Study Of Point Cloud Segmentation Based On Clustering Method

Reverse Engineering(RE) is a method to build a CAD model. This method acquires the existing product model point cloud through digital measuring equipment. A CAD model can be rebuilt by segmenting and fitting the point cloud data. Now, RE has been widely applied to many engineering domains, such as industrial inspection, autonomous navigation, and protection of historical relics, et al. With the constant development of digital precision, the 3D laser scanner and CT scanner acquire a lot of point cloud data, and the traditional method to segment is time-consuming, so it’s very important of solving the problem that realizes the segmentation of point cloud by Computer-Aided Design. Therefore, this dissertation aims to study a method to realize point cloud segmentation based on Gaussian mapping and clustering. This method is mainly segmenting point cloud of mechanical objects.Firstly, this dissertation improves the method of vector product, which can find six points which have the nearest distance from the given point by the k-nearest neighbor algorithm, and it can get the unit normal vector of the point by the improved vector product. The Gaussian map can be acquired by Gaussian mapping on the unit sphere. We can obtain the number of clusters and cluster centers by introducing the concept of measure, then a Gaussian map can division different areas by the improved of fuzzy C- means clustering algorithm on the unit sphere. We introduce correction coefficient of membership matrix and cluster center which can accelerate the speed of segmentation and reduce the effects of isolated points. The point cloud data can be segmented according to the relationship between point cloud data and Gaussian map.The corresponding interface can be compiled based on Visual C++ 6.0 and Open GL software. Gaussian mapping and clustering can segment mechanical objects and combination objects to get the corresponding segmentation results. Analyzing the results of experiment influence of parameters 1e and2e, and we can get the optimal value of those parameters. Experimental results show that the algorithm which consists of Gaussian mapping and adaptive fuzzy C- means clustering algorithm can realize robust segmentation of point cloud data.