TARGET CLASSIFICATION AND LOCALIZATION IN A HABITATMONITORING APPLICATION

1
TARGET CLASSIFICATION AND LOCALIZATION IN A
HABITAT-MONITORING APPLICATION

• Hanbiao Wang
• LECS, CS, UCLA

2
OUTLINE

• Goal
• Challenges
• System design
• Testbed implementation

3
REAL DEMAND

• Bio-acoustic community have struggled for years
  to build a system for animal call recognition and
  localization
• Current status
• collect data in the field with a few recording
  units synchronized by GPS
• process data off-line in the lab

4
GOAL

• We want a sensor network
• Recognize and locate a specified type of animal
  calls, one call at a time
• On-line real-time processing inside the network
• Long lifetime although powered by batteries

A beautiful dream
5
CLASSIFICATION ALGORITHM

• Spectrogram correlation
• Unique pattern for each type of animal calls
• Widely used in bio-acoustic community

6
LOCALIZATION ALGORITHM

• Time-difference-of-arrival (TDOA) based
  beamforming
• Known sensor locations, sound speed
• One target a time
• By professor Yao et. al.

TDOA
7
CHALLENGES

• Fine-grained time synchronization
• cm localization need 30 µs synch
• Difficult for NTP to achieve such sync in
  wireless network
• Reference Broadcast Synchronization (RBS) by
  Jeremy, a few µs in testbed

8
CHALLENGES

• Real-time on-line processing
• Spectrogram correlation and beamforming are very
  time-consuming
• Specified type of calls dont occur all the time
• Staged event-driven processing

9
CHALLENGES

• Long lifetime of battery powered system
• Communication is a major energy consumer
• Data reduction/compression before data transfer

10
SYSTERM ARCHITECTURE

• Tiered hardware platform, multiple micro nodes
  cluster around a macro node.
• Macro node for central coordination and coherent
  processing
• Micro nodes for distributed sensing and
  preprocessing

11
STAGED EVENT-DRIVEN PROCESSING

• Three stages, from fast to slow
• Signal intensity monitoring
• Target classification
• Target localization
• Not move to next stage unless pass the current
  stage

12
DATA REDUCTION

• Time series of sample signs identify signal
  cross-zero points, thus contain most timing
  information
• Correlation of two time series of sample signs
  indicates the same TDOA as that of original
  signals
• Each sample sign can be coded in 1 bit

13
TESTBED



Powered by
14
CLIENT SERVER MODEL

• One server on each sensor node
• Continuously sample signal, time-stamp samples,
  buffer the last few seconds of data (audio server
  by Lew)
• When receiving data request from cluster head,
  fetch data with specified starting time and
  duration, filter and reduce data, then send data
  to cluster head

15
CLIENT SERVER MODEL

• A daemon on cluster head
• Continuously monitor signal intensity
• When intensity exceeds threshold, classify signal
• When signal is the specified type of call,
  requests data from sensor nodes
• When all data are available, locate target using
  TDOA based beamforming

16
TESTING RESULT

• Spectrogram of observed frog call
• Reference spectrogram
• Cross-correlation coefficients

17
TESTING RESULT

• Raw waveform with wind noise
• Filtered waveform
• Reduced waveform segment

18
TESTING RESULT

• TDOA using filtered waveforms
• TDOA using reduced waveforms

19
TESTING RESULT
20
TESTING RESULT

• Signal intensity monitoring
• For every 2 ms data
• Takes 785 /- 5 µs
• Target classification
• Call duration 0.2 s
• Takes 361 /- 1 ms
• Target localization
• 5 sensor nodes
• Takes 500 ms

21
CONCLUSION

• Staged event-driven processing effectively enable
  system real-time
• Data reduction effectively reduces data volume by
  a factor of n, where n is raw sample size.

22
ALGORITHM REFERENCES

• D. K. Mellinger and C. W. Clark, Recognizing
  transient low-frequency whale sounds by
  spectrogram correlation, J. Acoust. Soc. Am.,
  vol. 107, no. 6, pp. 3518-3529, June 2000.
• H. Wang, L. Yip, D. Maniezzo, J. C. Chen, R. E.
  Hudson, J. Elson, and K. Yao, A wireless
  time-synchronized cots sensor platform, part II
  applications to beamforming, in Proc. IEEE CAS
  Workshop on Wireless Communication and
  Networking, Pasadena, CA, USA, September 2002.