LINEAR FEATURES IN PHOTOGRAMMETRIC ACTIVITIES 
A. Habib, M. Morgan, E.M. Kim, R. Cheng 
Department of Geomatics Engineering, University of Calgary, Calgary, 2500 University Drive NW, Calgary, AB, T2N 
1N4, Canada - (habib@geomatics.ucalgary.ca, mfmorgan@ucalgary.ca, emkim@ucalgary.ca, rwtcheng@ucalgary.ca) 
PS ICWG II/1V: Automated Geo-Spatial Data Production and Updating 
KEY WORDS: Analysis, Reliability, Automation, Feature, Digital, Photogrammetry 
ABSTRACT: 
Photogrammetric manipulation of imagery has been, for the major part, a point-based operation. The utilization of points remains to 
be convenient since few manually digitized points can be accurately obtained to carry out various photogrammetric orientation 
procedures (e.g., relative and absolute orientation as well as photogrammetric triangulation). On the other hand, the constant 
evolution of digital photogrammetry calls for the use of other primitives since distinct points fall short when attempting to derive 
higher level/semantic information from the input imagery. As a result, there has been a tremendous interest by photogrammetric 
researchers in utilizing linear features in various photogrammetric activities. This interest is attributed to the fact that the extraction 
of linear features from the image space is easier to automate than distinct points. On the other hand, object space linear features can 
be directly derived form terrestrial Mobile Mapping Systems (MMS), GIS databases, and/or existing maps. Moreover, automatic 
matching of linear features, either within overlapping images or between image and object space, is easier than that of distinct points. 
Finally, linear features possess more semantic information than distinct points since they likely correspond to object boundaries. 
Such semantics can be automatically identified in the input imagery to facilitate higher-level tasks (e.g., surface reconstruction and 
object recognition). This paper summarizes the authors’ prior research using linear features, which might be represented by 
analytical functions (e.g., straight-line segments) or irregular (free-form) shapes, in photogrammetric activities such as automatic 
space resection, photogrammetric triangulation, camera calibration, image matching, surface reconstruction, image registration, and 
absolute orientation as well as in medical applications. Current progress, future expectations, and possible research directions are 
  
d 
SAU: S 
TOR 
  
discussed as well. 
1. INTRODUCTION 
The majority of manual orientation as well as map compilation 
processes rely heavily on point-based features, which are well 
determined in two-dimensional imagery. For these features, the 
human operator has to store their semantics for later use. After 
the introduction of softcopy workstations and digital image 
compilation and analysis, the majority of recent research 
activities have been focusing on the automation of various 
point-based procedures in photogrammetry (Forstner, 1986; 
Haala and Vosselman, 1992; Gülch, 1994: Drewniok and Rohr, 
1997). The main difficulty in this type of research is the 
automatic extraction of useful semantics for the identified point 
primitives. 
Recently, more attention has been focused on using higher-level 
features (eg, linear and  areal features) in various 
photogrammetric operations. The emergence of digital images 
coupled with well-developed image processing tools motivated 
this direction of research. Also, the research has been propelled 
by the fact that automatic extraction of linear features is easier 
than distinct points (Kubik, 1991). The reliability of the 
extracted features is another factor to be considered. The 
reliability stems from the relationships among the sub-entities, 
which yields a robust extraction process that is not susceptible 
to inherent noise in the input imagery. Moreover, such features 
increase the system's redundancy and consequently improve the 
gcometric strength and robustness in terms of the ability to 
detect blunders. Therefore, matching ambiguities can be 
resolved, occluded areas can be casily predicted, and/or changes 
can be detected. Moreover, higher-level features, which can be 
automatically extracted, possess more semantic information 
regarding the object space. This constitutes an important factor 
for facilitating subsequent processes such as surface 
reconstruction and object recognition. In the object space, linear 
features can be easily derived from existing maps, GIS 
databases, or terrestrial Mobile Mapping Systems (MMS). 
There has been a substantial body of work dealing with the use 
of linear features that can be represented by analytical functions 
(such as straight lines and conic curves) in photogrammetric 
orientation (Mulawa and Mikhail, 1988; Heuvel, 2000). On the 
other hand, few research attempts have addressed the use of 
free-form linear features (Smith and Park, 2000). The main 
objective of the presented research in this paper is to establish a 
model capable of incorporating linear features, whether 
represented by analytical functions or free-form shapes, in 
images captured by frame cameras as well as scenes acquired by 
linear array scanners (line cameras). The implementation should 
allow for possible incorporation of these features in multi- 
sensor photogrammetric triangulation and other 
photogrammetric and medical activities (such as single photo 
resection, camera calibration, relative orientation, scene/image 
registration, and absolute orientation). 
= 
In the next section, a short discussion regarding the motivation 
for using linear features is introduced. Section 3 deals with 
various alternatives for image and object space representation ol 
linear features as well as the corresponding perspective 
transformation. The main applications of these primitives in 
various photogrammetric operations are briefly discussed in 
Section 4. Finally, conclusions and recommendations for future 
work are presented in Section 5. 
610 
Intei 
—— 
Feat 
be € 
phot 
obje 
vari 
obje 
2.1 
[may 
be € 
feat 
feati 
hom 
the 
cons 
or t 
requ 
2.2 
Obje 
line 
volu 
capa 
appr 
phot 
phot 
imag 
elem 
spac 
Corre 
imag 
repre 
(IOP 
inter: 
space 
can I 
far, 
solel 
word 
space 
leads 
suital 
and 
How 
calle 
its du 
2.3 
With 
photc 
scanr 
proce 
proce 
allow 
than | 
the e: 
often 
Systel 
space 
reseau 
line o