Section: New Results

Visual tracking

3D model-based tracking

Participants : Antoine Petit, Eric Marchand.

This study focused on the issue of estimating the complete 3D pose of the camera with respect to a potentially textureless object, through model-based tracking. We proposed to robustly combine complementary geometrical and color edge-based features in the minimization process, and to integrate a multiple-hypotheses framework in the geometrical edge-based registration phase [53][52][68][11] .

Pose estimation through multi-planes tracking

Participants : Bertrand Delabarre, Eric Marchand.

This study dealt with dense visual tracking robust towards scene perturbations using 3D information to provide a space-time coherency. The proposed method is based on a piecewise-planar scenes visual tracking algorithm which aims at minimizing an error between an observed image and reference templates by estimating the parameters of a rigid 3D transformation taking into acount the relative positions of the planes in the scene. Both the sum of conditional variance and mutual information have been considered[40]  [67] .

Pose estimation from spherical moments

Participant : François Chaumette.

This study has been realized in collaboration with Omar Tahri from ISR in Coimbra (Portugal) and Youcef Mezouar from Institut Pascal in Clermont-Ferrand. It was devoted to the classical PnP (Perspective-from-N-Points) problem whose goal is to estimate the pose between a camera and a set of known points from the image measurement of these points. We have developed a new method based on invariant properties of the spherical projection model, allowing us to decouple the pose estimation in two steps: the first one provides the translation by minimizing a criterium using an iterative Newton-like method, the second one directly provides the rotation by solving a Procrustes problem [65][26] .

Structure from motion

Participants : Riccardo Spica, Paolo Robuffo Giordano, François Chaumette.

Structure from motion (SfM) is a classical and well-studied problem in computer and robot vision, and many solutions have been proposed to treat it as a recursive filtering/estimation task. However, the issue of actively optimizing the transient response of the SfM estimation error has not received a comparable attention. In the work [64] , we studied the problem of designing an online active SfM scheme characterized by an error transient response equivalent to that of a reference linear second-order system with desired poles. Indeed, in a nonlinear context, the observability properties of the states under consideration are not (in general) time-invariant but may depend on the current state and on the current inputs applied to the system. It is then possible to simultaneously act on the estimation gains and system inputs (i.e., the camera velocity for SfM) in order to optimize the observation process and impose a desired transient response to the estimation error. The theory developed in [64] has a general validity and can be applied to many different contexts: in [64] it is shown how to tailor the proposed machinery to two concrete SfM problems involving structure estimation for point features and for planar regions from measured image moments.

3D reconstruction of transparent objects

Participant : Patrick Rives.

This work has been realized in collaboration with Nicolas Alt, Ph.D. student at the “Technische Universität München” (TUM).

Visual geometry reconstruction of unstructured domestic or industrial scenes is an important problem for applications in virtual reality, 3D video or robotics. With the advent of Kinect sensor, accurate and fast methods for 3D reconstruction have been proposed. However, transparent objects cannot be reconstructed with methods that assume a consistent appearance of the observed 3D structure for different viewpoints. We proposed an algorithm that searches the depth map acquired by a depth camera for inconsistency effects caused by transparent objects. Consistent scene parts are filtered out. The result of our method hence complements existing approaches for 3D reconstruction of Lambertian objects [30] .

Pseudo-semantic segmentation

Participants : Rafik Sekkal, Marie Babel.

This study has been realized in collaboration with Ferran Marques from Image Processing Group of the Technical University of Catalonia (Barcelona). We designed a video segmentation framework based on contour projections. This 2D+t technique provides a joint hierarchical and multiresolution solution. Results obtained on state-of-the-art benchmarks have demonstrated the ability of our framework to insure the spatio-temporal consistency of the regions along the sequence.

Augmented reality

Participants : Pierre Martin, Eric Marchand.

Using Simultaneous Localization And Mapping (SLAM) methods becomes more and more common in Augmented Reality (AR). To achieve real-time requirement and to cope with scale factor and the lack of absolute positioning issue, we proposed to decouple the localization and the mapping step. This approach has been validated on an Android Smartphone through a collaboration with Orange Labs [46] .

Dealing with AR, we have proposed a method named Depth-Assisted Rectification of Patches (DARP), which exploits depth information available in RGB-D consumer devices to improve keypoint matching of perspectively distorted images [44] .