Date of Award




Document Type


Degree Name

Doctor of Philosophy (PhD)


Department of Computer Science

Content Description

1 online resource (iv, 77 pages) : illustrations (some color)

Dissertation/Thesis Chair

Siwei Lyu

Committee Members

Ming-Ching Chang, Neil V Murray


3D video, event summarization, immersive visualization, large-scale 3D reconstruction, surveillance, Image reconstruction, Three-dimensional imaging, Three-dimensional display systems, Three-dimensional modeling, Image processing, Motion perception (Vision)

Subject Categories

Computer Sciences


The Structure-from-Motion(SfM) method of constructing a 3D model from 2D images reached maturity in the last decade. However, the computational time of pairwise image feature matching dominates the whole pipeline of SfM, and grows quadratically with the number of images. So the scaling issue is still a challenging problem. This dissertation introduces an efficient framework for large-scale 3D reconstruction from a large set of photos following the SfM paradigm with divide-conquer and fusion. The main novelty is to ensure commonality from overlaps between image sets corresponding to their reconstructed models, which facilitates effective stitching and fusion. Specifically, such commonality is ensured by selecting a set of duplicated images (which are termed {\em anchor images}) in adjacent image sets prior to the 3D reconstruction. The anchor images can assist accurate fusion of the 3D models. An efficient RANSAC scheme is developed for pairwise stitching. This method is intuitively scalable to large site reconstruction via subdivision and fusion following a graph construct. Another RANSAC algorithm is developed to improve loop closure in the sequential fusion approach. As an application in visual surveillance, a new visualization system based on the large-scale 3D reconstruction is developed to render steerable 3D views of tracked targets onto a reconstructed 3D site representation.