A FLEXIBLE AND AUTOMATIC 3D RECONSTRUCTION METHOD
Shunyi Zheng" , Zonggian Zhan“, Zuxun Zhang”
* School of Remote Sensing Information Engineering, Wuhan University, 129 Luoyu Road, Wuhan 430079, China
syzheng@263.net zzq09@?263.net zxzhang@supresoft.com.cn
EN
KEY WORDS: 3D Reconstruction, Relative Orientation, Consecutive Model Connection, Image matching, Texture mapping
ABSTRACT:
This paper proposes a flexible and automatic technique of 3D modeling from images captured by a handheld camera. The camera used
here is a metric digital camera. This method is well suited for use without specialized knowledge of 3D geometry or computer vision.
The technique only requires the camera to take photos around the object to be measured. The camera can be freely moved. But, the
distance of neighboring image station should not be too far. Then the taken photos are processed and the procedure consists of relative
orientation and connection of consecutive photos and a rigid bundle adjustment, followed by a robust image matching technique and
TIN generation and texture mapping. At this time 3D model is produced. Several sets of real data have been used to test the proposed
technique, and very good results have been obtained. Compared with classical techniques which use some special equipment or have
some special requirement, the proposed technique is easy to use and flexible, even can be automatic. It advances 3D computer vision
one step from laboratory environments to real world use.
1 INTRODUCTION
Obtaining three dimensional (3D) models of scenes from
images has been a long lasting research topic in
photogrammetry and computer vision. Many applications exist
which require these models. In traditionally robotics and
inspection applications accuracy was often the main concern.
Nowadays however more and more interest comes from the
multimedia and computer graphics communities. The evolution
of computers is such that today even personal computers can
display complex 3D models. Many computer games are located
in large 3D worlds. The use of 3D models and environments on
the Internet is becoming common practice. This evolution is
however slowed down due to the difficulties of generating such
3D models. Although it is easy to generate simple 3D models,
complex scenes are requiring a lot of effort. For existing objects
or scenes, the effort required to recreate realistic 3D models is
often prohibitive and the results are often disappointing.
A growing demand exists for systems which can visualize
existing objects or scenes. In this case the requirements are very
different from the requirements encountered in previous
applications. Most important is the visual quality of the 3D
models. In addition, there is an important demand for easy
acquisition procedures.
As described above, 3D reconstruction has been a long lasting
research topic and its wide application attracted research interest
of many researchers and scholars. And many different methods
have been developed and used in reconstruction of ancient
buildings, preservation of historical relics, medicine
reconstruction, industry measurement, human face
reconstruction, etc. All these technologies have their own
advantages, but requirement of expensive hardware or special
environment and control field and profound knowledge of
photogrammety and complex operation process limit their wide
use. For example, Fabio acquired human body image sequences
for reconstruction by using camera and special control field; F.
Paul Siebert reconstruct human body and human face by using
single baseline stereo and obtained good results, but his method
should confine to special shooting environment; Nicola
obtained high-accuracy human face model by using special
control field; Yongjun Zhang and Ulas Yilmaz measured
machine parts and reconstruct common objects’ face using
rotatory platform and made good results; Carlo carried out
experiments on the surface reconstruction of building using
expensive laser scanning device and obtained accurate point
cloud, but lacks of texture information and other information.
There are many other experiments like these, in some of which
good results have been made, but simplification, automation
and being practical are still urgent problems which need to solve
before wide application.
In this paper we present a new 3D technology in which a robust
image matching method has been used. Our goal is to
automatically reconstruct 3D model of object with image
sequence taken by hand-held camera. In this process, only
image acquisition process is done by people and all other
process is executed by computer automatically until 3D model
is produced. The technology applies to surface reconstruction of
many objects, no matter large or small, outdoor or indoor. It is
simple, fast and automatic. Its detail process is described in
following sections.
2 IMAGE ACQUISITION AND CALIBRATION
2.1 Image acquisition
The first step of this 3D reconstruction process is image
acquisition. There is no special demand when taking photos and
you can do it as you do when you want to get a scene picture.
But in order to reduce the difficulty of automatic image
matching, the distance between neighbouring image station, that
is the baseline, should not be too long. If the baseline is short,
the image data amount will be large, but this is not a serious
problem because post-processing is done by computers. As
shown in Fig. 1, in the measurement process of a coal pile, 83
photos were taken and the distance between neighbouring
image station was only about 1-2 meters.
2.2 Calculating of parameters of photos
In this paper, our goal is to visualize existing objects or scenes,
not to measure its actual size. So we only need to retrieve
objects’ shape which has same shape but possible not same size
as origin object. Therefore, no any control point is needed and a
free network model has been use. In addition, a metric camera is
Internat
used in
the who
same à
automat
point),
construc
with bi
baseline
automat
model
photogr
paramet
is the s
with ph
just forr
Objects
S n
3.1 Mi
Image r
reconsti
still vei
images:
