Visual place recognition techniques for pose estimation in changing environments
Year: 2016
Type: conference-paper
Abstract: This paper investigates whether visual place recognition techniques can be used to provide pose estimation information for a visual SLAM system operating long-term in an environment where the appearance may change a great deal. It demonstrates that a combination of a conventional SURF feature detector and a condition-invariant feature descriptor such as HOG or conv3 can provide a method of determining the relative transformation between two images, even when there is both appearance change and rotation or viewpoint change.
Authors: Stephanie Lowry, Henrik Andreasson
Institutions: Queensland University of Technology, Örebro University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2615698988
