MCI-Professor Bernhard Hollaus from the Department of Medical & Health Technologies and Head of the Health Tech at MCI recently presented his research findings on machine learning in cross-country skiing at the European College of Sport Science (ECSS) Congress 2022 in Seville, Spain.
Bernhard Hollaus on his visit to the ECSS Congress in Sevilla:
The ECSS Congress was a great success, as every year. Thousands of researchers from all over the world discussed research questions in the field of sports over several days, attended presentations or held presentations themselves.
The use of Artificial Intelligence in sports was one of the most discussed topics during the congress. I represented the MCI at the ECSS with my research on Machine Learning in Cross-Country Skiing. In professional sports, the analysis and evaluation of data are becoming more and more significant. No matter if it is talent acquisition, optimal training or the performance analysis of athletes. Based on the data obtained, it is possible to improve every aspect as well as the success of the athletes. That is especially true for cross-country skiing.
During my project, I extracted key images from training videos of different perspectives that match a reference motion pattern of the Skating 1-1 technique. For this, I defined five major poses from the Skating 1-1 technique that are used for the technical analysis. These 5 poses are extracted automatically from the videos. A series of algorithms were developed to analyse and assign these five poses.
Therefore, I used Yolo3 to analyze each frame of the entire video to extract the athlete within the respective frame. The result was a range of coordinates for each frame. After extraction with the algorithm "Open Pose", it was possible to determine 25 coordinates of the athlete's pose.After concatenating the data of a frame into a vector, a trained convolutional neural network is able to predict whether the frame is likely to be of pattern 1, 2, 3, 4, 5, or none of them. By doing this for all images on the video, the algorithm can assign the five most suitable images to five poses, allowing optimization in all aspects of cross-country skiing.
I was delighted that over 50 people attended my presentation to listen to my research on machine learning.
For more information on my research and its findings, take a look at the paper or visit our Health Tech website.
Impressions from the ECSS Congress 2022. ©Hollaus
