Self-localization Using Visual Experience Across Domains

— In this study, we aim to solve the single-view robot self-localization problem by using visual experience across domains. Although the bag-of-words method constitutes a popular approach to single-view localization, it fails badly when it's visual vocabulary is learned and tested in different domains. Further, we are interested in using a cross-domain setting, in which the visual vocabulary is learned in different seasons and routes from the input query/database scenes. Our strategy is to mine a cross-domain visual experience, a library of raw visual images collected in different domains, to discover the relevant visual patterns that effectively explain the input scene, and use them for scene retrieval. In particular, we show that the appearance and the pose of the mined visual patterns of a query scene can be efficiently and discriminatively matched against those of the database scenes by employing image-to-class distance and spatial pyramid matching. Experimental results obtained using a novel cross-domain dataset show that our system achieves promising results despite our visual vocabulary being learned and tested in different domains. I. INTRODUCTION In this study, we aim to solve the problem of cross-domain single-view robot self-localization. For solving SLAM and other similar problems in mobile robotics, visual localization is crucial [1]–[3], [5]. While self-localization can be done either by using prior knowledge of the problem domain [1] or without them [2], we deal with applications and scenes where a collection of visual images from different domains, termed cross-domain visual experience, is available as prior knowledge. At the same time, we require the localization algorithm to be extremely fast (to work in a fast robot navigation) and to recognize the place from a single frame [1] (i.e., without temporal tracking [7] and visual sequence measurements [8]). One of most popular approaches to address the problem of single-view localization is bag-of-words methods [1], [5], [10], wherein a collection of local invariant visual features is extracted from an input image, and each feature is translated into a visual word by using a pre-learned library of vector-quantized features. Consequently, an input scene image is described compactly and discriminatively as an unordered collection of visual words (" bag-of-words "). However, as argued by several authors, the bag-of-words method fails badly when learned and tested in different domains; the main reasons include the following: (1) Because the bag-of-words method ignores all information about the spatial layout of the features, it limits the descriptive ability …