Connecting the dots: Constructing spatiotemporal episodes from events schemas

Abstract

This paper introduces a novel framework for deriving and mining high-level spatiotemporal process models in in-situ sensor measurements. The proposed framework is comprised of two complementary components, namely, hierarchical event schemas and spatiotemporal episodes. Event schemas are used in this work as the basic building model of spatiotemporal processes while episodes are used for organizing events in space and time in a consistent manner. The construction of event schemas is carried out using scale-space analysis from which the interval tree, a hierarchical decomposition of the data, is derived. Episodes are constructed from event schemas using by formulating the problem as a constraint network, in which spatial and temporal constraints are imposed. Consistency is achieved using a path-consistency algorithm. Once created, possible episodes can be derived from the network using a shortest-path search. © 2009 Springer-Verlag Berlin Heidelberg.