Outlier Detection And Model Reconstruction Of 3D Point Cloud Data

With the rapid development of science and technology,3D point cloud data have been widely applied to many application domains, such as computer vision, scientific calculation, and virtual reality. The role of 3D point cloud is becoming more and more important. Outlier detection and model reconstruction are two key techniques of the 3D point cloud processing techniques, and they are the basis of follow-up processing:outlier detection of 3D point cloud is the precondition of geometry information estimation, multi-view registration and feature identification; model reconstruction is most important in the application of 3D point cloud and reverse engineering. It is significant to detect the outliers in an efficient way and find a robust model reconstruction method in the practical application of 3D point cloud. Therefore, this paper will carry on an in-depth study in outlier detection and model reconstruction. The main research content in this paper is described as follows:1. The existing outlier detection methods have their corresponding applicable scopes: some methods require point cloud follows established distribution; some detect global outliers; and some methods have high time cost. Therefore, a method with wide range of application, simple principle, and without special requirement for point cloud has been the urgent needs of the researchers. Therefore, this paper introduces a method which is based on the principal component analysis. This method firstly finds out the point set with maximum consistence from neighbor of a point, and then uses box-plot to detect outliers. It has simple principle, less calculation amount, and does not have more requirement for the point cloud. Experimental results can prove the precision of the outlier detection and the consistence of the geometry information estimated from the cleared point set with the real condition.2. The mesh reconstruction of uneven 3D point cloud has been a difficulty of 3D mesh reconstruction. For 3D point cloud, this paper proposes a density self-adaptive 3D mesh reconstruction algorithm. It is based on the Ball-pivoting algorithm (BPA), and take full advantages of BPA’s advantages of applicability for large-scale data, efficiency, and robustness. The new algorithm takes a rolling ball to traverse all points and connect each suitable point into mesh. The radius of the ball is dynamically determined by the condition of local density. In this way, the bad influences of uneven points such as missed points and holes can be avoided, and a better mesh reconstruction can be expected. Experimental results show good performances both on even and uneven point cloud.