Semiorder model for temporal composition of activities from events in multisensory information management systems

Many issues in the composition of a semantic activity from semantic events in a multi-sensory information management system are discussed in this dissertation. Events are atomic semantic units that are defined as complex spatio-temporal fusion of multi-sensory information. The temporal composition of semantic activities, which represent higher levels of spatio-temporal fusion, from semantic events is the main focus of the research presented here. In multi-sensory environments, an event occurs with a bounded interval of temporal uncertainty (D). Association of bounded intervals of uncertainties with event occurrences results in intransitive concurrency between events that is not modeled by ubiquitous sequential and weak order compositions. A proper subclass of partial orders, semiorders, models such compositions when the bound D is fixed for all events in an event set.; The linear representation and canonical tree-like structure of semiorders are used to design efficient algorithms and heuristics for the morphism class of operations for certain subclasses of labeled and unlabeled semiorders. Efficient algorithms for these semiorder operations are currently not known. Results include design of algorithms for the suborder isomorphism problem for unlabeled and labeled semiorders for the general and the special case when there are no crossovers. For unlabeled semiorders, the time complexities are O(n) and O(H! W n2) for the special and the general cases respectively, where n is the maximum size of semiorders, H is the height, and W is the width of the non-serial source semiorder. The corresponding results for labeled semiorders are O(n log n) and O(H! W n2 log n).; The architectural issues revolve around the design and implementation of two main components: a transducer for the detection of atomic events, and a semiorder database to store events and their temporal order. The semiorder data model, a corresponding query language, and a pattern definition language are the main database components discussed in this thesis. The data model extends the relational model with semiorder ordering of unordered relational tuples, and it provides many classes of nonrelational operations including classes of morphism and ordered aggregation operations. Finally, the implementation of architectural components is briefly discussed and results of a few illustrative queries from the domain of visual surveillance and monitoring are provided.