Indexing Images for Visual Memory by Using {DNN} Descriptors -- Preliminary Experiments
Result description
Visual memory in mobile robotics is important to make the local- ization of a robot robust to situations, when GPS or similar localization methods are not available. Unlike many conventional approaches us- ing local features, we use a holistic method that employs deep neural networks (DNNs) to calculate a global descriptor of the whole image. We consider a scenario in which a robot equipped with an omni-directional camera calculates and stores DNN descriptors of images together with the positions as itmoves in the environment. When the position is unknown to the robot, the algorithm estimates it for a given omnidirectional image by matching it with the most similar database image. We compared our approach with a recently tested GIST-based approach onthe same dataset and we found out that the DNN-based approach yields better results. The experiments also show that the DNN-based algorithm is quite robust to partial occlusion, rotation and changes in lighting conditions.
Keywords
The result's identifiers
Result code in IS VaVaI
Result on the web
http://cmp.felk.cvut.cz/pub/cmp/articles/svoboda/Derner-TR-2014-25.pdf
DOI - Digital Object Identifier
—
Alternative languages
Result language
angličtina
Original language name
Indexing Images for Visual Memory by Using {DNN} Descriptors -- Preliminary Experiments
Original language description
Visual memory in mobile robotics is important to make the local- ization of a robot robust to situations, when GPS or similar localization methods are not available. Unlike many conventional approaches us- ing local features, we use a holistic method that employs deep neural networks (DNNs) to calculate a global descriptor of the whole image. We consider a scenario in which a robot equipped with an omni-directional camera calculates and stores DNN descriptors of images together with the positions as itmoves in the environment. When the position is unknown to the robot, the algorithm estimates it for a given omnidirectional image by matching it with the most similar database image. We compared our approach with a recently tested GIST-based approach onthe same dataset and we found out that the DNN-based approach yields better results. The experiments also show that the DNN-based algorithm is quite robust to partial occlusion, rotation and changes in lighting conditions.
Czech name
—
Czech description
—
Classification
Type
Vsouhrn - Summary research report
CEP classification
JD - Use of computers, robotics and its application
OECD FORD branch
—
Result continuities
Project
—
Continuities
S - Specificky vyzkum na vysokych skolach
Others
Publication year
2014
Confidentiality
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů
Data specific for result type
Number of pages
16
Place of publication
Praha
Publisher/client name
Center for Machine Perception, K13133 FEE, Czech Technical University
Version
—
Result type
Vsouhrn - Summary research report
CEP
JD - Use of computers, robotics and its application
Year of implementation
2014