Multiview Shape-From-Shading for Planetary Images

We propose and validate an algorithm for shape-from-shading that works for planetary data, with multiple input images, realistic camera models, non-Lambertian reflectance, variable albedo, uncertain camera positions and orientations, low angles of illumination, and shadows. The algorithm begins with images and existing terrain data (provided by stereogrammetry, lidar data, or even just a flat surface) which is then optimized. We show that shape-from-shading is able to recover more detail than obtained from stereo alone while eliminating stereo numerical artifacts and improving the terrain accuracy. Our software implementation is released under an open-source license as part of the NASA Ames Stereo Pipeline.