U.S. Army Research, Development and Engineering Command
PVT WORKSHOP
16 June 2009
Sachi Desai & Shafik Quoraishee
U.S. Army RDECOM ARDEC
UNCLASSIFIED
Acoustic & Networked Sensors Division
Classifying Quadrupeds/Bipeds Employing
Ultrasonic Transducers
By: Sachi Desai & Shafik Quoraishee
Contact: [email protected]; 973.724.4314
US Army RDECOM-ARDEC
Acoustic & Networked Sensors Team US ARMY Acoustic Center or Excellence
Outline
• Objective of Project
• Past Work
• Sensor Modality/Data Collection
• Overview of Data
• Algorithm Process
• Current / Future Work
• Summary
Purpose
• The fidelity of an ultrasonic (active/passive)
sensor as a classifier.
• Investigate ultrasonic sensing modalities for
discrimination of mammals based
movements.
• Identify tell-tale characteristics within signal.
• Develop a classifier capable of identifying
classes of bipeds and quadrupeds.
• Integrate as part of a multi-modal sensor
solution to increasing detection confidence.
Motivation / Requirements
•
Call 1: Processing, Analysis and Reporting of Intelligence Information
– Tracking of ground objects from moving target indicator, radar, and detection,
and geospatial location of objects by acoustic, seismic, magnetic sensors and
change detection.
•
Call 2: The Ability to Observe and Collect Information Worldwide
– Find, fix, classify, and track friendly, enemy, and neutral fixed and moving
equipment and people.
•
DHS – Secure Border Initiative Network:
– Detect an entry when it occurs.
– Identify the entrant (i.e. human, quadruped).
– Classify its level of threat (who the entrant is, what the entrant is doing, how
many, etc.).
– Respond effectively and efficiently to the entry, and bring the situation to the
appropriate law enforcement resolution.
A current security need for person of interest detection.
Background of Work
• Dr. Alexander Ekimov & Dr. James Sabatier in
2006 published work utilizing ultrasonic Doppler
sensing modalities for human motion detection.
• Showed high frequency signals can capture
movement of footsteps with a high-resolution.
• Investigated against other types of mammals
movement(i.e. horses, dogs).
Another disparate sensor modality for personnel detection.
Initial: Equipment / Layout
Courtesy of NCPA
Person Walking In Hall Way
(NCPA Background Research)
• Active ultrasonic detection of walker.
–
–
–
–
Courtesy of NCPA
NCPA Spectral Plot
Carrier frequency, 40KHz, 96KHz sample rate, 16KHz FFT
Doppler based movement detection.
Walker direction clearly seen (dependent on walking direction)
• Illustrates the potential for detection & classification
Steps to Classification
• Identify automatically this perturbations
around the carrier signal that relate to
physical motion.
• Develop preprocessing techniques to
accelerate identification of event.
• Implement JTFA techniques to extract
features capable of characterizing the
event.
Biological Inspiration For
Mammal Discrimination
Web Extracted Image
• Bat echolocation provides the
ability to detect prey.
• The echolocation acts as a
mapping function that
augments poor vision.
• The active ultrasonic Doppler
sensor sends on a “standing
wave” to detect movement.
• A bio-mimetically incorporation
of other sensing modalities
increase event identification.
Active Ultrasonic Transducer
60o FOV
Biped Event
Abstract of Representation of
Domain of Interest
Standing
Wave
Activity of Interest
40.5-42KHz
Further
Extremities
from Body
Torso
Large Quadruped Raw Output
Large Quadruped Walking
Lobe Represents Distinct Period
of Motion
Large Biped
Complete Movement Cycle
(Gait Complete)
e
cy
n
e
qu
e
r
F
Fre
que
nc
Ti
m
y
Intensity
Intensity
Joint Time Frequency Analysis
Time
Process Map
Star
t
Ultrasonic Data
Process Diagram for Ultrasonic
Based Discrimination/Classification
Wait To
Process
Ultrasonic Data
Isolate Bands of
Interest
Waterfall Plot
Analysis
Calculated
Energy
Spectral Analysis
Biped /
Quadruped
Small /
Large
Small /
Large
End
End
Define Region of Interest in
Event
Reduction Process
Isolated Event
Reduced into manageable size for classification
Calculate Event Intensity of
Given Component
Develops envelope that is parsed for evaluation.
Isolating Events of Interest
Identified Series
of Events
Biped Event
Small Biped
Large Biped
Energy
Value
Classification of Ultrasonic
Movement
Large Biped
(Black)
Small Biped
(Green)
Instances
Quadruped
October Test Layout
• Addressing the issue of terrain
variability.
• Variable mammal speed.
• Variable mammal distribution.
• Increase variety of actors.
• Increase variety of movements
Layout 1
Layout 2
Layout 3
Data Collection Results
Region of Error
Isolation Of Particular Elements
• Visualize each of the sections of the
biped as a moment that has a specific
range of associated frequencies based
on motion.
• Isolate the particular frequency range
based on activity.
60o FOV
Joint Sensor Modality Data
Collection (June 2009)
• Day 1 (Indoor: Treadmill)
– Isolated various body parts
(Arms/Torso, Legs)
• Day 2 (Outdoor: Trail)
– Co-located ARDEC seismic
sensor and NCPA ultrasonic
Doppler sensor.
• Goal to conduct data
collection in development
of a preliminary multimodal fused solution and
characterizing further
activity.
Courtesy of NCPA
Isolated Body Section
Entire Biped On Treadmill
Current/Future Work
• Continue analysis of more
challenging data
environments.
• Analysis multiple actors
with current technique for
robustness of feature
extraction including
heterogeneous scenes.
• Data fusion and shared
signal processing with
other modalities.
• Utilize signal processing
methods for other data
analysis.
Sensor Fusion
Increase Confidence in SA
Relevance and Transition Plan
RELEVANCE:
PRODUCT/PAYOFFS:


Provides enhanced situational
awareness and minimizes false
alarms

Ability to early warn
infrastructure security personnel
with a higher degree of
accuracy and greater
confidence.

Cost reduction – given
increased confidence in sensor
outputs, personnel can be used
more effectively to respond to
notifications of activity.


Current increase in the use of
remotely deployed UGS for
border security
Present need to reduce sensor
false alarms from indigenous
wildlife due to limited personnel
response capabilities
This added information can be
integrated into a knowledge
management system.
 Allow higher probability of
detection given a lower false
alarm rate.
 Increase likelihood of enemy
capture.
POTENTIAL TRANSITION:

DOE, DHS - Remote and
Tactical UGS for infrastructure
monitoring and border security.
Summary
• Ultrasonic sensors are a viable modality for
discrimination and classification of various target
taxonomies for a subset of environments.
• The algorithm against the current data is proven
robust.
• JTFA is applied in a novel way to develop a map of
personnel
movement
that
provide
valuable
information.
• The techniques developed can be utilized to provide
valuable information in other sensing modalities.
A novel application of frequency intensity for identification.
Questions / Acknowledgements
• I would like to acknowledge contributions and
support to this effort by my colleagues at NCPA.
• Dr. James Sabatier, Dr. Alexander Ekimov, & Mr.
Richard Mack
• Questions?