Online Illumination Planning for Shadow-Robust Photometric Stereo

Hirochika Tanikawa, Ryo Kawahara, Takahiro Okabe

Abstract: Photometric stereo is a technique for estimating normals of an object surface from its images taken under different light source directions. In general, photometric stereo suffers from shadows, because almost no information on surface normals is available from shadowed pixels. In this paper, we propose an illumination planning for shadow-robust Lambertian photometric stereo; it optimizes the light source directions adaptively for an object of interest, because cast shadows depend on the entire shape of the object. More specifically, our proposed method iteratively adds the optimal light source for surface normal estimation by taking the visibility and linear independence of light source directions into consideration on the basis of the previously captured images of the object. We implemented our illumination planning with a programmable light source in an online manner, and achieve shadow-robust surface normal estimation from a small number of images.