In
spy novels and superhero films, the ability to see through walls has always
been a handy — not to mention, impressive — trick. And now, this tech could be
available to people in real life, with smartphone cameras that can help detect
moving objects even if they are hidden around corners, according to a new
study.
This futuristic-sounding tech could
one day help vehicles see around blind corners, the researchers said.
"We may eventually be able to use this idea to alert
drivers to pedestrians or cars that are about to dart out from behind buildings
into a driver's path. Perhaps a few seconds of notice could save lives,"
said study lead author Katie Bouman, an imaging scientist at the Massachusetts
Institute of Technology's Computer Science and Artificial Intelligence
Laboratory.[Mind-Controlled Cats?! 6
Incredible Spy Technologies That Are Real]
"Search and rescue, or helping to
understand what is going on behind a wall in a hostage situation, are also
potential applications," Bouman added.
Researchers have taken many different approaches in trying to
make the "superpower" of seeing around corners a reality. For
example, in 2015, researchers showed they could use lasers to see objects
around cornersby firing light pulses at surfaces near the items.
Those surfaces could act like mirrors, scattering the laser pulses onto any
hidden objects. By analyzing the light that was reflected off the objects and
other surfaces back onto the scanners, researchers could reconstruct the shapes
of the hidden items.
Although most strategies for seeing around corners "are
really great ideas," they also "usually require complex modeling [or]
specialized hardware, or are computationally expensive," Bouman told Live
Science. The 2015 study's technique, for example, required both extremely
fast lasers and
extraordinarily sensitive cameras.
But Bouman and her colleagues' method for
seeing around corners simply uses a smartphone camera.
"We use light naturally in the scene and
do not have to introduce our own light to probe the hidden scene," Bouman
said. "This allows us to use common consumer cameras and not specialized
equipment to see around corners."
The new system, known as CornerCameras, analyzes light that is
reflected off objects hidden around corners and
that falls on the ground within the line of sight of the camera. This light is
called the "penumbra."
The system works by
analyzing light at the edge of walls, which is impacted by the reflections of
objects around the corner from the camera.
Credit: MIT CSAIL
The system analyzes this penumbra over several
seconds, stitching together dozens of distinct images, according to the study.
This data helps the system measure the speed and trajectory of objects around
corners in real time. (It does not see any identifying details about those
objects — just the fact that they are moving.)
"I think the biggest surprise was that
the system worked well in situations that I would not have expected,"
Bouman said. "For instance, once, during filming, it started raining. This
caused big raindrops to start appearing on the ground, changing the color of
the concrete floor."
Because CornerCameras is trying to analyze
light signals that are just 0.1 percent of the total brightness of the ground,
"I thought these raindrops would wipe out any signal we had," Bouman
said. However, CornerCameras analyzes the data of a scene across dozens of
images, so "the effect of the raindrops was essentially averaged
out."
One current limitation of CornerCameras is that it requires a
stationary camera that's held very steady. "In many situations, such as in
a collision-avoidance system on a
car, you do not have the luxury of a stationary camera," Bouman
said. The researchers are now focused on getting the system to work first on a
moving wheelchair and eventually on a moving car, she said.
Future research will also aim to make
CornerCameras work in a variety of lighting situations, or in changing lighting
conditions, such as when clouds overhead constantly move in front of the sun.
"Getting the system to work in these scenarios would open up the
possibility of it being able to be used by a person with a handheld
smartphone," Bouman said.
Bouman and her colleagues will detail their
findings on Oct. 25 at the International Conference on Computer Vision in
Venice, Italy.
Called from: www.livescience.com
Post a Comment