A series of experiments that began a year ago and are continuing have demonstrated that it is possible to detect that someone intends to do something bad by picking up on physiological and behavioral clues using sensors and complex software algorithms, the officials running the program for the Department of Homeland Security (DHS) tell TR2.
The ability to detect “indicators” of malintent on individuals using the various technologies is “statistically significant,” that is, much better than a 50-50 chance, says Robert Burns, the program manager for the Future Attribute Screening Technology (FAST) effort run by the Human Factors Division within DHS Science and Technology (S&T). That fact that the protocols for detecting deception prior to the intended action ever taking place is already “cutting edge,” he says.
“We did a series of protocols with 200 people to see if our concepts of, ‘Can we detect signals or cues off of people, either physiological or behavioral, measure them in a repeatable way and analyze them to detect malintent?'” Burns says. “It actually seems to be working.” Those tests were done a year ago. Since then, between 600 and 700 people have participated as subjects in the experiments.
The FAST program was begun about two years ago, with the experimentation phase involving human subjects kicking off last year. While the protocols for the experiments have evolved, essentially people go through a checkpoint that has a security official and is also monitored by one or more sensors, such as a millimeter wave camera, thermal imager, or a BioLidar light detection and ranging system. Excluding the thermal imager, which is operating as a standalone sensor for now, the data is collected from the other sensors and then fused using sophisticated software algorithms to prompt the guard whether a person should be taken aside for additional screening.
As a person passes through the checkpoint, he or she is asked questions by the security official, creating the opportunity for a given individual to have a reaction, that while potentially masked to the human eye, is picked up by a sensor, or sensors, and analyzed using the software to determine if that person may be considering or planning a harmful action.
As a primary screening device, FAST would provide an automatic alert to security personnel that an individual should be pulled aside for additional questioning and maybe even screening.
The ongoing experiments have been taking place at three locations in Massachusetts, Montana and Virginia.
In the first experiments DHS S&T used contact sensors (and sometimes non-contact) to do the measurements on people. That made it somewhat easier in that those types of sensors typically require the subject to be still, which makes data collection easier, Burns says.
Non-Contact Sensors
As the experiments evolved, S&T transitioned to non-contact sensors, which has created a big challenge because “now I have people walking, I have people swaying, I have people turning their heads,” which makes it tougher to collect the data and make sense of it, Burns says. That has led to the use of tracking mechanisms to make sure a sensor is locking on to someone’s face to measure the “right spots” or to deal with different types of clothing or different body parts, he adds.
In the experiments, the sensors are looking at eyes, cardiac, respiration, skin signature and gross body movements. Other things will be looked at over the next year, Burns says.
Each of these features can give off a signal that may indicate malintent, such as a rapid breathing, rising skin temperature, or how someone’s eyes are moving. From the time the sensors begin tracking a person the software that is processing the data begins looking for changes. Every person is his or her own baseline and “what we’re looking at is your personal change,” Burns says.
“Another thing that we discovered is that no single cue is going to work for everybody,” Burns says. “If I’m looking at pupils and someone has really dark sunglasses on and I can’t have them remove their glasses, I need to have other things to look at. So we want to make sure that we have a broad grouping of signals to look at because I may lose one of them at any one time. The other thing is that no one signal will be the Holy Grail. So we have to look at the additive effect. Some of the cues, whether it be cardiac respiration or pupils, some of them are giving a stronger indicator.”
The accumulated sensor inputs are analyzed using data fusion software being developed by one of S&T’s partners for FAST, Draper Laboratory, a non-profit research and development laboratory. This software has been developed over the past year and remains relatively basic but the work itself is “high end,” Burns says. It is fusing data from three signals and providing outputs to advise for secondary or pass-through, he says.
Another bit of progress has been that so far the theory of Malintent is “supportable,” Burns says. The theory is the foundation of the FAST program and basically suggests that there are various cues that can be given off by a person, such as cardiovascular, respiratory, eye movements and others, which may be nearly invisible to the naked eye but can indicate if someone may be considering threatening action.
In addition to assisting screening officers in being able to more accurately identify who should be sent to secondary screening, another purpose of the project is to not take people aside who aren’t planning something bad so that throughput isn’t hindered, and instead may even be accelerated, Burns says.
Support for Theory
Last spring the FAST program was reviewed by a panel of experts throughout the country representing various disciplines such as deception detection, cognitive thought, human behavior, legal and operational DHS components. Burns says the overall they believed the theory of Malintent was being supported but they still made suggestions to improve the protocol. He says the program has worked to better create an environment where the human subjects “believe they are actually trying to smuggle [something] in” past the security guard. “Even though this is a laboratory-based experiment, the laboratory has been reconstructed to represent an operational scenario,” Burns says. “So when people come in, they don’t realize they’re going into a lab, they think they are still going into a technology fair.”
This week in Cambridge, Mass., the FAST program will be holding a demonstration for media. After that, it will be back to business.
Burns says that over the next two years more work needs to be done on how acquiring the signals from the sensors, improving the data fusion algorithm, which he says will still be based on an “interactive process” that involves someone like a security guard asking questions, and to make the experiments “more operationally based.” If the program is successful on those accounts then another phase would involve adjusting the environment to help us bring malintent out to the surface,” he says.
In addition to Draper Labs, some of other contractors assisting on FAST include Battelle, which is doing systems engineering, the Army Night Vision Directorate, which is doing work with the thermal imagers, Kai Sensors, which is providing the millimeter wave technology for cardiac and respiratory monitoring, and Veridical Research and Design Corp., which is doing the eye-related work.