Zuxun Zhang
SEMI-AUTOMATIC BUILDING EXTRACTION BASED ON LEAST SQARES
MATCHING WITH GEOMETRICAL CONSTRAINTS IN OBJECT SPACE
Zuxum ZHANG Jianging ZHANG
Wuhan Technical University of Surveying and Mapping, P.R.China
zxzhang@supresoft.com, jianging@supresoft.com
KEY WORDS: Building, Semi-automatic, Extraction, Least Squares Matching, Geometry, Object-space
ABSTRACT
Building extraction is one of the critical problems in digital photogrammetry. This article describes a semi-automatic
way based on the object space. According this method the building extraction is converted into a procedure of leaner
template matching with geometrical constrain. For it’s potential use in the DPW the approximate positions of the
building which are needed for least adjustment are pointed out by operator on the aerial photos. The space coordinates
of roof's corners can be calculated by forward intersection. The error equations of least squares image matching are
created between image and templates of straight line, and the unknowns are the corrections of the space coordinate of
roof's corners. The geometric constraints include right angle. parallel. same height (horizontal) condition and etc.
This method is suitable for the houses with flat or gabled roof. From the experiments the method proposed in this paper
has been proved to be very accurate and quite robust enough to the noisy.
1. INTRODUCTION
As the result of the great progress of Digital Photogrammetry within last decade, many commercial digital
photogrammetric workstations (DPW) have been developped and applied into practical production; and they are
progressively replacing the disappearance of the traditional photogrammetry equipment. Until now, with the
development and consummation of the theory and algorithm of image matching, most of the functions of digital
photogrammetric workstations, such as orientation (mainly inner orientation and relative orientation), DEM generation,
orthophoto creation and mosaic, pass point transformation and measurement in areatrianglation, have reached high
automation (Zhang, 1998). But there is still rather a long way for us to accomplish in the practice application of
automatic and semi-automatic man-made feature extraction. For the sake of application, we shall select an ideal “semi-
automatic” mathematical model firstly, and then expand its “semi-automatic” part steadily step by step, and finally
achieve the automatic man-made feature extraction. That means: an ideal “semi-automatic” mathematical model should
be open and expansible. According to our experience and understanding, automation of man-made feature extraction
could be divided into three parts. Q) Recognition and segment, divided the feature we would extract, for example,
houses, from its background, and then distinguish its type, (such as houses-- the types of their roofs); @Set the initial
value, for example, set the seed points of road, or corner point of houses; (S)Accurate positioning. And we may find out
that automatic recognition is rather easy for people, especially to recognize from the aerial image under the condition of
stereo measurement; and on the other hand, it is rather difficult and accurate positioning is relatively easy for computer.
Although there has been a great deal of study in automatic house extraction, for example, using digital surface model
(DSM) to recognize or to set the initial value of houses (Balstavias, 1995; Huertas, 1989; Kim, 1996; Trinder, 1995),
but it is still rather difficult in the city block crowded with buildings. In order to give prominence to emphasis and for
the sake of possibility of application, this paper would be centralized in the discussion of accurate positioning and
mathematical model of houses, which have been manually recognized and based on manually setting type of house and
initial value of roof boundary.
The mathematical model of house accurately positioning is the least squares matching (Heipke, 1992) with the
geometric constraint conditions in the object space. Namely, the beeline template and boundaries are used in the least
squares matching in order to get the space coordinates of house-corners directly, under certain geometric constraint
conditions.
2. MATHIMATICAL MODEL
2.1. Least squares Adjustment model
1022 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.