Map building and localization using Global-Appearance Descriptors applied to panoramic images
F. Amorós, L. Payá, O. Reinoso, L. Fernández
Journal of Computer and Information Technology (Spring 2012)
Ed. Publishing Services LLC ISSN:2161-7112 - Vol 2 (1), pp. 55-71
Global-Appearance techniques represent a very promising alternative in image processing applied to the tasks of robotic map building and localization. Their inherent characteristics present advantages over classical features descriptors, especially in unstructured environments where landmarks can be difficult to extract. However, when we apply them to real-time navigation, we have to take into account some restrictions that might limit their application since mapping and robot navigation requires specific advisable characteristics when building a descriptor.
This article makes a review and comparison of different methods based on global-appearance to create descriptors of panoramic scenes in order to extract the most relevant information, studying several characteristics and parameters, as invariance against rotations of the robot on the ground plane, computational requirements and accuracy in localization. For this purpose, we have carried out a set of experiments with panoramic images captured in real lighting conditions in indoor environments to demonstrate the applicability of the different descriptors to robotic navigation tasks and to measure their goodness in localization recovery and their computational requirements. The experimental part lets us to validate them and to make an analysis and comparison of each technique.