407
SEMI-AUTOMATIC CAMERA CALIBRATION AND IMAGE ORIENTATION
USING THE CIPA REFERENCE DATA SET
Frank A. van den Heuvel
Delft University of Technology, Department of Geodesy
Thijsseweg 11, 2629JA Delft, The Netherlands
E-mail: F. A. vanden He u ve I 7/ eeo .tudelft.nl
KEY WORDS: Architecture, Line-photogrammetry, Camera Calibration, Image Orientation, Object Reconstruction, Automation
ABSTRACT
The goal of this paper is to demonstrate the applicability of line-photogrammetric methods by applying them to a subset of the
images of the CIPA reference data set. This data set is set up by CIPA (The ICOMOS / ISPRS Committee for Documentation of
Cultural Heritage) in 1999 and consists of images of a historic building: the old city hall of Zürich. Although object reconstruction is
regarded as a final goal of a photogrammetric recording, this paper focuses on two tasks in architectural photogrammetry: camera
calibration and image orientation. The line-photogrammetric methods applied for these tasks have been developed in the past years.
These methods have in common that they require the extraction of image lines and information on the orientation of the related
object lines. The methods are not discussed in detail in the paper.
For camera calibration manually and automatically extracted straight image lines of five images were used. The required parallelism
and orthogonality information was obtained automatically by vanishing point detection. Correspondence between line features in
different images is not required. Camera parameters and their precision are estimated in a least-squares adjustment. The results
correspond well with the camera parameters provided in the CIPA reference data set itself and those provided in the literature.
The semi-automatic method for image orientation relies on line extraction and vanishing point detection as well. Furthermore,
coplanarity of object lines that reside in a façade is assumed. Fully automatic relative orientation is not possible for all four image
pairs used in the experiments. Depending on the image configuration and settings of a few parameters, one or two image points have
to be measured for successful orientation. These points are chosen on the corners of the building and thus in the overlap of
consecutive object models. This allows the scale to be transferred from one model to the next. Then this method for image
orientation results in a partial reconstruction of the building.
Figure 1: The subset of used images. Images 8, 9, 10 (top row), 11, 6, 3, and 16 (bottom). Images 8, 9, 10, 11, and 6 have been used
for calibration. The images in the bottom row are used for the image orientation experiments.
1. INTRODUCTION
1.1. Line-Photogrammetry
In the past years line-photogrammetric methods were developed
for the main tasks in architectural photogrammetry, i.e. camera
calibration (Heuvel, 1999a), image orientation (Heuvel, 2002),
and object reconstruction (Heuvel, 1999b). These methods have
in common that they require the extraction of image lines
(manually or automatically) in combination with information on
the orientation of the related object lines. The goal of this paper
is to briefly introduce these methods and to demonstrate their
applicability by applying them to images of the CIPA reference
data set.
1.2. The images of the CIPA reference data set
The CIPA reference data set consists of two sets of images taken
with two different digital cameras. Here, only a subset of the
images taken with the Olympus Cl400 is used. Due to the
manual interaction required for the reconstruction of a
structured object model, one of the goals of this research is to
use only a minimum number of images in order to make the
modelling more efficient. Therefore, only four images taken
from the corners of the building are selected for image
