shanmo/orcvio

The monocular version of OrcVIO, which reconstructs object using both semantic keypoints and bounding boxes

OrcVIO

This repo implements OrcVIO: Object residual constrained Visual-Inertial Odometry

Dependencies

The core algorithm depends on Eigen, Boost, Suitesparse, Ceres, OpenCV, Sophus.

Usage

Non-ROS version

cd ORCVIO
mkdir build
cd build
cmake -D CMAKE_BUILD_TYPE=Release ..
make

and then

cd YOUR_PATH/ORCVIO/
./build/orcvio ~/YOUR_PATH_TO_DATASET/MH_01/mav0/imu0/data.csv ~/YOUR_PATH_TO_DATASET/MH_01/mav0/cam0/data.csv ~/YOUR_PATH_TO_DATASET/MH_01/mav0/cam0/data ~/YOUR_PATH/LARVIO/config/euroc.yaml

eg

cd YOUR_PATH/ORCVIO/
./build/orcvio /media/erl/disk2/euroc/MH_01_easy/mav0/imu0/data.csv /media/erl/disk2/euroc/MH_01_easy/mav0/cam0/data.csv /media/erl/disk2/euroc/MH_01_easy/mav0/cam0/data /home/erl/Workspace/orcvio_cpp/config/euroc.yaml

to debug using VS code, define VS_DEBUG_MODE, then run

cd build
cmake -DCMAKE_EXPORT_COMPILE_COMMANDS=YES ..
make

this will generate the compile_commands.json in the build folder

ROS version

• how to build

cd YOUR_PATH/ORCVIO/ros_wrapper
catkin_make

and then

source YOUR_PATH/ORCVIO/ros_wrapper/devel/setup.bash
roslaunch orcvio orcvio_euroc.launch

• how to setup front end refer to wiki
• we also have an OrcVIO-lite version which only uses bounding boxes, no keypoints, this mode can be enabled by changing the flag in the launch file
• visualization

trajectory is more accurate with object residual update trajectory is less accurate without object residual update

Evaluation

unit test (Non-ROS version)

The tests are compiled by default and have to be run from top of the project directory

$ mkdir -p build && cd build
$ cmake -DCMAKE_BUILD_TYPE=Debug -DBUILD_TESTING=True ..
$ make
$ cd .. && ./build/orcvio_tests ; cd -

Or directly run make test from build directory and the test logs are saved in build/Testing/Temporary/LastTest.log.

VIO

1. evaluation on EuRoC dataset

• cd into python_scripts/trajectory_eval/ folder and python batch_run_euroc.py
• comparison with LARVIO
• comparison of left and right purbation in VIO

2. trajectory evaluation using KITTI dataset error metric

• cd into python_scripts/trajectory_eval/ folder and python traj_eval.py

Object mapping

• object map result is saved under cache/object_map
• cd into python_scripts/object_map_eval/ folder and python unity_object_iou_eval.py to get the 3D IOU results when using Unity simulation; to enable debug mode use python3 unity_object_iou_eval.py
• cd into python_scripts/object_map_eval/ folder and python3 kitti_object_iou_eval_single_sequence.py to get the 3D IOU results for one sequence in KITTI, python kitti_construct_pr_table_all_sequences.py to get the PR table for KITTI

Citation

@inproceedings{orcvio,
  title = {OrcVIO: Object residual constrained Visual-Inertial Odometry},
  author={M. {Shan} and Q. {Feng} and N. {Atanasov}},
  year = {2020},
  booktitle={IEEE Intl. Conf. on Intelligent Robots and Systems (IROS).},
  url = {https://moshanatucsd.github.io/orcvio_githubpage/},
  pdf = {https://arxiv.org/abs/2007.15107}
}

License

MIT License
Copyright (c) 2020 ERL at UCSD

References

• LARVIO
• MSCKF
• OpenVINS
• OpenVINS evaluation

orcvio's Issues

Can deep learning networks be integrated into C++?