Research On The Issues In Three Dimensional Reconstruction Of Mass Data Points

With the three-dimensional measurement technology's rapid development and wide application, the ability of spatial data access, accuracy and speed has been an unprecedented increase, which provides a solid foundation to achieve three-dimensional objects. Three-dimensional reconstruction is to obtain the actual existence of the object that its shape in the computer technology. The three-dimensional shape is the basis of the nature object's processing, operation and analysis, and is the key of the reverse engineering and interactive visualization. In the CAD/CAM, engineering design, scientific visualization, digital city and the museum, it has wide application areas, and is also an important basic theory and hot spots at present.To achieve high-precision three-dimensional reconstruction of objects in many practical applications, it needs some key processing such as data acquisition, noise removal, registration, holes repair and data compression. This paper will study the four key issues, and provide a new theoretical and simple, efficient algorithm for high-precision three-dimensional reconstruction. Specifically, ours study contents are the following four points:(1) Sampling and data denoisingThe first step of the three-dimensional reconstruction is data acquisition and noise reduction, which subsequently decided that all behind algorithms'accuracy. Because of the factors caused by the surface of the measured object (such as surface roughness, the ripple and other defects) and by the measurement system itself (such as measuring equipment accuracy, resolution, vibration, etc.), Point cloud data measured by laser scanner will inevitably be some noise. Results show that the measured point cloud data has 0.1% to 5% noise points to be processed. To extract the features more accurately and get a better surface model, it is necessary denoising the data points. (2) High-precision alignmentData alignment and surface reconstruction is the key issues and difficult problems of three-dimensional representation. The accuracy of data alignment determines the quality of surface reconstruction, especially for the reconstruction of the surface of fine materials (such as precision mechanical components and sophisticated national heritage, etc).It includes many work:isolated point division, the initial position, the choice of matching points, alignment correction and alignment of multiple views at the same time and so on. In essence, the data alignment is an optimization problem; the goal is to find suitable Euclidean transformation parameters to a set of data with another set of data matching. The main disadvantage of existing methods rely on the experience too much inference, need more accurate initial conditions, and lack of optimization of the algorithm is based on rigorous analysis of the theory, such as the convergence to be ignored. Although the convergence speed is very slow, Besl and McKay's ICP (iterative closest point) algorithm is still widely used. There is urgent need to give stable fast and effective algorithm.(3) Hole repairingDuring the data measurement, the measurement principle, or human factors will lead to lack of measurement data, and make the surface measurement data obtained have some "holes". The holes'repairing can greatly reduce the surface distortion. In addition, in reverse engineering if repairing worn or partial damage for some precious part or mold, we can greatly reduce production costs; thereby repair of model has its important significance.(4) Data compressionThe data obtained by three-dimensional scanning is usually a very large amount, while 3D graphics technology and the Internet combine to produce more and more interesting applications. But the network bandwidth limitations severely hinder the spread of this medium. Only by increasing investment in hardware is not enough to solve this problem, we need to study three-dimensional geometry data compression coding technology. Effective compression technology has very important significance to reduce the amount of three-dimensional geometric data storage and reduce its bandwidth required for Internet transmission.The specific work and major contributions of this paper are the following:(1) Point cloud denoising based on swarm intelligenceWe present a smoothing method for point cloud based on ant colony algorithm, and it can remove the noise efficiently and preserve the sharp features of the models. We propose an algorithm of kernel function clustering based on ant colony algorithm to analyze the point cloud data., and then adjust the direction of normal. Also taken kernel function method, we measure the similarity of curvature and normal in high-dimensional feature space. For normal adjustment, we take the method of interclass variance to get the threshold.(2) Registration based on local feature surfaceA new registration method for point clouds is presented. Building an Octree for point cloud, the matching of public areas is get according to the Octree comparison to determine. Gauss map is taken for the data of public regional, and a preliminary registration is making in the unit Gauss sphere. The objective function is given in the Gauss sphere, using geometric hash method to match in the four-dimensional vector space consist of spin axis and the rotation angle. Algorithm extracts feature point and constructs local characteristics surface, and gets an accurate match through point-to-surface distance to compute the objective function.(3) Repair based on fractalA hole repair algorithm based on fractal technique is proposed, which make full use of the data self-similarity. First, the possible boundary points are judged through the "angle". Then the boundary loop holes is extracted by continuity of the boundary points. For larger holes, some discrete point is inserted. Fractal interpolation surface is constructed in the triangle regional, and new data points are generated by the fractal interpolation for selected feature points.(4) Progressive compression based on SurfaceletA progressive mesh compression method based on surfacelet transform is proposed. It can take full advantage of the surfacelet transformation that effectively capture and express high-dimensional singular surface characteristics of the signal, and can better code and reconstruct the characteristics. Surfacelet transformation can carry out the direction filtering, which can effectively solve the problem that wavelet transform can not locate the edge and contour position for high-frequency signals.At present, the three-dimensional reconstruction of the object can not satisfy the needs of practical application, which become the bottleneck in reverse engineering, human-computer interaction design and visualization technology. It urgently needs high accuracy digital technology in CAD/CAM, engineering design, scientific visualization, and digital cities.