HSO

This released code is the VO part of HSO based on SVO and PL-SVO.

Authors: Dongting Luo, Yan Zhuang and Sen Wang.

Video: https://youtu.be/AchJQ2u8K50 or https://v.youku.com/v_show/id_XNDgyMjY4ODc4OA

0. Advertising

I am looking for a PhD position. If you think my work and research direction meet the needs of your Lab, please contact me (luo_dt@vip.163.com).

1. Related Publications

Dongting Luo, Yan Zhuang and Sen Wang. Hybrid Sparse Monocular Visual Odometry with Online Photometric Calibration. In International Journal of Robotics Research

2. Required Dependencies

boost

c++ Librairies (thread and system are needed). Install with

	sudo apt-get install libboost-all-dev

Eigen3

Linear algebra.

	sudo apt-get install libeigen3-dev

OpenCV 3.0 or later

Dowload and install instructions can be found at: http://opencv.org. We tested the code on OpenCV 3.2.0.

Pangolin

Used for 3D visualization & the GUI. Dowload and install instructions can be found at: https://github.com/stevenlovegrove/Pangolin. If you encounter difficulties in compiling the latest version of Pangolin, I put an old version in the thirdparty folder, which is easier to complie.

Sophus, fast and g2o (Included in thirdparty folder)

The three libraries will be compiled automatically using script build.sh.

3. Build and Compile

We have tested the code on Ubuntu 14.04, 16.04, 18.04 and 20.04. A newer Ubuntu platform may also be easy to install.

	cd hso
	chmod +x build.sh
	./build.sh

This will create the executable test_dataset in the bin folder.

4. Run code

EuRoC Dataset

Download the dataset from http://projects.asl.ethz.ch/datasets/doku.php?id=kmavvisualinertialdatasets and uncompress it.
Go to the test folder and open the script euroc_batch.sh.
Change pathDatasetEuroc variable to point to the directory where the dataset has been uncompressed and sequenceName variables to the name of each sequence in your computer.
Execute the script: ./euroc_batch.sh
The result of keyframes trajectory will be saved in the result folder using TUM format.

ICL-NUIM Dataset

Download the dataset from http://www.doc.ic.ac.uk/~ahanda/VaFRIC/iclnuim.html (select ICL-NUIM PNGs) and uncompress it.
Go to the test folder and open the script icl-nuim_batch.sh.
Change pathDatasetICL variable to point to the directory where the dataset has been uncompressed and sequece_name variables to the name of each sequence in your computer.
Execute the script: ./icl-nuim_batch.sh
The result of keyframes trajectory will be saved in the result folder using TUM format.

TUM MonoVO Dataset

Download the dataset from https://vision.in.tum.de/mono-dataset and uncompress it.
Uncompress the images.zip in each sequence.
Go to the test folder and open the script tum_monoVO_batch.sh.
Change pathDatasetTUM variable to point to the directory where the dataset has been uncompressed and sequece_num variables to the name of each sequence in your computer.
Execute the script: ./tum_monoVO_batch.sh
The result of keyframes trajectory will be saved in the result folder using TUM format.

Your Own Sequence

The minimal usage is

	cd test
	./single_sequence.sh

Change pathImageFolder in the script to the folder containing images, and cameraCalibFile to the geometric camera calibration file. HSO now support 3 types of camera model, please see below. The result will be saved in the result folder.

5. Camera Files Format

Calibration File for Pinhole model

Pinhole fx fy cx cy k1 k2 p1 p2
width height
true / false

Calibration File for FOV model

FOV fx fy cx cy omega
width height
true / false

Calibration File for Equidistant camera model

EquiDistant fx fy cx cy k1 k2 r1 r2
width height
true / false

When the image is taken with a fisheye camera, then fill in "true" in the third line, otherwise "false". Some camera file examples are saved in ./test/cameras.

6. Command Line Argument

If you use single_sequence.sh to run HSO, there are several commandline options you can use:

Requried

image=X where X is the folder containing images.
calib=X where X is the geometric camera calibration file.

Optional

start=X: Start at frame X. The default is 0. HSO’s monocular initialization is weak. When the initialization fails or there is a problem with the initial scale, we recommend skipping some beginning frames (e.g., start=450 in MH04).
end=X : End at frame X. The default is the number of the images the folder contains.
times=X: The timestamp file of the images. Its number of lines must be equal to the number of images in the folder. HSO now support 4 types of timestamp file format, please see the file ImageReader.cpp. The default is "None".
name=X : The file name of the result. The default is "KeyFrameTrajectory.txt".

7. License

The source code is released under a GPLv3 licence.

If you use HSO in your academic work, please cite:

  @article{Luo2022,
  title={Hybrid Sparse Monocular Visual Odometry with Online Photometric Calibration},
  author={Dongting Luo, Yan Zhuang and Sen Wang},
  journal={International Journal of Robotics Research},
  year={2022}
 }