3D Model Creation Using Self-Identifying Markers and SIFT Keypoints

Abstract

3D object modeling can be accomplished using fiducial markers and/or feature detectors. Fiducial markers provide high reliability of detection, however, it is undesirable to cover an object to be modeled with markers. Feature detectors can find correspondences between images but they cannot always be relied on to be usable for camera localization. A method is shown that uses the strengths of both to automatically create 3D models of object as well as simultaneously calibrating the camera. Self-identifying fiducial markers are used in arrays to localize the camera pose for each image and SIFT features are used to find and match object features between images. Tetrahedrons formed by Delaunay triangulation of the 3D SIFT points are carved to the model. A system is shown where 3D models are generated automatically of an object placed on a marker array simply by capturing a set of images from uncontrolled locations from a camera with unknown intrinsic parameters