KANTeNET Knowledge Enabled Sensor Network Middleware
Overview
1. Application Scenario
2. Case Studies
3. Middleware Architecture
Application Scenario
Hit and Run
Application Scenario
Witness places telephone call to police with description of suspect vehicle.
Police place query to system with description of suspect vehicle.
Application Scenario
Sensors scan environment for vehicles matching description.
Several vehicles spotted and identified as possible matches.
Application Scenario
Vehicle coordinates distributed to ground level sensors and Google Maps.
License plate number and/or driver image captured and sent to DMV and watchlist databases.
Application Scenario
Police patrol coordinates matched against suspect vehicle coordinates and interception orders distributed, along with:
• Continually updated Google map
• Info about vehicles
• Info about drivers
Application Scenario
Result of Hit and Run
Application Scenario
Stand off Staring
Close in Staring
Surface & Near-Surface Staring(SUAV, Bldg Sensors, Taggants, UGS)
(“Cupid Fire”)ATR-Driven Small UAV on Steroids
Sensor Aided Vigilance
• GWOT requires ability to operate seamlessly across layers to sense and track asymmetric threats.
• Puts increased demands on novel concepts for establishing and exploiting netted persistence and empirical phenomenal data.
• Key role for revolutionary taggant materials and advanced data management, all within an “integrated solutions” framework
SAVig
Goal Continuously track dismounts and vehicles in complex urban environments.
Objective Fingerprint, Detect, ID and Track
Dismounts
Vehicles
Fortifications
Ordinance / Weapons / IEDs
Approach Hyperspectral Sensors and Imaging (HSI)
Offer High Spatial and Spectral resolutions
Understand and Exploit HSI phenomenology for detection and tracking of urban targets.
Architecture
Recognition
Detection
Tracking
Identification
Recognition
Detection
Tracking
Identification
Recognition
Detection
Tracking
Identification
Storage
Storage
Storage
Storage
Storage
Raw Data
Specific Information
Action Action Action
Registration
Collection
Control
Sensor
Filter
Registration
Sensor Data
Processing Hierarchy
Data Management
Overview
1. Application Scenario
2. Case Studies
3. Middleware Architecture
Case Studies
1. GSN
• Global Sensor Network
• Digital Enterprise Research Institute (DERI)
• http://gsn.sourceforge.net/
2. Hourglass
• An Infrastructure for Connecting Sensor Networks and Applications
• Harvard
• http://www.eecs.harvard.edu/~syrah/hourglass/
3. IrisNet
• Internet-Scale Resource-Intensive Sensor Network Service
• Intel & Carnegie Mellon University
• http://www.intel-iris.net/
4. SNSP
• Sensor Network Services Platform
• University of California, Berkley & DoCoMo
• http://chess.eecs.berkeley.edu/
GSN
1. Global Sensor Network
2. Sponsored by DERI
3. Open Source - http://gsn.sourceforge.net/
Conceptual Data Flow in a GSN Node
GSN Container Architecture
Overview
1. Application Scenario
2. Case Studies
3. Middleware Architecture
Architecture
Gateway Node Gateway Node
Acquisition
Annotation
IntegrationSensor Data Manager
Live Data/Metadata Stream Archive Repository(Data/Metadata)
External Integration(sensor web, web services, etc)
Interface Manager / API
Gateway Node
SemanticModel
ProducerLayer
OperationLayer
ConsumerLayer
Query Manager Context Aware Repository
Semantic Enhancement
Data Provenance
Suppose a sensor network detects a shark swimming in the ocean.
How can we verify that the identified object is a shark?
We must look at the data used in the identification.
Identification is a complicated process where the data can be continuously altered.
Tracking data through such a workflow is notoriously difficult.
Our solution is to annotate the sensor data throughout its life-cycle, from acquisition to response, so that transformations and analysis can be processed without losing contact with valuable intermediary data.
Semantic Enhancement
Contextual Query
• Again, suppose a sensor network detects an object in the ocean, but cannot determine whether the object is a shark or a submarine?
• If we can access knowledge about the surrounding environment, then may be able to determine that the particular coordinates represent a known shark infestation.
• Now we may be able to reasonably make a determination using knowledge external to the sensor network.
Our solution is to provide a suite of domain aware ontologies containing knowledge of not only the sensor network but also the deployed environment.