Ontology-driven scene interpretation based on qualitative spatial reasoning

Abstract

In this paper we present an approach to interpret composite objects in a scene and label them as members of a particular ontological feature class according to the spatial arrangements of their components. As these arrangements are often indicators of the functional role of the corresponding components, they can be viewed as spatial signatures for such features, and their analysis offers a strong lead for scene interpretation and labeling. In order to pursue this objective, we need efficient quantitative metrics to describe the spatial arrangements, and for that purpose we adopt the Histogram of Forces technique and fuzzy Allen-derived descriptions of spatial relations. This leads to the generation of pairwise metric expressions describing the spatial relations among individual components of a composite object. Such pairwise expressions can be aggregated to describe the overall spatial arrangement of an object’s components. Such layout tables of different objects can be compared through an assessment of their normalized cross-correlation, in order to compare their spatial layouts and decide whether they belong to the same feature class. Through this approach we are merging principles from scene interpretation, similarity assessment, and ontology to advance our capability to understand complex scenes. In this paper we present the components of our approach and also provide an application using three different classes of airports, to demonstrate its performance.