How students help to improve campus parking
finding a parking spot on college campus
can be frustrating
not to mention it may lead to traffic
congestion
class delays and accidents
to alleviate this challenge a group of
computer engineering students
worked on equipping parking lots with
smart parking systems
at cal state san marino in this talk
i’ll present
our system that integrates iot sensors
and civilian cameras to count vehicle
at a parking structure
while various parking systems have been
deployed
the central problem is how to make the
trade-off
between cost accuracy and reliability
smart powering system can be divided
into three categories
the first one is sensor-based method
which use stationary sensors to detect a
vehicle’s presence
by sensing ultrasonic optical
magnetism or pressure changes
you might have noticed the black circles
and squares
on the road near intersections there are
so called
inductive loops when the vehicle passes
over
the car’s mental body reduce the
inductance of the wire loop
this change tells the sensor about the
presence of the vehicle
at intersections this information will
then be used
to adjust the signal timing the same
type of sensor
has been used to monitor parking lots
the second method for accounting vehicle
is by using cameras
this picture shows the license plate
radar from a nearby tow road
in camera based method computer
compromision techniques are used
to extract vehicles movement from the
camera footage
for plate raider it required a array of
specialized
high-speed cameras if it’s just for
counting vehicles
a consumer-grade camera can work
reasonably well
the last common method for parking
monitor is what it called
crowdsourcing in popular navigation apps
you’ve seen the popular time graph or
similar functions it relies on the users
voluntarily sharing their parting
location information
from their smartphone the the more
people that are using the app
the more accurate the estimation will be
each of these methods have some
limitations when it comes to tracking
the number of vehicles
coming in and going out of a parking lot
for sensor-based method it is hard to
distinguish
between two vehicles traveling closely
together
versus one slow moving vehicle
for camera-based method it’s challenging
to maintain a clear
line of sight at all times especially in
outdoor environment additionally
driving behavior such as the aggressive
and illegal ones in parking lots can
easily confuse
computer vision algorithm as to
crowdsourcing method
the performance heavily relies on the
adoption rate
of the smartphone app so crowdsourcing
are mainly used
for a rough estimation of the parking
lot occupancy
given these observations we built a
smart parking system
by integrating both sensor based
and camera based methods we installed
new iot sensors
at the entrances and utilized existing
civilian cameras
inside parking structure the pilot
the pilot side of the system is a
parking structure
with 745 parking spots
here’s a closer look at the sensor we
used it’s a
battery powered magneto resistive
sensing device
they are installed in holes cored in the
pavement
and are covered with epoxy this picture
was taken when we put them in
different from the traditional inductive
loops no wiring is required for this
sensor
vehicle detection data are wirelessly
transmitted using ieee
802.15.4 protocol
this picture on the left shows how we
install the access point
it is similar to the household wifi
antenna
the detection data is further related to
the id service room
inside the parking structure from there
the detection data is sent to our cloud
server
our system consists of three parts
wireless vehicle detector
civilian cameras and a web server
this floor plan shows the setup at one
of the entrances
there’s one camera pointing to the
entrance
the camera was installed by the campus
police to monitor security
before this project two sensors
were installed under each lane leading
towards the parking structure
now we’re looking from the civilian
camera’s perspective
our vehicle tracking algorithm process
the live
video footage the boundary box
shows the location of the vehicle
extracted from the background
by tracking the center of the boundary
box we can find if the vehicle is
entering or leaving the parking
structure
for the iot sensors they are triggered
when the vehicle are on top of them
integrating two sensing method it allows
us to achieve
good accuracy in our test
the vehicle counting accuracy is about
96 percent
our initial plan was to develop the
smart parking system
in two steps in the first year
we’ll conduct a pilot study at one
parking lot
then in the second year we’ll expand the
system to cover most of the campus
unfortunately the second step of the
project is currently on hold
due to the code 19 interruptions
so in the last few months we looked into
data collected by the pilot system
we’re interested in how to provide
parking prediction
and studied traffic patterns
the intuition behind parking prediction
is simple
we can estimate parking occupancy in the
next 15
or 30 minutes based on historical data
this figure were taken from our website
it shows the traffic pattern
is similar between two days in different
weeks
we can further improve our estimation by
looking at
daily and weekly traffic patterns
when we turn to data analysis we have
some interesting findings
our parking structure has two entrances
this figure shows
the hourly count of detected vehicles by
each line
from the two entrances each entrance has
two lines
one for enter and one for exit
it’s interesting to note the west
entrance
receives more traffic than the other
entrance
in this figure the count in blue is
significantly larger
than the counting gray this is
likely because the vast entrance is
closer
to other parking lots and major lecture
buildings
which makes students consider vast
entrance as the default entrance
to the park structure our study finds
it will be beneficial to add a parking
direction
at the east entrance this would help to
improve
traffic flow by balancing the use of
both entrances
and may reduce congestion during rush
hours
we also looked into weekly traffic
patterns
it may provide insight on long-term
traffic trends
well the last spring and the winter
quarters
are unconventional we can see
how wildfire induced power outage
and the kovale 19 pandemic affected the
university
generally speaking we can use the
traffic pattern
to study students movement behavior
this in turn can help to improve class
schedule
campus traffic management and plan for
emergency response
for future work there are a total of
7 300 parking spaces on campus
well the pilot system only covers 10
percent of them
we hope to expand the system to include
more parking lots
our ultimate goal is to provide parking
guidance
that is suggest parking location based
on parking lots occupancy information
so that we can direct the campus traffic
so far our students deployed the pilot
system
and designed a parking prediction
algorithm
all of this are student projects with
many support
from the university in the computer
science department
our objective has been providing
students
with hands-on experience solving
real world problems and to better
prepare them
with a data-driven mindset
and with that that’s how we define the
future
you