Method and system for sensing 3d shapes of objects with specular and hybrid specular-diffuse surfaces

Abstract

Surface normals and other 3D shape descriptors are determined for a specular or hybrid specular-diffuse object. A camera records an image of a smoothly spatially-varying pattern being reflected in the surface of the object, with the pattern placed at an initial position. The camera then records multiple images of the pattern undergoing a sequence of subsequent displacements to a final position distinct from the initial position. For a pixel in the images, the pattern displacement that corresponds to the minimum difference between the pixel value in the initial image and any of the final images is determined. The incident ray that strikes the surface of the object at the point being imaged by the pixel is then determined using the determined pattern displacement. The surface normal at that same surface point is then determined using the determined incident ray. Other 3D shape descriptors are determined from the surface normals of the object, such as principal curvatures, or the principal axes of a parametric surface, or the 3D shape, or the object pose.