Reverse Engineering Regular Objects: Simple Segmentation and Surface Fitting Procedures

Segmenting point clouds and fitting surfaces onto subsets of measured data are crucial elements of reverse engineering algorithms. Due to the variety in data acquisition procedures and the requirements concerning the representation of the model to be created there are several methods to approach this problem. Our current interest is to construct exact, 'watertight' boundary representation models of regular objects, based on laser scanned point clouds with relatively high density and high accuracy. After defining the term 'regular object', a non-iterative algorithm for direct segmentation is presented, where well-known techniques from computer vision are combined with new procedures for processing point and normal vector data. These include the separation of primary surfaces and transition elements, special filtering methods according to planarity and dimensionality and the detection of simple analytic surface types.