Multifloor tracking algorithms in Wireless Sensor Networks

1
Multifloor tracking algorithms
in Wireless Sensor Networks

• Devjani Sinha
• Masters Project
• University of Colorado at Colorado Springs

2
Why Location Tracking is Useful? Adapted from
Motetrack presentation

• Assist Firefighters in Search/Rescue inside
  building
• Often cannot see because of heavy smoke are
  unfamiliar with building
• Use wireless sensors (badge/beacon) GPS does not
  work in buildings
• Can greatly benefit from a heads-up display to
  track their location and monitor safe exit routes
  
• Chicago City Council all buildings more than 80
  feet tall must submit electronic floor plans
  Forefront, Fall 2003
• Incident commander can better coordinate rescuers
  from command post

3
Related work

• Motetrack (Harvard Lorincz and Matt Welsh)
• TinyOS/Mote based.
• 3D location tracking using radio signal
  information
• Distributed/reference signature based. Thus more
  reliable.
• No Multi floor implementation
• Spot-on (Washington, Jeffrey Hightower and
  Gaetano Borriello/XeroxParc, Roy Want)
• RFID, 3D location Tracking
• Requires customized special software
• centralized
• No Simulation yet for Multi Floor
• FRSN Location Tracking
• TinyOS/Mote based.
• Multi-Floor Simulation
• 3D location tracking using radio signal
  information

4
Research Goals

• Single Floor Location Tracking
• Use Jeff Rupp's Obstructed Radio Model (2D)
• 2D Hill climbing algorithm
• Multi Floor Location Tracking (3D)
• Extend Obstructed Radio Model to 3D
• Extend Hill climbing algorithm to 3D
• Analyze the performance and impact factors such
  as scaling, height, initial sensor sets
• Develop tool to visualize the results.

5
Why Motes/TinyOS seems to be the right platform

• MOTES are small in size
• Easy to embed in environment and equipment
• MOTES can operate off of battery it is low
  power
• Resilient to infrastructure failure
• TinyOS is a well established platform
• Used by over 150 research groups worldwide
• Easy to integrate new sensors/actuators
• Mica2 mote

6
Modeling and Simulation

• TinyOS mote operating system
• TOSSIM - Simulate TinyOS mote network
• TinyViz visual TOSSIM
• Standard Java application
• Uses a plug-in architecture to allow for
  expansion
• Wide array of existing plugins
• Easy to expand

7
Obstructed Radio Model Plugin

• Authored by Jeff Rupp, UCCS
• Plug-in is based in the Radio Model done by
  Nelson Lee at Berkeley
• Assumes 60dB equates to a maximum bit error rate
• Radio signals are obstructed by varying amounts
  by different materials
• Loss in free space over distance
• walls presented low attenuation, about 3-12dB

8
Multi Floor model assumptions

• For sake of simplicity, the following assumptions
  were made
• The layout of each floor is identical.
• Every floor is setup with equal number of Beacon
  nodes 10ft above the floor. The mote layout is
  identical for each floor.
• The floor height is set at 10 ft.
• The attenuation of the floor/ceiling is assumed
  to be 20dB.
• Cubicle attenuation is assumed to be 15dB
• Outer Wall attenuation is assumed to be 35dB

9
Multi floor Setup in the GUI
symbol is beacon sensor node. The label is
sensor ID. Here small rooms has one sensor, large
room has two. The hallway has 6 sensors. The top
one is the sink node which collecting the sensor
data.
10
Hill Climbing Algorithm
LegendRed square is actual target location. 4
purple/grey dots are sensors with strongest
signals.
11
Hill Climbing Algorithm
Based on the initial sensor set, an estimated
location, x, is computed. Through perturbation,
four neighboring locations from x is calculated
and the one with closest estimated signal
strengths will be chosen for next round.
x
12
Responder Position in GUI
Here the red squares are randomly generated
firefighter locations. The overlay green squares
are estimated locations.
13
Performance Effect of Scaling Factors
Single Floor
Multi Floor

• Identical results for SF1 and SF2

• SF2 results in error and variance in tracking

14
Varying Z value for responder

• Marginal Differences

15
Top4 vs. Top3 motes

• Top3 results in error and variance in tracking

• Top3 results in zero convergence issues

16
Conclusions

• This concept can be developed using small,
  inexpensive and low-power devices
• Using radio signal information alone, it is
  possible to determine the location of a roaming
  node at close to meter-level accuracy.
• First Responder Sensor Network software provides
  an attractive solution to the critical problem of
  indoor location tracking.
• The multi floor model is quite robust to
  variations in z co-ordinate of responder.
• Using top 4 beacon motes in the algorithm gives
  more accurate results

17
Future Work

• Incorporate Java 3D API in TinyViz
• 2D Mote Network conversion to 3D
• Multi Floor display of Responder Positions
• Implementation of Multi Floor FRSN

18
Key References

• Konrad Lorincz and Li Li, MoteTrack A Robust,
  Decentralized Approach to RF-Based Location
  Tracking, Proceedings of the International
  Workshop on Location and Context-Awareness (LoCA
  2005) at Pervasive 2005, May 2005.
• MoteTrack An Indoor Location Detection System
  for Sensor Networks, Konrad Lorincz and Li Li,
  Harvard University. (http//www.eecs.harvard.edu/
  konrad/projects/motetrack/)
• Radio Signal Obstruction Plug-in for TinyViz by
  Jeff Rupp, CS526 from UCCS CO 80933-7150, Fall
  2003.
• TOSSIM A Simulator for TinyOS Networks by
  Philip Levis and Nelson Lee, (Version 1.0 - June
  26, 2003), September 17, 2003.

19
Questions?