Robust automatic coarse registration of specular free-form surfaces

Many industrial applications of optical 3D sensors require “registration”. The alignment of different 3D data sets in one common coordinate system is necessary, for example, in order to compare a measured object shape with the design data or to combine several partial 3D measurements of an object into one complete 3D view. Methods for automatic registration typically try to find corresponding characteristic features on both surfaces and then map these features onto each other. However, for smooth surfaces like eyeglass lenses or parts of painted car bodies registration is a real challenge: These objects are inherently “poorly featured” meaning that there are no simple features like edges or corners available for a registration. We present a method based on the Hough transformation that is able to overcome these problems by employing higher-order geometrical surface properties. The method is robust against small shape deviations, outliers, and noise, and it does not require user interaction. So we are able to align partial measurements of a smooth object requiring only small overlaps.