GEOMETRIC CALIBRATION OF A VISIBLE-NIR VIDEO CAMERA
J. L. Lerma*, L. A. Ruiz, F. Buchön, R. Pons, M. Galindez
Departamento de Ingenieria Cartogräfica, Geodesia y Fotogrametria, Universidad Politecnica de Valencia,
C° de Vera s/n, 46022 Valencia, Spain - jllerma@cgf.upv.es
KEY WORDS: Calibration, Orientation, Video, Multispectral, Close Range, Architecture
ABSTRACT:
Methodologies of camera calibration ranging from self-calibration with additional parameters, on-the-job calibrations, laboratory
calibrations till simple methods for estimating the inner orientation parameters are well known and reported in the literature.
However, the estimation and reliability of inner parameters under new solid state image sensors with a priori unstable intrinsic
parameters, for instance, video cameras with zoom lenses and multispectral bandwidth, are not sufficiently tested, though some of
these cameras are the future in specific close-range and multispectral applications. In architecture and archaeology, for instance,
these kind of cameras can be used for the identification and characterisation of materials and damages on architectonic facades,
paintings recovery on both hand-made walls and caves, deterioration of external constituent materials on indoor/outdoor cultural
heritage, and other pattern recognition processing tasks.
This paper reports the results of several studies that were aimed at modelling and calibrating geometrically digital B/W bandwidth
video cameras using the DLT with additional parameters. The approach considers only the radial distortion parameters K; and K,.
The results show systematic and non-systematic variations in both interior and exterior orientation parameters depending more on
the frame than on the spectral wavelengths (visible and near infrared).
1. INTRODUCTION 2. EQUIPMENT AND TEST SITE
Non-metric cameras are used in heritage documentation tasks 2.2 Equipment
because of their simplicity, low weight, low cost and
functionality. The availability of zoom cameras makes them Available for use in this study was the Hitachi KP-F2A
mostly suitable for most of the expected and unexpected progressive scan, near infra-red monochrome charge-coupled
situations. Digital cameras in close-range work offer well-
proven benefits to the whole imaging community, journalists,
conservationists and restorers, engineers, metrologists,
photogrammetrists and computer vision specialists. Particularly,
photogrammetrists will always inquire for the metric
requirements in the final products, mainly maps, rectified
images, orthoimages or just 2D/3D co-ordinates. Whenever the
maximum accuracy is expected, the inner orientation
parameters of all the images must be fulfilled, i. e. principal
distances, principal point co-ordinates, radial and tangential
distortion parameters. However, neither all the surveying jobs
nor documentation tasks require metric accuracy, nor all the
cameras are metric ones.
Fortunately, both computer vision specialist and
photogrammetrist have developed their own ways of camera
calibration (Fraser, 1997; Gruen et al, 2001), which prepare
non-metric (analogue or digital; video, still-video) cameras
suitable for moderate accuracy requirements. This study used
the Direct Linear Transformation (DLT) because of the
simplicity of its mathematical form and the a priori instability
of the zoom visible-NIR video camera. The goal was to
evaluate the stability of a multiespectral video camera from the
set of orientation parameters.
* Corresponding author.
device (CCD) camera. This CCD camera has a focal plane of
approximately 4.87 mm by 3.67 mm, which correspond to an
aspect ratio of 4:3 for a standard RS170 video signal. The focal
plane consists of 658 horizontal by 496 vertical pixels. Each
pixel has an exterior dimension of 7.4 um (H) by 7.4 um (V).
Rainbow. 8 mm to 45 mm zoom lens was used. Each B/W
image has a quantization level of 8 bits/pixel. The output image
has a resolution of 640 by 480 pixels.
The spectral response of the video camera ranges from 400 nm
to 1200 nm, with a maximum sensitivity at 760 nm (Figure 1).
Schneider B+W filters were used for the extraction of the
spectral bands, in particular, filters 081 (blue), 080 (bright
blue), 061 (green), 092 (dark red and near-infrared) and 093
(near-infrared).
1
0.8
0.8
2 0.7
$ 05
E 04
$ Q3
m
0.2
0.1
a ^ ^ gt , = "P 7
400 500 800 700 800 2900 1000 1100 1200
Wave length (nm
Figure 1. Spectral sensitivity response.
—400—
22.3
Exper
Depar
Engin
distrit
Were
stand:
The €
where
co-orc
param
param
signif
were |
param
The e
indep:
The s
positi
five ii
093)
distan
repres
Figur
