Registration for outdoor systems for Augmented Reality (AR) cannot rely on the methods developed for indoor use (e.g., magnetic tracking, fiducial markers). Although GPS and the earth's magnetic field can be used to obtain a rough estimate of position and orientation, the precision of this registration method is not high enough for satisfying AR overlay. Computer vision methods can help to improve the registration precision by tracking visual clues whose real world positions are known. We have developed a system that can exploit horizon silhouettes for improving the orientation precision of a camera which is aligned with the user's view. it has been shown that this approach is able to provide registration even as a stand-alone system, although the usual limitations of computer vision prohibit to use it under unfavorable conditions. This paper describes the approach of registration by using horizon silhouettes. Based on the known observer location (from GPS), the 360 degree silhouette is computed from a digital elevation map database. Registration is achieved, when the extracted visual horizon silhouette segment is matched onto this predicted silhouette. Significant features (mountain peaks) are cues which provide hypotheses for the match. Several criteria are tested to find the best matching hypothesis. The system is implemented on a PC under Windows NT Results are shown in this paper.
