In the fields of photogrammetry and computer vision, three-dimensional (3D) shape recovery has remained an active research topic over the past decades. Accompanied by the boost in the development of sensor technology, considerable efforts have been expanded to image-based shape recovery. However, most of the approaches still rely on calibrated cameras with known orientations to establish the transformation between the object space and image space, keeping them from practical when the camera interior and exterior parameters are unavailable.
In this research, a novel approach is developed for the recovery of 3D object shape using uncalibrated multiple-view images. The approach is based on the assumption that the 3D projective space is composed of 2D discrete projective subspaces. In the designed framework, a 2D subspace corresponds to a set of hypothetical planes which creates cross-sections by slicing the objects in the scene. The images of such cross-sections are obtained by planar projective transforms which are estimated from the points defined on these hypothetical planes. A stack of these cross-sections provides a projective recovery of the 3D object shape. The resulting recovery becomes an affine or metric recovery when stack of cross-sections are transformed to an affine or ortho-rectified image respectively, or when absolute ground information is provided.
In this dissertation, two possible formations of subspaces are proposed, and several experiments using image sets of different characteristics are conducted to evaluate the performance of the proposed method. Generated 3D shapes, when qualitatively examined or quantitatively compared against ground-truth, show promising recovery performance.