I'm a full time researcher at the LASMEA-UMR 6602 UBP/CNRS
(chargé de recherche au CNRS, HDR) working on Computer Vision.
My research topics are (in the chronological order on the whole)
Image Matching: Quasi-Dense Matching, Occlusion Handling,
Automatic Methods for Image Pairs
(used tools: Harris Points, Correlation, Uniqueness and 2D Disparity Gradient
Constraints, Best-First Propagation, Fundamental Matrix Estimation, RANSAC)
Image-Based Rendering: Warping, Occlusion Handling, Face Morphing,
Automatic Methods for Image Pairs
(used tools: Joint View Triangulation or JVT, Constrained Delaunay
Triangulation, RANSAC, Image Mosaic and Matching)
3D Reconstruction for Visualization: Quasi-Dense Reconstruction,
Surface Reconstruction from 3D points, Uncalibrated Vision, Image-Based
Modeling, Automatic Vision System for Still Image Sequences
(used tools: Projective and Euclidean Bundle Adjustments, Level Set,
Pinhole Cameras, Projective Geometry, Auto-Calibration, Texture Merging,
RANSAC, Image Matching)
3D Reconstruction for Robotic: Structure from Motion, Localization,
Simultaneous Localization and Mapping or SLAM, Automatic Vision Systems for
Videos
(used tools: Hierarchical/Global and Incremental/Local Bundle Adjustments,
Calibrated Pinhole and Fish-eye Cameras, SIMD Instructions of Pentium
MMX/SSE/SSE2, Key-Frame Selection, Uncertainty Ellipsoids, RANSAC,
Image Matching)
Omnidirectional Vision: Central and Non-Central Cameras, Structure
from Motion, Generic Camera, Image-Based Modeling,
Automatic Vision System for Still Image Sequences
(used tools: Image and Angular Bundle Adjustments, Uncertainty Ellipsoids,
Robust Method by RANSAC, Catadioptric Cameras, Image Matching)
More details are given in the research summary,
publications and demos below.
For all topics, the main difficulty is to obtain fully automatic methods
involving low cost hardware (1 camera+1 PC) for uncontrolled environments
(no beacons/turntable/friendly background or highly finished light control).
I tried to write this summary for both non-experts and experts of Computer
Vision.
More details on publications are given in the
complete list below.
First, there is the Image Matching problem: given two images of a scene
taken at different view points, estimate all pixel pairs (a,b) such that
(1) a and b are pixels in the first and second image respectively
(2) a and b are projections of a same 3D point of the scene.
I proposed an automatic method which starts from a few pixel pairs and
propagates continuously the pairwise relation to a maximum of pixels in the
images.
The initial pairs are provided by a standard method (interest point matching).
The method deals with occluded areas: a pixel corresponding to a 3D point
which is visible in only one image should not be used in a pair.
Also, the method deals with deformable scenes (and sometimes with two similar
but different objects like faces).
These results are possible by enforcing simultaneously in the propagation the
uniqueness constraint (each pixel is used by zero or one pair) and a 2D
disparity gradient limit (continuity of the function between images defined by
the pixel pairs).
However, pixels in low textured areas are not matched due to the ambiguity
problem.
The most complete publication on this topic is TPAMI'02.
TPAMI'02,
Match Propagation for Image-Based Modeling and Rendering,
M. Lhuillier and L. Quan. ICPR'00
(also in french: RFIA'00),
Robust Dense Matching Using Local and Global Geometric Constraints,
M. Lhuillier and L. Quan. CVPR'99,
Image Interpolation by Joint View Triangulation,
M. Lhuillier and L. Quan. BMVC'98,
Efficient Dense Matching for Textured Scenes Using Region Growing,
M. Lhuillier.
Second, there is the Image-Based Rendering problem: given two different
images, generate an image sequence which interpolates the two images.
I proposed an automatic method which starts from
image matching, then constructs two adequate and
constrained Delaunay triangulations in images, and finally warps the
triangulations to obtain the image sequence.
The method deals with deformable scenes (sometimes for different faces) by
approximating the matched area (usually textured) and interpolating matching
information in unmatched areas (usually uniform areas).
Also, the method deals with occluded areas assuming that the scene is not
deformable by explicitly separating occluded and non occluded areas.
A correspondence between vertices of the two Delaunay allows the generation of
intermediate images (one way is the simple linear interpolation of
coordinates).
The most complete publication on this topic is TCSVT'03.
TCSVT'03,
Image-Based Rendering by Joint View Triangulation,
M. Lhuillier and L. Quan. CVPR'00,
Edge-Constrained Joint View Triangulation for Image Interpolation,
M. Lhuillier and L. Quan. CVPR'99,
Image Interpolation by Joint View Triangulation,
M. Lhuillier and L. Quan. VI'99,
Joint View Triangulation for Two Views,
M. Lhuillier.
The next topic is the 3D Reconstruction for Visualization.
The problem is: given a still image sequence of a scene taken with a hand-held
standard (perspective) camera, estimate a surface (and its texture) which
approximates the scene.
Thus, the obtained 3D model may be seen from other view points.
The two steps of the method are (1) estimate 3D points of the scene and camera
parameters from the image sequence (2) estimate a surface from the 3D points
and sometimes additional data like silhouettes or colors in images.
The locations, orientations and focal length of the camera are estimated
assuming that the scene is rigid (non deformable).
Step 1 is automatic and uncalibrated (scene reconstruction up to a projective
transformation, then upgrade to an euclidean reconstruction by
auto-calibration).
Step 2 approximates the recovered cloud of points by a deformable surface
assuming that the object has a smooth surface (using a variational method and
level-sets).
The image matching above is used and provides many more
points than the standard approach (matching interest points).
It improves robustness and uncertainty of step 1 and it is obviously better to
recover the surface in step 2.
The most complete publication on this topic is TPAMI'05.
TPAMI'05,
A Quasi-Dense Approach to Surface Reconstruction from Uncalibrated
Images,
M. Lhuillier and L. Quan
(short version in french: RFIA'04). ICCV'03,
Surface Reconstruction by Integrating 3D and 2D Data of Multiple
Views,
M. Lhuillier and L. Quan. ECCV'02,
Quasi-Dense Reconstruction from Image Sequence,
M. Lhuillier and L. Quan.
An other topic is 3D Reconstruction for Robotic.
This is mainly the work of three PhD students starting from my coming at the
LASMEA in 2002.
Here the camera is calibrated: the relation between image pixels and the
corresponding rays in the camera coordinate system is known.
The scene is assumed to be rigid, although some minor deformable parts are
allowed.
A first problem is the localization of a vehicle using a representation of the
environment.
The representation is a complete 3D reconstruction of the scene (camera motion
parameters and a sparse cloud of scene points in 3D) which is obtained from a
reference video sequence.
More precisely, the first step is the offline reconstruction for a sub-sampling
of the given video into key-frames (calculations on key-frames are more
robust and accurate than calculations on all video frames).
The second step is the inline (real time) localization of the vehicle given a
request frame, thanks to the closest key-frame and its reconstructed 3D
points.
This method is applied to the automatic replay of the reference trajectory by
the vehicle, without human assistance.
The most complete publication on this problem is
IJCV'07/IJRR'07.
A second problem is the real-time 3D reconstruction (or visual SLAM),
given a monocular and calibrated video sequence taken by a vehicle.
Now, both 3D reconstruction (3D point cloud and key-frame parameters) and
localization of the current frame are simultaneously done when the camera is
moving in the scene.
A very reduced set of reconstruction parameters (at the current sequence end)
are optimized for each new frame selected as a key-frame.
This greatly accelerates the results of the previous 3D reconstruction
approach, with a similar quality (accuracy and robustness).
The most complete publication on this problem is
IVC'09.
Also real-time uncertainty estimation for visual SLAM is described in
CVPR'09.
CVPR'09,
Error Propagations for Local Bundle Adjustment,
A. Eudes and M. Lhuillier. IVC'09,
Generic and Real-Time Structure from Motion using Local Bundle Adjustment,
E. Mouragnon, M. Lhuillier, M. Dhome, F. Dekeyser and P. Sayd. IJCV'07
(also in french: TSI'06),
Monocular vision for mobile robot localization and autonomous
navigation,
E. Royer, M. Lhuillier, M. Dhome and J.M. Lavest. BMVC'07
(also in french: RFIA'08),
Generic and Real Time Structure from Motion,
E. Mouragnon, M. Lhuillier, M. Dhome, F. Dekeyser and P. Sayd. CVPR'06,
Real-Time Localization and 3D Reconstruction,
E. Mouragnon, M. Lhuillier, M. Dhome, F. Dekeyser and P. Sayd. CVPR'05
(also in french: RFIA'06),
Localization in urban environments: monocular vision compared to a
differential GPS sensor,
E. Royer, M. Lhuillier, M. Dhome and T. Chateau. BMVC'04,
Towards an alternative GPS sensor in dense urban environment from visual
memory,
E. Royer, M. Lhuillier, M. Dhome and T. Chateau.
The last topic is Omnidirectional Vision started in early 2004.
My goal is to reconsider all topics above for a catadioptric camera.
A such camera is cheap and obtained by mounting a convex mirror in front of a
standard (perspective) camera.
The resulting field of view is very large: 360 degrees in a plane and about
100-110 degrees in the orthogonal direction.
Here, the camera calibration is approximatively known.
A first problem is Structure from Motion: given a still image sequence of a
scene taken by a hand-held catadioptric camera, estimate 3D points of the
scene and camera parameters (poses and calibration parameters) for both
central and non-central camera models.
A central model is such that all back-projected rays intersect a single point
in space (called the centre).
This approximation simplifies all calculations, especially if the mirror
profile is not a conic (e.g. for equiangular or radial uniform resolution
cameras).
The most complete publication on this problem is
CVIU'08.
A second problem is 3D reconstruction for visualization: estimate a textured
surface which approximates the scene and use it for interactive walkthrough
in the scene.
The three steps of the automatic method are
(1) Structure from Motion with the central model
(2) estimation of many local 3D models by (quasi)
dense stereo and
(3) selection in the global 3D model of the parts of local models with the
smallest uncertainties.
Each local model is reconstructed by a few images of the sequence (e.g. 3
consecutive images).
In step 3, a part (a patch) of a local model is retained in the final and
global model if there are no other local models which can reconstructs the
same patch with a smaller uncertainty.
Actually, the most complete publication on this problem is
CVPR'07.
Also there are alternative solutions using a generic camera model, for both
Structure from motion
(ICPR'06,BMVC'07)
and 3D reconstruction for visualization (CVPR'08).
CVPR'08
Toward Automatic 3D Modeling of Scenes using a Generic Camera Model,
M. Lhuillier. CVIU'08,
Automatic Scene Structure and Camera Motion using a Catadioptric System,
M. Lhuillier. CVPR'07
(also in french: RFIA'08),
Toward Flexible 3D Modeling using a Catadioptric Camera,
M. Lhuillier. ICPR'06,
Effective and Generic Structure from Motion using Angular Error,
M. Lhuillier. OMNIVIS'05
(also in french: RFIA'06),
Automatic Structure and Motion using a Catadioptric Camera,
M. Lhuillier.
Shuda Yu is doing a Ph.D thesis (2010-2013) about
"Real-time 3D Scene Modeling".
Alexandre Eudes is doing a Ph.D thesis
(2007-2010) about "Odometry for driving help in urban areas".
Etienne Mouragnon was a Ph.D student (2004-2007) about real-time 3D
reconstruction.
Now, Etienne is working at AIGL Developpement (firm on photogrammetry).
Eric Royer was a master student ("DEA" in french) during 2003,
worked on real time matching and robust geometry estimation on PC hardware
(using the MMX/SSE/SSE2 instructions under Linux).
Also, he was a Ph.D. student (2003-2006) about the localization of vehicles in
urban environments.
Now, Eric is lecturer at University Clermont 1.
IJCV'11,
A Generic Error Model and its Application to Automatic 3D Modeling of Scenes using a Catadioptric Camera,
Postscript,
PDF (see also the
IJCV web site).
M. Lhuillier,
International Journal of Computer Vision, 91(2):175-199
(DOI 10.1007/s11263-010-0374-2).
TOG'09,
Image-based Street-side City Modeling,
PDF,
J. Xiao, T. Fang, P. Zhao, M. Lhuillier and L. Quan,
ACM Transaction on Graphics, 28(5), 2009
(also in proceedings of SIGGRAPH ASIA'09).
IVC'09,
Generic and Real-Time Structure from Motion using Local Bundle Adjustment,
Postscript,
PDF,
E. Mouragnon, M. Lhuillier, M. Dhome, F. Dekeyser and P. Sayd,
Image and Vision Computing, 27:1178-1193, 2009.
CVIU'08,
Automatic Scene Structure and Camera Motion using a Catadioptric System,
Postscript,
PDF,
M. Lhuillier,
Computer Vision and Image Understanding, 109(2):186-203, 2008.
IJCV'07/IJRR'07,
Monocular vision for mobile robot localization and autonomous navigation,
Postscript,
PDF,
E. Royer, M. Lhuillier, M. Dhome and J.M. Lavest,
International Journal of Computer Vision, 74(3),
(special joint issue on vision and robotics,
with the International Journal of Robotics Research).
TS'06,
Localisation par vision monoculaire pour la navigation autonome,
Postscript,
PDF,
E. Royer, M. Lhuillier, M. Dhome and J.M. Lavest,
Traitement du Signal, 23(1), 2006.
TPAMI'05,
A Quasi-Dense Approach to Surface Reconstruction from Uncalibrated
Images,
Postscript,
PDF, (these files include large figures and
appendix)
M. Lhuillier and L. Quan,
IEEE Transactions on Pattern Analysis and Machine Intelligence,
27(3):418-433, 2005. Experimental data are also available for this
paper.
TCSVT'03,
Image-Based Rendering by Joint View Triangulation,
Postscript,
PDF,
M. Lhuillier and L. Quan,
IEEE Transactions on Circuits and Systems for Video Technology,
13(11):1051-1063, 2003.
TPAMI'02,
Match Propagation for Image-Based Modeling and Rendering,
Postscript,
PDF,
M. Lhuillier and L. Quan,
IEEE Transactions on Pattern Analysis and Machine Intelligence,
24(8):1140-1146, 2002.
Conferences
MIRAGE'11,
Surface Reconstruction of Scenes using a Catadioptric Camera,
PDF,
S. Yu and M. Lhuillier,
Lecture Notes in Computer Science, Volume 6930, Springer Verlag, 2011
(also in the conference on Computer Vision/Computer Craphics Collaboration
Techniques and Applications, October 2011).
CVPR'11,
Fusion of GPS and Structure-from-Motion using Constrained Bundle
Adjustments,
Postscript,
PDF,
M. Lhuillier,
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,
Colorado Springs, Colorado, June 2011.
PCV'10,
Environment Modeling from Images taken by a Low Cost Camera,
Postscript,
PDF,
M. Lhuillier,
International Archives of Photogrammetry, Remote Sensing and Spatial
Information Sciences, vol. XXXVIII part 3A, 2010
(also in proceedings of PCV'10).
BMVC'10,
Weighted Local Bundle Adjustment and Application to Odometry and Visual SLAM Fusion,
Postscript,
PDF,
A. Eudes, S. Naudet-Collette, M. Lhuillier and M. Dhome,
Proceedings of the 21th British Machine Vision Conference, Aberystwyth, United
Kingdom, September 2010.
ICPR'10,
Fast Odometry Integration in Local Bundle Adjustment-based Visual SLAM,
Postscript,
PDF,
A. Eudes, M. Lhuillier, S. Naudet-Collette and M. Dhome,
Proceedings of the IAPR International Conference on Pattern Recognition,
Istambul, Turkey, August 2010.
RFIA'10,
Propagations d'erreur pour l'ajustement de faisceaux local,
Postscript,
PDF,
A. Eudes, M. Lhuillier, S. Naudet-Collette,
Proceedings of the Congrès francophone AFRIF-AFIA de reconnaissance des
formes et d'intelligence artificielle, Caen, France, January 2010.
CVPR'09,
Error Propagations for Local Bundle Adjustment,
Postscript,
PDF,
A. Eudes and M. Lhuillier,
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,
Miami, Florida, June 2009.
CVPR'08,
Toward Automatic 3D Modeling of Scenes using a Generic Camera Model,
Postscript,
PDF
(see also the proof appendix Postscript,
PDF),
M. Lhuillier,
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,
Anchorage, Alaska, June 2008.
RFIA'08,
Vers la modélisation 3D flexible avec une caméra
catadioptrique,
Postscript,
PDF,
M. Lhuillier,
Proceedings of the Congrès francophone AFRIF-AFIA de reconnaissance des
formes et d'intelligence artificielle, Amiens, France, January 2008.
RFIA'08,
Reconstruction 3D générique et temps réel,
Postscript,
PDF,
E. Mouragnon, M. Lhuillier, M. Dhome, F. Dekeyser and P. Sayd,
Proceedings of the Congrès francophone AFRIF-AFIA de reconnaissance des
formes et d'intelligence artificielle, Amiens, France, January 2008.
BMVC'07,
Generic and Real Time Structure from Motion,
Postscript,
PDF,
E. Mouragnon, M. Lhuillier, M. Dhome, F. Dekeyser and P. Sayd,
Proceedings of the 18th British Machine Vision Conference, Warwick, United
Kingdom, September 2007.
CVPR'07,
Toward Flexible 3D Modeling using a Catadioptric Camera,
Postscript,
PDF,
M. Lhuillier,
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,
Minneapolis, USA, June 2007.
ICPR'06,
Monocular Vision Based SLAM for Mobile Robots,
Postscript,
PDF,
E. Mouragnon, M. Lhuillier, M. Dhome, F. Dekeyser and P. Sayd,
Proceedings of the IAPR International Conference on Pattern Recognition,
Hong Kong, China, August 2006.
ICPR'06,
Effective and Generic Structure from Motion using Angular Error,
Postscript,
PDF (including corrected appendix),
M. Lhuillier,
Proceedings of the IAPR International Conference on Pattern Recognition,
Hong Kong, China, August 2006.
CVPR'06,
Real-Time Localization and 3D Reconstruction,
Postscript,
PDF,
E. Mouragnon, M. Lhuillier, M. Dhome, F. Dekeyser and P. Sayd,
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,
New-York, USA, June 2006.
ICRA'06,
Uncertainty ellipsoids calculations for complex 3D reconstruction,
Postscript,
PDF,
M. Lhuillier and M. Perriollat,
Proceedings of the IEEE International Conference on Robotics and Automation,
Orlando, USA, May 2006.
ICRA'06,
3D Reconstruction of complex structures with bundle adjustment: an
incremental approach,
Postscript,
PDF,
E. Mouragnon, M. Lhuillier, M. Dhome, F. Dekeyser and P. Sayd,
Proceedings of the IEEE International Conference on Robotics and Automation,
Orlando, USA, May 2006.
RFIA'06,
Reconstruction 3D automatique avec une caméra
catadioptrique,
Postscript,
PDF,
M. Lhuillier,
Proceedings of the Congrès francophone AFRIF-AFIA de reconnaissance des
formes et d'intelligence artificielle, Tours, France, January 2006.
RFIA'06,
Localisation par vision monoculaire pour la navigation autonome d'un robot mobile,
Postscript,
PDF,
E. Royer, M. Lhuillier, M. Dhome and J.M. Lavest,
Proceedings of the Congrès francophone AFRIF-AFIA de reconnaissance des
formes et d'intelligence artificielle, Tours, France, January 2006.
OMNIVIS'05,
Automatic Structure and Motion using a Catadioptric Camera,
Postscript,
PDF,
M. Lhuillier,
Proceedings of the 6th Workshop on Omnidirectional Vision, Camera Networks and
Non-classical cameras, Beijing, China, October 2005.
BENCOS'05,
Performance Evaluation of a Localization System Relying on Monocular
Vision and Natural Landmarks,
Postscript,
PDF,
E. Royer, M. Lhuillier, M. Dhome and J.M. Lavest,
Proceedings of the ISPRS Workshop BenCOS (Towards Benchmarking Automated
Calibration, Orientation and Surface Reconstruction from Images), Beijing,
China, October 2005.
IROS'05,
Outdoor autonomous navigation using monocular vision,
Postscript,
PDF,
E. Royer, J. Bom, M. Dhome, B. Thuilot, M. Lhuillier and F. Marmoiton,
Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and
Systems, Alberta, Canada, August 2005.
CVPR'05,
Localization in urban environments: monocular vision compared to a
differential GPS sensor,
Postscript,
PDF,
E. Royer, M. Lhuillier, M. Dhome and T. Chateau,
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,
San Diego, USA, June 2005.
GIAE'04,
Recent Methods for Reconstructing Surfaces from Multiple Images,
,
G. Zeng, M. Lhuillier and L. Quan,
Lecture Notes in Computer Science, Volume 3519, Springer Verlag, 2005
(revised selected papers of GIAE/IWMM'2004).
BMVC'04,
Towards an alternative GPS sensor in dense urban environment from visual
memory,
Postscript,
PDF,
E. Royer, M. Lhuillier, M. Dhome and T. Chateau,
Proceedings of the 15th British Machine Vision Conference, London, United
Kingdom, September 2004.
ECCV'04,
Surface Reconstruction by Propagating 3D Stereo Data in Multiple 2D
Images,
Postscript,
PDF,
G. Zeng, S. Paris, L. Quan and M. Lhuillier,
Proceedings of the 8th European Conference on Computer Vision, Prague,
Czech Republic, Volume 1, pages 163-174, May 2004.
Also in Lecture Notes in Computer Science 3021, Springer Verlag.
ACCV'04,
Fast Segmentation-based dense stereo from quasi-dense matching,
Postscript,
PDF,
Y. Wei, M. Lhuillier and L. Quan,
Proceedings of the Asian Conference on Computer Vision, Jeju Island,
Korea, January 2004.
RFIA'04,
Reconstruction quasi-dense et modèles 3D à partir d'une
séquence d'images,
Postscript,
PDF,
M. Lhuillier and L. Quan,
Proceedings of the Congrès francophone AFRIF-AFIA de reconnaissance des
formes et d'intelligence artificielle, Toulouse, France, January 2004.
ICCV'03,
Surface Reconstruction by Integrating 3D and 2D Data of Multiple
Views,
Postscript,
PDF,
M. Lhuillier and L. Quan,
Proceedings of the 9th International Conference on Computer Vision, Nice,
France, October 2003.
ICPR'02,
Structure from Motion from Three Affine Views,
Postscript,
PDF,
L. Quan and M. Lhuillier,
Proceedings of the IAPR International Conference on Pattern Recognition,
Quebec, Volume 4, August 2002.
ECCV'02,
Quasi-Dense Reconstruction from Image Sequence,
Postscript,
PDF,
M. Lhuillier and L. Quan,
Proceedings of the 7th European Conference on Computer Vision, Copenhagen,
Denmark, Volume 2, pages 125-139, May 2002.
Also in Lecture Notes in Computer Science 2351, Springer Verlag.
ICCV'01,
Concentric Mosaics, Planar Motions and 1D Cameras,
L. Quan, L. Lu, H. Shum and M. Lhuillier,
Proceedings of the 8th International Conference on Computer Vision, Vancouver,
Canada, 2001.
CVPR'01,
Relief Mosaics by Joint View Triangulation,
Postscript,
PDF,
M. Lhuillier and L. Quan,
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,
Hawaii, USA, December 2001.
ICPR'00,
Robust Dense Matching Using Local and Global Geometric Constraints,
Postscript,
PDF,
M. Lhuillier and L. Quan,
Proceedings of the IAPR International Conference on Pattern Recognition,
Barcelone, Spain, Volume 2, pages 968-972, September 2000.
ICPR'2000 Piero Zamperoni Best Student Paper Award.
CVPR'00,
Edge-Constrained Joint View Triangulation for Image Interpolation,
Postscript,
PDF,
M. Lhuillier and L. Quan,
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,
Hilton Head Island, South Carolina, USA, Volume 2, pages 218-224, June 2000.
RFIA'00,
Appariement dense robuste a l'aide de contraintes geometriques locales et
globales,
Postscript,
PDF,
M. Lhuillier and L. Quan,
Actes du 12eme Congrès Francophone AFRIF-AFIA de Reconnaissance
des Formes et Intelligence Artificielle, Paris, Volume 3, pages 215-223,
February 2000.
CVPR'99,
Image Interpolation by Joint View Triangulation,
Postscript,
PDF,
M. Lhuillier and L. Quan,
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,
Fort Collins, Colorado, USA, Volume 2, pages 139-145, June 1999.
VI'99,
Joint View Triangulation for Two Views,
PDF,
Postscript,
M. Lhuillier,
Proceedings of the 12th Conference on Vision Interface, Trois Rivieres, Quebec,
pages 360-367, May 1999.
Also in INRIA technical report RR-3619 (title:
Towards Automatic Interpolation for Real and Distant Image Pairs ),
February 1999,
Postscript,
PDF.
BMVC'98,
Efficient Dense Matching for Textured Scenes Using Region Growing,
PDF,
Postscript,
PDF',
M. Lhuillier,
Proceedings of the 9th British Machine Vision Conference, Southampton,
United Kingdom, Volume 2, pages 700-709, September 1998.
Also in INRIA technical report RR-3382, Mars 1998,
Postscript,
PDF.
Click on these pictures to obtain some results.
If you encounter problems to play the movies or the 3d models, you may use
ffplay or
vrmlview, respectively.
Surface reconstruction using omnidirectional camera
[ CVIU'08, MIRAGE'11 ]
A top view of the reconstructed/learned scene and the current localized camera
are on the left.
The key frame used for localization and the current frame to be localized are
respectively shown on the top right and bottom right corners.
Thanks to B. Achermann (University of Bern, Switzerland) for providing the
input images.
These faces are ordered in the alphabetical order and automatically morphed.
