Hybrid Point Cloud Registration approach for global localization

Table of Contents

• 1. Introduction
• 2. Requirements
• 3. ROS API
• 4. Running ROS package with docker
• 5. Running ROS package without docker
• 6. Running ROS package with prerecorded dataset
• 7. Getting Started without ROS

1. Introduction

Hybrid approach of Point Cloud Registration consists of main parts: downsampling cascade, WHI feature descriptor, fast multidimensional nearest neighbors graph-search, Max Clique Inlier Selection, transformation estimator of Fast Global Registration algorithm. It also include options to use FPFH feature descriptor, transformation estimator of Teaser++ and Plane-to-Plane ICP cascade for registration refinement. This approach allows for indoor global localization of devices that build a spatial map.

2. Requirements

• ubuntu 20.04
• ros noetic

3. ROS API

Subscribed Topics

• /global_map (sensor_msgs/PointCloud2): Topic of global map.
• /local_map (sensor_msgs/PointCloud2): Topic of current map.

Published Topics

• /tf2 (geometry_msgs/TransformStamped): Tf2 publish frames: frame_id -> child_id

Services

• /relocalize (std_srvs/Trigger): Service to manually run relocalization.

Parameters

№	Name	Type	Default value	Description
1	tf_pub_rate	float	200	Frame rate of TF publiser
2	parent_frame_id	string	map	Name of parent TF
3	child_frame_id	string	map_slam	Name of child TF
4	mode_pub_by_request	bool	true	True means performing transform and publication only by a request. False means performing transform and publication automatically when receive a new data. The node publishes TF with fixed rate in any mode.

4. Running ROS package with docker

The easiest way to play with the package is to use docker. You can either build and run it manually, or use docker.sh script.

Typically, you need to run one of following commands with this script:

• ./docker.sh build to build a docker image
• ./docker.sh run to run it
• ./docker.sh rviz to run it with rviz
• ./docker.sh help to get additional info

5. Running ROS package without docker

Your installation process might be base don Dockerfile.

The core of the package is pc_global_localization_node node. To use it, you should load ros parameters.

The best way to use this node is to either use launch/run_pc_global_localization.launch file or to write your own.

6. Running ROS package with prerecorded dataset

Let's say you have 2 pcd files in your folder named Map.pcd and GlobalMap.pcd.

You could play with this package by running following commands in different terminals:

./docker.sh rviz

rosrun pcl_ros pcd_to_pointcloud Map.pcd 1.0        _frame_id:=map cloud_pcd:=/local_map __name:=my_node2

rosrun pcl_ros pcd_to_pointcloud GlobalMap.pcd 1.0  _frame_id:=map_slam cloud_pcd:=/global_map __name:=my_node1

7. Getting Started without ROS

7.1. Dependencies installation on Ubuntu/Debian:

1. Installation of Point Cloud Library (PCL):

$ sudo apt install libpcl-dev

2. Dependence for external WHI feature library:

$ sudo apt install libopencv-dev

3. Support to use multithreading:

$ sudo apt install libopenmpi-dev

7.2. Build on Ubuntu/Debian

From root folder:

$ mkdir build 
$ cd build
$ cmake ..
$ make

7.3. Running on Ubuntu/Debian

From build folder:

$  ./PCRegistrationAlgorithmTest <target_point_cloud> <source_point_cloud>

• <target_point_cloud> and <source_point_cloud> are the point cloud files of pcd format.