On-line visual vocabularies for robot navigation and mapping

Abstract

Detecting already-visited regions in vision-based navigation and mapping helps reduce drift and position uncertainties. Inspired from content-based image retrieval, an efficient approach is the use of visual vocabularies for measuring similarities between images. In this way, images corresponding to the same scene region can be associated. The state of the art proposals that address this topic suffer from two main drawbacks: (i) they require heavy user intervention, generally involving trial and error tasks for training and parameter tuning and (ii) they are suitable for batch processing only, where all the data is readily available before data processing. We propose a novel method for visual vocabulary navigation and mapping that overcomes these shortcomings. First, the vocabularies are built and updated online, during robot navigation, in order to efficiently represent the visual information present in the scene. Also, the vocabulary building process does not require any user intervention.