Active visual sensing of the 3-D pose of a flexible object

This paper presents an active visual method for determining the 3-D pose of a flexible object with a hand-eye system. Some simple and effective on-line algorithms to overcome various exceptional situations in chaincoding or determine the object pose more precisely are developed. The pose of a flexible object can be easily changed because of the flexible nature and the prediction of the pose is almost impossible. A new sensing pose is computed by using the image coordinates of the points on the border line of an image window and the current pose of a hand-eye system for the cases that the flexible object is extended outside the window. In a case of exceptional overlapping, a new sensing pose is computed by using the image coordinates of four extreme image points and the current pose of the hand-eye system. Through a chaincoding process on the skeletonized images, the stereo matching problem of two images is transformed into the matching of the curvature representations of the two skeletonized images. The 3-D pose of a flexible object is computed by using the results of this matching and the camera and hand-eye parameters calibrated beforehand. The initial sensing results are used in computing a new sensing pose to determine the object more precisely.