Criterion Validity Evidence Of Energy Expenditure Predicted By Ai Computer Vision Based Equations: Preliminary Study

PURPOSE:To validate the accuracy of energy expenditure (EE) estimation equations calculated using distance, speed, and acceleration data in soccer settings obtained by Artificial Intelligence (AI) Computer Vision (CV), which is a video-based physical data collection system. METHODS: A total of 10 males (age: 29.40 ± 3.03) performed four types of 200 m linear runs at speeds classified by FIFA standards: Walking(0-7 km/h), Jogging(7-15 km/h), Low-Intensity Running(15-20 km/h), and Sprint(>25 km/h). It was also including dribbling, shooting, and various movements in a real soccer match (e.g., jumping, zigzag running etc.). The whole sessions were recorded by the Bepro Camera System (KR), which utilizes AI CV to estimate player positions as well as to calculate speeds and accelerations. The accuracy of the data generated by the Bepro Camera System has been validated by FIFA’s official certification (FIFA, 2022). The prediction equation using distance, velocity, and acceleration values were driven to estimate EE (Osgnach et al., 2009). For validation of the equation, K5 (Italy) was employed as a criterion measure. Percentage difference ratios, which is (actual EE-estimated EE)/ average of actual + estimated EE, was calculated to determine the error rates. Also, paired t-tests between two EEs were applied with Type I error = .05. RESULTS: The results of estimated EEs were as follows: K5: 139.68 ± 21.20 kcal vs. AI CV: 157.13 ± 23.00 kcal. There was a statistically significant difference of EEs between K5 and AI CV (t = 4.371, p = 0.002), in which the AI CV overestimated EEs during the soccer settings. The calculated percentage difference ratio was 0.89 ± 0.08. It indicates that multiplying the estimated value by 0.89 could yield more accurate prediction of the actual value of EE. CONCLUSIONS:This study revealed that the existing EE estimation equation, which relies on distance, speed, and acceleration, overestimates EEs by approximately 13%. As this is a preliminary study, further research will be necessary to develop the new formula for increasing the accuracy of EE. Supported by: National Research Foundation of Korea[NRF-2021R1A2C2013651]