Real-time camera pose in a Room

Abstract

Many applications of computer vision require camera pose in real-time. We present a new, fully mobile, purely vision-based tracking system that works indoors in a prepared room, using artificial landmarks. The main contributions of the paper are: improved pose accuracy by subpixel corner localization, high frame rates by CMOS image aquisition of small subwindows, and a novel sparse 3D model of the room for a spatial target representation and selection scheme which gains robustness. © Springer-Verlag Berlin Heidelberg 2003.