The localization of an object within a certain environment requires the selection of an appropriate reference object and the establishment of adequate spatial relations between the objects. The decision, which reference object to choose, demands consideration of all applicable spatial relations, which therefore have to be computed before the final decision can be made. However, the calculation of all applicable spatial relations for all potential reference objects is a computationally intensive task. In order to increase performance, it is proposed that the set of spatial relations to be tested can be reduced in two respects without any loss of quality in the generated answer. The two ways in which reduction can be achieved are by taking into account the type of the localization --- approximate or precise --- and the type of the reference object. 1 Introduction Object localization is a task in which the integration of language and vision is a required necessity per se. The localizatio...

Abstract not found