A novel approach is proposed that relates the classical two-dimensional Hough space to a different Hough space embedding 3D information about the poses of planar objects in a single gray-level image. The Hough transform is used to detect rectilinear segments that, suitably grouped into a bounded figure, constitute a planar surface. Then, a pure geometrical mechanism is used to map the numerical Hough space representation of the image into a similar representation in a reference system that is fixed with respect to the surface. The object pose to be estimated is computed by comparing the numerical representations of the test and model images (usually in a fronto-parallel view) in the same space invariant to the object pose.
