Towards an Optimal Display of Superimposed Avatars for Motor Feedback

Abstract Visual motor feedback can support users doing exercises for fitness or rehabilitation. One of the most prevalent visual cues for motor feedback are superimposed human avatars. This work presents novel methods and valuable guidelines for optimizing the display of motor feedback using superimposed avatars. As a foundation, we explain how superimposed human skeleton-like avatars have to be registered according to the performed exercise. This not only helps to raise the quality of technology-supported feedback but also improves the general understanding of feedback provided by superimposed avatars. These foundational guidelines can be used as a base for implementation or further research. Given two registered avatars, we propose a novel method for real-time viewpoint selection. This method uses principal component analysis (PCA) to calculate viewpoints considering feedback (here superimposed avatars). The resulting camera movement is continuous, smooth, and faster than methods found in literature. This allows for real-time use or the viewpoint selection for superimposed avatars in exercise videos. A user study with 39 participants was conducted, verifying our basic assumptions and showing that our algorithm was preferred over the methods found in the literature. Together, the registration and view selection methods provide a powerful resource for optimizing the display of superimposed avatars. Furthermore, each of these methods can be used individually, providing additional value.