Quantitative Analysis of Athlete Performance in Artistic Skating using IMU, and Machine Learning Algorithms.

Abstract

This experimental study aims to analyze the inertial sensor-based measurements in artistic skating to perform quantitative analysis using statistical modeling and machine learning techniques. The primary motivation is to bring accountability and transparency for the artistic skating jury during competitions using the Inertial Measurement Unit (IMU) sensors and real-time analytics. For data collection, four athletes performed various skating patterns with the IMU attached to the ankle with the support of an elastic band. Athletes are categorized as Beginner level, intermediate level, and expert level, based on their previous experience. The visualization of the plotted gyroscope sensor data with three-axis line plots indicated a clear difference between the performance of the expert and other athletes. As part of the quantitative analysis, statistical features that highlighted the variations among the athletes were identified. To categorize the levels of the athletes, supervised and unsupervised machine learning techniques were used and we obtained good accuracy. This methodology helps the coaches to obtain a quantitative score of the skater’s performance in real-time and to provide effective training as well.