matsuda2021 - Flipbook - Page 1
2021 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops) | 978-1-6654-0424-2/21/$31.00 ©2021 IEEE | DOI: 10.1109/PERCOMWORKSHOPS51409.2021.9431076
3HU9HKLFOHUG,QWHUQDWLRQDO:RUNVKRSRQ3HUYDVLYH&RPSXWLQJIRU9HKLFXODU6VWHPV
A System for Real-time On-street Parking Detection
and Visualization on an Edge Device
Akihiro Matsuda1 , Tomokazu Matsui1 , Yuki Matsuda1,2 , Hirohiko Suwa1,2 , Keiichi Yasumoto1,2 ,
1
Nara Institute of Science and Technology, Ikoma, Nara 630-0192, Japan
E-mail: (matsuda.akihiro.lr2, matsui.tomokazu.mo4, yukimat, h-suwa, yasumoto)@is.naist.jp
2
RIKEN Center for Advanced Intelligence Project, Tokyo 103-0027, Japan
Abstract—In recent years, street parking in prohibited areas
has become a serious social problem, particularly in metropolitan
and tourist areas where there are many on-street parked cars. In
addition, because on-street parking can cause traffic congestion
and accidents, real-time detection is considered necessary. In
previous studies, fixed-point cameras have been mainly used for
traffic control; however, a major disadvantage of these systems is
their limited detection area. In this study, we developed a system
that can detect and visualize on-street parking in real-time using
video data captured by dashboard cameras, which have become
widely used in recent years. We created a learned model to detect
on-street parking and to recognize cars using an edge device. By
displaying the location information of on-street parked cars on a
map, their location can be visualized. This system could be used
to obtain statistical data for crackdowns on on-street parking
and to identify areas where on-street parking occurs.
Index Terms—street parking, edge device, real-time sensing,
dashboard camera, object detection
I. I NTRODUCTION
In recent years, street parking in prohibited areas has
become a serious social problem, especially in metropolitan
and tourist areas. According to a survey on street parking
conducted by the Traffic Bureau of the National Police Agency
(Japan) in 2019, the number of on-street parked vehicles in
prohibited areas in the special wards of Tokyo was approximately 52,700 [1]. Illegal street parking is not only hazardous,
but it can also cause a variety of other traffic problems, such
as traffic jams and rear-end collisions.
To deal with on-street parking and its related problems, the
extension of restricted parking zones and the implementation
of time-limited zones are being considered. However, these
measures are not a sustainable solution and they may complicate existing problems. In addition, they require large-scale
cooperation from the government and administrative agencies,
which is very costly and time-consuming [2]. A previous
study proposed a system that uses multiple fixed-point cameras
installed in the city to detect on-street parking. However, fixedpoint cameras can only cover a limited area.
In this study, we sought to develop a system capable of
detecting on-street parked cars more efficiently and widely
than fixed-point cameras in real-time using dashboard camera
video systems, which are increasingly being used in general
vehicles [3]. A dashboard camera records every situation in a
city and stores an overwhelming amount of information. Also,
it is more efficient than traditional traffic monitoring methods
because it is not limited by area. On the other hand, uploading
all the collected videos to the cloud for analysis is problematic
in terms of the loads placed on communications networks,
communication cost, and real-time performance. To reduce
communication costs, we sought to determine whether or not
the vehicle has been parked illegally in real-time using an edge
device that has been installed in the vehicle and only uploads
parking-related information to the cloud. In this study, we also
propose a real-time visualization system for determining the
location of on-street parked cars using a video-learning model
and a dashboard camera. This information is captured and
processed on an edge device mounted on a vehicle.
The remainder of this manuscript is organized as follows.
In Section 2, research related to the detection of on-street
parking is reviewed and the problems associated with existing
methods are summarized. Section 3 defines on-street parking
and describes the issue of on-street parking in Japan. Section
4 provides an overview of a system for real-time detection
and visualization of on-street parking. In Section 5, a method
for on-street parking detection on an edge device is described.
In Section 6, we describe the accuracy of instances of onstreet parking by the model, and the speed required for
recognition and processing on the edge device. In Section 7,
a method for visualizing the location information of on-street
parked vehicles is described. Finally, Section 8 presents the
conclusions of this thesis and discusses future directions of
this research.
II. R ELATED R ESEARCH
In this section, we review studies and existing technologies
related to on-street parking. Although numerous studies have
been conducted on on-street parking to date, few have examined how on-street parking can be detected [4] [5] [6].
Wen et al. proposed a system for vehicle recognition,
location detection, and classification using Mobile Laser Scanning (MLS) point clouds [7]. Briefly, their method involved
segmenting the recognition targets into point clouds and then
fitting a vehicle recognition model to each target. The acquired
information, such as position and orientation, was then compared with data acquired at different times to estimate the
consistency of the features with their durations. The system
was highly accurate in recognizing and classifying vehicles
and detecting changes, but recognition was limited by the
reach of the MLS and requires scanning through cars parked
k,(((
227 on June 30,2021 at 18:34:04 UTC from IEEE Xplore. Restrictions apply.
Authorized
licensed use limited to: BOURNEMOUTH UNIVERSITY. Downloaded