Design and Implementation of an Object Recognition Modularity Framework

Kurzfassung

Modularity as a software architecture principle has the goal to create independent parts which can be combined to a bigger system and allow or easy exchange or addition of single modules. For truly independent modules a framework is required for managing these modules and their interconnections. In research contexts, comparing different solutions for one problem and focusing on a very specific sub-problem are common situations. A modularity framework can improve the efficiency of the research process and make it more comfortable by allowing to reuse and rearrange the solutions of sub-problems. The object recognition software developed at the German Aerospace Center (DLR) is mostly implemented in C++ on Linux computers. Thus, a modularity framework for this environment is required. In this bachelor thesis, the topic modularity is explored theoretically, existing state of the art frameworks are evaluated and the requirements for an object recognition modularity framework for the DLR are defined. Because no existing framework can be used due to some DLR specific constraints, a custom framework is designed, implemented and evaluated. It is shown, that advanced modularity concepts can also be implemented using C++ and POSIX functions so that the performance benefit of a native programming language can be maintained without having to relinquish the advantages of modular development.