Surveying the road ahead
According to an August 2010 article in WIRED magazine, coalition forces in Iraq suffered more than 13,000 improvised explosive device (IED) attacks between 2006 and 2009. IEDs have now become the principal killer of U.S. troops in Afghanistan — responsible for up to 75 percent of casualties in some areas — and more than 3,000 additional IED incidences occurred in other parts of the world in 2009.
Detection and demobilization of landmines and explosive devices is a priority for the U.S. Department of Defense, though it is a daunting task since insurgents either adapt to military counter strategies or evade them, almost before they are in place.
Seeing potential in his work to eventually equip a successful mobile detection unit, the U.S. Army has repeatedly funded the research of James Keller, a University of Missouri curators’ professor who holds the R. L. Tatum professorship in electrical and computer engineering. The Leonard Wood Institute also has recently provided funds for his detection research.
Working under the umbrella of the College of Engineering’s Center for Geospatial Intelligence, Keller is collaborating with Dominic Ho, a professor of electrical and computer engineering, to tackle the challenge of landmine and explosive hazard detection in partnership with the Army and researchers at Duke University. The pair’s approach involves data garnered from a variety of sources, including video, forward-looking ground-penetrating radar (FLGPR) and infrared and color cameras.
“We have a goal of finding a sensor suite to locate these buried objects,” said Keller. “It’s something that’s larger than the sum of the parts.”
The Army’s Night Vision and Electronic Sensors Directorate (NVESD) furnishes controlled data for the research.
Keller said that the data collected from the FLGPR radar is being used as a cueing device although the technology involved with forward-looking radar is not as mature and results aren’t as good as those recorded by downward-looking systems.
Ho has successfully used downward looking radar for a past project implemented by the military in a handheld device for field use. But for obvious reasons, there can be limitations to using downward-looking radar in the case of landmine detection with a moving vehicle.
With FLGPR, the returning signal is weaker because it radiates forward over a large geographic area. “The target response may be buried in noise and may be difficult to see without any processing,” said Ho, who has extensive research background in signal processing, an expertise he is contributing to the project. Both men say that forward-looking radar has improved since they first began working with it ten years ago.
“We’ve tried a lot of things,” said Keller. “We are also looking at data with various bands of the infrared spectrum. Different objects have different temperatures and signatures. Disturbed earth behaves differently.
“We believe we can find 100 percent of the buried objects and use the other sensors to get rid of false positives,” he said.
Keller emphasizes that data collected from the radar are signals, not images. An image must be “constructed” from it before it can be registered with the camera imagery. They are experimenting with various pattern recognition approaches in each sensor domain and on the fusion of the results using fuzzy set theory.
“We are looking at ways to combine images with fuzzy sets,” said Keller, who has received national recognition for his work in these precise mathematical techniques that describe vague or imprecise events. “It gives you a way to look at data — not whether or not something happens, but how much it happens. It’s not probabilistic,” Keller said. “You can’t always say yes or no. You collect evidence and build your confidence.”
To describe the fuzzy logic process in simple terms, Keller uses the analogy of someone going to the doctor because they are ill, and the doctor diagnosing the patient as having the flu, based on his confidence in the evidence.
The MU research team plays a lead role in the IED detection research integrating all of their collaborators’ algorithms. Though the project is still in testing mode, within the next year they will report their answers to the military and they will be scored.
“The Army has looked at many technologies to solve this puzzle,” said Keller. “What we are doing is very applied research. They want to build a road-clearing vehicle.”
Keller and Ho meet weekly with the team assisting in the investigation, which includes Mihail Popescu, an assistant professor of health management and informatics who also is an adjunct in computer science. Popescu’s expertise is in the area of pattern recognition.
“Most of my work is in preventive medicine, which in a way also applies to this work,” Popescu said, adding that he enjoys the extremely harmonious culture of collaboration that exists in the Electrical and Computer Engineering Department.
In addition, many students have worked as part of the team.
“Tsaipei Wang was one of the first working on forward-looking radars,” said Keller, explaining that Wang, who already had earned a doctorate in physics, also earned a doctorate in computer engineering and computer science, with Keller as his mentor “He pioneered our look at sub-band frequency reconstruction.”
Another of Keller’s doctoral students, Tim Havens, worked on beam forming, detection algorithms and fusion of FLGPR to forward-looking infrared and color. After serving as the lead on a successful proposal to the Leonard Wood Institute earlier this year, Havens is now doing an NSF Computing Innovation Postdoctoral Fellowship at Michigan State.
Keller said that master’s student Kevin Stone has done most of the registration of image sequences to ground truth, built many of the algorithms and performed fusion for the project. This year, a former student who is now a research assistant professor, Derek Anderson, is taking the responsibility of guiding undergrads.
“I look to make my impact in the area of feature, algorithm and platform-level fusion using fuzzy sets for forward-looking ground penetrating radar, infrared and visible spectrum imagery, said Anderson.
Master’s students Chris Spain, Justin Farrell and David Lewis are part of the current team, and Keller said that three undergraduates, Brad Calhoun, Don Scharzman and Mark Busch, have made sensor contributions to the research.
“As researchers, we are interested in solving some challenging problems and satisfying our own curiosity,” said Ho. “This research has potential to help the U.S. government — especially our soldiers — to reduce the casualties involved. It is a good piece of work.”
Data detectives’ other projects
Dominic Ho’s signal processing expertise and James Keller’s proficiency in fuzzy logic and image processing provide frameworks for many different applications that are related to their work detecting explosive hazards.
Ho just received a large award from the U.S. Army to continue his successful work with IED detection using ground penetrating radar.
In the past, Ho has been funded by the Air Force Office of Scientific Research to monitor birds in a project aimed at deterring them from entering air space around airports that could potentially create bird strikes. These same signal processing techniques — recording and separating various signals — also have been applied to another federal agency-funded project to process and extract the sound of a single person from a mix of speakers. Additionally, Ho is looking into separating voices and weapon noises, potentially with microphones placed in soldiers’ helmets. “This is more challenging,” he said.
Research that expands on these techniques, but that is applied in an entirely different situation is Ho’s work with Electrical and Computer Engineering Professor Marge Skubic’s eldercare projects aimed at unobtrusive monitoring of seniors that guards their privacy but cues caregivers to potential problems, a project to which Mihail Popescu also contributes his pattern recognition skills.
“They are walking, watching TV, but if they fall, it will be a different sound. The aim is to separate the sound signals and develop an acoustic signal-detection method for a person falling,” he said.
Popescu’s work looks at pattern sensor data aimed at detecting anomalies in seniors’ daily activities that might signal something is amiss. He is also collaborating with Lori Popejoy, an assistant professor in the Sinclair School of Nursing, to identify and address challenges in discharging people from the hospital.
Keller is working on an extensive project with the National Geospatial Intelligence Agency to solve the problem of integrating vast amounts of data that have been acquired by a variety of sensor systems and human intelligence: matching spatial information with temporal linguistic descriptions.
In addition, Keller lends his expertise to Skubic’s eldercare research and to many additional computational intelligence problems and projects.