Electrical and Computer Engineering ETDs

Publication Date

Fall 11-21-2018

Abstract

Motion estimation has been dominated by time domain methods such as block matching and optical flow. However, these methods have problems with multiple moving objects in the video scene, moving backgrounds, noise, and fractional pixel/frame motion. This dissertation proposes a frequency domain method (FDM) that solves these problems. The methodology introduced here addresses multiple moving objects, with or without a moving background, 3-D frequency domain decomposition of digital video as the sum of locally translational (or, in the case of background, a globally translational motion), with high noise rejection. Additionally, via a version of the chirp-Z, fractional pixel/frame motion detection and quantification is accomplished. Furthermore, images of particular moving objects can be extracted and reconstructed from the frequency domain. Finally, this method can be integrated into a larger system to support motion analysis.

The method presented here has been tested with synthetic data, realistic, high fidelity simulations, and actual data from established video archives to verify the claims made for the method, all presented here. In addition, a convincing comparison with an up-and-coming spatial domain method, incremental principal component pursuit (iPCP), is presented, where the FDM performs markedly better than its competition.

Keywords

frequency domain methods, decomposition of digital video, segmentation of moving objects in video, fractional pixel/frame motion analysis

Document Type

Dissertation

Language

English

Degree Name

Computer Engineering

Level of Degree

Doctoral

Department Name

Electrical and Computer Engineering

First Committee Member (Chair)

Marios Pattichis

Second Committee Member

Balasubramaniam Santhanam

Third Committee Member

Christos Christodoulou

Fourth Committee Member

Constantinos Pattichis

Share

COinS