Projective invariants have become one of the main issues in the field of computer vision recently. This paper presents a new method of finding correspondences in the sequence of images of arbitrarily shaped 2-D objects using the simple five-point 2-D invariant and its application for the active vision system.
Hand-eye calibration can be done without reference objects by using invariant graph of the object contour. In this application we found a new method of finding 3 FOE's of 3 mutually perpendicular translations. And the relative position of object to camera can be determined by deliberate motion stereo also by using invariants of the corresponding points.
Time-to-contact object and the surface normal of the object can also be determined by using the lines constructed by corresponding points.
Experiments show that the corrspondences found by projective invariants can be succesfully applied to the calibration of the hand-eye orientation and the intrinsic parameters of camera. In simulation, when the perfect correspondence is known, we obtained very accurate translation between the hand and camera. In real experiments, the resulting translation showed a relatively large discrepancy. For the method to be useful in the calibration of hand-camera translation and in the motion vision, more accurate correspondence is needed.
