Bundle adjustment is an essential part in most modern multi-view geometry systems and is mainly used in the context of Structure from Motion (SfM) and Simultaneously Localization and Mapping (SLAM). Bundle adjustment jointly refines camera parameters and 3D-points by minimizing the reprojection error across multiple views. It is used as final refinement of approximate initial scene estimates, or serves to remove drift when invoked multiple times in the course of an incremental reconstruction. Bundle adjustment is still considered the gold standard for the non-linear refinement of camera parameters and 3D-points which occurs in many geometric computer vision problems. However, for very large scale reconstructions, the requirements bundle adjustment poses on memory and computational resources are just too high to be realized on a single machine. Memory as the limiting factor for a very large problem size can be bypassed by distributing the bundle adjustment algorithm across mutliple machines. In this thesis, we implement two consensus-based distributed bundle adjustment approaches: Point Consensus and Camera Consensus. Our implemented Camera Consensus extends the idea proposed in the literature by an automated update of certain parameters. Furthermore, we replace the averaging used in the consensus computation by a geometrically & stochastically driven update. We propose a splitting strategy to partition the original problem into sub-blocks that is tailored to the prevailing acquisition patterns of aerial imagery. The implemented consensus approaches are evaluated on real world data and on noisy synthetic data. We can show convergence behaviour of the distributed approaches on the used datasets similar to that of global bundle adjustment.