An Empirical Study of Collaborative Acoustic Source Localization

1
An Empirical Study of Collaborative Acoustic
Source Localization

• Andreas M. Ali
• Kung Yao
• EE, UCLA

Travis Collier Charles E. Taylor Daniel T.
Blumstein Ecol. and Evolutionary Biology, UCLA
Lewis Girod CSAIL MIT
2
Contribution

• Online marmot detector based on Constant False
  Alarm Rate (CFAR)
• A centralized marmot call localization system
  based on Approximate-ML (AML) bearing likelihood
  estimation
• A field study on real animals in natural habitat

Yellow Bellied Marmot Rocky Mountain Biological
Lab (RMBL)
3
Outline

• Motivation to study Marmot alarm calls
• Acoustic ENS Box Platform and Methods
• Analysis and Results
• Summary

4
Motivation for Using Marmot Alarm Calls
Rocky Mountain Colorado Lab (RMBL)

• Biologically important
• Infrequent
• Difficult to observe
• 30 identified by observation

Marmot at RMBL
5
Acoustic ENS Box Platform
Acoustic ENSBox V1 (2004-2005)

• Wireless distributed system
• Self-contained
• Self-managing
• Self-localization
• Processors
• Microphone array
• Omni directional speaker

V2 (2007)
6
DOA Based Localization System
CFAR
J1
JN
J2

• Constant False Alarm Rate (CFAR) based event
  detection
• Combine AML Direction of Arrival (DOA) likelihood
  estimate to produce the source location estimate

7
Satellite Picture of Wide Deployment

• Rocky Mountain Biological Laboratory (RMBL),
  Colorado
• 6 Sub-arrays
• Burrow near Spruce
• Wide deployment
• Max range 140 m
• Compaq deployment
• Max range 50 m

8
CFAR Detector Performance
9
Source Directionality Test
10
Aligning Node Position with GPS Coordinate

• Compact Deployment
• GPS (red)
• Self-Loc (blue)

11
Pseudo Log-likelihood Map

• Compaq deployment
• location estimate
• spruce location
• Normalized beam pattern
• Collective result mitigate individual sub-array
  ambiguities
• Marmot observed near Spruce

12
Aligning Node Position with GPS Coordinate

• Wide Deployment
• GPS (red)
• Self-Loc (blue)

13
Location Estimates Scatter Plot for Wide
Deployment
Meter
Meter
Meter
Meter
6 sub-arrays
4 sub-arrays
14
Summary

• Non-intrusive acoustic source localization is
  practical and tractable
• Self-localization and synchronization feature of
  Acoustic ENS Box is essential for practical field
  deployment
• Combining multiple AML based DOA likelihoods from
  multiple nodes effectively overcomes ambiguity
  individual node suffers
• Redundancy can be used to identify and exclude
  sub-arrays which have especially poor data due to
  reverberations, multi-path, or other practically
  unavoidable problems.

15
Yellow Bellied Marmot
Thank you!
16
Generate Pseudo Log-likelihood Map from AML

• From a position, sum all the log-likelihood
  values of the bearing estimates
• Beam crossing

17
Deployment Collection Method
Node n

• Self-localization stage
• Acoustic ENS Box
• Detection/recording stage
• Each node
• Runs CFAR event detector
• Silent recording
• Each node estimate bearing via AML (can be run in
  fusion center)
• Fusion centerGenerate a Pseudo Likelihood map to
  estimate source location
• Deployment size (Compaq or Wide)

Self-loc
Store
Event?
CFAR event detector
AML
Yes
No
Fusion center
Pseudo Log-likelihood
Estimate
18
Performance Impact on Sub-array Size