IEEE Technology and Society Magazine - September 2020 - 64

access point ID maps in order to both infer velocity of a
user (as they drive or walk) as well as parking location.
Although all the above cited systems were able to
achieve reasonably high detection accuracies (approximately 80% true positive rate) they often relied on some
form of user tagging and large data collection campaigns
to achieve such levels of reliability. However, early adopters of any of the above-mentioned systems are not likely
to see much benefit given the granularity of the data,
i.e., since each user effectively only provides a single
data point: when and where they parked/un-parked. To
obtain real-time data about parking occupancy across a
neighborhood in this manner requires nearly universal
adoption by users.
To address some of these issues, the ParkUs phone
app, the subject of this paper, was developed with a high
parking/un-parking detection accuracy while requiring
less energy expenditure per detection. Overall, it
achieved a higher true positive detection accuracy (over
98% for parking/un-parking events) than previous
attempts, while utilizing less energy [11]. To aid real-time
parking occupancy data, ParkUs was enhanced [12] to
detect when drivers are searching for a vacant parking
space. This was achieved with the understanding that
every road or row (in a car park or garage) traversed by
the user in their search for a vacant parking space indicates that all preceding spaces would, most likely, have
been occupied. In practice, it is also likely that at least
some of the drivers may cruise past vacant parking spaces for a variety of other reasons (including passing spaces not suitable for larger vehicles or a space dedicated to
specific types of users such as disabled or family parking
spaces). To counter this, road segments were labeled as
occupied only if more than one user has cruised there.
Even this adjustment would (in theory) allow for far richer data gathering with fewer users.
To evaluate ParkUs in a real-world setting, we developed a pilot trial with the University of the West of England (UWE) located in the city of Bristol, U.K. This trial
allowed the functionality of the solution to be tested, as
well as providing an opportunity to elicit feedback from
participants on usability and effectiveness. All activities
with participants took place in accordance with the UWE
code of ethics under appropriate ethics permissions. It is
worth noting that ParkUs does not require the user to
have the smartphone mounted to the vehicle or placed in
a particular fashion. Furthermore, the user is not required
to interact with the smartphone while driving, thereby
ensuring compliance with a vital motor regulation in most
nations, that is, not using a phone while driving.
This article builds on earlier work of this author team,
which described the cruise detection algorithm and
some results from a preliminary evaluation [12]. We take
this work forward along several dimensions: 1) We

64

further develop and then trialed the ParkUs app incorporating the cruise detection algorithm in a real world trial,
2) We show a positive impact on cruising time when
ParkUs shows parking availability information, 3) We evaluate and show the generalization capability of the cruising detection algorithm, 4) We perform a qualitative
study of parking behaviors focusing on usability and performance aspects of the app, and present this alongside
a statistical analysis of the trial data, 5) We lay out a
future roadmap for the app.

ParkUs Overview
ParkUs is a parking-availability predictor which uses an
Android app to both display current availability and
obtain data to populate its model. The availability is displayed using a road network overlaid with colored segments to indicate which roads are likely to have spaces.
The ParkUs architecture is illustrated in Figure 1. The
main research questions that we set out to address are:
■■
■■
■■

■■
■■

How to detect if a user is cruising without having to
manually label parts of the journey?
Does ParkUs provide an effective proxy for cruising
and does it work in practice?
Having detected cruising, how can that information
be fused with data from other users to label parking
availability information on different road segments?
Will ParkUs be used by drivers, and if so will it have a
positive effect on parking behaviors?
How might ParkUs functionality be best deployed to
achieve beneficial outcomes?

As illustrated in Figure 1, by logging various sensors
and location data and using these to train a cruising classification model (see "Cruising Direction" section for
more detail), it is possible to detect cruising behavior,
which can be used as a proxy for non-availability of parking on road segments where the vehicle cruises.

Parkus Operation
Prior to setting off on their trip, users are asked to enter
their desired destination postal code or street name. The
requested destination is displayed (red pin, as shown in
Figure 2) on a map along with the parking availability
within a 100 m (328 ft) radius. The radius is user configurable, to allow adjustments based on users' willingness
to walk from where they have parked their vehicles. Gray
colored road segments indicate no current data available. This is due to not many users using ParkUs in the
requested area. As ParkUs users appear in the area, the
gray segments change to shades of Green/Yellow/Red
depending on the availability of parking.
Once a user has decided on their destination and
which car parks or streets to target by tapping the "Go"

IEEE TECHNOLOGY AND SOCIETY MAGAZINE

∕

SEPTEMBER 2020
IEEE Technology and Society Magazine - September 2020

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - September 2020
