Experiences in Building a Visual SLAM System from Open Source Components

Christoph Hertzberg, René Wagner, Oliver Birbach, Tobias Hammer, and Udo Frese


This paper shows that the field of visual SLAM has matured enough to build a visual SLAM system from open source components. The system consists of feature detection, data association, and sparse bundle adjustment. For all three modules we evaluate different libraries w.r.t. ground truth.

We also present an extension of the SLAM system to create dense voxel-maps. It employs dense stereo-matching and volumetric mapping using the poses obtained from bundle adjustment, both implemented with open source libraries.

Apart from quantitative comparison we also report on specific experiences with the various libraries.


The paper as submitted to ICRA 2011 is available here (PDF).



Data Sets and Source Code

Here you can download the source code (5.5MiB) and data sets (994MiB).

If you have any problems building the program or running the data sets, do not hesitate to contact us.