1. Emerging Sensor Networks Can Make Sense for Your Business Tech 2003, May 7, 2003 Dr. Lisa Ann Osadciw from DREAMSNet – ( D evelopment and R esearch in E volutionary A lgorithms for M ultiSensor S mart N etworks) http://www.ecs.syr.edu/research/DREAMSNet/
2.
3.
4. Sensor Network System View Observation Space: nature, people, etc. Information Space: aircraft detection and location, person identification, etc. Sensors Sensor Manager (AI) Data Fusion Processing Sensor Status Controls Measurements
5. Current Research Focus Areas DREAMSNet Database Information Acquisition Trusted Data Network Sensors Intelligent Processing Control and Response Resource Management
33. Sensor Management Blocks Mission Manager Sensor Manager Sensor Information Sensor Models Assessment Data from Existing System Any Human Operator Information Operator Sensor Status Information
35. Normal Surveillance Support Threat Identification Surveillance in Harsh Weather ,6 ,3 ,2 ,5 ,3 ,1 ,2 ,4 ,4 ,5 ,1 ,4 ,47 ,22 ,31 System Subfunctions Pd Time Delay Accuracy Flexible Performance Measures Regional Coverage Low RCS Detection Track Quality Pd in weather Missions Performance Parameters
36.
37. Evolutionary Program Approaches * Technique Date Published Probability of Optimality (local optimal) Reachability (global optimal) Computation Time Complexity Traveling Salesman Problem (classic NP complete) 1930 5 (worst) 1 (best) 5 (worst) 5 (worst) Genetic Algorithm 1975 1 (best) 5 (worst) 2 2 Simulated Annealing 1983 2 4 3 3 Particle Swarm Optimization 1995 3 3 1 (best) 1 (best) Ant System 1995 4 2 4 4
38. Illustration of Optimization Process with Swarms Sensor 1 Sensor 2 Sensor 3 P global V Sensor Selections Check Performance in the Volume 3D Space Sensor Parameter Choices
47. LAN WAN ENROLLMENT SUBSYSTEM AUTHENTICATION SUBSYSTEM SECURITY FIREWALL READER STATION
48. LAN WAN ENROLLMENT SUBSYSTEM AUTHENTICATION SUBSYSTEM SECURITY FIREWALL READER STATION
49. LAN WAN ENROLLMENT SUBSYSTEM AUTHENTICATION SUBSYSTEM SECURITY FIREWALL READER STATION FUSION ACCEPT REJECT FUTURE BIOMETRICS
50.
51. Functional Diagram of ABMF Algorithm Bayesian Decision Fusion Biometric Sensor 1 Biometric Sensor 2 Biometric Sensor N Particle Swarm Optimization 2 2 N Possible fusion rules for N sensors Costs for False Acceptance and False Rejection Optimum Fusion Rule Cost Manager User Constraint Security State Accept/Reject Accept/Reject Decision Accept/Reject Decision Accept/Reject Decision
52.
53. AMBF – The Particle Swarm Optimizer Random Initialization of Particles Velocity and Position Updates Cost Evaluation Save the best solution so far Update Particles Memory i<n PSO parameters C FA Sensor Models Output the best solution To Fusion Processor
54. Example Performance Improvement Using Multimodal Biometrics 1 Sensor F AR = 0.000001% F RR = NA% 2 Sensors F AR = 0.000001% F RR = 62% 3 Sensors F AR = 0.000001% F RR = 85% Improvement F AR = 0.000001% F RR = 23%
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.
63.
64.
65.
66.
67.
68. Results and Analysis – Experiment I C FA = 1.9 Minima Achieved = 0.0102 Fusion Rule = AND rule Imposter Distribution Genuine Distribution Region of False Rejection Region of False Acceptance
69. Results and Analysis – Experiment II C FA = 1.8 Minima Achieved= 0.0138 Fusion Rule= OR rule
70. Results and Analysis – Performance of Particle Swarm Optimization C FA =1.8 C FA =1.9