Machine learning in cycling investigations into classification, computer vision and language modelling in cycling aerodynamics

This thesis pursues four areas of novel research, investigations into Shuffling, the classification of rider position on the bike, the estimation of a rider's pose on the bike through the use of computer vision and the application of Language Modelling within the domain of cycling. Each area of research is of interest; however, the motivation for these studies was inspired by the Body Rocket system, with the overarching goal of lowering the barrier of entry at the professional level. The Body Rocket system is a novel drag-force measurement device from Body Rocket Ltd (Sussex, UK) and comprises force sensors at the contact points of a bike, enabling direct measurement of a rider's drag force alongside other performance metrics. As a result, the first study aimed to validate one of these performance metrics: a rider's position on the saddle. After validating a rider's position on the saddle, the second study applied machine learning to all Body Rocket data to determine an athlete's full position on the bike. The third study took the second study a step further, moving away from the Body Rocket system as the data acquisition tool and using cameras to investigate continuous identification of a rider's pose on the bike using computer vision. The final study has the potential to integrate all of these studies, as each alone could require interpretation by a coach or professional. As a result, the final study aimed to investigate whether, using Language Modelling, a domain-specific chatbot for cycling aerodynamics could be created and evaluated. All of these studies demonstrate the potential of these applications and are preliminary works.