1: \begin{abstract}

2: A core component of all Structure from Motion (SfM) approaches is

3: bundle adjustment. As the latter is a computational bottleneck for

4: larger blocks, parallel bundle adjustment has become an active area of

5: research. Particularly, consensus-based optimization methods have been

6: shown to be suitable for this task. We have extended them using

7: covariance information derived by the adjustment of individual

8: three-dimensional (3D) points, i.e., ``triangulation'' or

9: ``intersection''. This does not only lead to a much better convergence

10: behavior, but also avoids fiddling with the penalty parameter of

11: standard consensus-based approaches. The corresponding novel approach

12: can also be seen as a variant of resection / intersection schemes,

13: where we adjust during intersection a number of sub-blocks directly

14: related to the number of threads available on a computer each

15: containing a fraction of the cameras of the block. We show that our

16: novel approach is suitable for robust parallel bundle adjustment and

17: demonstrate its capabilities in comparison to the basic

18: consensus-based approach as well as a state-of-the-art parallel

19: implementation of bundle adjustment. Code for our novel approach is

20: available on GitHub: \url{https://github.com/helmayer/RPBA}

21: \end{abstract}

22: