Self calibrating lenticular markers and articulated pose estimation

摘要

Chromo-coded lenticular markers have a color that depends on their viewing angle, which gives stronger constraints than typical fiducial markers for pose-estimation. In this paper, we work to make chromo-coded lenticular markers practical to use on a wide scale by solving their self-calibration problem: given freehand pictures taken of markers placed on an object by hand, solve for their relative position, orientation, and the colorto-angle relationship. We pose this as an optimization problem and show that standard tools give good solutions with naive initialization conditions. Using this calibration with a physical prototype, we show substantial improvements for pose estimation of rigid and simple articulated objects over standard fiducial markers.