PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2014
151
Extracting Parking Lot Structures
from Aerial Photographs
Liang Cheng, Lihua Tong, Manchun Li, and Yongxue Liu
Abstract
A new method for the automatic extraction of parking lot
structures from an aerial orthophoto is proposed which
has two major stages. Line features of a parking lot are first
extracted by using a line extraction method with principle
orientation constraints which offers a reliable basis for the
extraction of parking lots structures by emphasizing on the
extraction of accurate parking line information. The structure
of the parking lots are then determined from the extracted
parking lines through a maximum intersection orientation
method, parameter-estimating procedure, and a self-adap-
tive growth method. A comparative experiment is shown and
the result indicates that the proposed method can effec-
tively extract parking lot structures with high correctness,
high-quality completeness, and good geometric accuracy.
Introduction
Parking lots, a necessity of the automobile society, have
spread across downtown and suburban areas (Wang and
Hanson, 1998; Onishi
et al.,
2010). With rapid urbanization,
the number of parking lots is dramatically increasing; in
many places, the capacity of the parking lots surpasses that
of the buildings they serve (Davis
et al.
, 2010a and 2010b).
The parking lot has become a crucial factor in land-use and
planning.
Parking lots are required to meet the minimum parking
requirements (MPRs), which are attracting increasing concern
(Franco
et al
., 2010). While the public is concerned with the
quantity of parking space, parking lots are also associated
with environmental problems such as vehicle pollution (Hahn
and Pfeifer, 1994; Scoggins
et al
., 2007), the heat island effect
(Xian and Crane, 2006; Weng
et al
., 2011), and rainfall runoff
(Jakle and Sculle, 2004). Balancing the MPRs with the number
of parking spaces is becoming a focus of research (Davis
et
al
., 2010), and requires statistical information about existing
parking lots and their distribution. This situation calls for the
acquisition of accurate parking lot structures.
The detection of vacant parking spaces is an important
activity for parking guidance and information systems (
PGIS
).
Various techniques have been tried, all of which require
information regarding parking lot structure usage (Yamada
and Mizuno, 2001). Autonomous driving technologies require
accurate knowledge of final parking positions (Gómez-Bravo
et al
., 2001; Li and Chang, 2003; Amarasinghe
et al
., 2007;
Seo and Urmson, 2009).
Although a parking lot structure plays a significant role in the
above domains, parking lots are usually identified manu-
ally in aerial photographs. Many researches related to road
extraction, sign extraction, and building extraction from aerial
imagery or other data have been reported (Youn
et al
., 2008;
Mohammadzadeh and Zoej, 2010; Cheng
et al
., 2011). Meng
et
al
. (2012) gave an object-oriented approach to detect residen-
tial land-use of buildings from aerial photography, lidar data,
and road maps for urban land-use analysis. Yang
et al
. (2012)
introduced an approach to extract road markings from mobile
lidar point cloud. Although parking lots are et al. increas-
ing awareness, there is little literature available about them
(Lee and Lathrop, 2006). Some studies address the quantity
of parking spaces. For instance, Shoup (2005) estimated
that there are three to four spaces per vehicle in some urban
settings. Delucchi (1997) multiplied building parking require-
ments with total building areas and determined that there
were 125 to 200 million off-street parking spaces in America
in 1991. Davis
et al
. (2010b) digitized parking lot orthopho-
tos and estimated parking space quantity using a function of
parking lot size, finding approximately 2.2 parking spaces per
registered vehicle. However, the above studies are limited to
statistical estimation methods and regional areas; this merely
provides an approximate understanding of an area and ne-
glects the distribution and quantity of regional parking lots,
which significantly guide parking lot planning. Moreover,
there is little previous work on the extraction of parking lot
structure.
Wang and Hanson (1988) tried to extract parking lot
structures using stereographs. They used intensity maps to
identify the layout of individual parking spaces and elevation
maps generated from stereographs to provide bumpy regions
around vehicle locations. This method is suitable for densely
populated parking lots; however, there are still some draw-
backs. The generation of good feature correspondence for the
elevation map remains difficult and unstable, and the struc-
ture extraction result is unsatisfactory for a parking space
with few vehicles and poor-quality line markings.
Seo and Urmson (2009) proposed a hierarchical analysis
approach for parking lot structure extraction using a single
aerial image. In their method, canonical parking spaces are
detected through low-level analysis; then, interpolation, ex-
trapolation, and block prediction are used to identify parking
lot structures. Finally, a filtering method is used for extracting
parking spaces. The filtering strategy improves the extraction
precision to some extent; however, problems occur for images
with multiple principal orientations, high-contrast illumina-
tion, significant occlusions, and poor-quality line markings.
Despite these challenges, it is worth studying how to extract
parking lot structures automatically and precisely.
In order to extract and determine the parking lot structure,
a basic understanding about it is indispensable. As shown in
Jiangsu Provincial Key Laboratory of Geographic Information
Science and Technology, Nanjing University, 163 Xianlin
Avenue, Qixia District, Nanjing 210023, China
).
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 2, February 2014, pp. 151–160.
0099-1112/14/8002–151
© 2013 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.2.151
