Verarbeitung und Analyse von Positionsdatenströmen

The DFG research group 1508 (BATS) develops a sensor network consisting of static and mobile sensors to monitor bats. This thesis elaborates techniques and notations to process and analyze position data streams.

Abstractions for position data were elaborated: events and activities. Events are further divided into continuous queries interpreted as events and event patterns. Hence, the approach combines data stream processing and complex event processing. UML state charts were extended to be suitable for modeling activities.

The definition of activities exploits Dynamic Bayesian networks to predict behavior. Adaptive predictions are proposed to increase precision of predictions if several hypotheses of dependencies are specified.

Several state-of-the-art analyses, e.g. heatmaps and local convex hulls, as well as new analyses are implemented in the prototype. Techniques to predict activities additionally support inference on available Conditional Probability Tables as well as Conditional Tables for characteristics of activities. A possible definition of personality of animals is proposed.

Finally, the approach and prototype is evaluated against the objectives as well as reliability of predictions, influence of data quality on analyses and expressiveness of presented notations.