How to apply kalman filter on rotation matrix?

Can someone translate this paragraph into french without using an online diccionary?

• Hi, I want to traslate this two paragraph into french ( je voudrais traduiser ces 2 paragraphe vers le franaçais ) et parce que je ne suis pas forte en angalis je vous demanderais votre aide : The global head motion can be estimated by model-based 3D face tracking algorithms. Some of them used simple geometric head models such as a cylinder, , an ellipsoid (Basu et al. 1996), or a head-like 3D shape (Malciu and Preteux 2000). They recover the global motion by minimizing the difference of texture or optical flow between observation and their model. Vacchetti et al. (2004) used multiple key frames and feature point matching to estimate the motion of their model under large pose variation. These approaches assume that the 3D shape of the object does not change during tracking, which means that they do not have shape parameters. On the other hand, some researchers tried to track the deforming shape and global motion at the same time. Strom et al. (1999) used feature point tracking result with structure from motion in Kalman filter framework. DeCarlo and Metaxas (1996) used a deforming face model whose fitting algorithm integrated optical flow and edge information. However, these algorithms are not adequate for tracking of large head pose change and the quality of the estimated 3D shapes cannot be guaranteed. The global head motion can be represented by a rigid motion, which can be parameterized by 6 parameters; three for 3D rotation and three for 3D translation. Therefore, the number of all the model parameters are only 6. The low dimensionality of the parameter space results in robust tracking performance when compared to the high dimensionality of the AAMs. In addition, these methods do not require any learning stage, which means that they are person independent. Moreover, they are robust to a large pose change because they use the whole area of the head in the image instead of a specific part. However, the rigid head model cannot provide detailed information such as the local movement of the facial features; opening mouth, closing eyes, raising eye brows, and so on. Among three different geometric head models, we take the cylinder head model because it is the generally applicable and simplest method. The cylinder model is more appropriate for approximating the 3D shape of the generic faces than the ellipsoid model. Also, the head like 3D shape model requires a large number of parameters and its fitting performance is very sensitive to their initialization **************************************… Please help me ( pourriez vous m'aider SVP)

• Answer:

  Le mouvement global de la tête peut être estimé par les algorythmes du suivi des points de la face basés sur un modèle 3D. Certains d'entre eux utilisent de simples modèles de tête géométriques comme un cylindre, un ellipsoïde ou une forme de tête en 3D http://it.answers.yahoo.com/question/index;_ylt=An_m9WWj04BMKPjmxXcL.4grAgx.;_ylv=3?qid=20100523021456AAhptg0