Thesis (M. Phil.)--Hong Kong University of Science and Technology, 2003. / Includes bibliographical references (leaves 60-63). Also available in electronic version. Access restricted to campus users.

Thesis (Ph. D.)--Carnegie Mellon University, 2003. / "October 2003." Includes bibliographical references.

Thesis (M. Phil.)--Hong Kong University of Science and Technology, 2003. / Includes bibliographical references (leaves 64-66). Also available in electronic version. Access restricted to campus users.

Thesis (Ph.D.)--Hong Kong University of Science and Technology, 2003. / Includes bibliographical references (leaves 122-127). Also available in electronic version. Access restricted to campus users.

Includes bibliographical references (p. 59-66).

Thesis (M. Phil.)--Hong Kong University of Science and Technology, 2003. / Includes bibliographical references (leaves 123-127). Also available in electronic version. Access restricted to campus users.

published_or_final_version / Electrical and Electronic Engineering / Doctoral / Doctor of Philosophy

The task of multi-body motion segmentation refers to segmenting feature trajectories in a sequence of images according to their 3D motion affinity without knowing the number of motions in advance. It is critical for understanding and reconstructing a dynamic scene. This problem essentially consists of two subproblems, segmenting features and detecting the number of motions. While the state-of-the-art LBF algorithm achieves segmentation accuracy as high as 96.5%, it is still disturbed by a phenomenon called over-locality. A novel mode seeking algorithm with an adaptive distance measure is proposed to avoid this problem, and improves the accuracy to 98.1%. The LBF algorithm is incapable of detecting the number of motions itself. A randomized version of the mode seeking algorithm is presented, which could detect the number as well as preserve satisfactory segmentation accuracy. To detect the number of motions, a kernel optimization method locates it via kernel alignment. However, it suffers from over-locality and over-detects the number of motions. An intersection measure and two mutual information measures are presented to solve this problem. Using these measures, the proposed clustering consensus framework recasts the motion number detection problem to a clustering consensus problem. It extends the kernel optimization method from two-clustering consensus to multiple-clustering consensus. A large number of experiments and comparisons have been done, and convincing results are obtained. / published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy

published_or_final_version / Electrical and Electronic Engineering / Doctoral / Doctor of Philosophy

This thesis introduces novel methods for estimating the poses of a reference plane from its reflections for specular surface recovery. Traditional methods for specular surface recovery usually introduce a second camera and an auxiliary calibration pattern to calibrate the poses of the reference plane with respect to the camera. The calibration procedures are comparatively tedious. Auto-estimating the poses of the reference plane is therefore an appealing problem. In the first part of this thesis, two novel and practical methods are proposed to recover the poses of a moving reference plane from its reflections produced on the specular surface given its initial position. As for the first approach, the reference plane is constrained to undergo an unknown pure translation. By observing the reflections of the moving reference plane produced on the specular surface, a closed form solution is derived for recovering the unknown translation. Degenerate cases in which the proposed method fails are studied. As for the second approach, the constraint on the motion of the reference plane is removed, and the reference plane can move freely. Again, a closed form solution is derived for the unknown motion defined by a rotation matrix and a translation vector, and the degenerate cases are also presented. The thesis finally considers the problem for auto-calibrating the reference plane against the camera for specular surface recovery without prior knowledge of the reference plane’s initial position. By only observing its reflections produced on the specular surface, the poses of the reference planes can be recovered in two steps. First, by applying a collinearity constraint to the specular correspondences, a simple closed form solution is derived for recovering the poses of the reference plane relative to its initial pose. Second, by applying a ray incidence constraint to the incident rays formed by the specular correspondences and the visual rays cast from the image, a closed form solution is derived for recovering the poses of the reference plane relative to the camera. The shape of the specular surface then follows. / published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy