International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
PLANE MATCHING WITH OBJECT-SPACE SEARCHING
USING INDEPENDENTLY RECTIFIED IMAGES
H. Takeda * *, Y. Akamatsu *, Y. Minami*
“ KOKUSAI KOGYO CO.,LTD, R&D Division, 2-24-1 Harumi-cho, Fuchu-shi, Tokyo 183-0057 JAPAN
htake@kkc.co.jp , {yukio_akamatsu , yoshihiko_minami} @kk-grp.jp
Commission III, WG III/4
KEY WORDS: 3D city model, multi-image matching, rectification, object-space searching, building wall, multi-view images
ABSTRACT:
In recent years, the social situation in cities has changed significantly such as redevelopment due to the massive earthquake and
large-scale urban development. For example, numerical simulations can be used to study this phenomenon. Such simulations require
the construction of high-definition three-dimensional city models that accurately reflect the real world. Progress in sensor
technology allows us to easily obtain multi-view images. However, the existing multi-image matching techniques are inadequate. In
this paper, we propose a new technique for multi-image matching. Since the existing method of feature searching is complicated, we
have developed a rectification method that can be processed independently for each image does not depend on the stereo-pair. The
object-space searching method that produces mismatches due to the occlusion or distortion of wall textures on images is the focus of
our study. Our proposed technique can also match the building wall surface. The proposed technique has several advantages, and its
usefulness is clarified through an experiment using actual images.
1. INTRODUCTION
In recent years, the social situation in cities has changed
significantly such as redevelopment due to the massive
earthquake and large-scale urban development. For example,
cities are plagued by complex environmental problems such as
urban heat island and are required to formulate disaster
prevention methods such as anti-earthquake procedures; it is
becoming extremely difficult to investigate the causes of these
problems. One of the solutions to describe environmental
problems is the use of numerical simulations. However, to
create detailed numerical simulations of urban space, it is
necessary to construct a wide-area, high-definition, three-
dimensional (3D) city model. Although the demand for spatial
information, including 3D maps and city models, has increased,
it still largely depends on manual plotting. On the other hand,
sensing devices using aerial surveying cameras have evolved
significantly over the past ten years. Because of digital
photography, developing a film has become unnecessary.
Furthermore, the bundle-adjustment process has become
unnecessary or has been simplified by the evolution of the
gyroscope sensor. As technology advances, photographic
processes are reduced and highly overlapped multi-view images
can be obtained easily at a low cost and with less labor. It
appears there are many advantages to using multi-view images
for multi-image matching. However, many aspects are not
covered by the current multi-matching methods. In this research,
we aim to develop a multi-image matching method that can
efficiently build an exact 3D city model with consistent quality.
2. SUBJECT OF THIS RESEARCH
In the field of spatial-data modeling, a quality 3D city model
requires not only the reality of expression, such as computer
graphics, but also efficiency in data processing and accuracy in
position measurement. The following issues are of concern in
existing multi-image matching techniques.
* Corresponding author.
[Issue #1] The multi-ray method is complicated in its
calculation process, which requires feature searching with
alignment along the viewing direction in the 3D coordinate
system based on camera position. We propose an improvement
in efficiency.
[Issue #2] In case of the edge matching process, the extraction
of line segments on a roof footprint is difficult because an aerial
image of a real-world scene has complicated edge features in
addition to the true building footprint. We propose an
improvement in stable processing.
[Issue #3] In the case of image template matching, mismatching
occurs due to occlusion where one image of an image pair is
invisible, or there is distortion in the aerial image caused by the
presence of a wall texture or other elements. We propose an
improvement in accuracy.
3. PRINCIPLE OF THE PROPOSED TECHNIQUE
We developed an algorithm for the extraction of building walls
using the proposed technique of independent rectification. This
technique can extract the candidate line segments of a
building’s footprint, which consists of the top horizontal edge
of the building wall, using direction-limited edge matching in
the object space. In this paper, we demonstrate the
improvements made to the system, which in addition to the
building wall includes extracting the rooftop and the ground-
level area beside the wall. The data-processing flowchart of the
proposed technique is shown in Figure 1.
3.1 Generation using independent rectification
The conceptual diagram shown in Figure 2 indicates the method
for the independent rectification image (IR image) generated by
the independent rectification method (called IR method in this
paper).
> a a au ages uc
assssmnes
Firs
cor:
trans
syst.
forn
exte
syst
COO!
the
num
indi
