Remote Detection of Covert Tactical Adversarial Intent of Individuals in Asymmetric Operations
by Ann Bornstein, Thyagaraju Damarla, John Lavery, Frank Morelli, and Elmar Schmeisser
ARL-SR-197 April 2010
Executive Summary
The ability to identify covert intent of individuals who may be hostile would significantly
improve asymmetric counter-insurgency and peace-keeping operations. Such individuals are
generally embedded in extensive “clutter” of neutral and friendly human beings and various
physical objects. At present, covert adversarial intent is identified through judgment of Soldiers
and close-range sensing and searching, which often entail significant danger and possibly high
false-positive and false-negative rates. Approaches to checkpoints (before a person gets close
enough to blow up the checkpoint) and remote screening of people on patrol missions are
defense scenarios where remote determination of adversarial intent is needed. Determining
covert adversarial intent will help shift the balance in operations, mission planning, training, and
simulation from more costly and dangerous sweeping operations toward much safer pinpoint
operations based on refined estimates of people from which danger may come. Dual-use civilian
benefits will be in crowd control and antidrug, anticrime, and border security.
The fundamental principles that allow remote (i.e., at 3–50 m) identification of covert adversarial
intent based on externally observable physical information are not known. The goal of this
report is to design a first-order road map for modeling research to bridge the scientific gap
between observations from physical sensor networks at 3–50 m on the one hand and
determination of covert tactical adversarial intent of individuals with deception and in extensive
clutter on the other. Although empirical observations and experiments will play large supporting
roles in this research, the main emphasis is on discovery of theoretically justified quantitative
predictive principles (models) and their implementation in tractable analytical and computational
procedures. To be successful, the research needs to integrate components from kinesiology,
neurophysiology, psychology, cognitive science, sociocultural anthropology, and information
science.
An important and often overlooked concept is measuring the problem. Metrics for cognitive
phenomena and for how well detection systems work are needed. In addition to being practically
useful, the metrics need to be computationally feasible (not combinatorially expensive) and
mathematically justified. In cases where the computational cost of the desired metric(s) is too
large, approximate ersatz metrics need to be developed. Whatever metrics or ersatz metrics are
proposed should be justified not based on traditional use of the metrics in other areas, successful
as that use may be, but rather on the basis of human goals in the remote detection of covert
tactical adversarial intent.
One major issue is the development of data sets. Can “method acting” (or any other school of
acting) provide sufficient verisimilitude on all scales, including emotive/biochemical (sweat,
breath, body habitus, kinesiology) to permit its use as a surrogate for “real data? If so, the
creation of data sets, while still expensive, will be less expensive. One methodology for
developing valid empirical data sets is to design experimental scenarios so that behaviors of
interest are likely to be expressed. If enacted experiments cannot provide data that matches data
of “real” situations, the expense and uncertainty will be larger.
Sensing will require utilizing as many different measureable indicators of intent as possible and,
thus, integration of multiple sensor modalities. Potential indicators of adversarial intent include
posture, posture rigidity, heartbeat waveform, heart rate, breath rate (volume approximation,
patterns, anomalies), wheezing, coughing, gasping, blood pressure trends (waveform shape and
transit time), pulse-wave velocity (beat-by-beat approximation of blood pressure), movement
(fidgeting, remaining still, shaking, shivering, having spasms), body stiffness, muscle tension,
resonant frequency of body movement, voice stress analysis, voice onset timing, gastrointestinal
distress, bowel sounds, reluctance to engage socially (distance from others, response to attempts
to engage verbally), observation tendencies of subject (eye-glancing, head turning, situational
awareness), people whose actions are coordinated or who are actively avoiding each other,
exposure to bomb-making materials/chemicals, hyperthermia from stress (generally expressed in
the face, palms of the hands, and soles of the feet), gait as indicators of stiffness (stress) or
carrying a load or wearing protective clothing, breath biochemistry, and microbiology. Sensing
technologies that may be able to measure relevant data include visible bandwidth imagers,
thermal imagers, hyperspectral imagers, laser Doppler vibrometry, E-field, radar, ladar, gas
chromatography, interrogators of the genetic signatures of prokaryotic microorganisms, chemical
sensors (laser-induced fluorescence, laser-induced breakdown spectroscopy, Raman
spectroscopy), photoacoustic sensors, retroreflection sensors, seismic sensors, and magnetic
sensors.
Fusion of the information from multiple sensors will be required to achieve accuracy. The Joint
Directors of Laboratories (JDL) Data Fusion Model is the most widely used method for
categorizing data-fusion-related functions. The JDL model is composed of levels of abstraction,
with level 0 being the lowest or the minimally processed information level and with level of
abstraction increasing in levels 1–4. Although there are many criticisms of the JDL model and
many competing models, the JDL model has, in general, withstood the test of time, and most of
the fusion community has accepted the JDL fusion levels. The JDL fusion framework is a
suitable (but not the only) framework in which fusion of the output of many different sensors
could take place.
Successful identification procedures may need to include actively (but unobtrusively) perturbing
the situation in which the sensing takes place in order to elicit specific responses (an abstract
analogue of putting speed bumps in approaches to checkpoints so that the oscillation of cars can
be observed and one can infer whether the car is carrying a heavy load).
Comprehensive U.S. Department of Defense, Department of Homeland Security, Intelligence
Advanced Research Projects Activity, and Federal research and development (R&D) programs
are required to promote rapid progress. Specific recommendations are as follows:
• The Federal Government should fund R&D with the objective of producing a theoretically
founded prototype system for remote detection of covert tactical adversarial intent of
individuals in asymmetric operations within 5 years and a working operational system
within 10 years.
• The Federal Government should continue to provide broad support for academic and
industrial efforts both in remote detection of adversarial intent and in areas (such as linkage
of these systems with databases, media, and human input) that are useful for larger systems
of systems.
• The strong interdisciplinary nature of remote detection of adversarial intent should be
reflected in all efforts supported by the Federal Government.