DYNAMIC EVENT INTERPRETATION AND DESCRIPTION FROM VISUAL SCENE BASED ON COGNITIVE ONTOLOGY FOR RECOGNITION BY A ROBOT

Y. Wakuda,∗ K. Sekiyama,∗∗ and T. Fukuda∗∗

Keywords

Event description, scene analysis, cognitive vision, concept structure,attention

Abstract

Automatic detection and description of events, particularly human behaviour, is one of the most challenging issues, as event interpretation is highly dependent on the target of attention, which is not uniquely specified. To tackle this problem, we propose the concept of a “cognitive ontology as a framework for a system that can automatically decide the attention focus and describe the events. A cognitive ontology is structured with conceptual units which are Entities and Relations, and these units enable robot endogenous attention fixation and jumps based on a networked cognitive ontology. In addition, we introduce exogenous attention fixation based on how the observed targets differ from a predicted pattern. In this process, the corresponding target of attention is updated and assigned to an event description buffer that consists of two Entities and one Relation. In this paper, we have developed and experimented with this holistic event interpretation process taking into account endogenous attention, exogenous attention and determined event description.

Important Links:



Go Back