The number of jumps figure skaters perform daily has never been formally quantified, though it has been suggested that skaters perform 50-100 jumps per training day. The magnitude of force, high loading rates, and frequent repetitions likely contribute to the high injury rate of competitive skaters. Monitoring the number of jumps performed may help decrease risk of injury, similar to the institution of pitch counts in youth baseball.
Activity monitors that are commonly used for activities such as walking and running record many false positives during figure skating jump quantification due to the variety of skating movements that generate similar acceleration profiles. Previously, we developed an algorithm that successfully counted 39 of 40 jumps performed during the competitive routines of 7 local skaters whose isolated jumps were used to create the algorithm.
Purpose: To test the performance of the algorithm on an independent sample of skaters of varying skill levels.
Methods: 18 healthy competitive figure skaters participated in this study (ages 8-26y, 12 female). Each skater wore an IMU affixed to the lower back while they performed a variety of jumps, spins, and footwork. A high speed video camera recorded all trials for validation purposes. Custom software was used to analyze the IMU data to quantify the number of jumps performed with >1 rotation.
Results: Analysis of the videos showed that we recorded a total of 200 jumps with >1 rotation. The algorithm correctly quantified 94.5% of the jumps in this dataset (189 successful jumps). It also identified 11 jumps with ≤1 rotation.
Conclusions: These results show that this algorithm can be successfully applied to a unique dataset. Many of the jumps with ≤1 rotation that were counted were “popped” jumps, where a skater intends to perform a multi-revolution jump prior to take-off, but perform a single instead. Multi-revolution jumps that were not counted included falls and those with too much rotation that occurs on the ice prior to take-off. Finally, this dataset showed that the algorithm may need to be customized for smaller and/or low-level skaters as it failed to identify 7 of 12 jumps performed by a small, beginning level skater (8y, 122cm, 23.6 kg). Further improvements may be made by using machine learning algorithms to differentiate types of jumps as well as jump count.