08.
and
rks.
cial
SUS:
age
the
etry
nal
‘nal
and
EE
ited
03.
lity
)02.
ous
AAI
and
ity
nal
for
ean
tion
lof
cent
lof
istic
A,
2ory
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
AN INNOVATION APPROACH FOR DEVELOPING A 3D MODEL BY REGISTERING A MONO IMAGE
ON A DTM
Amir Saced Homainejad
Independent Research Group on Geospatial, Tehran, Islamic Republic of Iran, (saeed.homain@gmail.com)
Commission VI, WG V1/4
KEY WORDS: 3D model, Visualisation, Image registration, Object extraction
ABSTRACT:
This paper is reflecting a study on developing and reconstructing a 3D model from registering an aerial image on a point clouds or a
DTM. The point clouds has been chosen as the main data for reconstructing a 3D model. For achieving the aspects of the study, all
objects were individually detected and extracted from the point clouds and captured in a particular layer in a CAD environment. Then
cach object will be converted to a raster format and will be registered on the image. This process is called the reverse registration in
this study. Since the object segmentation and detection from a digital image is a complex process, the reverse registration is
implemented in order to assist the process of the object detection from the image. This paper will discuss two methods of object
detection from point clouds for the reverse registration. These methods were proposed and implemented for this study. Also, the
paper will discuss the reverse registration and how this method improves the process of the object detection and extraction from the
image. Discussion of reconstructing a 3D model from registering the digital image on a DTM or DSM (both of which developed
from the point clouds data) is another goal of this paper.
1. INTRODUCTION
Reconstructing of 3D model from acrial images specifically,
and in generic term from any type of images is now within the
grasp of a vast variety of businesses. There are great benefits
with rendering a 3D model for analysis and interpreting of the
object, and consequently it has attracted the interest of the most
businesses, government bodies, and developers and research
groups. Due to implementing a reliable method for
reconstructing a 3D model from images, a computer enhanced
with a powerful graphic card is essential. Especially with
development of computer graphic, the computer has been
become easier to interact with, and better understanding and
interpreting any type of data. Developments in computer
graphics have had great impact in most of computer's
applications as well as digital photogrammetry and remote
sensing. Reconstructing a 3D model from a stereo image using a
stereoplotter or a stereoscop is a well known technique in
photogrammetry. In conventional fashion, a stereo model was
reconstructed within a stereoplotter and coordinates of the
interest points on the object or the terrain were extracted
manually; however, in modern digital stereoplotters the process
has been implemented automatically or it is better to say semi-
automatically. Modern stercoplotters are able automatically to
develop a Digital Terrain Model (DTM) or Digital Surface
Model (DSM) from sterco images; nevertheless, there are other
approaches for reconstruction a 3D model from non-optical
sequence images such as those approaches has been
implemented by Brahim et al. (2010) or Wei et al. (2009). Also
there are some approaches for reconstruction a 3D model from
mono image with help of the auxiliary data (Zhang and Tsui,
1998, Qian, 2010, or Chen and Kohatsu, 2007). In recent years
with development of computer graphics, there is an interest
towards reconstructing 3D model in computer for visualising,
rendering, and objects analysing. In such a context, all platforms
for developing a 3D model have to meet the requirements of the
3D modelling. Therefore, majority of the techniques have been
developed based on integration of two or more sensors, while
cach sensor provides a particular data from the terrain or the
object and with combining those data finally a comprehensive
3D model can be reconstructed. For example, Sportouche et al.
(2009) reconstructed a 3D model with registering high
resolution optical image on a SAR image. In their approach, two
dimensional data were extracted from a mono image with
implementing a number of constraints; they registered the
extracted data on a SAR image. Tupin and Roux (2003) were
developed a method that was very close to previous method.
They also extracted planar data from images and registered on a
SAR image. The main aspect of both methods was to detect
buildings from SAR image which matches with their
corresponding which already were detected from images due to
reconstructing a 3D model. Dammann et al. (2006)
reconstructed a 3D model by registering an optical image on a
generated 3D model from Chirped Amplitude Modulation (AM)
LADAR image. The purpose of registration of optical image on
the constructed 3D model was to provide a texture to the 3D
model. Homainejad (2011 a, 2011 b, and 2010) implemented an
approach for reconstructing a 3D model by registering mono
images on a generated DTM or DSM or 3D model. The
approach consists of the following steps:
1- Dividing the image to sub area,
2- Each sub area was registered on its corresponding in
the DTM or 3D model. In this step all pixel will be
transformed to the 3D model space and will be
converted to points. Each point included X, Y, and Z
coordinates and intensity value which inherited from
the image.
3- The output from this approach is a 3D model;
however, it has an ortho-image characteristic as well.
This paper will discuss a new version of Homainejad's
approach for reconstructing a 3D model by registering an image
on a 3D model which was developed from the point clouds. The
organisation of this paper is: the background and proposal will
be given in the next section; study area will be given in section
3, the methodology, the result and analysis will be given in
section 4, and section 5 will give a conclusion.
2. BACKGROUND AND PROPOSAL
The author is participating in a project on urban classification
and 3D building reconstruction which has been proposed by
ISPRS — Commission III (Working Group III/4). The core of
179
