/plane_loc_2

A ROS package with PlaneLoc2 - an open source project that provides a probabilistic framework for global localization using planar segments.

Primary LanguageC++

PlaneLoc2 ROS

A ROS package with PlaneLoc2 - an open source project that provides a probabilistic framework for global localization using planar segments.

Prerequesties:

  • Boost
  • Eigen
  • PCL 1.8
  • OpenCV >= 3.0
  • CGAL

v0.2

Improvements:

  • Changed descriptor distance to be relative to the length of the descriptor. This gives better generalization to different scenes.
  • Using area of segments expressed in pixels rather than square meteres when selecting best views.
  • Bugfixes.

v0.1

Described in the paper PlaneLoc2: Indoor Global Localization Using Planar Segments and Passive Stereo Camera.

Paper

If you find PlaneLoc2 useful in your academic work please cite the following paper:

@article{wietrzykowski2022,
  author={Wietrzykowski, Jan},
  journal={IEEE Access},
  title={{PlaneLoc2: Indoor Global Localization Using Planar Segments and Passive Stereo Camera}},
  year={2022},
  volume={10},
  number={},
  pages={67219-67229},
  doi={10.1109/ACCESS.2022.3185732}
 }

Building:

Tested on Ubuntu 20.04 and ROS Noetic.

  1. Install dependencies:

    sudo apt install libboost-system-dev libboost-filesystem-dev libboost-serialization-dev libopencv-dev libpcl-dev libcgal-dev ros-noetic-cv-bridge ros-noetic-sensor-msgs ros-noetic-pcl-conversions ros-noetic-tf2 ros-noetic-tf2-ros ros-noetic-tf2-msgs ros-noetic-tf2-eigen ros-noetic-image-transport libeigen3-dev

  2. Clone libnabo package:

    cd ~/catkin_ws/src
clone https://github.com/ethz-asl/libnabo.git
cd ..

  3. Clone and build PlaneLoc2:

    cd ~/catkin_ws/src
clone https://github.com/LRMPUT/plane_loc_2.git
cd ..
catkin catkin build plane_loc_2 -DCMAKE_BUILD_TYPE=Release

Installing bindings

cd ~/catkin_ws/src
pip install ./plane_loc

Usage

To run PlaneLoc2:

  • Process dataset with Stereo Plane R-CNN and export detection and descriptors.
  • Pack exported information into a rosbag using pack_rosbag_planercnn.py script.
  • Adjust paths in MapperBagNode.cpp to a proper rosbag path and choose between building a global map and creating a rosbag with local maps.
  • Select a value of acc_duration in mapper.launch. Use 2 seconds for local maps and infinity for a global map.
  • Build a global map and create a rosbag with local maps.
  • Adjust paths in LocalizerBagNode.cpp to a proper local maps rosbag.
  • Adjust a global map path and camera parameters in settings.yml.

Dataset

The SceneNet Stereo dataset can be downloaded here.

The TERRINet dataset can be downloaded here.

Nodes

mapper_node

Mapper node processes detected planar segments from consecutive frames and merges them into a larger local map using odometry transformation from TF2.

Subscribed topics

  • objs_topic (plane_loc/Serialized) - detected planar segments as a serialized vectorObjInstance C++ type. There should be TF2 transformation available from odom frame to frame specified in the message header at the time specified in the header.

Published topics

  • map (plane_loc/Serialized) - accumulated maps as a serialized Map C++ type.

Parameters

  • settings (string, default: settings.yaml) - path to the settings YAML file.
  • objs_topic (string, default: /detector/objs) - name of the topic with detected planar segments.
  • acc_frames (int, default: 50) - number of frames that are accumulated before publishing the map.
  • merge_map_frames (int, default: 50) - number of frames after which additional check of all planar segments is performed and those that fulfill conditions for merging are merged.

localizer_node

Localizer node subscribes accumulated local maps and performs global localization against the map specified in the settings YAML file. It evaluates performance using TF2 transformations from map frame.

Subscribed topics

  • map_topic (plane_loc/Serialized) - accumulated local maps as a serialized Map C++ type.

Published topics

  • pose (geometry_msgs/TransformStamped) - computed pose of the sensor at the time specified in the header of map message. Published only if the transformation was considered as correct.

Parameters

  • settings (string) - path to the settings YAML file.
  • map_topic (string, default: /mapper/map) - name of the topic with detected planar segments.
  • score_thresh (float, default: 1.0) - minimal value of score to consider a transformation as correct.
  • score_diff_thresh (float, default: 0.0) - minimal value of score difference between the best and the second best transformation to consider the first one as correct.
  • fit_thresh (float, default: 0.07) - maximal value of fitness test to consider a transformation as correct.
  • distinct_thresh (int, default: 6) - minimal number of distinct matched planar segments to consider a transformation as correct.
  • pose_diff_thresh (float, default: 0.16) - maximal value of squared SE3 logarithm of difference between computed and ground truth pose to count the result as positive (used only for evaluation).

Launching

The package can be launched using the provided launch file:

roslaunch plane_loc plane_loc.launch

Make sure you slow down rosbag play using r flag, so the system can keep up with computation:

rosbag play /mnt/data/datasets/PUT_Indoor/2017_04_04_test1/data.bag --pause -r 0.02

CLion configuration

CMake options:

-DCATKIN_DEVEL_PREFIX:PATH=/home/lrm/catkin_ws/devel

env variables:

CMAKE_PREFIX_PATH=/home/lrm/catkin_ws/devel:/opt/ros/noetic;PYTHONPATH=/home/lrm/catkin_ws/devel/lib/python3/dist-packages:/opt/ros/noetic/lib/python3/dist-packages

Run configuration arguments:

__name:=mapper

env variables:

ROS_NAMESPACE=/;LD_LIBRARY_PATH=/home/lrm/catkin_ws/devel/lib:/opt/ros/noetic/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64;CMAKE_PREFIX_PATH=/home/lrm/catkin_ws/devel:/opt/ros/noetic;ROS_PACKAGE_PATH=/home/lrm/catkin_ws/src/plane_loc:/opt/ros/noetic/share;DISPLAY=:1

To enable X server in debug mode add to ~/.gdbinit:

set environment DISPLAY :1