On the use of shadows in stance recovery

The image of an object and of the shadow it casts on a planar surface provi des important cues for three-dimensional (3D) stance recovery. We assume th at the position of the plane on which the shadow lies with respect to a pin hole camera is known and that the position of the light source is unknown, If the light source is sufficiently far away that parallel projection may b e assumed, then knowledge of two point correspondences between images of fe ature points and Images of their shadows is enough to determine the positio n of the object and the direction of the light source. If the light source is close enough that the shadow points are obtained via perspective project ion, then there is a one-parameter infinite family of solutions for the pos ition of the object and the light source. Determining the stance of an obje ct is highly sensitive to noise, so we provide algorithms for stance recove ry that take into account known information about the object. In our experi ments, the errors for the location of the 3D feature points obtained by the se algorithms are generally less than 0.2% times the error in pixels in the Image points and the errors for the 3D directions of the links is roughly 0.04 degrees times the error in pixels, normalized by the distance to the o bject from the camera and the length of the link. (C) 2001 John Wiley & Son s, Inc. Int J Imaging Syst Technol, 11, 315-330, 2000.