Online illumination estimation of outdoor scenes based on videos containing no shadow area

Real-time estimation of outdoor illumination is one of the key issues for ensuring the illumination consistency of augmented reality. In this paper, we propose a novel framework to estimate the dynamic illumination of outdoor scenes based on an online video sequence captured by a fixed camera. All existing approaches are based on two assumptions, i.e. there exist some shadow areas in the scene and the distribution of the skylight is uniform over the sky. Both assumptions greatly simplify the problem of illumination estimation of outdoor scenes, but they also limit the applicability as well as the accuracy of these approaches. This paper presents a new approach that breaks these two hard constraints. It recovers the lighting parameters of outdoor scenes containing no shadow area through solving a constrained linear least squares problem. By representing the skylight as a parameterized model incorporating an occlusion coefficient, the proposed approach can handle the dynamic variation of non-uniform skylight distribution. Experimental results demonstrate the potential of our approach.