Diploma Thesis

Biologically Inspired Model for
Visually Driven Navigation

Nicolai Waniek
Ulm University, Faculty of Engineering and Computer Science, Institute of Neural Information Processing
submitted March 2012


→ Prof. Dr. Heiko Neumann, Ulm University
→ Dr. Florian Raudies, Boston University


Acquiring precise self-motion estimates from monocular visual input is an import problem in many SLAM (Simultaneous Localization And Mapping) systems. In this thesis, I present a novel biologically inspired model to increase the accuracy of those estimates. The model fuses two different processing paths, each of which yields an individual estimate of the self-motion. One path uses a template model of cortical area MST, the other path the epipolar geometry. Fusion of the two estimates is carried out by a simple head direction cell network. Augmented by a prediction signal, the fused estimate is used as feedback to the individual paths. In simulations, I can show that the feedback is beneficial to the self-motion estimation process. Consequently, the model yields a higher accuracy when compared to results without feedback. The model can serve to inspire real-time capable algorithms in robotics, or to guide research in neuroscience.

↓ download PDF (2.3 MB)

License: Creative Commons Attribution-NonCommercial-ShareAlike 3.0 (CC-NC-SA)
Code: Due to some limitations, the source code of the diploma thesis will not be fully disclosed. For example, I have some very small parts of foreign code which I am not allowed to release, but which are required during some computations. In addition, not every submodule is in a publishable state yet. If you wish to have access to the code, or to the data sets which I generated for testing, please contact me directly. Already published software is listed here:

→ libcensure
→ exrutils
→ exrmat