SPI's core technology is in pattern recognition and its applications. Over
the past five years, we have developed real-time products and
prototypes in chemical agent detection and classification, multimodal
biometrics, speech processing, fault diagnostics and prognostics, and target
detection and classification. Some recent projects
at SPI are
briefly described below; other past projects not
listed here can be found in the Publications. Interested parties can
contact Dr. Chiman Kwan (240-505-2641 or
email@example.com) for more information about projects in each area.
1. "A Novel and High Performance Framework for
Chemical/Biological Agent Detection and Classification Using Fluctuation
Enhanced Sensing," 2007.
2. "Enhanced Partial Fingerprint Recognition Using Feature Level Fusion,"
3. " Condition Based Maintenance Using Wireless Sensors," 2008.
4. "Chemical Agent Fingerprinting Using Zero-Crossing Patterns," 2008.
5. "Multi-thread multi-core processing architecture for real-time applications," 2008.
6. "Fast and novel nonlinear change detection algorithms for hyperspectral image processing," 2008.
1. "Enhanced Partial Fingerprint Recognition Using Feature Level Fusion,"
VSTTM is a
new software tool that can generate highly compressed videos for reviewing and archiving.
Our approach is an event based method. Novel algorithms are used
to separate a long video into event segments and non-event segments.
For event segments, more frames will be retained; for non-event segments,
less frames will be retained. The determination of events and non-events
can be done in real-time. The algorithms are robust to illumination changes,
shadows, moving trees, the presence of multiple objects, changing traffic
lights, etc. The tool is suitable for condensing long surveillance videos into much shorter videos and yet without
losing any important information in the original videos. The conventional way of
reviewing videos is to have an operator to manually view the videos, which is
tedious and error prone. Having a much shorter video will reduce the operators'
work load and also make the reviewing process interesting and fun. Moreover, the
data storage can be much smaller. The video demo on the left shows the case where
a raw video with a length of 14 minutes is compressed to a video with ony 16 seconds,
a compression ratio of 52 times. The red markers at the bottom show the event
segments and the green markers indicate the non-event segments.
CAC-CETTM is a
new software tool that can classify and estimate concentrations of
different chemical agents. The tool is modular, flexible, and
suitable for real-time chemical agent classification applications,
including conventional and nano-sensors. It is low cost and
independent of commercial software packages such as Matlab, Labview,
A personal identification system using finger and
voice prints has been designed and implemented. This product uses low cost
that have complementary properties.
on a niche biometric market involving business executives, military
personnel, and marketing people who need to protect valuable
information in their desktops, PDAs, and laptops. Compared to
passwords, fingerprint and voiceprint biometrics are hard to steal
and more reliable. Although low cost (<$30) biometric devices will
enable more users to use biometrics, the performance of existing
software processing algorithm cannot meet the performance of these
users. For example, some low cost fingerprint sensors have small
area that can only capture a partial fingerprint and consequently
the recognition performance drops significantly. Our proposed
multimodal biometric algorithm will meet the above requirements by
using multimodal biometric fusion, as our preliminary results
already achieved 1% False Rejection Rate (FRR) at 0.2% FAR (False
A speech separation package that uses various
algorithms to separate, ID, and recognize voices in noisy
environments has been designed and implemented. The various modules
were all implemented in C and are independent of any commercial
software packages such as Matlab, Labview, etc. We target a niche
market where speech separation is needed before speaker
identification and speech recognition can start. Potential
applications include personal identification in cocktail party
environment, biometrics in multiple competing speaker environment,
hands free communications in cars, etc.
Fault Diagnostics and Prognostics
SPI has extensive experiences in
fault diagnostics and prognostics. One paper written by Dr. Kwan was
one of seven finalists for Kayamori Best Paper Award in the 2003
IEEE Conference on Robotics and Automation. Another paper on power
quality monitoring and control received the 2004
Industrial and Commercial Power Systems (I&CPS) Ralph H. Lee
Department Prize Paper Award. Recently, we developed real-time
products for monitoring critical power system components using
Past statistics showed that 40% of machine failures occur due to
bearing problems. Remaining life prediction of bearings will provide
ample time to operators to order new parts, schedule maintenance
activities, and prevent unnecessary machine downtime.
Multiple Target Tracking of Maneuvering and Non-maneuvering Targets
Accurate tracking of multiple targets is important
in military surveillance and civilian applications (air traffic