Distributed complex event detection for pervasive computing
Dan O’Keeffe · 2010
Events Katsiri et al. argue in [KBM04] that current event models are limited for use in a variant of pervasive computing called Sentient Computing [Hop00]. These limitations arise from the incomplete mapping from abstract top-level knowledge to low-level concrete data produced by sensors. They propose the concept of abstract events to model notifications about transparent changes in distributed state. They express abstract events using temporal first-order logic (TFOL). An abstract event detection service takes in TFOL abstract event definitions and creates detectors to notify the user on their occurrence. Thus, abstract events allow users to specify high-level events that correspond closely to real world concepts. However their detection algorithm is based on the Rete algorithm (see Section 2.3.3), which scales poorly in terms of memory usage, and is unsuitable for deployment on resource-constrained devices. Furthermore, the Turing-complete nature of TFOL makes it hard to reason about. A closed-world assumption, as required to support negation and quantification, makes abstract event detection difficult in open distributed environments. Sentient Objects Biegel and Cahill [BC04] propose the Sentient Object programming abstraction to support event-based situation analysis in sentient environments. Each sentient object interacts with sensors and actuators using an event-based communication infrastructure to handle mobility. A Sentient Object provides components for fusing sensor data using Bayesian networks, storage of context for efficient reasoning, and an inference engine for determining what course of action to take on the occurrence of an event within a context. Application developers can program Sentient Objects at a high level using graphical tools. A context acts as a dynamic filter over input events streams, in that only events of interest in a particular context must be delivered to the object. However Sentient Objects themselves are implemented in a centralised fashion.