Tech Briefs Magazine - October 2021 - 32

Imaging
from one pixel to the next end up looking
fuzzy and indistinct. Instead of taking
a low-resolution image and slowly
adding new detail, PULSE scours AI-generated
examples of high-resolution
faces, searching for ones that look as
much as possible like the input image
when shrunk down to the same size.
The team used a tool in machine
learning called a " generative adversarial
network " (GAN) - two neural networks
trained on the same dataset of photos.
One network comes up with AI-created
human faces that mimic the ones it was
trained on, while the other takes this
output and decides if it is convincing
enough to be mistaken for the real
thing. The first network gets better with
experience until the second network
can't tell the difference.
PULSE can create realistic looking
images from noisy, poor-quality input
that other methods can't. From a single
blurred image of a face, it can spit out
any number of uncannily lifelike possibilities,
each of which looks subtly like a
different person. Even given pixelated
photos where the eyes and mouth are
barely recognizable, the algorithm man -
ages to do something with it - something
traditional approaches can't do.
The system can convert a 16 × 16-pixel
image of a face to 1024 × 1024 pixels in
a few seconds, adding more than a million
pixels - akin to HD resolution.
Details such as pores, wrinkles, and wisps
of hair that are imperceptible in the lowres
photos become crisp and clear in the
computer-generated versions.
For more information, contact Erin Kramer
at erin.kramer@duke.edu; 919-660-4257.
3D Motion Tracking System for Autonomous Technology
The system could one day replace LiDAR and cameras in automated manufacturing,
biomedical imaging, and autonomous driving.
University of Michigan, Ann Arbor
real-time 3D motion tracking system
combines transparent light detectors
with advanced neural network
methods to create a system that could
one day replace LiDAR and cameras in
autonomous technologies. The imaging
system exploits the advantages of transparent,
nanoscale, highly sensitive graph -
ene photodetectors.
A
The graphene photodetectors were
tweaked to absorb only about 10% of the
light they're exposed to, making them
nearly transparent. Because graphene is
so sensitive to light, this is sufficient to
generate images that can be reconstructed
through computational imaging. The
photodetectors are stacked behind each
other, resulting in a compact system, and
each layer focuses on a different focal
plane, which enables 3D imaging.
The team also tackled real-time mo -
tion tracking, which is critical to a wide
array of autonomous robotic applications.
To do this, they needed a way to
determine the position and orientation
of an object being tracked. Typical ap -
proaches involve LiDAR systems and
light-field cameras, both of which suffer
from significant limitations. Others use
metamaterials or multiple cameras. Hard -
ware alone was not enough to produce
the desired results.
The team built an optical setup and
enabled a neural network to decipher the
positional information. The neural network
is trained to search for specific
objects in the entire scene and then focus
only on the object of interest; for example,
a pedestrian in traffic or an object
32
Cov
Focal stack images of a point object are simulated by focusing a green laser beam onto a graphenebased
transparent photodetector array. Graphene detectors produce surprisingly high photoresponse
while only absorbing a very small portion of light. (Image: Robert Coelius/Michigan
Engineering, Communications and Marketing)
moving into your lane on a highway. The
technology works particularly well for stable
systems, such as automated manufacturing,
or projecting human body structures
in 3D for the medical community.
The type of algorithms the team developed
are unlike traditional signal processing
algorithms used for long-standing
imaging technologies such as X-ray and
MRI. The team demonstrated success
tracking a beam of light as well as an actual
ladybug with a stack of two 4 × 4 (16www.techbriefs.com
ToC
pixel)
graphene photodetector ar rays.
They also proved that their technique is
scalable. It would take as few as 4,000 pixels
for some practical applications and
400 × 600-pixel arrays for many more.
While the technology could be used
with other materials, additional advantages
to graphene are that it doesn't re -
quire artificial illumination and it's environmentally
friendly.
For more information, contact Nicole
Casal Moore at ncmoore@umich.edu.
Tech Briefs, October 2021

http://www.techbriefs.com http://info.hotims.com/79418-936

Tech Briefs Magazine - October 2021

Table of Contents for the Digital Edition of Tech Briefs Magazine - October 2021