At the Institut für Informationsverarbeitung (TNT), methods for fully automated image analysis and 3D reconstruction are developed. In a variety of industrial applications, in collaboration with partners from various industries, practical methods of computer vision and digital image processing are researched and implemented close to production.
Undetected corrosion damage can result in incalculable economic and ecological costs in pipelines. Early detection of such damage (predictive maintenance) is practically conducted, for cost reasons, by passively operated robots that are sent through the pipeline system during ongoing operations.
These robots, however, record vast amounts of data. This project, therefore, involves adapting machine learning methods and developing novel techniques to allow the fully automated analysis of these sensor data.
Plagiarism presents major challenges for companies in the machinery and plant engineering sectors. They result in significant revenue losses and expensive warranty cases. Existing plagiarism detection methods require applied or embedded markers. However, machined components exhibit a clear, individual surface topography at the micrometer level. Using 2D signal processing methods, a signature can be extracted from this, which is stored in a database immediately after production. At a later date, such as during maintenance, this "fingerprint" can be determined again and compared with the one stored in the database. If the signature is not recognized, it is a forgery.
Pattern-based methods are researched for estimating camera lens parameters (including fisheye and wide-angle lenses), such as the multimodal calibration approach, which aims to transfer calibration parameters between multiple capture modes.
External camera parameters can be determined using our offset correction approach with periodic calibration patterns. Additionally, at TNT, calibration approaches that do not require an artificial calibration object are being researched.
(Multimode) Camera Calibration, Multiview 3D Reconstruction, Feature Points Refinement, Energy Minimization, Machine Learning, Digital Signal Processing
