Study On Measurement Method Based On Computer Vision And Technology Of Surface Reconstruction

As a series of analysis method and application technology in digestion, absorption and improvement of advanced technology, reverse engineering has become a hot research topic in advanced designing and manufacturing field, which has a good application prospect and application value. In this paper, considering existing problems of agricultural machine with special surfaces during design and development, combining the key technologies of reverse engineering—digital measuring, point cloud data processing and surface reconstruction, measurement method based on computer vision and surface reconstruction technology are deeply researched, aiming at providing more efficient method for technology absorption from advanced products or innovative design, and also providing theoretical foundation for new-style farmer machinery design and development. The main research contents are as follows.①This paper elaborates the key technologies of reverse engineering, including digital measurement, the point cloud data processing and surface reconstruction etc., summarizes the domestic and foreign situations, achievements and existing problems, and explains the necessity of studying the key technique of reverse engineering and exploring its application in the agricultural fields. On the basis of recent literature review, the research contents and methods of this paper are outlined.②3D digital measurement is studied. Due to larger deviation and incompleteness of boundary data commonly exiting in general measuring equipments, composite measurement method based on vision and measuring scheme is proposed, which combines both advantages of structured light and binocular stereovision measuring method. In this method, structure light measurement method is used to measure tested object surface, and binocular stereovision method is used to acquire boundary data. The influence of structure parameters of binocular vision system on the measurement accuracy is studied through theoretical analysis and experiments. A composite measurement system is built under the guidance of this research results, and its measurement range is given.③The camera calibration method in binocular stereovision is studied. This paper introduces the camera imaging models, summaries the advantages and disadvantages of typical camera calibration method by contrast and experiments. In consideration of control points image recognition error is bigger in Zhang's calibration method, a new calibration method with relative higher precision in contrast to that of Zhang is proposed, which proved less influenced by noise, and convenient for corresponding point match.④Boundary extraction algorithm in binocular stereovision measurement is studied. This paper contrasts and analyzes image edge detection algorithm commonly used and theirs characteristics through experiment. In order to obtain boundary of the tested object, mathematical morphology is used to fulfill image edge detection and linking, which enhances the border's authenticity and integrality, avoids influence of internal structure on boundary by filling the holes on image. A binocular stereovision matching method is proposed, which combines boundary corner features with regional matching under epipolar constraints, and this method increases the matching speed, meanwhile reduces the probability of mismatch. The 3D data of the tested object boundary contours are obtained by binocular stereovision measuring principle, and the feasibility of the proposed algorithm is verified through examples.⑤Point cloud data processing is studied. This paper studies multi-view point cloud registration method based on the three benchmarks, data consolidation of overlapping area is finished under the man-machine interactive mode, which improves the computational speed, and registration error analysis is carried out through tests. Data filtering smoothing methods are studied, and the advantages and disadvantages of some filtering methods are contrasted and analyzed through tests; on this basis, a filter method based on median and mean is proposed, and the feasibility of this method and effect are verified through specific data point examples. The simplification methods of point cloud data are studied, the curvature calculation based on color is used to realize point cloud simplification, and the effect of non-uniform curtailing is achieved where in big curvature data is intensive, sparse in small curvature.⑥The theory and methods of surface reconstruction based on point cloud are studied. The method of direct fitting for surface piece based on point cloud is discussed thoroughly, modeling method based on curve fitting, surface model quality evaluation method are also studied, which provide the direct-viewing manifestation of the fairness of a curve and surface. For the specific examples, surface reconstruction and model quality evaluation are realized in Imageware software.⑦Plough surface and lean-corrugation disc tool that used to break stubble in no-tillage seeding are selected as actual samples to evaluate accuracy and feasibility of the measuring system. The overall process from plough surface scanning to point cloud data and then surface reconstruction is performed with the measurement system set up in this research, moreover,, measuring accuracy and feasibility of the system as well as related methods are experimentally proved practical. With regard to lean-corrugation disc tool, the structural parameters identification is achieved by fitting approximating method and curvature calculation etc., thus, parameterized model be completed in the light of reverse way.