Name of the participant: Volker Knauthe
Description of the IT research project: The digitalization of 3D objects is a huge technological step for a multitude of industrial and social areas. It enables new opportunities for quality assessment, designing methods, digital preservation and is a key technology for transferring the real world into augmented and virtual realities.
While the digital reconstruction of opaque objects is well studied and ever progressing, the reconstruction of transparent structures is at most partly solved for very restricted cases and predominantly well-known controlled setups.
The challenge is quite understandable, as it is hard to even perceive and capture certain types of transparencies. A prime example are children that have their first encounter with a near perfectly clear glass door or a toy store window. Plainly spoken, there are multiple depth levels (with possibly different levels of inherent fascination and recognizable objects) that are hard to distinguish. As humans grow older, their vision seldom improves but they tend to optimize those transparency interactions based on prior experience and hunches where a headache may hide in plain sight. This short story preludes a research plan that focuses on different approaches to harness experience and knowledge for a better recognition of transparent structures and the generation of insights about the characteristics of the structures themselves.
The main contribution of this project is to search for new approaches, test their limitations, strengths and weaknesses and asses how applicable they are (stand-alone or in combination) for potential industrial use cases. This includes and combines classical methods derived from our vision like motion parallax, focus changes and pattern similarities, as well as machine learning approaches.
Software Campus partners: TU Darmstadt, ZEISS Gruppe
Implementation period: 01.01.2021 – 30.06.2022
