Quantitative Benchmarking of Inline Computational Imaging

Measurement systems play a crucial role in quality management within companies.Due to fierce competition in globalized markets, increasingly specialized applications,and the pursuit of zero-defect production, the requirements for such systems are continuously rising. Inline Computational Imaging (ICI) is an algorithm designed to create a rapid and precise digital 3D reconstruction of visually inspected objects.It combines multi-view stereo and photometric stereo methods and can be applied for inline defect detection and metrological measurements in quality assuranceprocesses.Since no comprehensive analysis of ICI has been conducted so far, this thesis aims to quantitatively benchmark its performance with respect to different objectcharacteristics and algorithmic settings. This is accomplished through a series of experiments in which measurements are performed under systematically varied input parameters. To ensure high repeatability, the ICI system has been recreated as a digital twin using rendering software, where the object of interest is modelled in accordance with measurement standards to serve as the measurand. Certain measurements, depending on and adapted to the investigated object geometries,are performed on the reconstruction result of the ICI algorithm. The evaluation involves analysing key indicators, processing time, measurement uncertainty, error,and noise, based on the reconstruction results.The findings enable more reliable predictions of the ICI system’s performance for specific inspection tasks across various object classes, thereby improving its efficiency. Furthermore, the proposed performance evaluation process can be applied to other measurement systems as well.