Visual position tracking using dual quaternions with hand-eye motion constraints

摘要

In this paper a method for contour-based rigid body tracking with simultaneous camera calibration is developed. The method works for a single eye-in-hand camera with unknown hand-eye transformation, viewing a stationary object with unknown position. The method uses dual quaternions to express the relationship between the camera and end-effector screws. It is shown how using the measured motion of the robot end-effector can improve the accuracy of the estimation, even if the relative position and orientation between sensor and actuator is completely unknown. The method is evaluated in simulations on images from a real-time 3D rendering system. The system is shown to be able to track the pose of rigid objects and changes in intrinsic camera parameters, using only rough initial values for the parameters. The method is finally validated in an experiment using real images from a camera mounted on an industrial robot.