Can machine learning help reveal the competitive advantage of elite beach volleyball players?

Norway has the best beach volleyball players in the world, winning gold medals at the Olympic Games and other global events. What makes their movements so effective, and how can they be distinguished from novices? In an ongoing study, Norwegian Olympic-level beach volleyball players, as well as a group of lesser-skilled rowers, played training matches, while we recorded the movement of both their upper body and arm with accelerometers, along with data on their respiration and heart rate. The recordings will be cleaned and tagged according to which part of the match they belong to (serving, receiving, setting, spiking, blocking, defending).

Bildet kan inneholde: har, ansikt, sportsuniform, spiller sport, shorts.

An initial exploration has been undertaken in a group project in the IN5490 course autumn 2023 which this master project should build on. Your task is to apply machine learning to investigate how the movement of Olympic-level athletes can be distinguished from that of the lesser-skilled players. Some possible research questions are:

  1. Based on our data from video and accelerometers (and possibly also heart rate/respiration data), can we distinguish between Olympic-level players and the lesser-skilled players?
  2. Is there a specific part of the game (serving, receiving, setting, spiking, blocking, defending) that most clearly separates the two groups of players?
  3. How precisely can we from the recorded data determine what action a player is currently performing?
  4. Can we identify specific features in the plays behaviour that distinguish the two groups of players?

Suggested Approach:

  1. Read up on related work, including how motion data can be used to train a Machine Learning model that categorizes/classifies the ongoing motion. Studies from sports would be especially relevant. The paper resulting from the IN5490 project autumn 2023 is also relevant.
  2. Set up a machine learning algorithm that learns a mapping from movement data to what we want to classify (type of player in question 1, part of game in question 2)
  3. Train the model and analyze the results: How well did it classify the movement data? Are there ways to improve it by extracting features differently or applying different types of ML algorithms?
  4. Explainability: Can we using the trained model say something about what motion, or what part of the game, most clearly separates the two groups of players?

References:

  • Claudino, J. G., Capanema, D. D. O., de Souza, T. V., Serrão, J. C., Machado Pereira, A. C., & Nassis, G. P. (2019). Current approaches to the use of artificial intelligence for injury risk assessment and performance prediction in team sports: a systematic review. Sports medicine-open, 5, 1-12. doi: 10.1186/s40798-019-0202-3
  • Orten, K.F., Chen, B., & Helgesen, S.E.M. (2022). Can machine learning help reveal the competitive advantage for elite athletes? IN5490 proceedings 2023 (ask your supervisor for access to this paper)
  • Rajšp, A., & Fister, I. (2020). A systematic literature review of intelligent data analysis methods for smart sport training. Applied Sciences, 10(9), 3013. doi: 10.3390/appxx010005
  • Wang, Y., Zhao, Y., Chan, R. H., & Li, W. J. (2018). Volleyball skill assessment using a single wearable micro inertial measurement unit at wrist. IEEE Access, 6, 13758-13765. doi: 10.1109/ACCESS.2018.2792220
  • Wenninger, S., Link, D., & Lames, M. (2020). Performance of machine learning models in application to beach volleyball data. International Journal of Computer Science in Sport, 19(1), 24-36. doi: 10.2478/ijcss-2020-0002


 

Emneord: Machine learning applied to motion data
Publisert 11. nov. 2023 08:43 - Sist endret 11. nov. 2023 08:44

Veileder(e)

Omfang (studiepoeng)

60