Em 
A NEW STEREO MATCHING APPROACH IN IMAGE/OBJECT DUAL SPACES 
Yaonan Zhang 
Centre for Computer Graphics and Mapping 
Thijsseweg 11, Delft University of Technology 
2629 JA Delft, The Netherlands 
Commission V 
ABSTRACT: 
The stereo matching (or correspondence) remains one of permanent problems in Photogrammetry and Computer Vision. 
This paper presents a new approach to solve the problem, which incorporates the image space based matching techniques 
with the high level knowledge about the objects. The low-level processing (edge detection, feature extraction) and 
candidate matching are carried out in image space, while the final matching is determined in object space as solving a 
consistent labelling problem which results from the integration of candidate matching, high level constraints of objects 
and other constraints of image matching. One of the innovative features in our approach lies in back-projecting (back 
tracing) the line pairs from candidate matching into the object (scene) space, and combining all the constraints in a 
unified process. We substitute the concept of "figure continuity" usually used in the image matching with the high level 
knowledge from the object space. 
KEYWORDS: Image processing, Image Matching, Line detection. 
1. INTRODUCTION 
The research on matching has been taken in computer 
vision and photogrammetry society for quite a long 
time. According to the space where the matching takes 
place, the existing techniques for solving the matching 
problem roughly fall into two categories: image space 
based and object space based. In the image space based 
matching, the primitives of one image are compared 
with ones on the another image. Many solutions to the 
matching have been proposed in the image space. The 
methods vary with different choice of primitives: area- 
based (intensity-based), feature-based and structure- 
based (relational matching). Recently, several articles 
are devoted to the object space based matching. This 
method emerged originally from the task of 
reconstructing digital terrain model from a pair of digital 
images, independently developed by  Wrobel' and 
Helavz?, etc. Helava used the concept of "groundel" as 
a unit in object space similar to the "pixel" in the image 
space. The image intensities corresponding to each 
groundel can be analytically computed, if all pertinent 
geometric and  radiometric parameters (including 
groundel reflectance, etc.) are known. A least square 
method is adopted to determine a set of unknown 
quantities or improvements to their approximate values 
used in the analytical prediction process. 
Although the progress is undoubtedly made, most of the 
algorithms are still task and domain dependent, many 
parts of the problem still need full exploration of our 
human intelligence. In solving this difficult problem, we 
propose a novel approach which unify the techniques in 
image and object space, combining the geometric 
knowledge of scenes. The motivation behind this is that 
534 
a general solution for ill-posed problem such as 
matching is to use additional constraints or knowledge 
to restrict possible solution?. The existing constraints 
used in the matching are the uniqueness, smoothness, 
ordering, figure continuity and camera geometry 
(epipolar geometry), etc. In this paper, we introduce a 
new concept of "general geometric constraints of 
scenes". In the problem of reconstructing digital terrain 
model from digital image, the smooth constraints of 
surface can be adequately assumed'?^^, But in the most 
application of computer vision, industrial robot vision 
and automated close-range photogrammetric system, the 
scenes are full of lines, shapes and structures, it is 
impossible to unify all the information in the object 
space in a straight-forward way. The reason is that the 
original digital image consists of only raster pixels 
which can not directly provide much structural 
information required heavily by later analysis. Without 
low-level processing in the image space, it is practically 
impossible to get more structural description of the 
images. In our approach, we back-project (back trace) 
the image primitives (from low-level processing) into the 
object space, and carry the matching in the object space 
combining the available knowledge from scene with 
other constraints. We implement the line-based matching 
in the image space in order to find the candidate line 
pairs and then project the these line pairs into the object 
space. The final matching became a consistent labelling 
or constrained satisfaction problem. We use a relaxation 
procedure to get the final lines in object space. 
The main idea described in this paper has been 
previously reported by author in 19917. This paper 
includes the description on line detection and grouping, 
back projection for horizontal image lines, as well as 
more experimental results. 
ext 
prc 
loc 
fea 
loc 
inc 
cor 
res 
Ed; 
Sot 
and 
det: 
glo 
met 
whi 
low 
bou