By David Goddard
On April 15, 2013, a pair of homemade bombs sent shrapnel flying into the crowd gathered on Boylston Street near the finish line of the Boston Marathon. In a matter of seconds, three people were killed and 264 were wounded.
Numerous photographs and videos taken by bystanders and security cameras were analyzed following the attack. Within hours, authorities had isolated images of the perpetrators and released them to the media.
But what if those same cameras could have helped prevent the tragedy? That is the motivation behind Hairong Qi’s research.
“Being able to build smart sensors that can not only acquire data but also automatically detect things and predict their intent is one of our goals,” said Qi, professor of electrical engineering and computer science.
Qi is an expert in the field of image processing—the automated analysis of a visual representation of a scene. She is also a leader in the development of smart camera networks.
While self-aware sensors aren’t here yet, Qi believes they are closer than ever to reality. “A smart camera network holds a lot of promise,” she said.
Such a network would be capable of gathering information from a series of sensors, analyzing the data for things such as potential threats, and determining the location, direction, and speed of the target in a crowd.
Collaborative Networks
Since 2000, UT has been awarded more than $2 million in funding from the National Science Foundation and the Defense Advanced Research Projects Agency to advance the technology.
Initially the project revolved around sensors and what could be accomplished by using hundreds of them in one setting to merely gather intelligence. Different teams were assembled to study various challenges, such as detection or networking.
As the research progressed, the focus shifted to implementing sensors into a broader network, one where they not only communicated with each other but could also act according to the information they collected.
“The sensors need to take care of the data end of things while at the same time be able to network,” Qi said. “Tackling that issue plays into our strengths.”
That’s why Qi and her team of experts are developing a light-weight algorithm that enables collaboration between the sensors and, just as critically, between the sensors and the people controlling them.
Although Qi’s group doesn’t build the sensing hardware, their software development allows it to function in a much more intelligent way, making a smart sensor network possible.
“In much the same way the Internet connects people to people, a sensor net connects people to the physical world,” Qi said. “Our algorithm is the core of that process, making the large amount of data collected by the sensors actionable without adding extra burden to the resource-limited network.”
Even after fifteen years of work, many challenges remain. For one thing, current sensors—about the size of the palm of your hand—don’t yet have adequate resolution or processing power. After all, a picture isn’t worth a thousand words if you can’t discern the image.
“That’s the main limitation we are working to overcome,” Qi said. “You can have a sensor able to run a high-end analysis, but the power it requires isn’t yet attainable at a realistic level. The solution is to have less powerful cameras running in a network so you can combine their abilities.”
Beyond the Pixel
Qi has also been investigating the use of hyperspectral imaging that could allow sensors to detect a person hiding behind an object or land mines buried underneath the ground.
Instead of being limited to just the red, green, and blue spectral bands used in commercial digital cameras, hyperspectral imaging can read hundreds of bands by tapping into the infrared spectrum. Much more information can be gathered from infrared light even though human eyes cannot see it.
Using subpixel analysis—literally analyzing pixels for objects that appear behind them—sensors could detect an item beneath a tree and determine whether it’s friend or foe, apple cart or missile launcher.
This technology also has potential in the health care industry. “Switching to hyperspectral imaging could improve the accuracy of breast cancer detection and help get away from the invasive, non-user-friendly mammography we use today,” Qi said.
In 2012, her hyperspectral imaging research earned Qi the IEEE Geoscience and Remote Sensing Society award for highest impact paper. She was recently named the Gonzalez Family Endowed Professor at UT.
Whether used for something as mundane as counting the number of people attending an event or as serious as alerting soldiers to hidden threats, the potential applications for Qi’s research are immeasurable. Hopefully eliminating catastrophes like the Boston Marathon bombing will be part of that legacy as well.