FlowAR: How Different AR Visualizations of Online Fitness Videos Support Flow for At-Home Yoga

Problem — a fixed screen breaks the flow of home yoga

Online fitness videos are a popular, affordable way to exercise at home, but they share one flaw: to keep the screen in view, you twist your body and break the continuous flow of movement — exactly the two things yoga depends on. Earlier research moved the display with the user or recreated the instructor’s first-person view, but those approaches need motion-captured instructors and offer only a narrow viewing window, so they don’t scale to the large library of fitness videos already online, or to complex full-body poses.

Solution — always-present AR screen overlays

FlowAR is an augmented-reality system that renders online fitness videos as virtual screen overlays around the user, so the tutorial is always in view without forcing the body out of position. The question then becomes: how should that screen be placed? We designed four layouts — two static and two dynamic.

① Front — a typical video-in-front layout (baseline).

② Circular — screens around the user, viewed selectively by posture.

③ User-anchored — the screen follows the user’s gaze like a head-up display.

④ Trainer-anchored — the screen sits where the virtual trainer is looking. The trainer can come from motion capture or from AI pose estimation on an ordinary video.

Two studies

Study 1 (motion-capture lab, 16 participants) compared all four layouts. We motion-captured both an expert instructor and each user, then mapped user motion onto the expert’s with dynamic time warping to measure timing and joint-angle errors. Static layouts pulled the gaze away and disturbed the flow; the two dynamic layouts produced fewer timing and posture errors, and all 16 users preferred them.

Study 2 (home-like lab, 12 participants) tested the two dynamic layouts using YouTube videos with motion generated by AI pose estimation (Google MediaPipe). Users performed equally well — about 10° mean angle error and half a second of timing error — and most didn’t even notice the estimation noise. Preference split by expertise: beginners liked the Trainer-anchored layout’s natural guidance, while experienced yogis preferred the User-anchored layout’s convenience.

Takeaway

Dynamic, always-present overlays let people follow ordinary online videos with better motion flow — and with AI pose estimation, the trainer-anchored approach works outside a motion-capture studio. There’s no single best layout; each trades off guidance against convenience.