Authors:
Ahmad Hasasneh
1
;
Emmanuelle Frenoux
1
and
Philippe Tarroux
2
Affiliations:
1
Paris Sud University and LIMSI-CNRS, France
;
2
LIMSI-CNRS and Ecole Normale Supérieure, France
Keyword(s):
Semantic Place Recognition, Restricted Boltzmann Machines, Deep Belief Networks, Bag-of-Words, Softmax Regression.
Related
Ontology
Subjects/Areas/Topics:
Autonomous Agents
;
Informatics in Control, Automation and Robotics
;
Mobile Robots and Autonomous Systems
;
Perception and Awareness
;
Robotics and Automation
;
Vision, Recognition and Reconstruction
Abstract:
This paper presents a novel approach for robot semantic place recognition (SPR) based on Restricted Boltzmann Machines (RBMs) and a direct use of tiny images. RBMs are able to code images as a superposition of a limited number of features taken from a larger alphabet. Repeating this process in a deep architecture leads to an efficient sparse representation of the initial data in the feature space. A complex problem of classification in the input space is thus transformed into an easier one in the feature space. In this article, we show that SPR can thus be achieved using tiny images instead of conventional Bag-of-Words (BoW) methods. After appropriate coding, a softmax regression in the feature space suffices to compute the probability to be in a given place according to the input image.