The next step for
Professor Fu’s Innovative
Transportation System
Solutions (iTSS) Lab is to
integrate machine-learning into
signal control.
This effort is being led by Fu’s
colleague, Matthew Muresan, a
PhD student who says that the
machine-learning algorithms
they are working with replicate
human behavior.
During weather events
drivers approach intersetctions
more slowly. “Machine-learning
is trying to capture that sort
of behavior and automatically
adjust the traffic signal based
on that human response,”
says Muresan.
To do this the algorithm
is fed various datasets such
as queue length, operating
speed and travel time. How
the technology makes choices
based on this data is not
entirely clear because of the
so-called ‘black box problem’ in
machine-learning. This refers
to the fact that while you can
see the inputs and outputs of a
machine-learning algorithm it
is otherwise a closed system.
Because of this, says Muresan,
“we don’t fully understand why
a machine learning algorithm
makes a specific decision. It’s
like if two different people
look at the same situation and
make two different choices. But
they’re both good choices.”
So far the machine-learning
tool has been tested in
computer simulations with
“very promising results”, says
Fu. He says the university is
in discussions with the city
authorities in Ontario to see
if they can road test their
technology on the city streets.
Because the model is “openended”
– in the sense that it
is designed around the traffic
parameters rather any specific
technology – it could function
with a range of different
sensing technology including
loop detectors, LIDAR or video
cameras, says Muresan.
That said, he thinks cameras
will prove the best fit. “Video
collects a lot of usable data –
it’s just a question of how to
extract it,” he says.
New learnings
Waterloo University is now looking to develop a completely new type of
signal control – augmented with machine learning
to the time of day. Every few
years the road transport authority
will re-survey traffic patterns
at the intersection to see if the plan
needs adjusting.
“This is the most basic level of
signal timing and you’ll find it in a
lot of small towns and cities,” he says.
The most cutting edge solution, by
contrast, is adaptive signal control
(ASC), a sensor-based system that
uses loop detectors or, more often
these days, cameras deployed at the
approach to the intersection to
measure traffic patterns and adjust
the signalling in real time.
Cities where ASC is already in use
may benefit less from this portion of
the Waterloo research since ASC
already has responsiveness built-in.
But as Fu points out the high cost of
installing ASC – which often requires
fiber optics – can be prohibitive.
In this case he suggests that the
easiest way for transport authorities
to adopt weather responsiveness
would be to enlarge their FTP system
to include plans for different weather
conditions. A snowstorm plan, for
example, could be held in reserve
“and easily switched on when this
weather event occurs,” he says.
The responsiveness of traffic
management systems is dependant
on the ability to extract good quality
data. Until now the metrics needed to
calculate the free-flow speed and the
saturation flow rate have been
measured mainly using loop sensors
embedded in the roadway.
But loop sensors can be
problematic. If something goes wrong
you have to stop traffic to
replace them. So many road
authorities are switching to
video. These newgeneration
cameras have the
We don’t fully understand why
a machine learning algorithm
makes a specific decision… two different
people might look at the same situation
and make two different choices
Matthew Muresan, researcher, Transportation
System Solutions Lab, Waterloo University, Ontario
advantage of providing non-intrusive
detection. As well as providing traffic
monitoring data they can also be
leveraged to detect weather
conditions on the road way, says Fu.
And while such technology may be
costly, a handful of weather stations
with smart sensors could in theory be
used in the future to inform multiple
signals of road conditions, which will
be cheaper than kitting out every
intersection with full ASC.
Seeing the snow
Finnish technology provider Vaisala
already has tools for detecting how
weather impacts the road surface.
The company’s road weather station
is used by transport authorities to
alert them to slippery roads.
Jason Weinberger, of Vaisala’s
North America branch, says the
20%
The modeled reduction
in delays achieved by
changing signal phase
and timing to match
weather conditions
Weather Responsive Intersections |
028 Traffic Technology International November/December 2019
www.TrafficTechnologyToday.com
/www.TrafficTechnologyToday.com