Non-uniqueness in C-arm calibration with bundle adjustment

Anastasia Konik (1) , Laurent Desbat (1) , Yannick Grondin

1 TIMC-GMCAO - Gestes Medico-chirurgicaux Assistés par Ordinateur

In this work, we consider the problem of C-arm geometric calibration with the bundle adjustment (BA) method. This method was initially developed and used in computer vision. It’s based on markers, but with unknown geometry. We don’t know the 3D positions of the markers from which we have to estimate the calibration parameters. Thus, the geometric calibration is only based on the positions of the projected markers on 2D projection images. Pollefeys et al. [1] have shown that such calibration from BA can be performed up to a similarity transformation. In our work, we present a numerical solution to the C-arm geometric calibration with the BA method. We show the non-uniqueness of the geometric calibration with the BA method for cone-beam tomography. Just like in computer vision, we can’t find the solution better than up to a similarity transformation.

Domaines: Imagerie médicale

Anastasia Konik, Laurent Desbat, Yannick Grondin. Non-uniqueness in C-arm calibration with bundle adjustment. 16th International Meeting on Fully 3D Image Reconstruction in Radiology and Nuclear Medicine, KU Leuven, Jul 2021, Leuven, Belgium. pp.299-303. ⟨hal-03419265⟩