Investigating the segmentation of freeform triangulated surfaces using self organising maps

Abstract

Freeform surfaces can be used to describe manufactured objects. These surfaces can be represented as point clouds, triangulated surfaces and range images. Before these objects can be analysed in any way they need to be broken down into their constituent parts. Using this description stamped parts can be indexed and retrieved to assist in determining how to manufacture a part that has similar properties. One means of performing this task is to segment the object based upon its surface properties. Curvature can be used to describe the behaviour of a surface. In order to use these metrics a single Self-Organizing Map is used to automatically categorise surface into regions of similar curvature. The SOM is first trained using a small number of simple shapes and curvature metrics. It is then used to segment an object that is a mixture of free form surfaces and planes. The combination of these metrics, shapes and the use of a SOM allows for the representation of many types of surfaces. The shapes and curvature metrics used to train the model determine how sensitive it is to different surface descriptions. This technique is successfully applied to a complex object that combines free form surfaces and planar surfaces using robust discrete curvature metrics.