Transcript Document

Localization and Tracking Noise
Sources with Autonomous Vehicles:
From Node Processing to Central
Command Fusion and Tracking
Dr. James Carneal, Prof. Marty Johnson, Dan Mennitt, and Philip Gillett
2006 NC-ASA Spring Meeting
April 6th and 7th
Ramada Blue Ridge, Raleigh, NC
Motivation
• The use of autonomous
unmanned vehicles (AUV)
to perform acoustic
surveillance removes
humans from dangerous
environments.
Microphone
Array
Autonomous
Vehicle
• Congress passed a mandate
requiring 1/3 of all ground
armed forces to be
autonomous by 2015 .
Typical Autonomous Vehicles
Aircraft
Land
based
Grand
Challenge
Underwater
Acoustic Surveillance: Approach
•
•
•
•
Small microphone arrays.
Limited local processing.
Broadcast over network.
Tracking system to
account for:
–
–
–
–
–
Sparse data sets
Positioning of AVs
Motion of AVs
Number of AVs
Vehicle self noise
• Experimental based
approach.
Noise Source
Tracking system
System Structure
Event
Intelligent nodes
(limited processing)
Environment
Environment
Environment
Event
Detection
1
2
Angle of
Arrival
Coarse
Classification
Event
Info.
System Level
Battlefield
model
Localization
Storage
Buffer
Detailed
classification
Request for
detailed Info
3
AV Array
Node Level Processing
System Level
Processing
Node Level Array Design
• Array Design
– Modular
– Vehicle mounted
– Redundancy and
robustness
– Free-field
characterization
GPS
Compass
Modular
• Self noise removal
– Design array mics for low
noise
– Use engine mounted
sensors to remove noise
(LMS)
– “Quiet mode” of operation
Battery storage
and generator
power
Vehicle
mounted
Magnitude (dB)
0
500
1000
0
0.02 0
Source
500
1000
1500
2000
Frequency (Hz)
Magnitude
0.02
Imp(x)
= 54
0
-0.02
0
0.2
0.4
0.6
Time (seconds)
2
1.5
Ton(X)
= 39
1
0.5
0.8
1
0
0
2000
4000
6000
Frequency (Hertz)
8000
0.2
2500
2500
30
0
50
20
100
10
3000
3500
0
150
200
Angle of Arrival (degrees)
3000
3500
Instantaneous Power
250 Running300
350
Average Power
8
6
0
500
1000
0.4
0.6
Time (seconds)
1500
2000
Frequency (Hz)
4
0.8
1
0
2500
3000
3500
Ton(X)
= 121
2
2.5
0
1500
2000
Frequency (Hz)
Instantaneous Power
Running Average Power
Imp(x)
= 5.5
-0.01
Magnitude
0
0
3
0.04
-0.04
20
Magnitude
0.005
Instantaneous Power
Running Average Power
0.04 20
Magnitude (dB)
0.01
-0.005
0.06
40
0
Magnitude
– Loudness
– Angle of arrival
(diffraction
compensated)
– Impulse detection
– Tonality
–…
Magnitude of PSF
• Coarse Classification
40
Magnitude (dB)
Node Level Processing
0
2000
4000
6000
Frequency (Hertz)
8000
System Level Processing
Sensor Nodes
Event
Data used to
respond to
events
Network
communication
Environment
model
Central
System
Event
:
:
:
:
Event # 134
Node #3
14.34Hrs
23.2sec
37.228804 N
80.427313 W
Az=1230
Impulsive
Event # 135
Node #10
14.34Hrs
23.8sec
37.228790 N
80.427353 W
Az=-320
Vehicle
:
:
:
:
Networked Data
Structure
:
:
System Level Details
• Fusion Center:
– Localization:
associate node
information to
events
– Classification:
merge events to
objects
– Tracking: merging
objects to paths.
Event t=4
same
Event t=6
System Level: Initial Simulation
• Noisy
estimates
with
moving
sensor
nodes and
sources
Kalman Filtering
Filters
estimate
Noisy
estimate
Moving
sensor
nodes
Experimental Data Collection
• Algorithms developed using real data
Noise source
Acquisition equipment
Arrays
Localization results
Three arrays
Camera
Jamie
Impacts
Durham Hall
Localization results
Bus
Three arrays
Camera
Durham Hall
Localization results
Bus
Camera
Durham Hall
Low cost nodal processors
Laptop
6-channel data
acquisition
GPS & Time
USB
Compass
&Temp
• Low cost: $1600/node.
• Allows fixed/moving array measurements over
large areas
Node Level Challenges
• Reduction of footprint integration.
• Improved signal processing:
– Characterization
– Beamforming
• Eventually, other sensor types.
System Level Challenges
Time reversal
• Event confusion due to
propagation delays
No Directivity
– Simplification using
event characteristics
• Variable node geometry
– Accuracy variations with
space
• Sparse/intermittent
data
• Sub-set detection
– Multiple “confidence”
levels
- Events
- Nodes
300 Directivity
Conclusions
• A system for Localization and Tracking
Noise Sources with Autonomous
Vehicles has:
– been developed
– experimentally demonstrated (in a realistic
environment).
• Next Step: Physically implement low
cost nodal processor system.
– Implementation on AV’s will further test
and improve the systems capabilities.