Cheese Factory

1
Cheese Factory projectDistributed Data Fusion
In Peer2Peer Environment http//tisu.it.jyu.fi/che
esefactory
InBCT 3.2
M.Sc. Sergiy Nazarko nazarko_at_cc.jyu.fi
2
Area of interests

• Data fusion algorithms which can be used for
  target tracking and identification
• Kalman Filtering
• Transferable belief model

3
Distributed data fusion

• Real World Simulator (trajectory generator)
• Creates scenario of target movements
• Feeds sensors with data via XML-files
• Situation display
• Shows the real trajectory of target
• Measured trajectory
• Estimated trajectory
• Predicted trajectory
• P2PStudio
• Generates network failures

Real-World Simulator
P2P Studio
Situation display
4
Track Simulator

• Observation space temporarily
• divided into 6 sectors
• Every sector may contain any number of sensor
  nodes
• Possibility to create
• scenarios of target movements
• Generated data is stored in XML-
• files

5
Example of data for one sensor
6
Distributed Data Fusion

• Situation display is impossible without Data
  Fusion
• Collects information from the sensor nodes and
  combines them to obtain clear picture of the
  whole observation space
• Classifies the observation data and uses this to
  make joint estimation between different sensors
• Works as request-reply application

7
Situation display

• Now situation display is combined with data
  fusion in one node
• Target trajectory is displayed dynamically, when
  the Kalman filter computation gives new set of
  points
• Displays various types of trajectory
• Real trajectory (stored in XML files)
• Measured trajectory (given by sensors)
• sensor data is corrupted by noise
• Estimated trajectory (Kalman Filter)
• Predicted trajectory (Kalman Filter)

8
System Description

• External Real world simulator
• One Chedar-application different features
• Sensor node
• DisplayData fusion node
• Possibility to run simultaneously few
  DisplayData fusion nodes to protect system from
  the failure of one of them
• At any time any Sensor node can become processing
  node by simple clicking on the button
• Tracking application based on Chedar and Real
  world simulator have been developed

9
System Weaknesses

• Request-reply architecture
• Every new measurement is requested by flooding
  the whole network with resource request (in real
  world sensor should send coordinates when they
  are received)
• System works not in real time but it has its own
  time (timestamp), always starting from one
• If only one situation display node is running,
  and it crashes, new situation display node will
  start with initial parameters of Kalman filter
• Need time for filter adjusting

10
Current configuration

• Classroom with workstations running 12 Chedar
  nodes, which act as Sensor nodes
• Workstation running Situation display
• Trajectory is already generated

Lea Pulkkinens hall
Classroom
11
Future Work

• Add one or more targets into simulation
• Transferable belief model will appear on scene
• Make distributed processing network
• Separate Data Fusion node from Situation Display
  node
• Put few processing nodes in every sector
• Store backup copies of Kalman filter parameters
  in other nodes
• Make information about target follow the target
• Discover quality of measurements by destroying
  sensor node
• Find how quality of measurements depends on
  number of sensors
• What is the optimal quantity of sensors in one
  sector?

12
Future Work (continued)

• Improve request-reply architecture
• Smart sensor should send information about target
  whenever the information is available
• Data fusion node sends request periodically (e.g.
  every 30 seconds), in case new sensor appears on
  the scene
• Create a sensor network that measures a real
  environment
• E.g. 10 microcontrollers with infrared sensing
  chip and a radio interface chip, distributed all
  over Agora
• 1 or more moving targets -)

13
Thank You!