master-thesis

Repository containing all code and stuff for my master thesis. For more specific documentation check the docs/ directory.

What is this

The ultimate goal of this project is to localize depth images in a known 3D model. To achieve this the depth images are converted into a derived feature image that allows classical computer vision algorithm to detect features, like with classical camera images. From this point on, all algorithms for optical localization are applicable to depth images. This model can either be a laserscan, a reconstructed mesh or a dense pointcloud.

All the science for that is written in my master thesis (english). The colloquium was done with these slides (german).

Examples

For in-depth usage examples of the provided tools check docs/examples.md.

$ ls
> .. a bunch of orthographic depth maps AND and intrinsic file ..
> depth_0000.png depth_0001.png depth_0002.png ...
> intrinsic.txt

$ cat intrinsic.txt  # example intrinsic for a kinect pinhole image
> 960 540                       # == <width> <height>
> 519.226 0.000000 479.462      # == <fx> ignored <cx>
> 0.000000 522.23 272.737       # == ignored <fy> <cy>
> 0.000000 0.000000 1.000000    # this line is ignored

$ depth2x flexion \
    --calibration intrinsic.txt \
    --input depth_{:04d}.png \
    --type pinhole-depth \
    --start 0 \
    --end 100 \
    --output flexion_{:04d}.png

$ ls
> .. still a bunch of orthographic depth maps and the intrinsic file ..
> .. with additional a bunch of flexion images ..
> flexion_0000.png flexion_0001.png flexion_0002.png ...

Getting the software

Docker

It is always possible to get the latest and greatest version as docker image.

$ # Getting access to the container registry
$ docker login git.informatik.tu-freiberg.de:5050

$ # Pulling Debian Based Image
$ docker pull git.informatik.tu-freiberg.de:5050/jtoth/master-thesis:latest
> Download log

$ # Alternative: Alpine Linux based image (musl and other hardened software)
$ docker pull git.informatik.tu-freiberg.de:5050/jtoth/master-thesis:alpine-edge
> Download log

Running the command requires mounting of software or creating a container based on that one.

Binaries

You can download binaries from the CI setup. The old-linux-most-portable build in the stage platforms is your best bet. It is a ubuntu 12.04 based build but only requires the default linux system libraries as dynamic dependencies. It should e.g. work on CentOS 7 and other non-ubuntu systems as well. The only requirement is an glibc>2.15.

See the Pipelines Page that provides a download link for the artifacts. If there is no artifact that works on your system you need to fall back to compiling on your own.

This allows for the most customization and most control over the final executable. This project tries to support many compilers and environment but is very Linux oriented. For more information on building see the Compilation Reference.

Please note, that artifacts from the platform might stage require libjpeg-dev and libpng-dev libraries as they are dynamically linked. On Linux system you can check ldd <binary-executable> if all dynamic dependencies are resolved. If yes, the binary should work, given the libc is new enough ;)

Using a docker container with ubuntu:18.04 should definitly work! More robust ways to distribute, e.g. packages, are implemented when necessary and time allows it.

Contributing

Right now, not so much. As this is part of my master thesis I need to develop it on my own. Bug Reports are of course always welcome!