Statistical Pose Averaging with Non-isotropic and Incomplete Relative Measurements

摘要: In the last few years there has been a growing interest in optimization methods for averaging pose measurements between set of cameras or objects (obtained, instance, using epipolar geometry estimation). Alas, existing approaches do not take into consideration that might have different uncertainties (i.e., noise be isotropically distributed), they incomplete (e.g., known only up to rotation around fixed axis). We propose Riemannian framework which addresses these cases by covariance matrices, and test it on synthetic real data.