What is this

This is my (Scott Saunders) collection of things regarding summer 2017 research.

This is the main-level work directory for me during the summer 2017 research project. The folders at the top level are for my commonly used programs/utilities/scripts/text files.

Most of the directories here are git submodules, linking to other gits.

How to run:

Initial Setup:

The script setup/setup.sh assumes the user is running Arch Linux, and will attempt to install/update all needed dependencies. (as of August 28, 2017)

The key dependencies for Elastic Reconstruction:

• PCL built with kinfu. (Requires Cuda, OpenNI1, VTK... and more)
• g2o (The default homebrew version of it for MacOSX doesn't quite work, building from source does though. Note: Currently put at a fixed version, as an updated caused segfaults)

Aside: if you have built/installed the dependencies already, there is a submodule ElasticReconstructionRelease, which is a smaller and simpler git, designed to build just the code-base for ElasticReconstruction. (See it's readme for more details)

Running:

• source setup_path.sh will setup the PATH environment variable for use with most utilities in the folder.
• source pipeline.sh will setup the PATH environment and bash functions to run the Elastic Reconstruction pipeline locally from this folder. (It is also sourced from setup_path.sh)

At this point, you need to change directory (cd) to a folder with data, where you can then run Pipeline (defined in pipeline.sh) to run the whole pipeline. Assuming everything works, and meshlab is also installed, you will be left a mesh.ply in a mesh folder in the current directory. Without meshlab, this folder will be empty, and the final results will be in integrate as either go_world.pcd (Always present) or fo_world.pcd (Only sometimes present)

• Pipeline.sh: Most of the actual ElasticReconstruction file-gluing between parts is done in pipeline.sh, which is a re-implmentation of the original demo.sh provided in the binary download. As-is, the arguments for a particular segment of the pipeline are hard-coded in this script. If one wants to change them, it should be relatively easy todo so, by either running the command manually in the corresponding folder and typing in the arguments, or simply editing them in the script.

Submodule Quick use:

New method:

• sh setup/submodules.sh

Old method:

1. cd ./someFolderThatIsAGitSubmodule
2. git submodule init ./
3. git submodule update --remote ./
4. git checkout master ./
5. Then use that directory as a regular git directory.

Highlights (files):

• todo.txt, is my todo,
• todone.txt, is what I have done.
• pipeline.sh, is how I run the elastic reconstruction pipeline.
• setup_path.sh is a bash source-able file, which sets up my $PATH.
• windows_bins.sh - Downloads the windows bins and examples and puts them in a winbin folder.
• testdata.sh - obsolete way of running the pipeline. (How I was doing it initially)

Sub-modules:

• pcl - my fork of PCL
• spcl - The Stanford PCL fork. (I also have one, but this does not reference it, mine only has a dos2unix run on it.)
• libfreenect2 - The same, but for libfreenect2
• openni2 - The same, but for Openni2...
• ER_port - git submodule of the Elastic Reconstruction port.
• ER_myfork - a reference to my fork of it (unused, but will be a clone of ER_port)
• GraphOptimizerTest - Eriks GO-tester.
• StSTK - Structure SDK
• 3D-re - Erik's 3D-reconstruction program (it just records data.)
• openCV_Traj - A simple program (and shell files) for using OpenCV to generate inital trajectories.

Directories:

• setup - my partal setup scripts.
• misc - misc programs/related projects.
• notes - misc notes.
• utils - misc utilities.
• freenect2 - incomplete attempt to apply image registration using the libfreenect2 library.
• oni_stuff - Has my own OpenNI utilities. ni2Recorder (What I use to record data.) ni1Recorder/ni1Reader/ni2Reader (Readers very simply try to access data) - Has a converter submodule, this is my Oni2 -> Oni1 converting library.

#Odd segfaults:

• compiling some of the programs with -march=native seems to cause segfaults. Internal to my fork of PCL, there is something that disables this in the pcl/gpu/CMakeLists.txt and/or pcl/features/CMakeLists.txt.

• g2o also seems to suffer with these optimization.

• A good way to find the source of segfaults is by gdb --args <programname> <args> and then running it r. When it segfaults it returns you to a prompt, which you can use info stack to get a printed list of the stack, usually with the filenames/lines/etc. However, if it is from a library, before even entering main(), they will all just be addresses. Using LD_DEBUG=ALL <programname> <args...> 2>ldd.txt and then looking through (bottom-up) of ldd.txt will allow you find what caused the issue.