Vision based localization in urban environments

Abstract

This paper describes a vision-based system that performs localization in outdoor urban environments by detecting building features and relating those features to an a priori map derived from an aerial image. Our approach is motivated by the desire to enable a human operator to intuitively direct an autonomous unmanned vehicle (AUV) to a specific goal location without the use of GPS. Implicit in this problem is the presumption that operator has sufficient knowledge of the environment to identify a desirable goal location based. This goal location would, we expect, be selected because it placed the robot's sensors and/or actuators in some particular relationship to features of interest in the environment.