Supervised machine learning for holes classification of three-dimensional free-form models

Abstract

Hole-filling task in tridimensional reconstruction process requires an expert user to select holes to be corrected (filled) in cases where there are presence of real holes in the surface of the object that is being reconstructing.  Generally, proposed works in hole–filling task, assume the surface of the object is continuous so that all holes must be corrected.  The foregoing is not true for many cases e.g. industrial parts and free-form objects.  In this work, it is proposed a method for automatic holes classification in tridimensional surfaces of free-form objects, in two categories: real or holes that must not be corrected, and anomalies or holes that must be corrected. For this purpose three characteristics of contour of hole are calculated: torsion, curvature and size, and subsequently are used in a supervised classification system.  Results show a classification rate over 90%.