Automatic complete polygon surface reconstruction from overlapped three-dimensional point clouds

This research develops an automatic approach to convert multiple overlapped 3D point clouds obtained from surface shape measurement into a single complete polygon surface. Such a conversion, as an important part of complete surface reconstruction, is traditionally achieved in a semi-automatic fashion and involves extensive user interactions. Through investigation of the key steps of polygon surface reconstruction process, improved automatic methods for point set registration and integration are proposed and tested. Raw 3D point clouds from measurement systems are pre-processed and represented with polygon meshes. A unique sculpture surface feature matching method is proposed to automatically obtain initial positions between overlapped polygon meshes for point set registration, in which surface features are extracted based on differential geometry properties calculated from local surface fitting. Iterative Closest Point (Surface Distance Minimization) methods for accurate alignment of polygon meshes are improved to ensure accurate final positions and computational efficiency. Different corresponding pair projection schemes for registration are developed and compared. To achieve automatic integration in case of arbitrary-shaped overlapping point sets, an improved integration method is also proposed that creates final polygon mesh with relatively even vertex distribution. The registration and integration program is build with object-oriented point set management, and simulated point sets are used to test the accuracy and efficiency of the polygon surface reconstruction approach.