| local
rior
, pp. 111-
nu. ACM.
ion, Ph.D.
No. 3, pp.
"), Pattern
Hee Ju Park
COLOUR IMAGE MATCHING FOR DTM GENERATION AND HOUSE EXTRACTION
Hee Ju PARK, Petra ZINMMERMANN
"Swiss Federal Institute of Technology, Zurich, Switzerland
Institute for Geodesy and Photogrammetry
heeju Q ns.shingu-c.ac.kr
petra? geod.baug.ethz.ch
Technical Session III-2
KEY WORDS: DTM/DEM/DSM, Image matching, Reconstruction, Urban Objects
ABSTRACT
Image matching plays a key role in automatic DTM generation and house extraction. For the house extraction from
large scale imagery, point matching and line matching complement one another: Line matching gives the 3 dimensional
line information which supports house reconstruction at ridge lines or roof boundaries; dense well distributed point
matching results contribute to the surface model generation of the remaining non-breakline regions. We propose a new
epipolar line equation, which is determined by orientation parameters and supports both epipolar line search and
epipolar imagery generation. The proposed matching process is divided into point matching and line matching. To
derive highly reliable results in point matching we include blunder suppression based on the positional relationship
between possible corresponding point pairs. Line matching is supported by the results of point matching to reduce the
number of possible corresponding line pairs. Regarding similarity comparison for line matching we use the line shape,
the flanking regions colour, information on positional relationship and connectivity between candidates for
corresponding lines and neighbouring points and lines. We tested the proposed method with a sample dataset and show
the results .
1 INTRODUCTION
Image matching plays a key role in automatic DTM generation and house extraction. Applications on DTM generation
can be found in many current commercial digital photogrammetry workstations, and methods and applications of house
extraction can be found in [Gruen et al, 1997; Henricsson, 1996]. For the extraction of houses from large scale imagery,
both point matching and line matching are necessary: Line matching gives the 3 dimensional line information which is
useful for both for breakline detection and house reconstruction; whereas dense well distributed point matching
contributes to the surface model generation of the remaining non-breakline regions. Many studies have been made
related to this issue, but still there is a general robust method missing.
The aim of this study is to find a matching method focussing on this particular problem. In this paper firstly we will
describe the new epipolar equation with geometrical proof. Secondly we will describe our study on a new matching
method. Finally the results of our matching methodwill be described and discussed.
2 PROPOSED EPIPOLAR LINE EQUATION
Epipolar line geometry is one fundamental principle in image matching domain [Paul R. Wolf, 1984]. Within an
overlapping imagery pair, the corr esponding point of one point lies on the epipolar line which is corresponding to that
point. Here we suggest a new epipolar line finding method. The basic principle of the epipolar line derivation is the
condition of coplanarity, the details are as follows:
Let there be a couple of cameras which capture an object at the same time. We call one as left camera with indices 1,
another as right camera with indices 2. Let O,, O, be the left and the right camera centre. And Let P be a point on
the object. By definition Or; O, , P are on the same plane called epipolar plane. Epipolar lines corresponding to P
are defined by the intersections of the epipolar plane and the image planes of the left and right camera. As the
intersection of two planes is always a straight line, the resulting epipolar lines are straight lines and there exist a couple
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 697