International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
PHOTOGRAMMETRIC NETWORK FOR EVALUATION OF HUMAN FACES
FOR FACE RECONSTRUCTION PURPOSE
Péter Schrott', Ákos Detrekói, Károly Fekete
Budapest University of Technology and Economics, Department of Photogrammetry and Geoinformatics
Müegyetem rkp. 3., 1111 Budapest, Hungary
schrott.peter @ fmt.bme.hu
KEY WORDS: Biometrics, Close Range, Convergent, Networks, Modelling
ABSTRACT:
Facial reconstruction is the process of reconstructing the geometry of faces of persons from skeletal remains. A research group (BME
Cooperation Research Center for Biomechanics) was formed representing several organisations to combine knowledgebases of
different disciplines like anthropology, medical, mechanical, archaeological sciences etc. to computerize the face reconstruction
process based on a large dataset of 3D face and skull models gathered from living persons: cranial data from CT scans and face
models from photogrammetric evaluations. The BUTE Dept. of Photogrammetry and Geoinformatics works on the method and
technology of the 3D data acquisition for the face models. In this paper we will present the research and results of the
photogrammetric network design, the modelling to deal with visibility constraints, and the investigation of the developed basic
photogrammetric configuration to specify the result characteristics to be expected using the device built for the photogrammetric face
measurements.
1. INTRODUCTION
The face reconstruction is the process of reconstructing the
mimetic muscles and the soft tissues, solely based on the
morphology, structure and texture of the cranial bones. The
muscles and the bones develop and change together during
lifetime, so the morphology of the skull gives an opportunity to
estimate the muscular system of the face. Additional aids for the
he soft tissue estimation are the so-called median soft tissue
datasets. These sets based on statistical measurements of the
average soft tissue thickness of the face in several
anthropological landmarks. (Kustár, 2004) This statistical
method however carries some uncertainty. A multidisciplinary
research study has been started in 2007 in the Budapest
University of Technology and Economics, which aims to
support human morphologic measurements by photogrammetric
methods. The goal of our research cooperation is to develop the
face reconstruction method and create a face reconstruction
software based on statistically sufficient samples (3D face and
skull models of persons) and guided by defined mathematical
correlations between the anatomy of the skull and the face
geometry.
The first phase of the project is targeted at finding the optimal
measuring technology to collect geometric data of the human
face. Medical imaging is a continuously developing science,
and nowadays we can choose from several possibilities. We
have reviewed some of them, and made an accuracy analysis of
them, the results were published in the ISPRS XXI. Congress,
Beijing, 2008. (Schrott, 2008.) These results suggested that for
obtaining cranial data the only viable option is to use CT scans.
Even though the soft tissues absorbs X-rays to a lesser extent,
the geometric precision of the 3D model is only slightly lower
than in the case of bony structures, therefore theoretically we
can gain 3D model of the face by CT. The reason we chose
another method was that all of the CT devices we know works
on lying persons facing upwards. This position causes the face
distorted by gravity which in this case takes effect in different
direction as usual. In special occasions (elderly or overweight
persons), the difference of the face of a standing or lying person
can be so large that the person is virtually unrecognizable.
Investigation of other possibilities result in the decision of using
photogrammetry, the high accuracy required by the
anthropologist experts can be achieved by this.
2. NETWORK DESIGN
Non-topographic applications of photogrammetry rarely uses
stereoscopic image pairs for evaluation, using a convergent
photogrammetric network can result in more accurate
measurements. Hence, several parameters (like the number of
cameras, the geometry and arrangement of them etc.) can be
adjusted more freely to optimize the measurement. The
development of the photogrammetric capturing device required
the calculation and the design of this subject-specific
photogrammetric network.
The 3D capturing of a human face’s geometry has its own
specificities and problems which should be taken into
consideration during the development.
The network design problem in photogrammetry is to achieve
the required measurement quality (reliability and precision) by
finding a suitable set of measurements with the least possible
cost. It is widely accepted that the First Order Design (FOD) of
the close range photogrammetric network is to adapt the
network for the network design constraints, followed by an
iterative simulation of the network. (Fraser, 1996) (Fekete
2006)
2.1 Incident and intersecting angle constraints
The viewing direction and the surface normal at the feature
highly affect the reliability of the measurement: perpendicular is
the optimal arrangement, from directions close to coplanar are
the worst. The other angle-specific problem is the intersecting
angles of the camera’s optical axles; each perpendicular to
another is desirable. One of the most widespread network
architecture in close range photogrammetry is a square-based
equilate
measure
arrange
decided
(Figure
Figure
2.2 Vis
One of t
the visi
framewo
object t:
subsets c
This con
the came
The nat
caused |
however
most sig
nose. Th
appear ii
accuracy
of the po
visibility
spatial ar
Visibility
The visil
faces. Ol
result of |
a photogi
but it is 1
Works for
V. Blanz
face mod
people. TI
parametei
like age,
Modeller
and we g
visibility
adjusted
(Blender)
point den