SEMI-AUTOMATIC MONOPLOTTING
ON A DIGITAL PHOTOGRAMMETRIC STATION
Peggy Agouris, Dirk Stallmann, Haihong Li
Institute of Geodesy & Photogrammetry
Swiss Federal Institute of Technology
ETH - Hoenggerberg
CH-8093 Zurich, Switzerland
Ph.: +41-1-6333054, Fax: +41-1-3720438
e-mail: peggy or dirk or li @p.igp.ethz.ch
ISPRS Commission ll, Intercommission II/III
KEY WORDS: Digital Photogrammetry, GIS, Automation, Monoplotting, Object Extraction, Matching
ABSTRACT
This paper deals with the subject of semi-automatic object extraction from digital imagery in monoscopic mode.
Addressed topics include conceptual, algorithmic and implementational issues, performed experiments, as well
as discussion of encountered problems and adopted solutions. In particular, we present the algorithm and ob-
tained experimental results for the semi-automatic extraction of road networks from SPOT imagery using wave-
let-transformed images. In addition, for larger scale images and various object types, we present the
comparison of two methods for object extraction we have applied and tested, namely least squares template
matching, and active contour models (Snakes). In the former case, edge locations are precisely identified
based on local gray value variations, while in the latter case global continuity constraints are enforced to pro-
duce meaningful results. Finally, we propose a novel algorithmic approach for semi-automatic object outline
detection which combines the strong points of the previous two methods. Least squares matching provides
again the mathematical foundation while at the same time global continuity is enforced through the introduction
of object-type-dependent shape constraints which ensure geometrically coherent results.
1. INTRODUCTION into two broad categories [Fischler et al., 1981]:
s : = Q type | operators, offering high reliability in prop-
The objective of semi-automatic monoplotting is the erly identifying classes of objects without particu-
extraction of objects from digital imagery in mono- larly dealing with precise outline determination,
scopic mode. Object extraction from images consists and
of the following phases:
Q identification of an object within an image, which
involves image interpretation, understanding and
object classification, and mations of the object location are available.
Q tracking the object by precisely determining its In an effort to optimize both measures, operators from
outline. these two classes can be combined in complex com-
putational strategies for object extraction [Suetens et
Q type Il operators, which do not aim at reliable
identification, but instead offer high precision in de-
tecting outlines, provided that adequate approxi-
It is well known that there exist no global edge detec-
tors which could be applied to a digital image function
to both identify and track edges with sufficient suc-
cess. Instead, one can witness a trade-off between
reliability, which expresses a measure of the qualita-
tive accuracy associated with identification, and pre-
cision which expresses a measure of the geometric
accuracy associated with tracking. According to
these measures, one can classify existing operators
al., 1992]. In this paper we present several methods
which can constitute the automated object extraction
core within a broader digital photogrammetric semi-
automatic monoplotting strategy.
2. SEMI-AUTOMATIC MONOPLOTTING
In semi-automatic monoplotting, the identification
146
task of a typ
single image
ule performs
More specifi
tify an object
age, select t
object belon,
rough appro
this approxin
screen node
for curviline:
quently, thes
necessary aj
edge positior
objects withi
cisely positic
according to
tor contributic
vided for suc
to the actual
Judging fron
grammetric c
extraction, si
broader obje
mal. Human
sly and almo:
optimizes acl
burden. At th
outline posit
shows to be
prone part of
formed auton
In the next se
foundation ai
lowing semi-:
Q road exti
imagery,
Q the use
outline exti
O least squ
Q globally :
ing.
3. ROAD
TR/
The semi-au
road network
let decompos
feature extra
gramming (Fi
Wallis filter |
available im:
