|
SFB 603, C2: Analysis, coding and processing of light fields, for acquisition of realistic model dataThe subproject C2
of the Sonderforschungsbereich 603
examines the field of "analysis, modeling and processing of light fields for
acquisition of realistic model data" together with the
Chair for Computer Graphics (LGDV).
The central data structure of this subproject, the light field, allows to
generate arbitrary views of a real scene by so-called image-based modeling.
The appearance of the scene is only known from a collection of image data.
The required
information about the parameters and pose of the camera are estimated
by "structure-from-motion" algorithms directly from image data, which is
usually an image stream from a hand-held camera. Other possible data are
the images of a camera mounted on a robot arm. The subproject C2 has been
funded by the Deutsche Forschungsgemeinschaft
(DFG) since 1998.In recent years, the consideration of solely static scenes, which is
commonly the case for light fields, was extended more and more to
dynamic scenes. The work on reconstruction of light fields from scenes
which contain moving but rigid objects was begun already in 2003 and
was continued in 2005 as well. After the segmentation of features in "background" and
"moving object", the two 3D reconstructions, which have been estimated
independently from each other, are now registered with
sufficient accuracy. From this, the object's own motion is calculated,
and similar object positions are assigned to one respective time step
using vector quantization. The visualization of the resulting
dynamic light field is done by so-called confidence maps, which
allow the masking of the moving object for time steps which are currently not
visible. A newly developed technique for evaluating the quality of a light field
which calculates the signal-to-noise ratio (SNR) allows an objective
comparison of different light field reconstruction procedures.
It was already employed for evaluating the estimation of intrinsic camera
parameters and for a comparison of different procedures for calculating
depth information. At the same time, local proxies, i.e., triangle meshes
which are valid locally for each image and which were applied
to light field visualization for the first time, were compared to
conventional depth maps. In addition to that, SNR was used to optimize
the depth maps of an existing light field by feeding back information from
light field synthesis. The applicability of the methods developed was shown in several cooperations
with other projects. In cooperation with the Graduate Research Center "3D Image Analysis and Synthesis"
the 3D reconstruction algorithms were applied for fast training in object
recognition. In subproject A7
of the Sonderforschungsbereich 603 these algorithms were used for
3D reconstruction of image sequences which were recorded using flexible
endoscopes. | Project manager: Prof. em. Dr.-Ing. Dr.-Ing. h.c. Heinrich Niemann, Prof. Dr. Günther Greiner
Project participants: Dr.-Ing. Ingo Scholz, Dr.-Ing. Christian Vogelgsang
Keywords: SFB 603, C2: Analysis, modeling and processing of light fields, for acquisition of realistic model data
Duration: 1.1.1998 - 31.12.2007
Sponsored by: DFG
Mitwirkende Institutionen: Lehrstuhl für Graphische Datenverarbeitung Erlangen
Contact: Scholz, Ingo E-Mail: i-uni@bellingo.de
| Publications |
---|
Scholz, Ingo ; Vogelgsang, Christian ; Denzler, Joachim ; Niemann, Heinrich: Dynamic Light Field Reconstruction and Rendering for Multiple Moving Objects. In: Katsushi Ikeuchi (Ed.) : Proceedings of the Ninth IAPR Conference on Machine Vision Applications (IAPR Conference on Machine Vision Applications Tsukuba Science City, Japan 16-18.5.2005). Tokyo, Japan : IAPR MVA Conference Committee, 2005, pp 184-188. - ISBN 4-901122-04-5 | Niemann, Heinrich ; Scholz, Ingo: Evaluating the Quality of Light Fields Computed from Hand-held Camera Images. In: Pattern Recognition Letters 26 (2005), No. 3, pp 239-249 | Scholz, Ingo ; Denzler, Joachim ; Niemann, Heinrich: Calibration of Real Scenes for the Reconstruction of Dynamic Light Fields. In: IEICE Transactions on Information & Systems E87-D (2004), No. 1, pp 42-49 |
Institution: Chair of Computer Science 5 (Pattern Recognition)
|
|
|