ABSTRACT
We address here the so-called static registration errors. Earlier systems used a priori calibration of the camera parameters and recovered the ob server’s pose with a magnetic tracker or infrared tracker with landmarks on the observer. Such systems are restricted to close-range environments without magnetic interferences. Vision techniques can alleviate these weak nesses and offer a kind of direct feedback; the image data used for visualiza tion and merging are also used for solving the registration problem. This fact was recognized early by Bajura [Bajura 95] and Mellor [Mellor 95]. Both of their approaches, as well as the more sophisticated one proposed by Roller et al. [Roller 97], use off-line calibration of intrinsic camera parameters and reference landmarks which are accurately distributed in space. Although these algorithms can perform very accurately, they suf-
fer from practical restrictions: they cannot improve the initial estimation of camera parameters and require a distribution of fiducial targets with known relative positions. In long-range, outdoor terrain applications, such a landmark configuration is difficult to achieve.
