Section: New Results

Motion, scene and camera reconstruction

Participants : Marie-Odile Berger, Srikrishna Bhat, Christel Leonet, Gilles Simon, Frédéric Sur.

• Enhancing the grid method for in-plane strain measurements

  This work is motivated by a problem from experimental solid mechanics. The grid method permits to measure the heterogeneous strains on the surface of specimens subjected to mechanical tests. Among full-field measurement techniques, the grid method consists in transferring a regular grid on the surface of the specimen and in taking images of the grid before and after deformation. Windowed Fourier analysis then gives an estimate of the surface displacement and strain components. In a collaboration with Institut Pascal (Université Blaise Pascal, Clermont Ferrand), we have shown that the estimations obtained by this technique are approximately the convolution of the actual values with the analysis window. We have also characterized how the noise in the grid image impairs the displacement and strain maps [18] . This study has allowed us to improve the metrological performance of the grid method with deconvolution algorithms. A numerical and experimental study can be found in [17] .

• Visual words for pose computation

  Visual vocabularies are standard tools in the object/image classification literature, and are emerging as a new tool for building point correspondences for pose estimation. Within S. Bhat's PhD thesis, we have proposed several methods for visual word construction dedicated to point matching, with structure from motion and pose estimation applications in view. The three dimensional geometry of a scene is first extracted with bundle adjustment techniques based on keypoint correspondences. These correspondences are obtained by grouping the set of all SIFT descriptors from the training images into visual words using transitive closure (TC) techniques. We obtain a more accurate 3D geometry than with classical image-to-image point matching. In a second on-line step, these visual words serve as 3D point descriptors that are robust to viewpoint change, and are used for building 2D-3D correspondences on-line during application, yielding the pose of the camera by solving the PnP problem. Several visual word formation techniques have been compared with respect to robustness to viewpoint change between the learning and the test images. Our experiments showed that the adaptive TC visual words are better in many ways when compared to other classical techniques such as K-means.

  More specifically, the work of this year has focused on improving pose estimation from visual words with respect to strong viewpoint changes. 2D-3D correspondences are actually difficult to establish if there are too large viewpoint changes between the image whose pose is sought and the images that yielded the visual words attached to 3D points. We assessed several viewpoint simulation techniques in order to enrich the visual word description of the 3D points.

• Acquisition of 3D calibrated data

  Christel Leonet joined the team in October 2010 as an Inria assistant engineer with the aim of building an integrated 3D acquisition system. More specifically, the objective of her work is to combine an IMU (Inertial Measurement Unit), a GPS receiver, a laser rangefinder and a video camera for ground truth data acquisitions of camera movements and scene structures. These data will be useful to validate several algorithms developed in our team. This year, a new visual pan tracking method has been designed and implemented. We considered spherical environments made of sparse video images instead of fully-covered environment maps which often suffer from geometric and photometric misalignments. The scanning process has been improved in order to increase the accuracy of the recovered polygons and allow for visual assessments of this accuracy. The 3D laser pointer has been validated in several indoor environments. Finally, the GPS has been integrated to the system and preliminary results have been obtained in outdoor environments.