A Protocol for Tracking Mobile Targets using Sensor Networks

1
A Protocol for Tracking Mobile Targets using
Sensor Networks

• H. Yang and B. Sikdar
• Proceedings of IEEE Workshop on Sensor Network
  Protocols and                 Applications , May
  2003
• Byun, Eun-kyu
• (ekbyun_at_camars.kaist.ac.kr)

2
Contents

• Introduction
• Challenges
• A distributed algorithm for predictive tracking
• Simulation result
• Conclusion

3
Introduction

• Tracking mobile target using sensor network
• Scenario military, civilian
• Issues
• Scalability
• Minimize the power consumption
• Distributed Predictive Tracking (DPT) algorithm
• Scalability and Robustness
• Using a cluster based architecture
• Power conservation
• Distributed mechanism for determining optimal set
  of sensors, others are in the hibernation mode
• Predictive mechanism to intimate cluster heads
  about approaching target

4
Challenges

• Scalable coordination
• Size of the network, the number of targets and
  number of active queries
• Tracking accuracy
• Reduce likelihood of missing a target
• Robust against node failure
• Ad-hoc deployment
• Computation and communication costs
• Minimize communication requirement
• Power constraint

5
Algorithm Overview

• Distributed Predictive Tracking (DPT) algorithm
• Does not require any central control point
• Assumes cluster based architecture for sensor
  network
• Ensure the sensor networks scalability and
  energy efficiency
• Sense predict communicate - sense
• Distinguished role between border and inside
  sensors

6
Algorithm Assumption

• Assumption of the DPT algorithm
• All sensors are have the same characteristics
• Sensors are uniformly distributed
• Sensors have high beam, low beam mode
• Sensors perform sensing according to its cluster
  heads requirements
• At lease 3 sensors to sense the target jointly
• Sensor density satisfy that probability that at
  least 3 sensors can sensing target is more than
  0.99
• Cluster head know location information of all
  sensors in cluster

7
Target Descriptor

• Target descriptor formulation algorithm
• Target Descriptor (TD)
• Target identity
• Targets present location
• Targets next predicted location
• Time stamp
• Upstream cluster head(CHi) send TDi to downstream
  cluster head(CHi1)
• Prediction mechanism
• Predict next location based on pervious n-1
  actual location
• Accuracy VS overhead
• Use linear predictor

8
Sensing

• Sensor selection
• 1. Cluster head choose 3 sensors such that their
  distances to the predicted location are less than
  the sensors normal beam
• 2. Choose insufficient sensors using high beam
  distance
• 3. Asks its neighboring cluster heads for help
• Wake up selected sensors
• Collect information
• Formulate TDi1

9
Failure Recovery

• Failure scenario
• Downstream Cluster head failure
• Prediction failure
• Recovery scheme
• First level recovery
• switch to high beam
• Second level recovery
• A group of sensors which are around r meters away
  from Li are activated
• Nth level recovery
• (2N-3)r meters away from Li are activated

10
Energy consideration

• Energy saving strategies
• Hibernation mode and prediction
• Sensors use normal beam whenever possible
• Communication cost of transmission of TD is
  insignificant
• TD has to be sent to sink
• The energy required for obtaining the TD for one
  location
• Keep pmiss small enough

11
Simulation result

• Simulation study concentrated on
• The miss probability under different situation
• The adaptability of the algorithm to targets
  different speeds
• Average energy consumed for tracking
• Simulation setup
• 600m X 600m 2-dim sensing area
• Normal beam 35m, high beam 55m
• The movement pattern random waypoint model

12
Simulation result

• Miss probability vs. Tracking resolution

13
Simulation result

• Miss probability vs. sensing radius/moving speed

14
Simulation result

• Energy consumption
• 15m/s
• 35m, 55m beam
• Tracking resolution 1s
• 59 misses over 2000 point
• E(Di)43.93, Var(Di)35.44
• Minimum 6000

15
Conclusion

• Distributed Predictive Tracking algorithm
• Prediction based on information of previous
  location
• Totally distributed and scalable
• Good performance at higher tracking resolution
• Aimed at minimizing the energy consumption
• Future Work
• More complicated prediction algorithm
• Interpolation mechanisms at cluster head when a
  sensors reports multiple target simultaneously
• Consider mobile sensors
• Drawback
• Energy consumption of border sensors
• Not good at very fast target
• Assuming Uniform distribution