An issue in AR applications with Optical See-Through Head-Mounted Display (OST-HMD) is to correctly project 3D information to the current viewpoint of the user. Manual calibration methods give the projection as a black box which explains observed 2D-3D relationships well (Fig. 1). Recently, we have proposed an INteraction-free DIsplay CAlibration method (INDICA) for OST-HMD, utilizing camera-based eye tracking [7]. It reformulates the projection in two ways: a black box with an actual eye model (Recycle Setup), and a combination of an explicit display model and an eye model (Full Setup). Although we have shown the former performs more stably than a repeated SPAAM calibration, we could not yet prove whether the same holds for the Full Setup. More importantly, it is still unclear how the error in the calibration parameters affects the final results. Thus, the users can not know how accurately they need to estimate each parameter in practice. We provide: (1) the fact that the Full Setup performs as accurately as the Recycle Setup under a marker-based display calibration, (2) an error sensitivity analysis for both SPAAM and INDICA over the on-/offline parameters, and (3) an investigation of the theoretical sensitivity on an OST-HMD justified by the real measurements.
各種検索
サポート
ヘルプ