2026-09-18T21:56:34Zhttps://dspace.mit.edu/server/oai/requestoai:dspace.mit.edu:1721.1/417522022-01-13T07:55:03Zcom_1721.1_7582com_1721.1_7581col_1721.1_131023
Deb Roy.
Robbel, Philipp
Massachusetts Institute of Technology. Dept. of Architecture. Program in Media Arts and Sciences.
Program in Media Arts and Sciences (Massachusetts Institute of Technology)
2008-05-19T16:14:36Z
2008-05-19T16:14:36Z
2007
2007
http://hdl.handle.net/1721.1/41752
226231368
Thesis (S.M.)--Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, 2007.
Includes bibliographical references (p. 127-132).
This thesis explores how state-of-the-art object recognition methods can benefit from integrating information across multiple observations of an object. Considered are active vision systems that allow to steer the camera along predetermined trajectories, resulting in sweeps of ordered views of an object. For systems of this kind, a solution is presented that exploits the order relationship between successive frames to derive a classifier based on the characteristic motion of local features across the sweep. It is shown that this motion model reveals structural information about the object that can be exploited for recognition. The main contribution of this thesis is a recognition system that extends invariant local features (shape context) into the time domain by adding the mentioned feature motion model into a joint classifier. Second, an entropy-based view selection scheme is presented that allows the vision system to skip ahead to highly discriminative viewing positions. Using two datasets, one standard (ETH-80) and one collected from our robot head, both feature motion and active view selection extensions are shown to achieve a higher-quality hypothesis about the presented object quicker than a baseline system treating object views as an unordered stream of images.
by Philipp Robbel.
S.M.
132 p.
eng
Massachusetts Institute of Technology
M.I.T. theses are protected by
copyright. They may be viewed from this source for any purpose, but
reproduction or distribution in any format is prohibited without written
permission. See provided URL for inquiries about permission.
http://dspace.mit.edu/handle/1721.1/7582
Architecture. Program in Media Arts and Sciences.
Exploiting object dynamics for recognition and control
Thesis
application/pdf
false
Publication
unknown
unknown
<Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="1c659e74-cae7-4dcf-853e-a9ae7d1785e7">
<Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type>
<Language>eng</Language>
<Title>Exploiting object dynamics for recognition and control</Title>
<PublishedIn>
<Publication>
</Publication>
</PublishedIn>
<PublicationDate>2007</PublicationDate>
<Authors>
<Author>
<DisplayName>Robbel, Philipp</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
</Authors>
<Editors>
</Editors>
<Publishers>
<Publisher>
<DisplayName>Massachusetts Institute of Technology</DisplayName>
<OrgUnit />
</Publisher>
</Publishers>
<License>http://dspace.mit.edu/handle/1721.1/7582</License>
<Keyword>Architecture. Program in Media Arts and Sciences.</Keyword>
<Abstract>This thesis explores how state-of-the-art object recognition methods can benefit from integrating information across multiple observations of an object. Considered are active vision systems that allow to steer the camera along predetermined trajectories, resulting in sweeps of ordered views of an object. For systems of this kind, a solution is presented that exploits the order relationship between successive frames to derive a classifier based on the characteristic motion of local features across the sweep. It is shown that this motion model reveals structural information about the object that can be exploited for recognition. The main contribution of this thesis is a recognition system that extends invariant local features (shape context) into the time domain by adding the mentioned feature motion model into a joint classifier. Second, an entropy-based view selection scheme is presented that allows the vision system to skip ahead to highly discriminative viewing positions. Using two datasets, one standard (ETH-80) and one collected from our robot head, both feature motion and active view selection extensions are shown to achieve a higher-quality hypothesis about the presented object quicker than a baseline system treating object views as an unordered stream of images.</Abstract>
<Access xmlns="http://purl.org/coar/access_right"
>
</Access>
</Publication>