Several unconventional ideas for viewer/camera pose estimation are discussed. The methods proposed so far advocate the use of advanced image processing for identification and precise location of calibration objects in the images acquired, and base pose recovery on the identification of the viewing dependent deformations of these objects. We propose to more fully exploit the freedom in the design of `space fiducials' or calibration objects showing that we can build objects whose images directly encode, in easily identifiable gray-level/color or temporal patterns, the pose of their viewer. We also show how to construct high-precision fiducials, which can determine a viewing direction quite accurately when it is known to lie within a relatively narrow range.
ASJC Scopus subject areas
- Computer Vision and Pattern Recognition
- Artificial Intelligence