An investigation looking into the application of Artificial Intelligence for live positional tracking using on-bike aerodynamic sensors. In this study data from the wind tunnel and outdoor conditions were collected using a system from Body Rocket Ltd. By applying a Gradient Boosted Machine to the force and moment data the discrete positons of a rider on a bike were successfully identified for a rider in the wind tunnel within the dataset it was trained on to 100% accuracy. When applied to blind data collected from the wind tunnel the models accuracy was limited with a performance of 45%, however, with a new model built around data collected outdoors the accuracy of this model was found to be 100%. Overall this study finds that with machine learning techniques it is possible to identify positions of a rider on a bike just from the raw force data and with further research there is potential to determine a continuous range of positions outside of the discrete positons investigated in this study.
© Copyright 2024 Journal of Science and Cycling. Cycling Research Center. All rights reserved.