Emerging Sensor Networks Can Make Sense for Your Business

1
Emerging Sensor Networks Can Make Sense for Your
Business

• Tech 2003, May 7, 2003
• Dr. Lisa Ann Osadciw from DREAMSNet
• (Development and Research in Evolutionary
  Algorithms for MultiSensor Smart Networks)
• http//www.ecs.syr.edu/research/DREAMSNet/

2
Objective

• Awareness of the new emerging sensor networks
• Provide you with a new understanding of how you
  may take advantage of these networks to improve
  your business

3
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses for Sensor Networks Today
• Research Groups for Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

4
Sensor Network System View
Sensor Manager (AI)
Sensor Status
Controls
Sensors
Data Fusion Processing
Measurements
Information Space aircraft detection and
location, person identification, etc.
Observation Space nature, people, etc.
5
Current Research Focus Areas
DREAMSNet
6
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses for Sensor Networks Today
• Research Groups for Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

7
Sensor Network Applications

• Defense Applications
• Medical Applications
• Patient wears sensors recording or sending data
• Monitor patients in the hospital
• Weather Applications
• Indoor Environmental Control
• Building Security
• More Applications will Emerge as Sensor and
  Network Technology Improves and Businesses

8
Problems Solved through Sensor Networks

• distributed sensing observations (e.g.
  temperature, air flow, movement of objects)
• Situation assessment based on observations
• Assessment accuracy is due to the set of
  observations rather than any single observation
• Sensors cover a large region
• Measurements must be collected over a significant
  time duration. For example
• More than one patient must be monitored
• Whole building must be covered
• Structure monitoring (airplanes, bridges, etc.)

9
Minimize Sensor Network System Cost

• Current research area
• 50-90 of sensor cost lies in the wire so
  wireless communication reduces cost.
• Micro-electromechanical systems (MEMS) will allow
  systems to be embedded into ceiling fans, ceiling
  tiles, furniture, etc. (use of such will be
  discussed later). Berkeley Center for Built
  Environments
• Placement, quantity, and type of sensor

10
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses for Sensor Networks Today
• Research Groups for Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

11
Just Some of the Many Advantages Possible Using
Sensor Networks

• Monitor manufacturing processes long distance and
  even over the internet anywhere in the world.
• Move the monitoring of any business function that
  is followed by a sensor or software to anywhere
  in the world.
• Add remote control to your business.
• Increase security without adding more keys or
  passwords. Increase it while removing keys
  through biometrics.
• Reduce operation costs through better heating
  control and more effective business operations.
• Track employees remotely for security or safety
  purposes.
• Improve employee satisfaction by increasing
  employee comfort as well as reduce illness.

12
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses Supplying Sensor Networks Today
• Research Groups Improving Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

13
Businesses

• Xsilogy software and hardware products to
  integrate with your own existing sensing devices
  (www.xsilogy.com)
• Network Associates, Inc. (NAI) battlefield
  network problems, SensIt - cryptography
  (robustness, weight, communications)
  (http//www.nai.com/common/media/nai/pdf/NAI-Labs-
  DSNS-1-5-01.pdf)
• Intel embedded sensors monitoring the
  well-being of a wide variety of things, which
  currently focus on grapes for winemaking
  (http//www.intel.com/labs/features/rs01031.htm)
• IBM T.J. Watson Research Center software,
  operating systems, protocols, biometrics
  (http//www.watson.ibm.com/)
• Lockheed Martin networked systems,
  architectures, defense applications
  (http//www.lockheedmartin.com/)

14
Businesses(cont)

• Crossbow licensed by Intel to sell the mote
  sensors (www.xbow.com)
• Microstrain- sells wireless sensors for strain,
  temperature, humidity type sensors
  (http//www.microstrain.com/)

15
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses for Sensor Networks Today
• Research Groups for Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

16
Active Research Groups

• Syracuse University DREAMSNet in
  CASE(Development and Research in Evolutionary
  Algorithms for Multisensor Smart Networks)-
  architectures, algorithms, and sensor control
  (http//www.ecs.syr.edu/research/DREAMSNet/)
• University of California - CITRIS (Center of
  Information Technology Research in the Interest
  of Society) collaboration of groups
  (graphics.cs.ucdavis.edu/okreylos/
  ResDev/SensorNetworks/ )
• University of Vermont Wireless Ad-Hoc Research
  Group mainly architectures and algorithms (
  http//www.emba.uvm.edu/jfrolik/uvmwan.htm)
• Center for Embedded Network, UCLA network
  security and protocols (http//cens.ucla.edu/)
• IrisNet Carnegie Mellon University protocols,
  database, operating system issues
  (http//www-2.cs.cmu.edu/srini/)

17
Active Research Groups

• Intel Research Laboratory at Berkeley mote
  sensors or tiny wireless sensors slightly larger
  than a penny collect light, temperature, humidity
  (http//intel-research.net/berkeley/features/tiny_
  db.asp)
• Sensor Computing Research (SCR) at University of
  Missouri- Rolla security application
  (http//web.umr.edu/mma882/Sensor.htm)
• Wireless Sensor Network Management Protocols
  (SNMP) Rutgers University (http//www.cs.rutgers
  .edu/bdeb/sensor_networks.html)

18
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses for Sensor Networks Today
• Research Groups for Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

19
DREAMSNet Research Summary

• Intelligent Sensor Network Control
• Mathematics POSets, Bayesian Belief Networks,
  Swarm Agents
• Architectures design software to be as flexible
  as possible
• Projects
• Sensor Management for Biometric Security
  buildings, any access control problem
• Face Recognition improving performance
• Sensor Management for Intelligent Defense Sensors
• Communication Protocols message routing based
  on information importance and resources

20
Intelligence Within Sensor Network

• Software architecture designed for flexibility
• Isolating the code that is application specific
  increases the softwares robustness

21
Design Objectives for Sensor Manager

• Problem As intelligent networks of heterogeneous
  sensors are created, automatically assessing and
  optimizing global performance for changing
  mission requirements easily overwhelms processors
  and operators.
• Objectives
• Adapt intelligent sensors for changing mission
  requirements.
• Design a mathematical framework based on global
  performance parameters that adapt to mission
  changes.
• Solve the resulting complex optimization problem
  with computational efficiency.

22
Technical Challenges

• Reduce contention for sensor resources.
• Prevent overwhelming communication network with
  sensor data.
• Enhance, not hinder, existing sensors operation
  and adaptability.
• Automatically compensate within the network for a
  sensor failure or degradation.

23
Sensor Networks Discussed in this Talk

• Built Environmental Sensor Networks
• Improve Occupants Comfort
• Reduce Heating and Cooling Costs
• Defense Sensor Networks
• Accurate Regional Surveillance
• Reduce Response Time
• Biometric Building Security Sensor Networks
• Improve Identity Verification Accuracy
• Reduce Cost of Maintaining System

24
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses for Sensor Networks Today
• Research Groups for Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

25
Built Environmental Sensor Network

• Wide Variety of Purposes or Missions
• Chemical/biological agent detection
• Fire and smoke detection
• Monitor occupant movement
• Wireless Communication cost savings
• More Efficient Use of Resources within Building
  (e.g. electricity, gas, light)

26
Why Consider Built Indoor Environmental Sensor
Networks?

• 90 of our time is spent here but 30 are
  uncomfortable
• 40 - 250 billion productivity loss due to poor
  Indoor Environment Quality (IEQ)
• 40 of total building energy consumption is for
  environmental control
• 110 billion annual economic loss due to air
  pollution in urban areas

EQS Center Overview by H. Ezzat Khalifa,
Syracuse University
27
Built Environment Sensor Network
28
Diagram of Built Environment Network
29
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses for Sensor Networks Today
• Research Groups for Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

30
Defense Sensor Network

• Network a heterogeneous set of sensor varying in
  complexity
• Objectives
• Solve resource contention problem
• Rapid response to changing mission
• Minimize communication bandwidth contention

31
Joint Directors of Laboratories JDL Fusion Levels
JDL 2 Situation Assessment
Sensor Observation Data
JDL 1 Object Detection and Feature Processing
JDL 4 Sensor Management
JDL 3 Threat Assessment
DREAMSNet
Sensors
32
General Approach

• Design a isolated block of software to handle the
  mission management needs.
• Design and isolate the interface to existing
  sensors and other fusion software packages.
• Optimize performance over separate geographic
  regions using a very new individual agent based
  approaches.
• Agents can actually physically move through the
  network where our software exists
• Agents make decisions concerning sensor operation
  using models we receive from the sensor
  manufacturer. (currently Lockheed Martin)

33
Sensor Management Blocks
Assessment Data from Existing System
Any Human Operator Information
Mission Manager
Operator
Sensor Manager
Sensor Information
Sensor Status Information
Sensor Models
34
New Segmented Geographical Approach
P2
P3
35
Flexible Performance Measures
Support Threat Identification
Normal Surveillance
Surveillance in Harsh Weather
Missions
,5
,3
,2
Regional Coverage
Low RCS Detection
Track Quality
System Subfunctions
,4
,1
,4
,2
,4
,6
,3
,5
,1
Accuracy
Pd in weather
Pd
Performance Parameters
Time Delay
,47
,22
,31
36
(No Transcript)
37
Evolutionary Program Approaches

38
Illustration of Optimization Process with Swarms
Sensor Selections
Sensor 1
Sensor 2
Sensor 3
Sensor Parameter Choices
Check Performance in the Volume
Pglobal
V
3D Space
39
Example Performance Evaluation
40
Choice 1 Boost Performance to More Uniformly
Cover Region
41
Choice 2 Sensor Manager Moves Sensors
42
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses for Sensor Networks Today
• Research Groups for Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

43
Biometric Security Network

• Control access
• Flexible and change security levels in real-time
• Improved accuracy over single biometric modality
• Minimize transaction time
• Serve all users (universal)

44
Biometric Terms

• Universality everyone is processed with equal
  ease and accuracy
• False Rejection Rate (FRR) rate at which the
  system does not allow access to someone who
  should be allowed, genuine user.
• False Acceptance Rate (FAR) rate at which the
  system accidentally allows someone in who should
  be restricted.
• Accuracy how well the system does on both FRR
  and FAR
• Failure to Enroll the system does not accept
  the initial biometric data for the individual.
• Failure to Acquire the initial set of data was
  accepted but the individual is unsuccessful in
  producing biometric data later. This may be due
  to a cold for voice recognition, injured hand for
  hand geometry or fingerprints, etc.

45
Why Choose Biometric Security?

• Improves authentication by automatically using
  physiological or behavioral characteristics to
  verify identity.
• Eliminates the problem of forgotten identifiers
  required for knowledge-based identity schemes.
• password problems account for between 40 to 80
  of all IT help desk calls. (Forrester Research,
  Inc. - http//www.forrester.com)
• resetting forgotten or compromised passwords cost
  as much as 340/user/year. (GartnerGroup -
  http//www.gartnergroup.com)
• Increases the difficulty and cost of forging the
  required identifiers.
• While fooling biometrics is extremely difficult,
  people can buy false driver's licenses and green
  cards for 50 in many American towns bordering
  Mexico. Most college campuses are awash in false,
  paper-based ID cards used to purchase drinks.
  (The Limits of Privacy by Amitai Etzioni)
• Eliminates the risk of guessing identifiers to
  gain access.

46
Bayesian Decision Fusion

• Each sensor decides to accept or reject a
  individual prior to fusion
• A global decision at the fusion center

FACE
VOICE
HAND
u1
u2
u3
FUSION
ug
47
WAN
ENROLLMENT SUBSYSTEM
AUTHENTICATION SUBSYSTEM
SECURITY FIREWALL
LAN
READER STATION
48
WAN
ENROLLMENT SUBSYSTEM
AUTHENTICATION SUBSYSTEM
SECURITY FIREWALL
LAN
READER STATION
49
WAN
ENROLLMENT SUBSYSTEM
AUTHENTICATION SUBSYSTEM
SECURITY FIREWALL
LAN
READER STATION
50
Current Technical Challenges Being Addressed

• Design a self-adapting biometric fusion
  processor.
• Addresses universality problem.
• Reduces failure to acquire with backup biometric
  data.
• Improve face recognition.
• Easier and less invasive biometric.
• Face recognition already exists through photo
  IDs, which can be used of individuals that are
  not enrolled.
• Advance smart card technology to increase
  communication distance.
• Ease of use.
• Individuals maintain their own biometric data
  set.
• Improves logistics with data already with the
  individual.

51
Functional Diagram of ABMF Algorithm
User Constraint
Security State
Biometric Sensor 1
Biometric Sensor 2
Biometric Sensor N
Cost Manager
Accept/Reject Decision
Accept/Reject Decision
Costs for False Acceptance and False Rejection
Accept/Reject Decision
Bayesian Decision Fusion
Optimum Fusion Rule
Particle Swarm Optimization
Accept/Reject
22N Possible fusion rules for N sensors
52
Characteristics of Resulting Adaptive Multimodal
Biometric Fusion (AMBF)

• Improves accuracy performance of authentication.
• Reduces the problem of a biometric identifier not
  being truly universal.
• Reduces problems enrolling individuals.
• Increases the systems tolerance to any temporary
  problems in collecting the biometric identifier.
• Provides the system with inherent adaptability to
  changing security needs.

53
AMBF The Particle Swarm Optimizer
PSO parameters
Random Initialization of Particles
iltn
Output the best solution
CFA
Cost Evaluation
To Fusion Processor
Velocity and Position Updates
Sensor Models
Update Particles Memory
Save the best solution so far
54
Example Performance Improvement Using Multimodal
Biometrics
1 Sensor FAR 0.000001 FRR NA
2 Sensors FAR 0.000001 FRR 62
3 Sensors FAR 0.000001 FRR 85
Improvement FAR 0.000001 FRR 23
55
Biometric Technology Market

• Finger-scan commands 50 of non-AFIS (Automated
  Fingerprint Identification System) biometric
  revenue.
• Facial-scan follows with 15.4 of the non-AFIS
  market.

56
Why focus on improving face recognition?

• Inexpensive biometric sensors.
• More universally accepted biometric.
• A simple drivers license may be used.
• Current techniques have poor performance.
• Currently, this is a modality that is easy to
  spoof.

57
EM Based Eigenfaces Example
58
Multiple Poses Are Better
59
Illustration of Approach More information in EM
Eigenfaces
Our EM Eigen faces a more interesting image
Eigen faces darker and more blurred
60
Performance Improvements with Single Image
Slight accuracy improvement
Larger time improvement
61
Improved Accuracy for 5 Training Faces EM
Eigenfaces contain more information
Very slight accuracy improvement
Significant time improvement 11 sec for 20
features
62
Reducing Energy Expended by Smart Card

• Adapt error control to the link quality and
  available power.
• Forward Error Control (FEC) Scheme with varying
  Reed-Solomon code rate.
• 53 savings over fixed rate of n6, m3
• 40 savings over fixed rate of n12, m3

63
Overview

• Introduce Sensor Networks
• Sensor Network Applications
• Problems Solved by Sensor Networks
• Minimizing Sensor Network Costs
• Business Advantages of Sensor Network Technology
• Businesses for Sensor Networks Today
• Research Groups for Future Sensor Networks
• DREAMSNet
• Environmental Sensor Network Concept
• Large Intelligent Defense Sensor Networks
• Biometric Security Sensor Networks
• Biometric Technology
• Adaptive Multimodal Biometric Fusion Algorithm
• Summary, Suggestions, and Questions

64
Summary

• Built Environmental Sensor Networks need to be
  tailored for the needs of the business. Much
  interaction required between customer and
  designer.
• Multimodal Biometrics required to support user
  diversity in biometric identifiers as well as
  changing physical characteristics over time.
• There are more similarities between the smaller
  networks and larger networks than expected.

65
Suggestions

• This is an explosive area of research now so its
  difficult to keep up with.
• Sensor Networks covers a large number of topics
  so make sure during discussions that both parties
  are discussing the same thing.
• Businesses will be the first to define new
  applications so contact between researchers and
  potential customers is essential.

66
Questions???

• Dr. Lisa Osadciw
• DREAMSNet
• 277 Link Hall
• Syracuse University
• Syracuse, NY 13244
• Visit us at www.ecs.syr.edu/research/dreamsnet

67
Results and Analysis Experimental Settings

• Gaussian models for sensors is assumed
• Swarm parameters

68
Results and Analysis Experiment I
Imposter Distribution
Genuine Distribution
Region of False Rejection
Region of False Acceptance
CFA 1.9 Minima Achieved 0.0102
Fusion Rule AND rule
69
Results and Analysis Experiment II
CFA 1.8 Minima Achieved
0.0138 Fusion Rule OR rule
70
Results and Analysis Performance of Particle
Swarm Optimization
CFA1.8
CFA1.9