Mobile robot global localization with non-quantized SIFT features

Presently, the mainstream approach to appearance-based localization with local features uses a quantized representation. In this work, the quantized and non-quantized representations are compared with respect to their discriminativity and information content properties. Having demonstrated the advantages of the non-quantized representation, the paper proposes a localization method based on it, and mechanisms to reduce the computational burden this approach would carry, if taken straightforwardly. This reduction is achieved with context cues provided by gist and by exploring two simplifying assumptions about the training data.