In order to realize the rapid measurement for large dimension work-pieces in industrial fields, a new measuring system based on a stereo vision was proposed and implemented. The key algorithms used in the system, such as feature recognition, camera orientation, stereo matching, object coordinate 3D reconstruction and multi-view point cloud registration, were studied. Firstly, an improved ‘CANNY’ edge detecting method was presented to achieve the sub-pixel localization,and the experiential knowledge was used to eliminate miss-recognized reference points.By fitting point centers of the position, the ring of the reference points was sampled in multi-time to obtain a median and to constructed its ID. Then, the homonymous coded points in each photo pair were searched according to their unique IDs,and the relative orientation and absolute orientation were carried out.Secondly, the multi-image epipolar constraint method was utilized to implement the un-coded point matching and to remove the mismatching.Furthermore, the 3D coordinates of all reference points were calculated by the spatial intersection,and the interior and exterior parameters, image and object coordinates are corrected by the bundle adjustment synchronously. Finally, a binocular structure-light scanner was constructed to acquire the surface detail information of the work-pieces. An improved camera calibration approach was introduced and its calibration accuracy was compared with those of traditional methods.The multi-view point cloud global registration algorithm was also discussed in detail. Experiment results indicate that the new measuring system is suitable for the large work-piece measurement with an accuracy of 0.112 mm/3 m,and it can satisfy the precision and efficiency demands of the large dimension measurement.