Date of birth: 18th March 1954 in Nantes, French
2006 Master in business administration MBA (Institut de gestion Rennes)
Dissertation: La maison des droits de l'Homme
2005 Master in religious studies (University of Lancaster UK)
Dissertation: Deconstructing the sacred with the theory of mediation
2001 DEA in human sciences (University of Rennes 2)
Equivalent Master in human sciences. Title: L'homme Internet
2000 Honors in linguistics (University of Rennes 2)
Title: Les droits de l'homme pour quel homme (Human rights for what human)
1999 Honors in French as a foreign language (University of Rennes 2)
Training period in Oxford (UK). A level pack: Brittany and Europe
1997 Diploma in Chinese (University of Rennes 2)
1995 Basic & intermediate level course of Chinese (Northern Jiaotong University, China)
1994 PhD in computer science (University of Rennes 1)
Title: Computer stereovision, geometry and experimentation
1994 Certificate of Astronomy (University of Rennes 1)
1990 Diploma of elementary notions of acupuncture in Esperanto
China Akademio de Sciencoj in Beijing (China)
1987 DEA in computer science (University of Rennes 1)
Training period in ERLANGEN (Germany) on the Parallelisation of algorithms
on the DIRMU multiprocessor.
1984 Licence LEA English & German (University of Rennes 2)
Equivalent B.A. in Applied Languages
1977 Diploma in physiotherapy (Rennes)
1973 Baccalaureat serie D. (equivalent to the A levels)
English (TOEIC:945), German, Esperanto : good conversation,reading &
Spanish, Chinese : intermediate knowledge
2005 - Manager of a teaching center in
computer science (Cyberlog, France)
2000 - 04 Teacher in computer science (Syntheses formations, Rennes, France)
1998 - 99 Teaching French in England (Westminster College, Oxford UK)
1997 - 98 Sabbatical year : trip around the world
(Santiago de Chile - Vancouver - Hong Kong - Rennes)
1994 - 95 Post-doctorate in the Northern Jiaotong University, Beijing (China)
1989 - 94 Assitant lecturer in the High School of Agricultural science in Rennes
(teaching and reseach)
1988 - 89 Humanitarian mission in the Lebanon
(teaching and organization of a rehabilatation center)
1982 - 87 Manager of a private limited company in the electrical goods trade
1978 - 82 Physiotherapist
Reports and publications
Computer stereovision is the calculation of the relief of a scene starting from a couple of stereoscopic images. The two problems of this research topic are the calculation of the points in space and the matching of points between the two images.
The main difficulty for the calculation of the points in space is the presence of three referentials that have to be brought back to only one. The two first are related to the cameras and the third belongs the scene observed.
In the traditional methods a preliminary stage consists in gauging the stereoscopic system, i.e. using a test object to define the relation which binds the referentials of the cameras. Once this relation is known, it is easy to calculate a point in space from its two projections on the images planes by the intersection of two lines in space in the referencial of one of the cameras. The weakness of this method is the great instability of calculations related to the brittleness of this calibration because of the very rigorous mechanical constraints which make the system not very reliable in the event of shock, change of temperature or focusing. In addition, in certain cases, the calibration by test object is impossible, for example in electronic microscopy.
Another type of approach is based on several points of space for the autocalibration the cameras. The first stage consists in calculating the epipolar geometry which induces constraints between the two images that simplify the problem. Algebraic methods calculate, from eight couples of paired points, a fundamental matrix which represents the change of referential between the two cameras. This allows to calculate the points in a projective space and to know their relative positioning. The passage in three-dimensional space related to the scene observed is based on the knowledge of five noncoplanar points in space and on each image.
Our method of the autocalibration type has the same constraints as the algebraic methods. Its originality lies in the geometrical approach. Initially, we calculate the projection of the homologous image on a virtual plane in the space (VIP) made up with the second projection on the image of reference. The virtual plane is defined by three not aligned points. We use their projections as projective bases on each image with their barycentre as unit point. The knowledge of five other couples of paired points makes it possible to calculate the epipolar geometry as well as the homography resulting from the double projection. The superposition of the homologous image recomputed on the reference image provides us information on the relative positioning of the points in space. Then, the calculation of the points in referential of the scene is a simple change of referential using the three-dimensional co-ordinates of five points of the scene. The advantages of this method are a better stability and a satisfactory behaviour for the borderline cases. The self-checkings during the treatment make it possible to validate the pairing of the points. The relative positioning of the points is known without using the three-dimensional co-ordinates of the points taken as reference, which defers the errors of measurement of these points at the end of the calculation. In addition, the transformation of the homologous image constitutes a first stage in the matching of the two stereoscopic images and the self-checkings are used to extract the eight paired points essential to the implementation of this method. One can plan the use of a model of deformation which minimizes the errors of the epipolar geometry, in order to correct the images plans and lead to true pinhole models of camera. Thus it could be possible to obtain an evaluation of their intrinsic parameters.
The matching of points between the two images initially consists in extracting a minimal set of paired points in order to calculate the epipolar geometry and to carry out the double projection passing by the virtual plane. A dynamic technique of programming calculating a way of minimal cost carries out the matching of contours between the two images. This matching is validated by the self-checkings of the VIP method and provides in the majority of the cases a robust set of paired points. Thus we obtain a co-operation between the extraction of the paired points and the double projection of the homologous image on the reference image. This technique can be used for pattern recognition starting from a model of the required object defined by its contours.
The properties of the superposition of the reference image with the recomputed homologous image enable us to obtain a facets model in a projective space from the segmentation of the two images in paired triangles.
This method approximates the shape of the object with a facets model from a reduced whole of paired points and enables the visualization in pseudo 3D. The precision of the shape of the object can then be improved by the subdivision of the facets. When a VIP facet corresponds a plan of the object in the space, the variance between brightnesses of the homologous triangles is low. This enables to detect planes in space, and thanks to simple statistical methods, to correct the variations of brightnesses due to the different points of view of the two cameras. It is thus possible to operate a photometric correction of the images thanks to a lighting model in order to homogenize the stereo pair and to extract new topological information. The analysis of several multispectral images allows a classification of surfaces. Lastly, the problem of the hidden parts due to the change of point of view can be partially solved by detecting the homologous triangles which show very different characteristics.
At the end of the treatment, the minimization of a disparity function of brightness or gradient between the two images perfects the matching.
The VIP Method is a very promising technique for it solves at the same time the strongly dependent problems of matching and autocalibration. This makes it possible to consider the automation of computer stereo-vision. Some applications of this technique can be the three-dimensional visualization of a scene in order to control a robot, industrial control, the calculation of digital models of ground, the teledetection or the measurement of three-dimensional objects in scanning microscopy.
The following stage is the calculation of a three-dimensional model of an object starting from different points of view which describe it completely.
Keywords: stereo-vision, epipolar geometry, relative positioning, matching, facets model, pattern recognition.