Determining a Depth Map using a Dual Photometric Stereo

Abstract

This paper describes a method for determining a depth map from a pair of surface-orientation maps obtained by a dual photometric stereo. A photometric stereo system determines surface orientations by taking three images from the same position under different lighting conditions, based on the shading information. A photometric stereo system can deter mine surface orientations very rapidly, but cannot determine absolute depth values. This paper proposes a dual photomet ric stereo system to obtain absolute depth values.

A dual photometric stereo generates a pair of surface-ori entation maps. Then, the surface-orientation maps can be segmented into isolated regions with respect to surface orien tations, using a geodesic dome for grouping surface orienta tions. The resulting left and right regions are compared to pair corresponding regions. The following three kinds of constraints will be used to search for corresponding regions efficiently: a surface-orientation constraint, an area con straint, and an epipolar constraint. Region matching is done iteratively, starting from a coarse segmented result and pro ceeding to a fine segmented result, using a parent-children constraint. The horizontal difference in the position of the center of mass of a region pair determines the absolute depth value for the physical surface patch imaged onto that pair. This system takes only a few minutes on a Lisp machine to determine an absolute depth map in complicated scenes and could be used as an input device for a bin-picking system.