Point Cloud Simplification Based On Affinity Propagation Clustering

Three-dimensional objects are widely applied in fields of reverse engineering, computer-aided design and computer graphics. It provides necessary conditions for rapid development of products. Recently, as the rapid development of digital measuring makes the acquirement of point cloud data from surface of objects easily, it leads to an increasing interest in processing and modeling point cloud data in computer graphics field.Nowadays, there are some problems in point cloud simplification, such as imprecision, slowly processing and so on. This thesis proposed a method of point cloud simplification based on affinity clustering and re-sampling for the uniform and non-uniform point cloud. In this thesis, the achievements are as follows:1. This thesis proposes a new algorithm integrating both methods of affinity clustering and uniformly re-sampling to overcome the shortage of storing huge similarity matrices, while it maintains the strongpoint of affinity clustering which regards every point in point cloud as potential exemplar.2. We adapt a covariance method to calculate the curvature of points which are chosed as preferences for affinity clustering. The algorithm increases the probability of points of high curvature so that more details of point cloud can be remained. For non-uniform point cloud, both density and curvature as preferences to choose exemplars. In order to retain more details of point cloud, we divide grid according to curvature value of point.3. The distance between the simplified and the original point cloud is computed by triangular mesh as the standard error for evaluating results of point cloud simplification.Finally, experiments show the proposed method improves the accuracy of the simplified point cloud, and it is able to adapt to different types of point cloud.