Case Study – surface reconstruction from point clouds for prosthesis production

Abstract

Currently, the production of a high-quality and highly aesthetic prosthesis is still mainly based on handwork and subjective comparison to get the proper shape and colour of the prosthesis, likely not by a first attempt. This article describes a case study whose goal was to investigate the possibilities of computer-aided surface reconstruction and supplementation to improve the quality and reduce the manufacturing time for an orthopaedic prosthesis. For this purpose, a model of human finger made of plaster had been scanned with the high fidelity laser triangulation scanner. The result was a set of very dense point clouds, representing the surface of the finger model in all its complexity. The main target of this work was to create a water-tight, high resolution computer model of a human finger, which would be ready for further manipulation, such as scaling, mirroring or stretching in arbitrary dimensions. A larger amount of such models could represent a virtual database of human shapes, which would be suitable for prosthesis production and many other (medical) purposes. The first steps after scanning were done in an attempt to reorient several scans, taken from different viewpoints, relative to each other in order to get the proper shape of a finger. This was done by applying an ICP algorithm, integrated in commercial software, and its comparison to the results of reorientation, based on information about finger's position transformation during the scanning process. This information proved to be vital for a fast and accurate alignment of the scans and successful surface generation. This paper also discusses the possibilities of avoiding the influences of geometric errors, generated by a triangulation scanner on surface alignment and the creation of a 3D model. The surface was created by applying Delaunay triangulation to the point cloud. Afterwards, it was followed by manual and automatic refining and reconstruction of a triangular mesh. The final result is a 3D computer model of a human finger with all its details, such as fingerprints and wrinkles. Additional measurements showed that the arithmetical average of deviation between a computer and a physical model was less than 0.3 mm, which is a good result for the desired purposes. The study also showed the possibilities for acceleration of scan alignment while the accuracy could be increased.