Optical preprocessing for automated inspection

It appears that a simple TV-camera connected to a computer will not produce a very good inspection system. The system is far too ‘anthropomorphic’ and not well enough adapted to the limitations of present-day computers nor to the preprocessing capabilities of present-day optical systems. The taxonomy of preprocessing, utilizing a priori knowledge about an object, is briefly discussed. One particular form of preprocessing is found to be specifically effective—the acquisition of three-dimensional data: 3D data are invariant under changes in illumination. Most often the real concern is over the 3D shape of an object, instead of the 2D distribution of the local reflectivity. Evaluation of 3D data is frequently much easier than that of 2D data and 3D sensors can disply a large dynamical (depth) range. Two new 3D sensor principles, developed at the University of Erlangen, are discussed.