Hierarchical Multi-Modal Place Categorization

In this paper we present an hierarchical approach to place categorization. Low level sensory data is processed into more abstract concept, named \emphproperties of space. The framework allows for fusing information from heterogeneous sensory modalities and a range of derivatives of their data. Place categories are defined based on the properties that decouples them from the low level sensory data. This gives for better scalability, both in terms of memory and computations. The probabilistic inference is performed in a chain graph which supports incremental learning of the room category models. Experimental results are presented where the shape, size and appearance of the rooms are used as properties along with the number of objects of certain classes and the topology of space.