The complexity of flight raises the acquisition cost for aerial datasets to prototype, test or evaluate airborne computer vision algorithms. A possible surrogate is the usage of virtual environments. However, it is unclear how results acquired in such environments transfer to real world situations. This thesis presents a general concept to identify performance differences of computer vision algorithm on synthetic and natural data. Further, it correlates these difference to image content differences to find causal relations. Lastly, different ways to parametrize the virtual environment are evaluated to identify rendering and modelling techniques reducing the algorithms performance difference. The results are eventually formulated as recommendations for modelling engineers and programmers to optimize their simulation environment.    «

The complexity of flight raises the acquisition cost for aerial datasets to prototype, test or evaluate airborne computer vision algorithms. A possible surrogate is the usage of virtual environments. However, it is unclear how results acquired in such environments transfer to real world situations. This thesis presents a general concept to identify performance differences of computer vision algorithm on synthetic and natural data. Further, it correlates these difference to image content differen...    »