Combining Positions and Photometric Stereo Normals in Linear Time with Belief Propagation

Abstract

To increase the accuracy of 3D scanners, photometric stereo scanning is combined with structured light with a new method that scales well with high-resolution cameras. This new method defines a Markov random field initially with the low-frequency data of the structured light scan and applies the data of the photometric scan locally. In an experiment, the resulting scans are found to be more accurate than the original scan. Also, a new method is proposed to calibrate the positions of lights for photometric scanning by using a rotating platform. The technique finds the reflections of the lights in reflective spheres and traces back the positions of the lights. It then rotates the platform and repeats the experiment. While the rotation is found to improve the result, the sequence in total is found to perform badly.