LDSO: Direct Sparse Odometry with Loop Closure

Related Publications

• LDSO: Direct Sparse Odometry with Loop Closure, X. Gao, R. Wang, N. Demmel, D. Cremers, In International Conference on Intelligent Robots and Systems (IROS), 2018.
• Direct Sparse Odometry, J. Engel, V. Koltun, D. Cremers, In IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2018
• A Photometrically Calibrated Benchmark For Monocular Visual Odometry, J. Engel, V. Usenko, D. Cremers, In arXiv:1607.02555, 2016

Dependencies

sudo apt-get install libeigen3-dev libgoogle-glog-dev libgtest-dev libsuitesparse-dev libopencv-dev libzip-dev libboost-all-dev

build DBoW3 in the thirdparty folder first.

cd thirdparty/DBoW3
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=$BUILD_TYPE ..
make

Other libraries

Compile and install Pangolin for visualization.

Compile

This will build the thirdparty library and also ldso library for you. You can also follow the steps in this script manually (will compile DBoW3 and g2o first, and the ldso).

Run

TUM-Mono:

To run LDSO on TUM-Mono dataset sequence 34, execute:

rosrun ldso ldso
    preset=0 \
    files=XXXXX/TUMmono/sequences/sequence_34/images.zip \
    vignette=XXXXX/TUMmono/sequences/sequence_34/vignette.png \
    calib=XXXXX/TUMmono/sequences/sequence_34/camera.txt \
    gamma=XXXXX/TUMmono/sequences/sequence_34/pcalib.txt

Notes

• LDSO is a monocular VO based on DSO with Sim(3) loop closing function. Note we still cannot know the real scale of mono-slam. We only make it more consistent in long trajectories.

• The red line in pangolin windows shows the trajectory before loop closing, and the yellow line shows the trajectory after optimization.

• If you are looking for code instructions, take a look at doc/notes_on_ldso.pdf and see if it can help you.

• Set setting_enableLoopClosing to true/false to turn on/off loop closing function. setting_fastLoopClosing will record less data and make the loop closing faster.

• If you need loop closing, please set setting_pointSelection=1 to make the program compute feature descriptors. If setting_pointSelection=0, the program acts just like DSO, and setting_pointSelection=2 means random point selection, which is faster but unstable.

• Some of the GUI buttons may not work.

• You might also want to have a look DSO's README: https://github.com/JakobEngel/dso/blob/master/README.md

License

LDSO, like DSO, is licensed under GPLv3. It makes use of several third-party libraries. Among other's it comes with the source of:

• Sophus (MIT)
• g2o (BSD, GPL, LGPL, ...)
• DBoW3 (BSD)