Registration-based Mapping of Aboveground
Disparities (RMAD) for Building Detection in
Off-nadir VHR Stereo Satellite Imagery
Alaeldin Suliman, Yun Zhang, and Raid Al-Tahir
Abstract
Reliable building delineation in very high resolution (
VHR
)
satellite imagery can be achieved by precise disparity informa-
tion extracted from stereo pairs. However, off-nadir
VHR
imag-
es over urban areas contain many occlusions due to building
leaning that creates gaps in the extracted disparity maps. The
typical approach to fill these gaps is by interpolation. Howev-
er, it inevitably degrades the quality of the disparity map and
reduces the accuracy of building detection. Thus, this research
proposes a registration-based technique for mapping the dis-
parity of off-terrain objects to avoid the need for disparity in-
terpolation and normalization. The generated disparity by the
proposed technique is then used to support building detection
in off-nadir VHR satellite images. Experiments in a high-rise
building area confirmed that 75 percent of the detected build-
ing roofs overlap precisely the reference data, with almost 100
percent correct detection. These accuracies are substantially
higher than those achieved by other published research.
Introduction
Overview and Motivation
Building detection is important for many applications. City
planning and management (Nielsen and Ahlqvist, 2014), city
modeling (Singh
et al.
, 2014), population estimation (Xie
et al.
,
2015; Zhu
et al.
, 2015), and urban growth monitoring (Sugg
et
al.
, 2014) are just a few examples where continuously updated
building information is required. The very high resolution
(
VHR
) satellite images with their large coverage and detailed
spatial information provide substantial information necessary
for mapping complex urban environments. Therefore, building
detection and mapping using
VHR
satellite images have become
an active research area in the remote sensing community.
VHR
satellite sensors usually acquire off-nadir images
with across-track and/or along-track angles using various
acquisition modes (e.g., target, corridor, and stereo acquisi-
tion modes). The readily available off-nadir satellite images
captured in a stereo mode allow generating elevation data to
support building detection at a lower cost than other sources
of elevation data such as lidar. However, off-nadir images al-
ways suffer from the compounded effect of tilt and relief dis-
placement for all elevated objects. This effect reveals building
façades, hides areas, and results in registration difficulty
between elevation surface models and the images.
Building detection methods based on optical images make
use of various types of information provided by
VHR
images.
These information types can be categorized broadly into two
classes: mono-based and stereo-based image information. The
mono-based image information includes spectral and spatial
(morphological, textural, and contextual) information that is ex-
tracted directly from
VHR
images. In contrast, the stereo-based
information comprises image-derived products such as digital
surface models (
DSM
s) that are generated after photogrammetric
processing of multiple overlapped images (Salehi
et al.
, 2012).
Following this taxonomy, building detection methods can be
divided into two categories: mono-based and stereo-based.
The most successful building detection methods from
VHR
images using mono-based image information are com-
prehensively evaluated by Khosravi
et al.
(2014). However,
mono-based image information poses some limitations for
building detection. Chen
et al.
(2014) reported a few chal-
lenges for building detection methods and stated that the
intra-similarity of most impervious surfaces in
VHR
images is
the most intractable difficulty for such methods. For instance,
it is very difficult to automatically distinguish building roofs
from parking lots with the same spectral and spatial (i.e.,
textural and morphological) properties. To overcome this de-
ficiency, Ghaffarian and Ghaffarian (2014) introduced a shad-
ow-based building detection method in
VHR
images. Although
the results are promising, they concluded that such methods
are still incapable of separating spectrally-spatially similar
objects. Therefore, when employing off-nadir
VHR
images over
dense urban areas, the challenges of mono-based building de-
tection methods will increase due to the existence of façades
and the possibility of buildings overshadowing others.
Considering that buildings are elevated impervious surfac-
es, stereo-based information provides the third dimension as
the key feature for accurate detection and reliable differentia-
tion from similar traffic areas. Heinl
et al.
(2009) found that the
accuracy of urban mapping is increased when the elevation
data are incorporated regardless of the technique implemented.
Krauß
et al.
(2015) and Tian
et al.
(2014) are just two successful
examples of stereo-based building detection approaches. The
third dimension in these examples was the surface elevation as
derived photogrammetrically from stereo images.
Alaeldin Suliman is with the Department of Geodesy and Geo-
matics Engineering, University of New Brunswick, 15 Dineen
Drive, Fredericton, NB, Canada, E3B 5A3 (
.
Yun Zhang is with the Department of Geodesy and Geomatics
Engineering, University of New Brunswick, 15 Dineen Drive,
Fredericton, NB, Canada, E3B 5A3 (
. Addi-
tionally, he is with the Institute of Remote Sensing and GIS at
Peking University, and MIT Media Lab at the Massachusetts
Institute of Technology.
Raid Al-Tahir is with the Department of Geomatics Engineer-
ing and Land Management, University of the West Indies, St.
Augustine, Trinidad and Tobago, and also with the Department
of Geodesy and Geomatics Engineering, University of New
Brunswick, Fredericton, NB, Canada, E3B 5A3.
Photogrammetric Engineering & Remote Sensing
Vol. 82, No. 7, July 2016, pp. 535–546.
0099-1112/16/535–546
© 2016 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.82.7.535
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2016
535
