VIGILANT: A semantic Model for Content and Event Based Indexing and Retrieval of Surveillance Video

This paper presents a semantic video-object model for e cient storage, indexing and content/event-based retrieval of real-time surveillance video without reverting to the constant re-interpretation of source and thus avoiding timeconsuming analysis of every video surveillance query. Based on the work on object tracking carried out at the Digital Image Research Centre (DIRC) at Kingston University [Orwell99a, Orwell99b], the model automatically extracts moving objects such as car, identi es their properties and labels events (e.g. a car entering the park) in a car park environment. A Video-Object (VO) is an abstraction of a Video Shot (a sequence of video frames lmed either from a xed camera position or using a coherent camera motion such as panning, rotation and zooming). A combination of object oriented modelling and description logic knowledge representation techniques will be used in the realisation of VIGILANT video object model. Object-Oriented Database (OODB) will model the static aspects of a video shot using the concepts of classes and relationships. A Description Logics (DLs) knowledge base will be employed to handle the dynamic and incremental content description of videos. This video object model will be used to index extracted features and recognised objects provided by the Image Processing module.