Advisor(s)

Octavia Camps

Contributor(s)

Mario Sznaier, Gilead Tadmor

Date of Award

2010

Date Accepted

4-2010

Degree Grantor

Northeastern University

Degree Level

M.S.

Degree Name

Master of Science

Department or Academic Unit

College of Engineering, Department of Electrical and Computer Engineering

Keywords

electrical engineering, computer engineering, Hankel rank minimization, video registration

Subject Categories

Automatic control, Global Positioning System, Roads--Inspection

Disciplines

Electrical and Computer Engineering | Engineering

Abstract

Knowing the precise location where data is collected is a key feature for automated road inspection, including pavement surface and subsurface condition evaluation. The accuracy of commercially available GPS systems (5 to 10 meters) is inadequate because data for road inspection is collected at 2.5cm or smaller intervals with sensors mounted on vehicles moving at 30 mph or faster. Video data recorded from a camera mounted on the vehicle can provide additional data registration to landmarks in the scene and previously recorded data. However, using video data poses additional challenges including the collection, processing and visualization of vast amounts of data, temporal and spatial registration among different cameras used at different times, natural clutter from unstructured environments, noise, and missing key data due to occlusion or dropped frames.

In this thesis, an image registration system composed of GPS, video tracking and tools from system theory is presented to register video sequences taken from moving vehicles achieving an accuracy of less than 15 cm. After narrowing down the searching scope with GPS, Scale-Invariant Feature Transform (SIFT) features are detected in the current and reference frames. Then, corresponding features are matched to obtain the projective transformation between each pair of frames where outliers are filtered by RANSAC. Finally, the current frame is registered to the reference frame whose geoinformation has been precisely known. In order to improve robustness and tracking accuracy, a dynamic system minimizing the rank of the Hankel matrix built by homography matrices in previous frames is introduced to predict the one in the following frame.

Document Type

Master's Thesis

Rights Information

copyright 2010

Rights Holder

Bo Li


View Full Text

Click button above to open, or right-click to save.

Share

COinS