Study On The Key Technologies Of3D Optical Measurement For Large-Scale Complex Surface Based On Computer Vision

With the rapid development of manufacturing industries, many products of trade are large-scale objects with curved surface. The high-speed, high-precision and convenient3-D measuring methods and equipment for the large-scale mechanical systems is a pressing need in industry process and quality control. In computer and reverse engineering and technology has developed rapidly in the background, by elevating efficiency and performance of the computer vision and three-dimensional optical measurements to get3D cloud points of a large scale complex surface in a fast, efficient way is of great reaching significance and practical value.This paper aims at the automatic high-accuracy3D shape acquisition of large-scale objects. The two measuring systems of overall feature points and curved surface have been established systematically, which have achieved the integration of close-range photogrammetry, structured light projection and stereo vision to get the3-D data of overall feature points, curved surfaces and registration of different curved surfaces according to the overall feature points. The research content include modeling and self-calibration of camera, extraction and3D realization of overall feature points, caption of image based on digital projection, phase shift, unwrapping phase and3D reconstruction of dense points of curved surface. At the same time, the direct measurement of some specific optical surfaces including objects with high dynamic range of surface reflectivity and large forgings in thermal state has been studied. The combined method of theoretical analysis and experimental research is adopted, and the research of key technologies of3D optical measurement for large-scale surface has been carried out systematically in the paper.The linear and non-linear models of camera are described at first in the paper. Several classical self-calibration methods of camera are analyzed. To solve this problem of big error of layered self-calibration method, the factorization based on multiple views is introduced to recover3D projective reconstruction. The accuracy and robustness of self-calibration of cameras are markedly increased since all the images are aligned in the factorization process.The principle of acquisition of3D feature points in the world coordinate is described, and the error propagation rules in3D reconstruction process have been deduced on the base of the traditional dual-view reconstruction method. We proposed a method that can control the accumulative errors by reducing the times of coordinate transformation with common view based dual-view reconstruction. This method involves constructing an image network composed of many image groups each of which contain a common view to orientate camera and reconstruct3D points. A baseline threshold method is introduced to remove the negative impact on the reconstruction precision due to the weak nature of the image combination. The method of the sums of or re-projection residual of all the common points is proposed to assess the validity of the solutions of the five-point algorithm of essential matrix. Experiments carried out with real images reveal that the proposed method can handle the accumulative error problem with robust and high accurate results especially for a long image sequence when a large-scale object is measured.The3D measuring method of curved surface has been systematically studied. This paper focuses on the temporal phase unwrapping algorithm to get point clouds of complex surfaces with reliable results when complex and unconnected surface involve.The measurement of objects with high dynamic range of surface reflectivity, especially specular surface and high contrast surface without spray painting has also been researched deeply. This method involves producing measurable high-quality points by integrating such techniques as multiple exposures, high dynamic range imaging and combination of camera-projector monocular and binocular stereo vision to match the measuring range of intensity of camera with the range of reflection intensity. Our experiments demonstrate that these techniques can obtain not only the high-accuracy and robust result of the binocular stereo vision, but also the complete data of the whole projector-illuminated objects when the monocular systems operated from different viewing angles fill in missing data of the binocular reconstruction.Some problems existed in online fast measurement of large forgings in thermal state have been studied fully based on structured light stereo vision. To solve the problem of the acquirement of structured light strip on the condition of thermal state, the method combined physical filtering and digital filtering was adopted to overcome the influence of red thermal radiation, and temporal phase unwrapping structured light technology was used to improve the optical noise immunity significantly. Multi-view auto registration techniques based on mark balls and the calibration of multiple structured light measuring head were put forward to measure large forging objects. The developed system has advantages of high speed, high precision, simplicity and reliablility, which can meet the requirements of online measurement of large forging in thermal state.At last, some large-scale objects whose surfaces have smooth properties and no smooth properties have also been tested with our developed system. The precision estimation of measurement has also been performed. The online measurement of large forging in thermal state has also been verified at industry application.