Path localization using Gabor-Gist

Learning and then recognizing a path is a challenging task for state of the art algorithms in computer vision and robotics. In this paper, we present a new approach to visual localization along a path. Classically visual paths have been described using keyframes, single images taken at specific locations. Our method uses all the images of a path segment, Gabor-Gist and, principal component analysis to represent a segment as segment specific principal components. Localization is achieved by comparing a query image descriptor to the segment's principal components using a new reconstruction similarity measure, choosing the path segment which best reconstructs the original query descriptor. Using two datasets of indoor and outdoor environments we compare our method to the same path represented using keyframes. While the feature-based keyframes perform poorly, the new method is able to correctly localized the robot 93% of the time.