Efficient range and image data processing - algorithms and software paradigms

The growing demand towards industrial automation and autonomous systems requires more flexible technologies in different but interdependent domains of engineering. This thesis introduces and discusses two important areas: Time-of-Flight (ToF) camera data improvement (algorithm development) and related model driven engineering techniques (software development). In the last decades, these areas have been extensively studied and important progress was made.

The first part of this thesis discusses the special challenges of data quality improvement on a very deep layer of ToF cameras. It deals with different challenges related to the working principle of this kind of sensor. A new method for a fast motion artifact compensation for ToF cameras is presented. It is shown that the algorithm gives good results for simulated as well as real data while providing real-time performance. The second proposed algorithm deals with the automatic integration time estimation of ToF cameras. An online integration time adaption algorithm that works on a per-pixel basis and uses knowledge gained from an extensive analysis of the underlying inherent sensor behavior is introduced. Finally, an industrial real-time 3D car reconstruction example is presented. It shows how the data of three PMD (Photonic Mixer Device) cameras has to be preprocessed and combined, using an extensive depth data processing and filtering pipeline.

The second part of this thesis addresses the challenges of data related model driven software engineering. It introduces and contributes the new domain specific language GU-DSL and two data and image processing related extensions: GPGPU (General Purpose Computation on Graphics Processing Unit)-programming and CBSE (Component-Based Software Engineering) principles. The presented GU-DSL GPGPU extension contributes a convenient combination and mixture of textual and graphical model- and dataflow-driven design. Using a code generator, GU-DSL code can be transformed into C++, compiled and executed. All the GPU related features are encapsulated into a C++ Heterogeneous Computing framework. The GU-DSL CBSE system introduces a concept for component based software engineering in the domain of data- and image-processing. It proposes several new concepts for component- and component-instance-diagrams in combination with class- and activity-diagrams. Using a newly developed Rich Client Platform supporting a plugin based extension system, it shows how the GU-DSL CBSE concept can be realized and used in practice using C++. Exemplary, a simple processing pipeline is implemented to demonstrate the new concepts.