Point Cloud Registration Tool

Automatically registers (aligns) and visualizes point clouds, or processes a whole bunch at once

Main Features:

Coarse-to-fine registration (correspondence-based followed by ICP refinement)
PLY/OBJ input support
GUI
Batch processing
Parallelized
Outputs to disk:
- Registered cloud with residual colormap
- Transformation matrix
- F-score
- Histogram of residuals (in GUI mode)

Demo:

Each point is colored proportional to its distance to the closest point on the target cloud.

Color range of the residual colormap: RED = 0cm, BLUE = maximum residual threshold (default: 10cm)

Up/down arrow keys adjust the: maximum residual threshold.

Install:

Ubuntu 14.04:

If you already have ROS installed:

sudo apt-get install libpcl-1.7-all-dev

Else:

sudo add-apt-repository ppa:v-launchpad-jochen-sprickerhof-de/pcl
sudo apt-get update
sudo apt-get install libpcl-all

Ubuntu 16.04:

sudo apt-get update 
sudo apt-get install libpcl-dev

MacOS:

brew install pcl

More info at: Install PCL on MacOS via Homebrew

Build:

mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make

In MacOS, you may need to use a different compiler than the one shipped with Xcode to build with openMP,

to compile without openMP parallelization add: -DWITH_OPENMP=OFF

Example Usage:

The source cloud is the one we want to register to the target cloud

./point_cloud_registration_tool --verbose --gui source_cloud.ply target_cloud.ply

Some Flags:

--verbose, print information along the way
--gui, show the GUI
--batch_file, path to a text file containing paths to point cloud pairs

Batch File Format:

source_cloud_1.ply,target_cloud_1.ply
source_cloud_2.obj,target_cloud_2.obj
source_cloud_3.ply,target_cloud_3.obj

you can adjust many parameters of the registration routine, just ask --help

Tested on:

Ubuntu (PCL 1.7)
MacOS (PCL 1.8)

Minimum PCL 1.7 required

License

BSD 2-Clause License

Tool crashes after computing descriptors

I am trying to use the tool to automatically align two point clouds that have a lot of overlap. Unfortunately, I cannot release the point clouds for privacy reasons, but I can perform tests that may help to track down the source of the bug.

The command output looks like this:

Executing in parallel
Loaded point cloud with 240283 points:
Scan 3.ply
Loaded point cloud with 849421 points:
Scan 2.ply
Computed source cloud normals (240283)
Computed target cloud normals (849421)
Subsampled source cloud from 240283 -> 204804
Subsampled target cloud from 849421 -> 722911
Computed 204804 descriptors on source cloud keypoints
Segmentation fault (core dumped)

I ran the tool in GDB and got the following backtrace:

(gdb) bt
#0  __GI___libc_free (mem=0x9600d4) at malloc.c:3103
#1  0x00000000080698cb in boost::detail::sp_counted_impl_p<pcl::PointCloud<pcl::ReferenceFrame>>::dispose() ()
#2  0x000000000803c19a in boost::detail::sp_counted_base::release() ()
#3  0x000000000806cb67 in pcl::SHOTEstimationBase<pcl::PointXYZ, pcl::Normal, pcl::SHOT352, pcl::ReferenceFrame>::~SHOTEstimationBase() ()
#4  0x00000000080613d5 in Registrator::computeDescriptors() [clone ._omp_fn.4] ()
#5  0x00007ffff6c96c2e in ?? () from /usr/lib/x86_64-linux-gnu/libgomp.so.1
#6  0x00007ffffefd76db in start_thread (arg=0x7fffde150700) at pthread_create.c:463
#7  0x00007ffff678188f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
(gdb)

I am running this on Ubuntu LTS 18.04 on Windows 10. I am running VcXsrv in case the tool's --gui mode is trying to connect to an X server. The nature of the error makes me suspect that this is a logic error in the code, and not something specific to what platform it is running on.

What is my next step?

drethage / pointcloudregistrationtool Goto Github PK