Implementation Conclusion and Future Work

The presented computational model was fully implemented and tested. For validation purposes diierent visualization methods (cf. (Gapp 94)) are used to get an idea of the spatial relations' region of applicability (Figure 4 and 6). The system proceeds from a geometrical scene description and answers questions in natural language about the constellation of the objects involved in a cognitively plausible manner. Using an object indexing method permits an immediate object access to accelerate answering partially instantiated queries. Presently every possible applicable spatial relation has been generated. The construction of a component which decides what spatial relation or compositions in a certain situation should be used is currently under investigation. The computational model is integrated in a work-bench which serves as a basis for the incremental event recognition component of the VITRA project. In this paper methods were developed for computing the basic meanings of spatial relations with respect to geometrical object properties. We have proposed an approximative algorithm which represents a reasonable compromise between simpliication and necessary exactness. The described methods for computing the relations between objects in 2D and 3D space consider the extension of the REFO and permit a plausible evaluation dependant on the user's perception. The evaluation procedures are applicable to 2D as well as 3D scene data. Even the application of a spatial relation to objects with diierent dimensions, e.g., a parking lot and a house, is supported. The next step is to acquire psychological evidence of the applicability of basic spatial relations' meanings. For this purpose, empirical studies have been designed. Subsequently further phenomena can be considered, e.g., the innuence of intervening objects or borders.