Baqer 2007 Pattern Recognition for Wireless Sensor Networks Mohamed Baqer 24 May 2007
2 M. Baqer Outline Sensor Networks Energy Conservation Patterns and Sensor Networks Application So What’s the Big Deal? Challenges of Event Recognition in Sensor Networks Event Recognition for Sensor Networks Voting Graph Neuron VGN Model Voting and Consensus Sleeping Mode Example SGSIA Summary
3 M. Baqer Sensor Networks Random vs. deterministic deployment Long term deployment Dynamic infrastructure Unattended operations Scale
4 M. Baqer Energy Conservation Scheduling-based –Operation mode (transmitting, receiving, idle, sleeping) In-network Processing-based –Aggregation –Compression –Beamforming –CSIP
5 M. Baqer Patterns and Sensor Networks Spatio-temporal event patterns Pattern collection –continuously –periodically –Even-driven –User-driven –hybrid
6 M. Baqer Application: Structural Health Monitoring SHM replace visual inspection Applied for – Predict – Detect – monitor structures for damages
7 M. Baqer So, What’s the Big Deal? Can’t centralised servers (base station / sink node) perform pattern recognition for sensor networks? –Geographically dispersed sensory data –Require global information –Communication overhead –Offline detection
8 M. Baqer Challenges of Event Recognition in Sensor Networks Global vs. local Constraint resources Dynamic infrastructure Energy efficiency Scalable
9 M. Baqer Event Recognition for Sensor Networks Template matching Distributed artificial intelligence Cooperative distributed problem solving
10 M. Baqer Voting Graph Neuron Model Storage Communication
11 M. Baqer VGN algorithm Votes vectors: –Local match Use: –Local processing, information exchange and decision fusion Consensus –Cooperatively negotiating by casting votes –Cast and rebuild vote vectors
12 M. Baqer Sleep Mode Committee members enter into sleep mode to conserve their energy Who may go into the sleep mode? –Committee members that already cast their vote –Committee members with identical votes When do identical vote vectors get created? –Initialisation stage –Negotiation stage
13 M. Baqer Example Input sensory patternCommittee negotiation process Colour map of the negotiation process
14 M. Baqer Comparison results of the difference in the pattern matching performance for committee storing random patterns and alphabet character patterns
15 M. Baqer SGSIA In-network Data Processing for Secure Grid-Sensor Integration Architecture Provide timely and accurate responses to data acquisition requests intended for WSNs Data processing at the sensor nodes to filter raw sensory data Optimal and selective forwarding of grid-generated queries to the appropriate sensor networks. Grid proxy: interface, QoS, cashing Gateway (base station): managing, fuse, translate
16 M. Baqer SGSIA
17 M. Baqer Summary Ambient intelligence Decentralised in-network pattern recognition Scalability Adaptability
18 M. Baqer Questions
19 M. Baqer Acknowledgment Zubair Baig And my supervisor Asad Khan