Section: New Results
We propose a new method for estimating the physical properties of complex objects from an image sequence. Actually, we want to approximate the mechanical properties of a deformable object directly from a real video. We have been working on a new algorithm being able to recover the parameters of a deformable solid simulated by a finite element method, directly from a real image sequence: a user applies a force on a deformable solid using a special device that we created for this purpose (see left picture of figure 6 ). This device has been achieved past year, and we have been realizing many experiments to validate it. Indeed, it appears that because of friction issues, we had to enhance some details for ensuring a valid estimation of the external forces captured by this device. At the same time, we have also developed a complete environment that includes real-time acquisition of video sequences, automatic calibration of cameras and rerendering of synthetic sequences.
For estimating the physical properties of a soft bodies, we capture the whole scene using a camera, and we project a specific pattern on the real object using a regular video projector. The whole scene is reconstructed in computer graphics and the physical parameters (the Young modulus and the Poisson coefficient) are now estimated from the real image sequence using an iterative technique minimizing the error between the real and the synthetic videos (see figure 7 ).
We use Christian Duriez's FEM simulator to reproduce the soft body in computer gaphics. Our main algorithm has been designed for automatically recovering the Young modulus and the Poisson Coefficient only using our device and a video sequence of the deformable body. Therefore we have achieved a full study of these two parameters and their impact on the generated image sequences. It appears that the Poisson coefficient does not vary very much for very different silicons even if their hardness is extremely different. That is why we have decided to extend our technique to other materials such as cleaning sponges for example. The error metric used in the comparison between real and synthetic images seems to work well but we definitely need to further validate it. Once all these techniques will have been validated on simple deformable solids, we will work on recovering the parameters of more complex objects such as real human organ.