Performance evaluation of patterns for image-based 3D model reconstruction of textureless objects

This paper evaluates the performance of patterns used to solve a very challenging problem in close range photogrammetry and computer vision when the surface of the object or scene is textureless. Three dimensional surface modeling from arbitrary viewpoints is an active field of research due to its wide range of applications. However, structure-from-motion, a common approach for surface modeling of the objects that are not well textured fails due to insufficient discriminative features in the images and hence results in incomplete and inaccurate three dimensional model of the surface. Mainly, two approaches have been used widely for 3D reconstruction of such kind of objects. First uses a structured light or coded pattern, and second uses a random pattern that provides artificial markers on the surface of interesting object. In this paper, second approach is implemented that helps point-based features, such as SIFT, to find discriminative features from arbitrary viewpoints taken from the same object surface. We evaluate the performance of patterns with respect to quality of reconstruction of the surface of a textureless object. At the end, a comparison scheme between reconstructed model and the ground truth data is also presented and results are evaluated.