Reconstructing atmospheric cloud particles from multiple fisheye cameras

In this article we address the problem of accurately reconstructing the 3D position and short-term motion of atmospheric cloud using passive instruments. The proposed system is based on multiple fish-eye cameras that capture daylight sky images in a synchronous manner. After introducing the inherent difficulties of the task, we propose novel methods for improving robustness and accuracy in the given application scenario. We emphasize the effectiveness of a multi-view (in our example a trinocular) camera system over a stereo camera pair with respect to reliability and precision of both 3D position and motion reconstructions. During the evaluation of the methods we address limitations and possible effective parametrization of the system, including the positioning of the cameras.