Topological Localization of Mobile Robots Using Probabilistic Support Vector Classification

摘要

Topologically localizing a mobile robot using visual information alone is a difficult problem. We propose a localization system that comprises Gaussian derivatives as raw local descriptors, a three-tier spatial pyramid of histograms as the image descriptor, and probabilistic multi-class support vector machines for classification. Based on the probability estimate, the proposed system is able to predict the unknown class which corresponds to locations that are not imaged in the training sequence. To exploit the continuity of the sequence, a smoothing procedure can be applied, which is shown to be simple yet effective.