Skip to content
This is a repo for my master thesis research about the Fusion of Visual SLAM and GPS. It contains the research paper, code and other interesting data.
C++ Shell Python CMake
Branch: master
Clone or download


This is a repo for my master thesis research about the Fusion of Visual SLAM and GPS. It contains the thesis paper, code and other interesting data.

The website that accompanies this research can be found at:

Master Thesis

The Master Thesis was written in LaTeX and is published here:


This video shows first tests with the setup running in a car: Watch the video


Two papers have been accepted and published related to this master thesis:

  1. VISAPP 2019: "B-SLAM-SIM: A novel approach to evaluate the fusion of Visual SLAM and GPS by example of Direct Sparse Odometry and Blender" by Adam Kalisz, Florian Particke, Dominik Penk, Markus Hiller and Jörn Thielecke. Link (VISAPP technical program):
  2. DICTA 2018: "Systematic Analysis of Direct Sparse Odometry" by Florian Particke, Adam Kalisz, Christian Hofmann, Markus Hiller, Henrik Bey and Jörn Thielecke. Link (IEEEXPLORE):

A third publication was accepted at VISAPP 2020 and will be published as soon as possible after the conference has taken place.


Code was written in C++ (main realtime implementation), Python (Blender Addon "B-SLAM-SIM" and Sensor data fusion in Blender), HTML5 (sensor recorder, track viewer, synchronization and live-demo tools).

This repository provides two tools that assist you evaluating your sensor data fusion approach. Firstly, the data generation can be performed using the "B-SLAM-SIM" Addon (02_Utilities/BlenderAddon/ in Blender. Secondly, the generated data can be fused via the Linear Kalman Filter (02_Utilities/FusionLinearKalmanFilter/ in Blender.

How to use the B-SLAM-SIM Blender Addon

The B-SLAM-SIM Addon helps you to generate datasets from within Blender for your evaluations. There is one version for the old Blender version 2.79 (02_Utilities/BlenderAddon/ and the recent Blender version 2.80 (02_Utilities/BlenderAddon/ The old Blenderversion will not be developed further, therefore please use the recent Blender version 2.80 addon if possible.


The usual way to install addons in Blender is to choose the addon file from the user preferences. For a quicker way you may want to switch to the "Scripting" workspace in Blender 2.80, create a new text block in the text editor, paste in the content of the file 02_Utilities/BlenderAddon/ and click on "Run Script". You should then see a new Panel available in Blender in the Properties Window under "Scene" Settings.

B-SLAM-SIM User Interface

The user interface is shown in the following two figures - this is the upper part:

and this is the lower part:

For a sample use case let us consider that we need to export the animated camera pose from Blender into a text file for general use. Due to the goal of this work to further process this information in the Direct Sparse Odometry (DSO) algorithm, the addon automatically transforms camera poses from Blenders' coordinate system to the coordinate system of the DSO. In case you need the camera pose in Blenders's coordinate system, the addon will create that for you as well. In order to export your camera, please follow these steps:

  1. Select where you want to export the data by specifying the "Output directory:" in the addon
  2. Select what camera you would like to export from the object list "Camera to use:" in the addon
  3. (Optional) If noisy camera poses are needed: Specify the stochastic errors for noisy measurements and the name of the noisy camera object (setting: "Altered camera:")
  4. Click on "Export camera poses and project files"

That was easy, wasn't it?! The generated ideal measurements might look similar to this:


And the noisy measurements might look similar like this:


The data is structured as follows:

Timestamp (optional, in milliseconds),Translation (x), Translation (y), Translation (z), Orientation Quaternion (w), Orientation Quaternion (x), Orientation Quaternion (y), Orientation Quaternion (z)

There are many more features in the addon that may help you generating datasets quickly. If you need assistance, please feel free to write me an issue on GitHub and I will explain the process in more detail.

How to recreate the Plots from the thesis

In order to recreate the plots from the thesis and papers you can use the attached python script. First, make sure you are using Python3 and have "matplotlib" installed. For the latter you can usually do:

pip install matplotlib

This repository contains a larger amount of .txt files which have been generated in the sensor data fusion step within Blender. The location of those files is "02_Utilities/FusionLinearKalmanFilter" (the sensor data fusion script and a .blend file are located there as well!). These .txt files can be converted to plots using the following command:


If you want to test your own sensor data fusion, you can use the Blender 2.79 (no 2.8 at the moment) "01_LinearKalmanFilter_allEvaluations.blend" file in "02_Utilities/FusionLinearKalmanFilter" and take a look at how sensor data fusion is done there. Please, if you have any questions related to that, feel free to contact me or just open an issue. I will respond as soon as possible! I have planned to create a video tutorial on how to use it, but didn't manage to finish it yet. This is still on my TODO list and will most likely make everything a lot clearer.

How to use the modified Direct Sparse Odometry (DSO) Code?

Please follow the Instructions on the DSO Github Page. As of October, 06th, 2019 I had no issues to compile the code on a Jetson Nano embedded board. However, during my Master's thesis I had quite a few annoying compiler crashes (VTable Bugs with Gnu GCC):

At that time, I had fixed them with generating additional diagnosis data, which you can see in my bug report here: GNU GCC Bugzilla. Maybe it will help you the same way as it did for me.

Once you have DSO compiled, you may want to copy my code (from "03_Application/dso") over the original one to get my modifications to the viewer (I am drawing the world grid, world axes and camera axes to help with coordinate system transformations), the DSO Code (I can import camera poses from Blender - however, fusing ground truth data did not work as expected still - and I export the camera poses in a wrapper. Furthermore I upload the cameraposes via HTTP via a POST request)

How to use the modified Robot Operating System (ROS) DSO Wrapper?

Once again, please follow the Instructions on the DSO ROS Wrapper Github page but be sure to checkout the "catkin" branch instead of the "master" branch using:

git checkout catkin

After this, compile using the following command in your catkin workspace (typically "~/catkin_ws":


You may now copy my code (from "03_Application/dso_ros") over the original code in order to have my working code to operate the real-time DSO via ROS using a Stereolabs ZED Camera with their official ROS Wrapper (see ZED ROS Wrapper Github page).

Why GPLv3 license?

I have modified code that is licensed under GNU GPL Version 3 (see below), so I have to make my changes available under the same license. I can provide a different license for my custom Blender Addon scripts if you prefer that. Just feel free to message me.


Direct Sparse Odometry (DSO)

This work uses the Direct Sparse Odometry Project by TUM (see: and License: GPL-v3.

Direct Sparse Odometry for ROS (ros-dso)

There is also a ros-wrapper for DSO, which is used for the real-time part of this work (see The "catkin" branch by Nikolaus Demmel was actually used here, as the original "rosmake" version did not work for me. License: GPL-v3.

You can’t perform that action at this time.