Abstract

Visual tracking has become an increasingly important topic of research in the

field of Computer Vision (CV). There are currently many tracking methods based on the

Detect-then-Track paradigm. This type of approach may allow for a system to track a

random object with just one initialization phase, but may often rely on constructing

models to follow the object. Another limitation of these methods is that they are

computationally and memory intensive, which hinders their application to resource

constrained platforms such as mobile devices. Under these conditions, the

implementation of Augmented Reality (AR) or complex multi-part systems is not

possible.

In this thesis, we explore a variety of interest point descriptors for generic object

tracking. The SIFT descriptor is considered a benchmark and will be compared with

binary descriptors such as BRIEF, ORB, BRISK, and FREAK. The accuracy of these

descriptors is benchmarked against the ground truth of the object's location. We use

dictionaries of descriptors to track regions with small error under variations due to

occlusions, illumination changes, scaling, and rotation. This is accomplished by using

Dense-to-Sparse Search Pattern, Locality Constraints, and Scale Adaptation. A

benchmarking system is created to test the descriptors' accuracy, speed, robustness, and

distinctness. This data offers a comparison of the tracking system to current state of the

art systems such as Multiple Instance Learning Tracker (MILTrack), Tracker Learned

Detection (TLD), and Continuously Adaptive MeanShift (CAMSHIFT).

Library of Congress Subject Headings

Computer vision; Automatic tracking--Data processing; Optical pattern recognition; Machine learning

Publication Date

5-2014

Document Type

Thesis

Student Type

Graduate

Degree Name

Computer Engineering (MS)

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Andreas Savakis

Advisor/Committee Member

Raymond Ptucha

Advisor/Committee Member

Sonia Lopez Alarcon

Comments

Physical copy available from RIT's Wallace Library at TA1634 .S736 2014

Campus

RIT – Main Campus

Plan Codes

CMPE-MS

Share

COinS