Project 3-1: Information Dissemination and Aggregation Under Mobility
Principal Investigator: Leonidas Guibas (Stanford)

Increasingly, military operations are carried out in urban environments. Often, military units must quickly set up their own sensor and communications networks to provide surveillance and information exchange among authorized users, rather than relying on existing civilian communications infrastructures.

Ad hoc sensor networks deployed in urban environments must be capable of providing timely, reliable information to multiple authorized users operating in that environment. Beyond merely providing streams of raw data, such networks must be capable of aggregating and performing in-network analysis of the data streams. This makes it possible to provide lightweight, distributed, highly specific information brokerage between the sensor network and the users.

Network nodes typically are constrained by available computing and communications resources. Fixed or mobile HPC resources deployed within the network can alleviate such resource constraints and facilitate network planning, protocol optimization, and sensor data analysis, as well as support real-time network operations and query processing.

AHPCRC researchers are analyzing user mobility and communication patterns in order to develop better algorithms and protocols for delivering information to users moving from place to place within the network. These efforts can assist in improving network layouts, determining the number of nodes and their placement, and methods for delivering data efficiently. Studies on network reliability and signal interference aim to reduce the effects of intermittent coverage and disruptions, including physical destruction of parts of the network.

Research on interlinked images (image webs) has produced useful insights into tracking the movements of people and vehicles within a sensor network. Image webs could also provide a means of orienting and guiding robotic vehicles, using recognizable features of the environment to provide information on location and direction. Image webs require significant computational resources to identify and correlate features and to identify changes when objects move within the field of an image.

Because communication requires even more resources than computation, efficiency requires identifying and extracting relevant image features at the site of each camera using a simple "image vocabulary" amenable to the very basic computing capabilities of individual sensor nodes. New "words" must be added to this vocabulary as needed, and image features must be ranked by importance to identify which features to relay to the network.

A small testbed camera sensor network has been implemented on the Stanford University campus to test the concepts and methods produced by the computational studies.

Read more about this project...