Contact
Email : Antoine.Guillaume.Petit@irisa.fr
Address : Irisa / Inria Rennes
Campus de Beaulieu
35042 Rennes cedex - France
Tel : +33 2 99 84 74 21
Fax : +33 2 99 84 71 71
Assistant : +33 2 99 84 22 52
(Céline Gharsalli)
I graduated from Supélec in 2010 as an engineer, with majors in Control and Signal Processing, and from Centrale Paris for a Research Master in Aerospace Engineering. Before joining the Lagadic team as a PhD student, I was intern in the team for 5 months, working on vision based tracking, with an application regarding Orbital Rendez-vous between spacecrafts, in partnership with EADS Astrium. I began my PhD in december 2010, under the responsability of Eric Marchand, within a project with EADS Astrium. I study improvements in model-based tracking algorithms.
Determining the complete 3D pose of the camera with respect to a target object is a key requirement in many robotic applications involving 3D objects. Based on the knowledge of the complete 3D model of a target with a complex shape, methods are proposed to adress the challenging tasks of detecting and localizing this target and then tracking it frame-by-frame.
The active removal of heavy space debris (typically larger than 1000kg) has been identified as a key development to control the growth in the debris population and to limit the risk for active satellites. In that context, Astrium has been working on optimization and implementation of sensors and navigation solutions onboard a Debris Removal Vehicle with the main objective to ensure high safety proximity maneuvers. In particular, special attention has been paid to the design of autonomous, vision-based navigation solutions for uncooperative rendezvous with space debris. For these reasons, our approaches, which allow fine estimation of chaser states with respect to a target spacecraft or debris, are particularly suited for these applications.
Efficient edge-registration using graphics process unitsSome classical methods achieve the tracking by relying on the alignment of projected lines of the 3D model with edges detected in the image. However, processing complete 3D models of complex objects, of any shape, presents several limitations, and is not always suitable for real-time applications. We propose an approach to avoid these shortcomings. It takes advantage of GPU acceleration and 3D rendering. From the rendered model, visible edges are extracted, from both depth and texture discontinuities. Correspondences with image edges are found thanks to a 1D search along the edge normals. Our approach addresses the pose estimation task as the full scale nonlinear minimization of a distance to a line Multiple-hypotheses frameworkIn order to improve the robustness of the pose estimation and to avoid problems due to ambiguities between edges, it is possible to consider and register different hypotheses corresponding to potential edges. They correspond to different local extrema of the gradient along the scan line. We choose the hypothesis which has the closest distance to the projected model 3D line during the minimization process. We also propose to take advantage of line primitives, by clustering the model edge points into lines and to register different hypotheses consistently with these primitives. |
|
The general idea is to combine in the global criterion to be minimized two complementary cues: the classical geometrical one, relying on distances between model and image edges, and a color-based one, relying on color features computed along projected silhouette edges of the model. For more robustness, we employ M-estimators for both objective functions, and we add temporal consistency for the color-based features. |
|
Application in Augmented Reality This method has been applied for Augmented Reality, with a complete 3D model of the object, reconstructed using a Kinect sensor. |
|
In order to servo a 6 DoF robotic arm and to simulate a space rendezvous with a mock-up of a telecommunication satellite, we propose to use 3D model based tracking algorithm within a 2 1/2 D visual servoing control loop. The maneuver has been divided into three phases: a first translation to drive the target into the center of the image, a fly-around phase to align to the docking port axis of the target, and a final translation until almost docking. |
Complete list (with postscript or pdf files if available)
|
| Lagadic
| Map
| Team
| Publications
| Demonstrations
|
Irisa - Inria - Copyright 2009 © Lagadic Project |
