Abstract: A 3D data registration method was proposed for vision measurement using mark points as global measurement control points. Mark points were attached to the measured objects and one or more baseline rulers were put in the measured area. The images of mark points and the baseline rulers were captured by a digital camera from different viewpoints, by which the coordinate frame transformations between viewpoints were estimated. Using the stereo vision model, the local coordinates of the mark points were determined, and then converted at the global coordinate frame, forming the global measurement control points. Scanning different parts of the object at different positions, the vision sensor obtained the coordinates of some mark points in the local coordinate frame of the sensor. The transformation from the local to the global coordinate frame was computed using at least three non-collinear mark points, thus the 3D data acquired by the sensor at different positions were integrated into the global coordinate frame. The proposed method is convenient and applicable in field measurements. Experimental results show that the registration error in x, y, z axis are 0.03mm, 0.02mm, 0.07mm respectively.