Illumination estimation based on human visual characteristics for AR

Realistic image synthesis that users do not feel the artificiality is important for AR(Augmented Reality) technology. Although illumination environment in the real world needs to be estimated in order to realize the photometric consistency, the estimation precision is enough to satisfy human perception. A questionnaire is taken from users to analyze factors which affect the users to feel artificiality for synthetic AR images in advance. In the result of the questionnaire, more than 73% of the users perceived photometric consistency to shadow deviation of 30 degrees or less. This paper proposes a simple illumination estimation based method on the human visual characteristics. A light source position is estimated by using a monocular camera with a cylindrical object. First, images with and without a cylindrical object are captured. Next, the cylindrical object region and its cast shadow region are extracted by using subtraction of these two images. Finally, the proposed method estimates the direction of the light source from the extracted regions. AR images are generated with the estimated light source direction. In the user survey, more than 70% of users feel the AR images consistent. Usefulness of the proposed method is presented by the user survey.