A markerless virtual coach for home-based exercise: movement guidance and training intensity feedback

Home-based exercise often lacks professional supervision, making it difficult for users to assess movement quality or regulate training intensity during practice. Existing research on home-based exercise guidance typically focuses on a single aspect of performance or requires additional equipment, which limits its applicability in everyday settings. In this work, we develop a markerless Virtual Coach for home-based exercise guidance. The system relies on a single standard camera and extracts kinematic features through human pose estimation to enable the simultaneous assessment of movement quality and training intensity. Assessment outputs are presented through visualization-based feedback, providing multiple forms of information and brief suggestions to achieve exercise guidance. For system performance, the movement quality scores generated by the model showed consistency with expert reference ratings, and the estimated training intensity values showed acceptable agreement with heart rate-derived reference values. Concerning system usability, a randomized controlled study involving 64 participants was conducted over eight weeks. The experimental group showed greater improvements in movement standardization and a descriptively higher regulatory efficacy score than the control group. The findings suggest the potential of a markerless, visualization-based framework for home-based exercise guidance.