Shape in a Box

Abstract

Many techniques have been developed in computer vision to recover three-dimensional shape from two-dimensional images. These techniques impose various combinations of assumptions/restrictions of conditions to produce a representation of shape (e.g. a depth/height map). Although great progress has been made it is a problem which remains far from solved, with most methods requiring a non-passive imaging environment. In this paper we develop on a variant of photometric stereo called “Shape from color” (SFC). We remove the restriction of known, direct light sources by exploiting mutual illumination; we simply take pictures of objects within a colourful box, hence “Shape in a Box”. We discuss the engineering process used to develop our set-up and demonstrate experimentally that our passive imaging environment recovers shape to the same accuracy as SFC. A second contribution of this paper is to benchmark our approach using real objects with known ground truth, including some 3D printed objects.