DSOL是一种适用于嵌入式芯片上的稀疏直接法的VO。相比官方的代码(ros)，我们用pangolin可视化。相比特征点法，直接法生成的三维点更稠密，更鲁棒。

run

# install dependencies

## install glog
git clone --depth 1 --branch v0.6.0 https://github.com/google/glog.git
cd glog
mkdir build && cd build 
cmake -DBUILD_SHARED_LIBS=TRUE ..
make -j6
sudo make install

## install fmt
git clone --depth 1 --branch 8.1.0 https://github.com/fmtlib/fmt.git
cd fmt
mkdir build && cd build 
cmake -DBUILD_SHARED_LIBS=TRUE -DCMAKE_CXX_STANDARD=17 -DFMT_TEST=False ..
make -j6
sudo make install

## Install Abseil
git clone --depth 1 --branch 20220623.0 https://github.com/abseil/abseil-cpp.git
cd abseil-cpp
mkdir build && cd build
cmake -DABSL_BUILD_TESTING=OFF -DCMAKE_CXX_STANDARD=17 -DCMAKE_INSTALL_PREFIX=/usr -DBUILD_SHARED_LIBS=TRUE ..
make -j6
sudo make install

## Install Sophus
git clone https://github.com/strasdat/Sophus.git
cd Sophus
mkdir build && cd build
git checkout 785fef3
cmake -DBUILD_SOPHUS_TESTS=OFF -DBUILD_SOPHUS_EXAMPLES=OFF -DCMAKE_CXX_STANDARD=17 ..
make -j6
sudo make install


## Install Google benchmark
git clone --depth 1 --branch v1.6.2 https://github.com/google/benchmark.git
cd benchmark
mkdir build && cd build 
cmake -DBENCHMARK_DOWNLOAD_DEPENDENCIES=on -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_STANDARD=17 -DBENCHMARK_ENABLE_GTEST_TESTS=OFF ..
make -j12
sudo make install 


## clone this codes
git clone https://github.com/lturing/dsol_pangolin
cd dsol_pangolin
mkdir build && cd build
cmake .. && make -j12

## run dsol stereo on kitti00
cd ..
./example/dsol_stereo_kitti

demo

• b站视频

todo

• add loop close
• add imu

---

🛢️ DSOL: Direct Sparse Odometry Lite

Reference

Chao Qu, Shreyas S. Shivakumar, Ian D. Miller, Camillo J. Taylor

https://arxiv.org/abs/2203.08182

https://youtu.be/yunBYUACUdg

Datasets

VKITTI2 https://europe.naverlabs.com/research/computer-vision/proxy-virtual-worlds-vkitti-2/

KITTI Odom

TartanAir https://theairlab.org/tartanair-dataset/

Sample realsense data at

https://www.dropbox.com/s/bidng4gteeh8sx3/20220307_172336.bag?dl=0

https://www.dropbox.com/s/e8aefoji684dp3r/20220307_171655.bag?dl=0

Calib for realsense data is

393.4910888671875 393.4910888671875 318.6263122558594 240.12942504882812 0.095150406

Put this in calib.txt and put it in the same folder of the realsense dataset generated by the python file.

Build

This is a ros package, just put in a catkin workspace and build the workspace.

Run

Open rviz using the config in launch/dsol.rviz

roslaunch dsol dsol_data.launch

See launch files for more details on different datasets.

See config folder for details on configs.

To run multithread and show timing every 5 frames do

roslaunch dsol dsol_data.launch tbb:=1 log:=5

Dependencies and Install instructions

See CMakeLists.txt for dependencies. You may also check our Github Action build file for instructions on how to build DSOL in Ubuntu 20.04 with ROS Noetic.

Disclaimer

For reproducing the results in the paper, place use the iros22 branch.

This is the open-source version, advanced features are not included.

Related

See here for a fast lidar odometry

https://github.com/versatran01/rofl-beta

https://github.com/versatran01/llol