jake emerson

jake emerson

Department of Spatial Information Science and Engineering, University of Maine
jakeemerson[AT]spatial.maine.edu

Detection and Routing Near an Event Boundary
Using Sensor Networks

As flood waters rise and a high-intensity rain cell moves east across the city, traffic grinds to a halt on the southeast side of downtown as a stream spills onto minor roadways. News reports give general information about the flood and traffic, but drivers on the boundary of the event remain uncertain of which route to take to avoid trouble. News reports, which are transmitted from repeater stations or web servers, are information bottlenecks – failure of one repeater affects many people at once. Additionally, these centrally controlled systems are unable to respond to all listeners' information needs. For example, a driver might need to know which exit to take to avoid the high water. Despite the perpetual recurrence of this and similar scenarios, where information is needed just outside the boundary of an event, we are only beginning to see the potential of a decentralized approach to data collection. A decentralized information system can be more robust against failure and adaptive to local information needs.


This research envisions a sensor network and associated information system which would address the problem of defining and detecting the boundary of a spatially extensive event, and support routing decisions in flow networks affected by dynamic phenomena. In this system the sensor network consists of immobile, passive sensor devices. These sensors record data from the environment as time-series data and communicate with their immediate neighbors and mobile agents (e.g., cars) in the flow network.

Certain occurrences in the time-series data (e.g., sharp drop in visibility, decrease in average distance between cars) are recorded at the sensor node as primitive events. Using an algorithm common among the sensor nodes they collaboratively establish a boundary. This boundary area defines a composite event. Boundary nodes then periodically transmit a message which travels away from the boundary, called a stigmergic trace. Stigmergic trace messages move from one sensor node to the next along a path which tends to curve toward locations holding older boundary messages, and away from newer messages. The result is a magnetic field-like set of traces surrounding the event that curve toward previous locations, thus encoding boundary movement. Mobile agents, which store a picture of the pre-event flow network, use the traces to route around event boundaries. By moving along routes that tend to be perpendicular to traces they will avoid the boundary. Routing parallel to a trace will generate the most direct path toward the nearby boundary.


The systems investigated in this research, along with boundary detection and routing algorithms, will be applicable to more than traffic routing. Logistical supply planning in disaster response and control of combined storm and sewer systems could use these techniques to guide the control of routes in the face of changing conditions. This research will further develop understanding of decentralized action in dynamic flow networks.