20
Tracking Mobile Sensor Nodes in Wildlife Francine Lalooses Hengky Susanto EE194-Professor Chang

Tracking Mobile Sensor Nodes in Wildlife Francine Lalooses Hengky Susanto EE194-Professor Chang

Embed Size (px)

Citation preview

Tracking Mobile Sensor Nodes in Wildlife

Francine Lalooses

Hengky Susanto

EE194-Professor Chang

Outline

Recap Tracking with Binary Sensor Network Distributed Predictive Tracking Algorithm Our Tracking Approach Future Work References

Recap

Sensors only monitor land animals Animals are tagged Sensors placed at certain location Better understanding of region/animal

relationship Not specific to animal size or velocity

Tracking with Binary Sensor Network

Assumptions: Sensors have only one bit of information Sensors broadcast bit to base station (BS) Proximity sensor requires one more bit

One bit of information gives accurate predictions about direction of motion Approaching (+) Moving away (-)

Simple broadcast protocol

Tracking a Moving Object with a Binary Sensor Network, Dartmouth College.

Binary Sensor Network Geometry

Future position lies: Inside plus sensor overlap (+) Outside minus sensor (-)

+ +

-

Tracking with a Proximity Bit

Simulation results Estimated trajectory (star – dashed line) Actual trajectory (triangle – line) Plus sensors (squares) Minus sensors (circles)

Object gets in range at time 3

Binary Sensor Summary

Advantages: Trajectory prediction error is low Broadcasting single bits over network feasible BS computation is fast

Disadvantages: Only tracks one animal at a time No consideration for energy efficiency No failure recovery model

What if . . .

Tracking Algorithm: Low Duty Cycle

Distributed Predictive Tracking Algorithm

No central point Cluster based architecture Assumptions:

Randomly distributed sensors Default to normal beam Hibernation mode

Predictive mechanism Cluster head activates appropriate sensors

before target arrives

A Protocol for Tracking Mobile Targets using Sensor Networks, RPI

DPT: Target Descriptor Formulation

Target descriptor (TD) consists of: Target’s identity

Unique Created when target first

detected Target’s present location

Sensor triplet triangulation Target’s next predicted location

Alerts CHs most likely approached

Linear predictor Time stamp

Time TD created

DPT: Sensor Selection Algorithm

Prediction: When CHi predicts target location, downstream

CHi+1 receives message CHi+1 has information of all sensors in cluster

Selection: CHi+1 locally decides sensor-triplet to sense

target and sends wake-up message Each sensor sends location message to CHi+1

CHi+1 formulates TDi+1

CH CHi+1 CHi+2

CH = cluster head TD = target description

downstreamupstream

DPT: Failure Recovery

Failure: If upstream CH does not receive confirmation

from downstream CH, assumes downstream CH is not available and target lost

Target changes direction or speed and moves away from predicted location

Recovery: Wake up all sensors

within area Calculated from target’s

previous actual location

DPT: Energy Considerations

Sensor-hibernation method Most sensors stay in hibernation mode Only chosen sensors become active

Energy for obtaining TD of one location

Keep pmiss small to minimize energy consumed for recovery

Energy consumed in failure recovery

Failures cause extra communication between clusters and sensors

Etotal = (1 - pmiss) Esuccess + pmiss Efailure

Efailure = Esuccess + 3EHBPHB + (1-PHB)(3EHB+C)

Our Tracking Approach Nature

Cluster based algorithm Hierarchical approach

Variables: distance = E[velocityrunning] * time

CH CH

d

d

CH

First level CH

Master CH

Our Tracking Algorithm

Calculation based on maximum hop and popularity

Variables: h = CH hop count

2d

1h

4h

2h

Lost region

Our Intuition

Accuracy finding lost target improves over time More information = better search boundary Error handling wakes up all nodes in region of

diameter 2*d Advantages over sweeping across region:

More energy efficient Less network traffic

Sweeping Across the Region

Perform increment layer outward from last seen position until lost target is found or reaches border layer

Only notifies their neighbor at outer layer When successful, the founder takes over target When target is not found, border sensors report to node in

charge Awake all nodes in region and flood network Running time is O(n)

Sensor node layers

Example of sweeping:

Future Work

Failure and recovery algorithm Further develop our algorithm Compare DPT with our algorithm

Performance Energy efficiency Error handling

References

A Protocol for Tracking Mobile Targets using Sensor Networks. H. Yang and B. Sikdar. RPI.

Tracking a Moving Object with a Binary Sensor Network. J. Aslam, Z. Butler, F. Constantin, V. Crespi, G. Cybenko, D. Rus. Dartmouth College, CSU Los Angeles.

Rumor Routing Algorithm for Sensor Networks. D. Braginsky, D. Estrin. UCLA.

The ACQUIRE Mechanism for Efficient Querying in sensor Networks. N Sadagopan, A Helmy. USC.

Distribute Online localization in Sensor Networks Using a Moving Target. A Galstyan, K Lerman, S Pattem. USC.

Distributed Target Classification and Tracking in Sensor Networks. R. R. Brooks, P Ramanathan, A Sayeed. Penn State University and University of Wisconsin.

Detecting Moving Radioactive Source Using Sensor Networks. D Stephens, A Peurrung. Pacific National Laboratory.

Questions