science

Background:

• Existing alpine skiing studies often use limited data or controlled environments.
• A functional sensor setup and algorithm are needed for real-world data collection and activity distinction.
• Detecting alpine skiing activities reliably is crucial for performance analysis and injury prevention.

Purpose of the Study:

• To develop and validate an unsupervised method for detecting alpine skiing activities using smartphone inertial measurement units (IMU).
• To assess the feasibility of this method for daily use across varied skiing conditions and skill levels.
• To compare the performance of different unsupervised learning algorithms for this specific application.

Main Methods:

• Collected data from full skiing sessions of novice to expert skiers in diverse conditions using smartphone IMUs.
• Applied a windowing strategy for feature extraction and Principal Component Analysis (PCA) for dimensionality reduction.
• Compared three unsupervised learning techniques: KMeans, Ward's method, and Gaussian Mixture Model (GMM).

Main Results:

• Unsupervised learning accurately distinguished alpine skiing activities from other daily activities.
• The method demonstrated independence from skier skill level and environmental conditions.
• The best performing model achieved an accuracy of 99.25%.

Conclusions:

• Unsupervised detection of alpine skiing activities using smartphone IMUs is feasible and highly accurate for daily use.
• The developed method offers a practical solution for collecting and analyzing skiing data in real-world settings.
• This approach has the potential to advance sports science research and athlete monitoring in alpine skiing.