Contributions à l'intégration vision robotique : calibrage, localisation et asservissement

摘要

The integration of computer vision with robot control is the concern of this thesis. This integration has many advantages for the interaction of a robotic system with its environment. First, we are interested in the study of calibration methods. Two topics are treated : <br /> i) hand/eye calibration and <br /> ii) object pose. <br /> For the first, we developed a nonlinear method that seems to be very robust with respect to measurement errors; for the second, we developed an iterative para-perspective pose computation method that can be used in real-time applications. <br /> Next we are interested in visual servo control and extend the well known method "image-based servoing" for a camera that is not attached to the robot being serv(o)(e)d. When performing relative positioning, we show that the computation of the goal features do not depend on an explicit estimate of the camera intrinsic or extrinsic parameters. For a given task, the robot motions are computed in order to reduce a 2D error to zero. The central issue of any image-based servoing method is the estimation of the image Jacobian. We show the advantage of using an exact image Jacobian with respect to the dynamic behaviour of the servoing process. This control method is used in automatic object grasping with a 6 DOF robot manipulator. All the methods presented in this thesis are validated with real and simulated data.