Master Thesis Defense 2013

Master Thesis Defense:
February 21st, 2011 Time: 14:55 – 15:55
Media Room, Department of Computer Science, Gunma University Kiryu Campus
*This paper has been accepted for publication in the Asian Conference on Pattern Recognition 2013 (in conjunction with IAPR–International Association of Pattern Recognition).
Thesis: Pdf file will be available soon

“Performance Evaluation of Feature Descriptors for Application in
Outdoor-scene Visual Navigation”

Abstract: During the past decades, scientists
have made numerous efforts in developing autonomous mobile robots.
One of the popular method to be applied for visual navigation is
scene matching algorithm. The idea is to localize the robot
position by finding out matching between the input running scenes
and a set of reference images. However, to achieve an appropriate
matching between the compared images is not an easy task. A
frequent change in the illumination intensity is one of the biggest
challenge during the experiment of scene matching. Unfortunately,
there is no quantitative evaluation of feature matching
performance, especially focusing in the problem of illumination
changes. This thesis presents an investigation of a number of
popular feature detectors and descriptors in matching the outdoor
environment and observe the performance in three different lighting
scenes, i.e., sunny, cloudy daytime, and cloudy evening, captured
within the same route by the autonomous mobile robot. To give an
equality comparison, we applied the Lowe’s matching procedure
[Low04] for all compared methods. The matching percentage and ROC
curve are used for the evaluation measurements. As for the
experimental results, the hybrid performance of FAST detector and
SURF descriptor gives the best evaluation measurement by showing
the largest value of area under ROC curve. In addition, FAST, and
SURF show advantages on their speed in extracting local image
features which is favorable for real time application. On the other
hand, SIFT achieves its stability in nearly all situations. ASIFT
presents the highest number of extracted keypoints although it
suffers from the problem of computation complexity and redundant
matches. Video

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: