GPU-assisted 3D Pose Estimation under realistic illumination

Abstract

This paper describes an approach which combines computer vision methods with techniques from the area of computer graphics. This method which is called analysis-by-synthesis explicitly seeks for consideration of environmental information in order to improve the resulting estimation of the 3D camera pose. In this paper, two different kinds of pose estimation will be presented. The first approach uses intensity-based methods and the second one is a feature point-based approach. The described approaches are based on a GPU-assisted rendering considering the real world illumination. These real world lighting conditions are captured using a HDR sampling technique. The results of this GPU-assisted approach are that both methods, the intensity-based as well as the feature point-based method, achieve better results in terms of a more robust and stable 3D camera pose under consideration of the real environmental information.