Adhoc Deployable FineGrained Localization for Wireless Sensor Networks

1
Ad-hoc Deployable Fine-Grained Localization for
Wireless Sensor Networks

• PhD Qualifying Examination
• Lewis Girod
• UCLA Computer Science Department
• girod_at_cs.ucla.edu

2
Wireless, Distributed Sensing

• Why Distributed Sensing?
• Closer to phenomena
• Improved opportunity for LOS
• 1/r4
• Why Wireless?
• Ad hoc deployment
• Remote locations
• Why Distributed Processing?
• Energy budget for comms
• Moores law brings down cost of local processing,
  does not affect radio propagation

3
Motivating Application

• Wildlife Habitat Monitoring
• Instrumented with cameras and microphones
• Task is to detect presence of bird and photograph
  it
• One approach
• Use microphones to detect birdcall and estimate
  location
• Then, select a camera that has the bird in field
  of view

Species Detection and Tracking
4
Relative Localization

• Sensors need to know relative locations
• Acoustic sensors need precise position/orientatio
  n for beamforming and signal localization
• Image sensors need to know FOV relative to each
  other and relative to acoustic sensors
• Tracking Task
• Compute trajectory of target relative to sensors
• Survey points reference to larger coordinate frame

5
Wildlife Localization Challenges

• Fine-grained
• Acoustic phase differences sub-cm
• Camera FOV varies with distance, as low as 10cm

6
Wildlife Localization Challenges

• Fine-grained
• Ad-hoc deployed
• Avoid explicit calibration or configuration
• Leverage density to increase opportunities for
  LOS
• Independent of environment
• Foliage, clutter, resilient to obstructions
• GPS

7
Wildlife Localization Challenges

• Fine-grained
• Ad-hoc deployed
• Energy constrained
• Communications energy
• High speed clocks
• Fast convergence time needed

8
Many Uses for Localization

• Collaborative Sensing and Tracking
• Calibration of sensor arrays
• Assessment of coverage density
• Guide placement of new nodes
• Selection of active nodes

9
Many Uses for Localization

• Collaborative Sensing and Tracking
• Routing
• Energy expenditure is a function of path
• Location is a natural namespace for physically
  motivated applications location to name
  endpoints
• Geographical routing
• Selection of next hop
• Metadata about coverage gaps (Holes)

10
Many Uses for Localization

• Collaborative Sensing and Tracking
• Routing
• UI
• Define tasks and results in context of location
• Physical UI
• Pointing, moving objects
• Selection of nearest phone or terminal
• Awareness of room boundaries

11
Review of Related Work

• RF Localization Systems (Time of Flight)
• GPS (1978)
• PinPoint Tags (5m) (Werb, et al, 1998)
• TimeDomain UWB Ranging (2000)

12
Review of Related Work

• RF Localization Systems (Time of Flight)
• RF Localization Systems (RSSI, Multipath)
• RADAR (802.11 - 5m) (Bahl et al, 2000)
• USC/ISI (Ricochet) (Bulusu et al, 2000)
• RadioCamera (US Wireless, 2000)
• Berkeley (Doherty et al, 2001)
• BWRC (Savarese et al, 2001)

13
Review of Related Work

• RF Localization Systems (Time of Flight)
• RF Localization Systems (RSSI)
• RF Localization Systems (Proximity)
• USC/ISI (function of density) (Bulusu et al,
  2000)
• Adhoc Positioning System (Niculescu et al, 2001)

14
Review of Related Work

• RF Localization Systems
• Acoustic Localization
• Active Bat (Infrastructure) (Ward et al, 1997)
• Cricket (Not fine-grained) (Priyantha et al,
  2000)
• AHLoS (In progress) (Savvides, et al, 2001)
• USC (Ranging only) (Girod, 2000)
• Characterization of NLOS conditions
• Cricket and Active Bat rely on beacon placement

15
Review of Related Work

• RF Localization Systems
• Acoustic Localization
• Adhoc Position Estimation Algorithms
• Convex Optimization (Doherty et al, 2001)
• Iterative Triangulation (Savarese et al, 2001)
• Iterative Multilateration (Savvides et al, 2001)
• No discussion of NLOS problem
• Real data vs simulation? Realistic environments?

16
Goal Environmental IndependenceApproach
Multimodal Localization

• Thesis
• Plan Validation through implementation
• Build a testbed for multimodal localization
  indoors
• Experiment by fusing acoustics and cameras
• Demonstrate improvement in accuracy and
  convergence time using existing position
  estimation algorithms

Any individual mode of sensing used for
localization (e.g. acoustic) will suffer from
unrecoverable ambiguities and undetectable
errors. In many cases, these errors are
persistent features of the environment and are
not readily eliminated statistically. A more
robust solution to this problem is to
cross-validate sensor data with data gathered
from alternate perspectives and from other modes
of sensing.
17
Problem Statement

• Consider a collection of sensors Si, with
  coordinate Xi and orientation ?i .
• Given a subset of Si, i lt k, are survey points,
  with defined values for Xi and ?i ,
• Given a set of measurements that relate the
  positions and orientations of Si,
• Estimate Xi and ?i .
• Design of position estimation algorithm depends
  on nature of constraints Nature of constraints
  depends on types of ranging.

18
Early Results with Acoustic Ranging

• Initial work has focused on implementation and
  characterization of an active acoustic rangefinder

19
Acoustic Ranging Error Model

• Our observations showed that we could express
  acoustic ranges as
• Rij Xi Xj2 nij
  Nij,
• where
• nij is a gaussian error term (?0,?1.3)
• Nij is a fixed bias present only when LOS blocked
• While nij can be reduced by repeated
  observations, Nij cannot because it is caused by
  persistent features of the environment, such as
  detection of a reflection.

20
NLOS Problems

• Magnitude of Nij is often on the order of true
  range
• usually the caused by a reflection
• results in poor performance from any algorithm
  that assumes gaussian-distributed error
• Conclusion
• For good position estimation, we must find ways
  to detect and eliminate non-LOS range data

21
Single-mode NLOS Detection

• Geometric consistency checks
• Triangle inequality
• Angle of arrival (if known)

22
Single-mode NLOS Detection

• Geometric consistency checks
• Good cluster analysis
• Cluster of ranges believed to be LOS (green)
• Inconsistency detected in ranges to 4th point
  (red)

23
Single-mode NLOS Detection

• Geometric consistency checks
• Good cluster analysis
• Motion analysis
• Record sequence of ranges from moving object
• Given model of track, deduce positions of sensors
• Discontinuities and inconsistencies suggest NLOS
  (similar to good cluster analysis)

24
NLOS not always locally detectable

• Obstructed LOS can introduce ambiguities that are
  not always locally resolvable based on acoustic
  range data

• For example, reflections can cause ambiguity.
  The simplest explanation (red nodes), is not the
  true explanation (green nodes).
• To solve this kind of problem, alternative
  hypotheses must be formulated.
• These hypotheses are not always decidable without
  global knowledge.. bad for scaling

25
Corner/Reflection Ambiguity

• Not locally resolvable
• Any range is potentially a reflection

• Locally resolvable with additional modes of
  sensing
• Camera reflections may not be correlated with
  acoustic reflections
• Cameras may be able to distinguish reflections
  from LOS
• Cameras may be able to gather more information
  about structure of scene (e.g. walls)

26
Multi-modal NLOS Detection

• Cross-validation across sensor modes
• Simpler, local algorithms
• Resolves ambiguities faster, more certainty
• Less communication overhead (faster convergence)
• Example Cameras and IR LEDs

• Camera sees LEDs (red dots)
• LEDs emit ID coded pulses
• Acoustic ranges estimated between all components

27
Multimodal Cross-checks

• Provides rich set of cross-validation
  opportunities
• Any LED seen by the camera is known to be LOS,
  therefore acoustic range between camera and that
  node is likely to be true
• LED images have known range (from acoustics)
  therefore depth map from only one camera
• With a depth map, the distance between two
  visible LEDs can be approximately determined,
  enabling further validation
• Stereopsis is simplified, because the coded LED
  pulses solves the correspondence problem

28
Work Plan

• Improvements to acoustic ranging (1 quarter)
• Development and characterization of image-based
  ranging (2 quarters)
• Experiment with cross-validation and test using
  position estimation algorithms (2 quarters)
• Writeup (2 quarters)

29
1. Improving Acoustic Ranging

• Reduce orientation dependent errors
• Currently incurs biases when sensor is pointed
  away from emitter, caused by signal diffracting
  around edge of sensor.
• Solve problem by having two microphones pointed
  in opposing directions, count earliest arrival.
• Locally determine angle of arrival
• Leverage existing H/W (2 microphones)
• Because we are aiming this at small form factors,
  we do not assume a long baseline (4cm).
• Expect error of /- 15

30
New Acoustic Data Collection

• Characterize performance of new 2-channel sensors
• orientation dependence and estimating AOA
• Multipoint indoor localization tests
• Use data to drive position estimation algorithms
• Collaboration with Sascha Slijepcevic
• Test with mixtures of LOS / NLOS data, evaluate
  algorithm performance as a function of NLOS data
  mixture

31
2. Development of Image-based Ranging

• We plan to develop an image-based ranging system.
  
• We plan to use the GALORE camera platform for
  this work.
• Camera software sends radio message instructing
  LEDs to be activated camera delays, then
  captures image
• Sequence of images is stored for post-processing
• Then analysed to detect LEDs

32
LED Detection Algorithms

• We plan to investigate a number of algorithms
• Early work looked at a naïve thresholding
  algorithm
• Did not work well, sensitive to environmental
  changes
• A technique based on mutual information looks
  promising (Viola97).
• Basic idea is sequence of frames is compared with
  sequence of frames generated by known LED pattern
• Translation and dilation selected to find best
  match
• Need to find effective search technique

33
3. Cross Validation Position Estimation

• Integrate several modes to try to detect NLOS
  conditions in realistic environments. Basic
  plan
• Use cameras to detect cases that are definite LOS
  or NLOS
• Use cameras to compute AOA for visible nodes
• Use cameras to estimate distance between visible
  nodes
• Try to find cases where cameras and acoustics
  both give the wrong answer
• How much does NLOS detection improve the results?

34
The End Thank you..
35
(No Transcript)