9-11 Nov. 1999
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
A NEW APPROACH FOR MATCHING SURFACES FROM LASER SCANNERS AND OPTICAL SCANNERS
Ayman Habib and Toni Schenk
Department of Civil and Environmental Engineering and Geodetic Science, OSU
habib.1@osu.edu, schenk.2@osu.edu
KEY WORDS: Surface Matching, Correspondence Problem, Hough Transform, Change Detection, DTM analysis, Cali-
bration, Fusion
ABSTRACT
Surfaces play an important role in diverse applications, such as orthophoto production, city modeling, ice sheet
monitoring, and object recognition. Surfaces are usually obtained by a sampling process. The raw sampled data
must be processed further. A frequently occurring task is the comparison of two surfaces. In the most general
case, the two surfaces are described by discrete sets of points, whereby the point density may be different as well as
the reference systems. We propose to compare two surfaces by computing the shortest distance between points in
one surface and locally interpolated surface patches of the second surface. This entails a correspondence between
points and surface patches. We describe a solution to this matching problem that is based on a parameter space
representation. After a brief problem statement we explain the proposed matching scheme by way of an example.
We then apply the method to determine the transformation parameters between the two surfaces. To arrive at an
operational solution, we reduce the n-parameter space to one dimension by an iterative solution. The feasibility
of our matching scheme is demonstrated with simulated data sets as well as real data. We show how a surface
determined by laser scanning can be compared with the same physical surface but established by photogrammetry.
As a natural extension, one can use the method for change detection.
1 Introduction
There is an increasing demand for the rapid generation
of digital surface models (DSM). The production of digi-
tal orthophotos as backdrops for GIS requires DSMs, for
example. More recently, city modeling is an application
that poses a challenge for generating DSMs. Surfaces
play also an important role in such diverse applications
as ice sheet monitoring and recognizing objects in aerial
and satellite scenes.
Surfaces are typically determined by a sampling process.
This is certainly true for airborne laser scanning and
stereo photogrammetry. The net result of data acquisi-
tion is a set of points that constitutes a discrete surface
description. In case of laser scanning, the point distri-
bution is irregular and the surface characteristics, for
example breaklines, are not explicitly encoded.
The set of points obtained during data acquisition is
hardly a useful end product. There are a number of ba-
sic operations that must be performed on surfaces. One
of the first steps usually involves what we may consider
a resampling process. The classical example is interpo-
lating the original set into a regular grid (gridding) be-
cause most every subsequent process assumes regularly
spaced data.
In one way or another, many processes involve the com-
parison between surfaces. Examples are abundant; cal-
ibrating data acquisition systems involves the compar-
ison between the observed surface and the known sur-
face (e.g. test field); change detection compares two sur-
faces sampled at different times; merging two or more
data sets for a combined surface (fusion) requires quality
control; a data set acquired in a local reference system
must be transformed into a differently registered set.
Surface comparison is usually performed by interpolat-
ing both data sets into a regular grid. Then, the compar-
ison is reduced to analyzing the elevations at the grid
posts. Not all applications allow this simple procedure,
however. Take the example of two irregularly spaced
data sets that are acquired in different reference systems
with unknown transformation parameters. We describe
in this paper a new approach for solving this general
problem.
Ebner and Strunz (1988) and Ebner and Ohlhof (1994)
describe a solution that is based on interpolating the
data to a grid, subject to a transformation with unknown
parameters, which are determined in an adjustment pro-
cedure whereby the elevation differences at the grid
posts are minimized. We propose to minimize the dis-
tance between the points of one set along surface nor-
mals to locally interpolated surface patches of the other
surface. As shown in Schenk (1999a) this makes a weak-
posed problem well-posed.
The next section provides a more detailed problem state-
ment and discusses solutions. We then concentrate on
the solution of the matching problem and illustrate the
proposed approach of using a voting scheme to analyze
the parameter space by an example. Finally, we present
experimental results obtained from synthetic and real
data sets.
2 Problem Statements and Solutions
2.1 Simple Case
Given are two sets of points that describe the same
surface. Let $1 — Ípi,p»,..., pa) be the first set and
S» = {q1,92,.-- ,qQm} the second set, n += m. Suppose
the points are randomly distributed (no point to point
correspondence). The problem is to determine how well
the two data sets agree describing the same surface.