Data modeling and querying in video databases

Multimedia databases have been the subject of extensive research for the last ten years. In particular, indexing and knowledge based representation of semantics associated with video databases are challenging tasks. This research focuses on addressing issues related to event modeling, query formulation, and query processing algorithms for video data. In order to develop viable solutions for content-based retrieval of video data, formal models are needed to capture and represent video events. In this thesis, we propose a Petri-net based formalism, known as Hierarchical Petri-net (HPN), to represent and index video data. HPN\u27s allow multi-level semantic abstractions of events with an arbitrary degree of complexity. We elaborate on how HPN\u27s can capture video data semantics succinctly, and propose algorithms to build a novel video browsing technique that seamlessly integrate low level video semantics such as object movements to higher level semantics representing complex scenarios. Another key contribution of this research is a Petri-net based formalism for content-based video query formulation and associated query processing algorithms. A major issue in this context is handling of inherent imprecision in query specification and data representation of video contents. We elaborate on different ways of specifying video queries and analyze their expressive power. Accordingly, we propose and analyze different techniques for processing video queries in terms of two key performance parameters, namely precision and recall