ABSTRACT
Our objective in this work is to conceive a semantic representation language that enables complete specifi cation and modeling of visual events. Th ese event models can then be used to automatically retrieve the visual events from a video database or in a real-time video broadcast. Th is modeling language is also used for the validation of the MPEG-7-based description of video sequences. In addition to the expression of spatial and temporal constraints, this language takes into consideration uncertainty, off ers diff erent levels of granularity in description, and is independent from storage format in order to assure interoperability between description systems. Based on fi nite automata formalism, the language allows the description of state sequences. Each state is associated with a situation occurring within the event, and each situation is characterized using fuzzy conceptual graphs to describe its spatial composition. By enabling the description of complex concepts, this formalism can be used, for example, to monitor behaviors in a video-surveillance context, to detect important highlights in a sport context, and to validate MPEG-7-based descriptions.
