Polarisationsbildgebung in der industriellen Qualitätskontrolle

In this thesis the usage of polarization imaging in the field of automated optical inspection is investigated. Starting from the classical reflection models of computer vision the polarization parameters are modelled with a special focus on materials with strong volume scattering. Thereby the dependence of the polariza-tion angle on the alignment of the sample can be modelled quantitatively, whereas the behaviour of the intensity and the degree of polarization can be described only qualitatively due to the surface roughness. The influence of this roughness is studied using numerical simulations. Furthermore the measurement setup is enhanced for the industrial application. In particular the well known shifts between the raw images can be effectively compensated using a new optical flow based approach. Moreover the image quality is particularly dominated by the noise of the intensity sensor. Based on mathematical statistics, the lower bound (Cramer Rao Bound) for the measurement precision is derived. Since the dependence of the grey value variance on the grey value is taken into account for the first time, a good agreement between the theoretical consideration and the experimental data is achieved. An experimental setup which was optimised with respect to image acquisition time and stability was integrated in a production line and showed a considerable improvement in detecting defective samples while at the same time showing its suitability for series production.