Localized Edge Detection in Sensor Fields

1
Localized Edge Detectionin Sensor Fields

• Krishna Kant Chintalapudi, Ramesh Govindan
• University of Southern California
• Ad-hoc Networks Journal, 2003. 2
• Presentation Kim Sang-Won

2
Contents

• Introduction
• Edge Detection
• Assumption, Models and Terminology
• Metrics
• Three Approaches
• Statistical Approach
• Image Processing Approach
• Classifier-based Approach
• Simulation Results
• Conclusion

3
Introduction
Sensor Field

• Sensor network can be used for detecting
  large-scale phenomena
• Energy-efficient sensor network
• stores the detections of the phenomena within the
  network
• provides a query interface
• The boundary information is sufficient and more
  energy-efficient
• A key component of energy-efficient boundary
  finding algorithm is a localized edge detection
  scheme

4
Introduction

• Previous Works
• No Previous Works in Sensor network field.
• Edge detection in other fields
• Digital image processing.
• Filtering can be applied to localized edge
  detection
• ButIn Sensor network, there is no spatial
  regularity
• The efficacy of edge detection based on digital
  image filtering is unclear
• Consider two other scheme, statistical scheme and
  a classifier-based scheme

5
Edge Detection

• Assumptions, Models and Terminology
• Sensor nodes
• can be arbitrarily deployed, but each such node
  knows its location
• Can determine whether it belongs to the
  sub-region covered by the phenomenon or not
• can make measurement errors
• Edges
• Boundary of the phenomenon
• Edge sensor, Tolerance radius, Thickness of edge

6
Edge Detection

• Metrics
• Robustness
• Low intrinsic error
• Robust to reasonable levels of sensor calibration
  error
• Insensitive to the threshold settings over a
  broad range of operating conditions.
• Performance
• Energy expended in communication
• Quality of of the result thickness of the edge
• Percentage Missed Detection Errors
• False Detection Errors
• Mean Thickness Ratio

7
Three Approaches

• BasicCommon scheme
• Each sensor gathers information(location,
  Values(1/0) from sensors in its neighborhood and
  independently tries to determine if an edge
  passes within its tolerance radius
• Probing radius Tolerance radius
• Energy accuracy trade-off.
• Three Approaches
• The Statistical Approach
• The Image Processing Approach
• The Classifier-based Approach

8
Statistical Approach

• Gather data and perform statistical analysis to
  decide whether or not the sensor is an edge
  sensor.
• Example
• Advantage
• Can be explicitly tailored to be robust to errors
  using threshold, if error characteristics are
  known.
• The choice of an appropriate threshold
  would determine the performance of the scheme.

n of 1s n- of 0s
9
Statistical Approach

• Choosing (threshold)
• Sensor density 15 sensors in area of
  tolerance
• Sensor error probability p0.05

edge
edge
Detect!!!
Detect!!!
Unwanted detection
Pure false detection
10
Image Processing Approach

• Using high pass filtering tech. of image
  processing literature
• Sensors may not exhibit pixel-like regularity in
  placement.
• W(xs, ys) are weights to compensate for the
  uneven weighing cause due to arbitrary
  positioning and variations in number of the
  sensors. In general, W(xs, ys) is a function of
  sensor locations and H.
• The Prewitt filer based scheme

11
Image Processing Approach

• Example
• A high value of
  would indicate an edge.

Interior sensor(1)
Exterior sensor(-1)
12
Classifier-based Approach

• Approach comes from the pattern recognition
  literature.
• Find linear-classifier that partitions the data
  into two subsets.
• Classifier explicitly encodes a notion of
  geography
• Classifier does not require any thresholds

13
Simulation Result

• Simulation Framework
• 200m x 200m area, 10m Radio range
• Tolerance radius r 10m
• With Optimal Threshold( )
• Data set
• Linear boundary data sets Dl ymxc0
• Elliptical boundary data sets De E(a,b, x0 , y0
  , ?)0
• Factors
• Density
• Sensor errors
• Parameters
• R/r 1, 1.5, 2, 2.5, 3

200m
200m
200m
14
Simulation Results

• Energy Accuracy Trade-off

Statistical
Image Processing
Classifier-based
15
Simulation Results

• False Detections and Unwanted detections

Classifier based scheme
Detect!!!
Detect!!!
edge
edge
Unwanted detection
Pure false detection
16
Simulation Results

• Detection Probability with different sensor
  density and error rate
• Threshold variation
• Statistical scheme varies widely with changes
  in R/r and p.(0.790.17)
• Image processing based varies widely with
  changes in R/r at higher sensor density

17
Simulation Results

. Exterior sensors Interior sensors o
Edge sensors
18
Conclusion

• Introduced the problem of localized edge
  detection in a sensor field.
• Discussed an edge and proposed metrics to
  assess edge detection algorithms
• Proposed three qualitatively different approaches
  to edge detection namely statistical, image
  processing based and classifier based approaches.
• Dynamically setting thresholds will be hard to
  do.
• The classifier based scheme seems to be the most
  promising for localized edge detection.
• No Threshold, Thinnest edges and better True
  detection