Research Point Cloud Hole Repair

With the rapid increase of the measurement accuracy of the3D scanning devices and the rapid development of scanning technology, through the3D scanner can easily get high precision and density the point cloud model, which makes point cloud become a new type of digital media data. However, due to the complex shape of the actual surface, or duo to the limitations of the methods of measurement and scanning equipment, the data acquired by the scanning device may have holes. Only a complete point cloud model can be to surface reconstruction, therefore, hole filling technology is essential pre-processing work of the surface reconstruction, and also is the key technology whether the virtual reality technology can achieve better effect or not. To study the existing hole filling algorithm, we presents a better robustness hole filling algorithm, the main contents include detecting the holes boundary point, extracting the holes boundary edge and filling the holes point with partial differential equations (PDE). The major contributions of this paper are:1. The methods for computing geometric attributes are proposed. Point cloud model using Kd-tree method to find the k-neighbor points, which building covariance matrix. Computing the point cloud the normal vector by using principal component analysis (PCA) and adjusting the normal orientation by using propagation algorithm. Besides, we use moving the least squares method to calculate the local surface fitting polynomial, which calculated the mean curvature, Gaussian curvature, maximum principal curvature and minimum principal curvatures of each point.2. We present an algorithm for extracting the curves based on investigating properties of point clouds and deriving criteria for automatic holes-detection. Our algorithm first collects k-neighboring points for each point in a given point-based model and then evaluates the normal of point-set surface at point. For each point, the algorithm combines several criteria, i.e., angle criterion, half-disc criterion and shape criterion respectively, into a weighted boundary probability. Finally, our algorithm constructs boundary curves of holes by searching the nearest point strategy and connects the boundary points to a poly-line to generate the boundary curve. Experiments show that our algorithm can run stably and robustly, and can extract various holes boundary curve from point cloud models. 3. Proposed a method which based on partial differential equations (PDE) to fill holes in point cloud mode. Our algorithm first pre-processing the extracted holes polygon to make the polygon more regular, then, equidistant parameterization of the new holes polygons in order to generate three new boundary curves, and constructed fourth-order partial differential equations based on these boundary curves. Finally, filling holes with the solution of partial differential equations in point cloud model. The experiments show that the algorithm can quickly fill holes, fill holes point can be smooth, natural blend with the original model.Besides, a proto system based on these algorithms is implemented in this thesis, and several typical hole-filling examples are given as well. The results of these examples indicate the correctness of the hole-filling method.