CCD: Locating Event in Wireless Sensor Network without Locations

Abstract

Event detection is a typical application of wireless sensor networks. The existing approaches of event detection usually employ certain event models that are constructed with prior domain knowledge. The resulting event detection processes appear to be cost-inefficient, which require either intensive data exchanges among neighboring nodes or caching large columns of history data. In this paper, we focus on the issue of locating event in wireless sensor network without locations. This involves two tasks, namely detecting an event and identifying an area in the network where the event occurs. Motivated by the real system, we propose a model-free approach for event detection called CCD (Coding Cost based event Detection). Coding cost is a metric that quantifies the diversity of a set of sensor readings. Incorporated into the inherent data collection mechanism, CCD passively constructs a gradient map of coding cost throughout the network. An event is then detected where a change point of gradient appears and identifies the event pattern automatically. CCD is fully distributed and does not incur apparent communication overhead. We implement CCD and evaluate its performance with extensive experiments and simulations. The results demonstrate that CCD is accurate, scalable, and applicable to a variety of sensor networks.