A PCA based method for image and video pose sequencing

Show full item record

Redirect: RIT Scholars content from RIT Digital Media Library has moved from http://ritdml.rit.edu/handle/1850/11991 to RIT Scholar Works http://scholarworks.rit.edu/theses/7258, please update your feeds & links!
Title: A PCA based method for image and video pose sequencing
Author: Massaro, James
Abstract: Problems exist in image sequence processing that require an ordered set of object views. In some cases, multiple angled images are acquired in random order and the angle of view information is not available. When this occurs, the poses have to be put into proper order. For example, in databases containing images of an object or scene taken over a period of time, each image pose or angled-view with respect to the camera or scene is unknown. This is important to achieve a complete or partial three-dimensional reconstruction. Other applications exist in photogrammetry, machine vision, computer-aided design, and military intelligence. The main contribution of this thesis is an automated method for ordering images of random object views. This method uses Principal Component Analysis (PCA) and a confidence metric in eigenspace. The confidence measure is based on local curvature and correlation of the estimated pose trajectory in a multidimensional manifold. The use of the confidence metric is for detecting areas in the manifold where poses appear similar and ordering becomes difficult. It has been extended for use with synchronized double and multiple camera system by providing a basis for camera selection, choosing the most salient camera view for pose ordering. By adding multiple cameras, a high pose estimation accuracy can be achieved. This thesis compares other classification and recognition methods such as the Scale Invariant Feature Transform (SIFT) and Laplacian Eigenmaps. The SIFT algorithm struggles with pose sequencing because it computes local feature spaces for each image and does not consider the entire set of images. Laplacian eigenmaps show better results for ordering, but close analysis show it is better suited for clustering poses than sequencing. Results for ordering many set of objects, theoretical development, and comparison of methods is presented in this research.
Record URI: http://hdl.handle.net/1850/11991
Date: 2010-02

Files in this item

Files Size Format View
JMassaroThesis2-2010.pdf 7.157Mb PDF View/Open

The following license files are associated with this item:

This item appears in the following Collection(s)

Show full item record

Search RIT DML

Advanced Search