Exploring Somersault Feedback Design using OpenPose: A User Centered Approach

Providing effective feedback is crucial for the improvement of athletic performance, particularly for complex movements like somersaults. Various technological tools are available, though complexity and costs renders them often inaccessible to amateurs. Rapid movements and complex poses from a challenging situation for feedback since they are difficult to observe for both humans and technological systems. This research explores the utility of a easy to use sports feedback system for a complex movement: the somersault. The system uses OpenPose, which is a marker less motion capture system that detects body parts from footage. A contextual exploration was performed both through literature and interviews with gymnasts. After, the performance of OpenPose on a somersault was evaluated qualitatively and quantitatively. Together with gymnastics trainers, a diverging and converging ideation phase led to the development of six low fidelity prototypes for feedback. The designs were reflected on with gymnasts and refined into four high fidelity prototypes, which again were evaluated. OpenPose performs poorly on somersaults, this was accounted for by using a data enhancement strategy as well as a user centered strategy which aimed to bypass the limitations of data by design choices. Altogether, this research provide three main contributions. Firstly, a quantitative analysis of OpenPose’s performance on somersaults is given. Additionally, design guidelines for sports technology are presented, including strategies of coping with incomplete or inaccurate data. Finally, four Hi-Fi prototypes were implemented and evaluated, forming a basis for novel somersault learning approaches. In conclusion, OpenPose is a promising system for developing feedback systems for sports.