I am interested in the filtering and search of data from networks and
real
time anomaly detection of high level information. Networks of sensing devices embedded in
our world (weather, traffic, environmental, seismic, temperature)
produce massive amounts of real-time information and require
new search methods. Rather than traditional database searches
based on parameter thresholds
or boolean queries, my thesis work develops a method to detect
and track high level physical processes. This approach develops
techniques from information theory, tracking theory, Bayesian
statistics, and HMMs (Hidden Markov Models.) One example of a
physical "process" is the release of a chemical or radioactive plume,
however the process detection method applies to any well understood
process. Other applications include physical target tracking,
computer network security, and social network analysis.
Plume Tracking In Sensor Networks
Given a network of simple
binary sensors capable of
detecting a chemical of radioactive substance, how can the network
deconstruct large observations sets to identify the number, time, size,
and location of re:lease sites in a region? Imagine a city
outfitted with thousands or tens-of-thousand of these low resolution
sensors: how can the data be made useful? This was the main focus
of my thesis work at Dartmouth. The approach treats the
plume sources as "targets" and then makes
likelihood assignments between sensor observations. Groups of
observations are partitioned into "tracks" where each track is believed
to consist of observations generated by the same plume source.
The forward plume simulation environment
with three chemical release plumes. The wind field was generated with
a 2D Markov model, where the transition probabilities were trained from
historical weather statistics.
Observations at binary sensor nodes (red)
for 2 plume sources with unknown locations.
Airborne
Plume Tracking With Sensor Networks. Glenn Nofsinger and
George Cybenko. In Proceedings
of the SPIE , Defense and Security Symposium - Conference 6231 -
Unattended Ground, Sea, and Air Sensor Technologies and Applications
VII, Orlando, FL,April
2006
Plume
Source Detection Using a Process Query System. Glenn
Nofsinger, George Cybenko, In Proceedings of the SPIE Vol. 5416, Defense
and Security Symposium - Chemical and Biological Sensing V,Orlando,
FL, 2004
