Visual location awareness for mobile robots using feature-based vision

Abstract

This thesis presents an evaluation of feature-based visual recognition paradigm for the task of mobile robot localization. Although many works describe feature-based visual robot localization, they often do so using complex methods for map-building and position estimation which obscure the underlying vision systems' performance. One of the main contributions of this work is the development of an evaluation algorithm employing simple models for location awareness with focus on evaluating the underlying vision system. While SeeAsYou is used as a prototypical vision system for evaluation, the algorithm is designed to allow it to be used with other feature-based vision systems as well. The main result is that feature-based recognition with SeeAsYou provides some information but is not strong enough to reliably achieve location awareness without the temporal context. Adding a simple temporal model, however, suggests a more reliable localization performance.