srinivasrama / map_merge_3d

ROS package for merging multiple 3D point cloud maps. Includes octomap occupancy map generation capability.

map_merge_3d

ROS package for merging multiple 3D point cloud maps. Includes octomap occupancy map generation capability.

Installing

The following ROS packages are required for this map-merge package:

• octomap_ros
• octomap_msgs
• fast_gicp

The package is released for ROS Melodic. Build it through catkin_make process. git clone it to your catkin_ws/src folder. After that, build the package through catkin_make

cd ~/catkin_ws/src
git clone https://github.com/leonardlohky/map_merge_3d
cd ..
catkin_make

Building

The package should build as a standard catkin package. Use rosdep to resolve dependencies in ROS. The package is intended for ROS Melodic and newer, it should build on all supported platforms of ROS Melodic. Most notably, the package depends on PCL >= 1.8.

Master branch is for the latest ROS.

Wiki

The package's original documentation by the original author can be found at ROS Wiki.

• map_merge_3d

Execution

Map Merger

The package contains two executable ROS nodes:

• map_merger_node: For merging of point cloud maps
• octomap_mapper_node: To generate octomap occupancy map from the merged point cloud

The map merger process used for this package is illustrated below:

A template launch file can be found under launch/map_merge.launch, modify it accordingly to suit the application. To execute the process, simply run the launch file. Remember to source for the workspace if you haven't.

roslaunch map_merge_3d map_merge.launch

Registration Visualization

The registration_visualization.cpp executable helps to visualises pair-wise transform estimation between 2 maps. It uses PCL visualiser for the visualisation and can be executed via the following line of code:

rosrun map_merge_3d registration_visualisation [--param value] map1.pcd map2.pcd

// Example
rosrun map_merge_3d registration_visualisation [--keypoint_type HARRIS --normal_radius 0.3 --filter_z_min 0.3 --filter_z_max 5.0 --keypoint_threshold 0.005] map1.pcd map2.pcd 

Parameters

Most of the parameter descriptions can be found in the original ROS wiki entry. Due to additional parameters not found in the original package, a complete list of parameters for this map merge package is seen below:

Parameter Name	Meaning	Values
robot_map_topic	Name of robot map topic without namespaces	string, default: map
robot_namespace	Fixed part of the robot map topic. Only topics which contain (anywhere) this string are considered for lookup	string, default: <empty string>
merged_map_topic	Topic name where merged map is published	string, default: map
world_frame	Frame id (in tf tree) which is assigned to published merged map and used as reference frame for tf transforms	string, default: world
compositing_rate	Rate in Hz. Basic frequency on which the node merges maps and publishes merged map	double, default: 0.3
discovery_rate	Rate in Hz. Frequency on which this node discovers new robots (maps)	double, default: 0.05
estimation_rate	Rate in Hz. Frequency on which this node re-estimates transformations between maps	double, default: 0.01
publish_tf	Whether to publish estimated transforms in the tf tree	bool, default: true
resolution	Resolution used for the registration.	double, default: 0.1
descriptor_radius	Radius for descriptors computation	double, default: resolution * 8.0
outliers_min_neighbours	Minimum number of neighbours for a point to be kept in the map during outliers pruning	int, default: 100
normal_radius	Radius used for estimating normals	double, default: resolution * 6.0
keypoint_type	Type of keypoints used. Possible values are SIFT, HARRIS, ISS, NARF	string, default: HARRIS
keypoint_threshold	Keypoints with lower response that this value are pruned	double, default: 5.0
descriptor_type	Type of descriptors used. Possible values are PFH, FPFH, RSD, SHOT, SC3D	string, default: FPFH
refine_method	Method to refine initial estimated coarse transform. Possible values are ICP, FAST_GICP, FAST_VGICP	string, default: FAST_VGICP
estimation_method	Type of descriptors matching algorithm used. Possible values are MATCHING, SAC_IA, NDT	string, default: MATCHING
correspondence_method	Method to find correspondence points. Only applicable for MATCHING estimation method. Possible values are KDTREE, RECIPROCAL	string, default: KDTREE
refine_transform	Whether to refine estimated transformation or not	bool, default: true
inlier_threshold	Inlier threshold used in RANSAC during estimation	double, default: resolution * 5.0
max_correspondence_distance	Maximum distance for matched points to be considered the same point	double, default: inlier_threshold * 2.0
max_iterations	Maximum iterations for RANSAC	int, default: 100
matching_k	Number of the nearest descriptors to consider for matching	int, default: 5
transform_epsilon	The smallest change allowed until ICP convergence	double, default: 1e-2
confidence_threshold	Minimum confidence in the pair-wise transform estimate to be included in the map-merging graph. Pair-wise transformations with lower confidence are not considered when computing global transforms	double, default: 10.0
output_resolution	Resolution of the merged global map	double, default: 0.05
reg_resolution	Resolution for selected refine method	double, default: 1.5
reg_step_size	Step size for selected refine method	double, default: 0.1
filter_z_max	Max point z-height cutoff filter for input pointcloud	double, default: inf
filter_z_min	Min point z-height cutoff filter for input pointcloud	double, default: -inf
reference_frame	Method to determine which node will be used as the global reference frame. Possible values are FIRST, AUTO. FIRST will use the very first node, while AUTO will try to determine the node automatically. Note that the map will jump if AUTO is used	string, default: AUTO

Troubleshooting

While running the package, there is a chance that the node will crash with the following error message:

[pcl::PFHEstimation::compute] input_ is empty!
[pcl::PFHEstimation::initCompute] Init failed.
[pcl::PFHEstimation::compute] input_ is empty!
[pcl::PFHEstimation::initCompute] Init failed.
[pcl::PFHEstimation::compute] input_ is empty!
[pcl::PFHEstimation::initCompute] Init failed.

This is most likely due to the lack of keypoints, or lack of descriptors. E.g., if your point cloud does not have a RGB value, then using SIFT as the keypoint type WILL NOT work and results in this error. Thus, a solution is to switch to HARRIS as the keypoint type.

Copyright

The package is licensed under BSD license. See respective files for details.

Acknowledgement

This package is based on the original map_merge_3d package by hrnr.