Novelty detection using graphical models for semantic room classification

Authors:
André Susano Pinto;Andrzej Pronobis;Luis Paulo Reis
Affiliations:
Dep. Informatics Engineering, Faculty of Engineering, University of Porto and Centre for Autonomous Systems, The Royal Institute of Technology, Stockholm, Sweden;Centre for Autonomous Systems, The Royal Institute of Technology, Stockholm, Sweden;Dep. Informatics Engineering, Faculty of Engineering and Artificial Intelligence And Computer Science Lab., University of Porto, Porto, Portugal
Venue:
EPIA'11 Proceedings of the 15th Portugese conference on Progress in artificial intelligence
Year:
2011

Citing 8
Cited 1

Contextual Priming for Object Detection

International Journal of Computer Vision
Novelty detection: a review—part 1: statistical approaches

Signal Processing
Factor Graphs for Region-based Whole-scene Classification

CVPRW '06 Proceedings of the 2006 Conference on Computer Vision and Pattern Recognition Workshop
TOOMAS: interactive shopping guide robots in everyday use - final implementation and experiences from long-term field trials

IROS'09 Proceedings of the 2009 IEEE/RSJ international conference on Intelligent robots and systems
libDAI: A Free and Open Source C++ Library for Discrete Approximate Inference in Graphical Models

The Journal of Machine Learning Research
A Probabilistic Appearance Representation and Its Application to Surprise Detection in Cognitive Robots

IEEE Transactions on Autonomous Mental Development
Factor graphs and the sum-product algorithm

IEEE Transactions on Information Theory
Exploiting probabilistic knowledge under uncertain sensing for efficient robot behaviour

IJCAI'11 Proceedings of the Twenty-Second international joint conference on Artificial Intelligence - Volume Volume Three

Review: A review of novelty detection

Signal Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper presents an approach to the problem of novelty detection in the context of semantic room categorization. The ability to assign semantic labels to areas in the environment is crucial for autonomous agents aiming to perform complex human-like tasks and human interaction. However, in order to be robust and naturally learn the semantics from the human user, the agent must be able to identify gaps in its own knowledge. To this end, we propose a method based on graphical models to identify novel input which does not match any of the previously learnt semantic descriptions. The method employs a novelty threshold defined in terms of conditional and unconditional probabilities. The novelty threshold is then optimized using an unconditional probability density model trained from unlabelled data.