Doable in real time with a compact pipeline and a small nonlinear solver per frame. I'll give a complete, practical approach you can implement right away:
-
assumptions I’ll make (don’t worry if some don’t hold; I include fallbacks):
- camera intrinsics (K) are known (focal length (f), principal point). If not, I show a fallback.
- each bead has known physical radius (R) and the known center-to-center spacing sequence (s_i).
- each bead has a unique color and the colors appear in the same sequence left→right in the string (you said sequence is known).
