Research On Point Clouds Preprocessing And Surface Reconstruciton With Fringe Projection

Point clouds preprocessing and surface reconstruction with fringe projection take fringe scanning as measuring ways, entities, samples as research objects. The main content of the research includes 3D data detecting, point clouds preprocessing, surface reconstruction and rapid manufacture. It is one of the most important techniques to realize rapid products development and manufacture. It is of wide application prospect in remodeling and creation design of automobiles, motors, airplanes, home appliances, molds, crafts, clothing, medical fields and so on.The relevant theories of phase measurement with fringe projection are studied. Phase demodulation principle of fringe projection measurement is explained, and the relation between object height and fringe phase is obtained. The phase demodulation effect of 1D FTP and 2D FTP methods is compared, and it shows that 2D FTP can precisely reflect the change along arbitrary directions. Phase demodulation principle with phase shifting profilometry is thoroughly analyzed, and phase solving method adopting 4-step phase shifting method is explained. Solving formula of deformation fringe is established. Wrapping phase and its effect on 3D surface measurement is analyzed, and minimum differential algorithm is adopted to implement phase unwrapping to realize real phase continuity.The registration and simplification of point clouds are studied in the paper. In the manual registration of the point clouds, inheriting and optimizing algorithm is proposed to realize precise point clouds registration. Rotating matrix and translation vector can be obtained after 25 iterations, while the registration error is 0.10mm. Compared to ICP algorithm, it has advantages of higher registration precision and less iteration times. In the automatic registration of point clouds, the sum of squares of Euclidean distance between surfaces is reduced by Gauss-Markoff model, and overlapping area is matched by the least squares algorithm to realize automatic point clouds registration without mark points. In the simplification of point clouds, various mesh sizes are chosen corresponding to object profile and point clouds curvature, and point clouds simplification is implemented by medium value filtering method to reduce redundant points. Not only can it inherit quick and effective merits of uniform mesh method, but also it can retain detail data of the object shape.Triangle division, quadrilateral division and NURBS reconstruction are studied in the paper. Triangle growth method based on LABN (least angle between normal) is adopted to implement the direct division of the scattered point clouds. Firstly, a threshold range is given to define points set of the extension side. Then the third point by counterclockwise triangle alignment is found out. Finally triangle meshes are optimized by the local optimization rules. Quadrilateral deformation factor is calculated out according to the quality of triangles to implement maximum weigh matching. Two adjacent triangles are merged to implement quadrilateral division. NURBS surface is obtained through triangle division, curvature detecting, contour lines construction, quadrilateral division, UV parameter lines construction and surface fitting.Surface-point clouds error, surface smoothness and their influence factors are studied, and the methods to improve surface quality are proposed. The number of control points has great effect on reconstruction precision. For the point clouds including about 1 million points, if the number of quadrilateral meshes is 100, it is suitable to choose 60 to 80 control points. The effect of surface order, span and smoothness on the error of surface-point clouds is analyzed. If the point clouds include 6082 points, when the surface order is 4×4, span is 5×5, smoothness is 0.7000, the maximum error of surface-point clouds is 0.0408mm. If the surface-point clouds error meets the requirement, the smoothness property of the surface can be improved by reducing redundant control points and adjusting the position of the control points.3D data detecting, surface reconstruction and rapid prototyping manufacture of the face are studied. Firstly, sinusoidal fringe is projected to the face by double fringe scanner, and fringe image is modulated by the face profile. Two separate point clouds of the face are obtained by the conversion of the deformation fringe. After the automatic registration of the point clouds, triangle surface of the face is obtained by triangle division and surface fitting. Irregular triangles around holes boundary of the face are eliminated, and holes are filled, relaxed and smoothed to form closed and smooth surface. Finally NURBS face surface reconstruction is implemented by quadrilateral meshes division and surface fitting. STL file obtained from reverse process is input to rapid prototyping equipment to implement the manufacturing of the face model to realize the entire procedure of reverse engineering.