Scene geometry from several partly cloudy days

Nathan Jacobs, Scott Workman, Richard Souvenir

Abstract: We describe new methods for estimating the geometry of an outdoor scene from either (1) a single calibrated camera or (2) a network of arbitrary, uncalibrated cameras. Our methods do not require camera motion nor overlapping fields of view. We use simple geometric constraints based on appearance changes caused by cloud shadows and combine constraints from multiple days. We describe a linear method for calibrated cameras and a nonlinear method for an uncalibrated imaging system. We define these geometric constraints, describe new algorithms, and demonstrate the features of these algorithms on real and synthetic scenes.