Reverse engineering involves digitizing a physical 3-D model or part using a tactile or optical sensor, and transforming the data into a standard CAD (computer-aided design) form. In a typical reverse engineering sequence, the physical model is first digitized with a laser scanner to produce a dense map of surface points. The natural faces and edges of the object are identified by the operator in a process termed segmentation. The segmentation is in the form of a patchwork of curves on the digitized surface and is constructed interactively within the computer. Mathematical surfaces are fitted to the patches defined by the curve network to create a concise CAD model of the part.

There are a number of difficulties with the reverse-engineering process. In the scanning phase, a number of range maps, each consisting of an array of (x,y,z) surface point positions, are acquired to digitize the surface. After multiple range maps have been acquired, many regions will contain excessive amounts of overlapping data, while other regions may still be lacking data. This unstructured, redundant data impedes visualization and makes the interactive segmentation process both tedious and error prone. In addition, the manually directed scanning and the interactive segmentation are both time-consuming processes.

There are other problems related to the fitting of mathematical surfaces. Current surface-fitting approaches are not capable of handling arbitrarily shaped wrap-around surfaces, even though such surface patches appear in many consumer products. A second problem appears after surfaces have been fitted. The seams between multiple fitted surface patches may not appear smooth, even though a smooth seam is indicated by the physical model.

In this work, solutions for these problems have been developed and tested. The scanning problems have been solved with an automated scanning approach, which is based on an intermediate model composed of orthogonal cross sections (OCS). The OCS model allows laser scan data from multiple views to be merged into a compact, easily visualized, wrap-around model. Steps toward solving the segmentation problem have been made with a semi-automated segmentation based on an active contour. Additionally, a surfacing approach permitting the fitting of wrap-around surfaces and a method of globally smoothing the seams of the fitted surfaces have been developed. In all the above cases, the proposed approaches have been implemented and tested on consum er products with satisfactory results.