Estimation of upper body posture while riding a bicycle based on load measurement

Abstract

Knowing one’s own movement and form in a sport, and getting close to the ideals, is essential for improvement in that sport. In cycling, pedaling motion and upper body posture are especially important for the rider to understand. Pedaling measurement methods are widely available. On the other hand, there is no easy way for riders to know their upper body posture. Therefore, we estimate upper body posture by using sensors that can be mounted on a bicycle. In this study, we measured the load on the handle and estimated the upper body posture based on the load. We created neural network models with the load as the input and the joint angle of the upper body as the output. The estimation accuracy was high when the model was based on personal data, but there was a problem with estimation accuracy when the model was based on other people’s data.