Localization Based on Building Recognition

Navigational capabilities of people in urban areas are to a large extent determined by their knowledge of current location. In addition to location information available by means of global positioning sensors, images can provide additional and often complementary information about relative position and/or viewpoint of the person with respect to some known landmarks. In order to enable such functionality, landmarks (e.g. buildings) have to be reliably and efficiently recognized. In this paper we describe a hierarchical approach for recognition of buildings. At the first stage, we use a novel and efficient representation named localized color histograms. This representation enables efficient retrieval of a small number of candidate matches from the database. At the second stage, the recognition is refined by matching descriptors associated with local image regions. Once the correct building is identified, the relative pose with respect to the building is recovered. The proposed approach is validated by extensive experiments, with images taken in different weather conditions, seasons and with different cameras.