Shadow-Based Lightsource Localization With Direct Camera–Lightsource Geometry

Accurate lightsource location or direction is important in shape from shading (SFS) and photometric stereo (PS). Tracking interested object by the lightsources on it is also a widely used positioning scheme. However, existing methods for lightsource localization are primarily designed for calibration. They are usually inferior in speed and accuracy, because of the intermedium geometry they introduced. Specifically, they must recover the camera–intermedium geometry before gaining the desired intermedium–lightsource or camera–lightsource geometries. In this article, we propose a lightsource localization method that can directly recover the camera–lightsource geometry. A sphere and its shadow are used as the intermediums, but their roles stay in the images. The parallax geometry of sphere is investigated to transfer the lightsource localization problem into solving the outer bitangents’ intersection of two visible conics. The solution cases, degenerated cases, error distributions, and other details are discussed and tested. The experiment has demonstrated that our method has significantly better accuracy and robustness than the state-of-the-art, while it is obviously low-cost and lightweight.