Semantic interpretation of novelty in images using histograms of oriented gradients

Authors:
Nicolas Alt;Werner Maier;Qing Rao;Eckehard Steinbach
Affiliations:
Institute for Media Technology, Technische Universität München, München, Germany;Institute for Media Technology, Technische Universität München, München, Germany;Institute for Media Technology, Technische Universität München, München, Germany;Institute for Media Technology, Technische Universität München, München, Germany
Venue:
ICIRA'12 Proceedings of the 5th international conference on Intelligent Robotics and Applications - Volume Part III
Year:
2012

Citing 8
Cited 0

Support-Vector Networks

Machine Learning
Histograms of Oriented Gradients for Human Detection

CVPR '05 Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) - Volume 1 - Volume 01
Image-Based Rendering

Image-Based Rendering
Co-occurrence Histograms of Oriented Gradients for Pedestrian Detection

PSIVT '09 Proceedings of the 3rd Pacific Rim Symposium on Advances in Image and Video Technology
Bayesian surprise and landmark detection

ICRA'09 Proceedings of the 2009 IEEE international conference on Robotics and Automation
Visual homing and surprise detection for cognitive mobile robots using image-based environment representations

ICRA'09 Proceedings of the 2009 IEEE international conference on Robotics and Automation
LIBSVM: A library for support vector machines

ACM Transactions on Intelligent Systems and Technology (TIST)
A Probabilistic Appearance Representation and Its Application to Surprise Detection in Cognitive Robots

IEEE Transactions on Autonomous Mental Development

Quantified Score

Hi-index	0.00

Visualization

Abstract

An approach for the semantic interpretation of image-based novelty in real-world environments is presented. We measure novelty using the concept of pixel-based surprise, which quantifies how much a new observation changes the robot's current probabilistic appearance model of the environment. The corresponding surprise maps are utilized as prior information to reduce the search space of a "Histograms of Oriented Gradients" object detector. Specifically, detection windows are scored and selected using surprise values. Several object classes are simultaneously searched for and learned from a low number of manually taken reference images. Experiments are performed on a human-size robot in a cluttered household environment. Compared to object detection based on a search of the complete image, a 35-fold speed-up is observed. Additionally, the detection performance increases significantly.