NICP (Normal Iterative Closest Point)
NICP is a novel on-line method to recursively align point clouds. This method exploits the 3D structure to determine the data association between the two clouds taking into account each point and its local features of the surface: normals and curvature. This method adopt a line of sight criterion to find the corresponding points between the two clouds to register. This, together with the efficient algorithms and data structures used by NICP, increase the speed of the method allowing real-time computation. NICP solves the alignment problem by casting a least squares formulation that minimizes an error metric depending on both the point coordinates and the associated normal. This renders the algorithm more robust and accurate, thus computing better transformation.
Tutorials and much more @ http://goo.gl/W3qXbE
System Requirements
Minimum
- CMake >= 3.5
- Eigen3 >= 3.2.0
- OpenCV >= 3.2.0
- Flann >= 1.7
Complete
- OpenGL
- Qt5
- QGLViewer
Building on Linux (Ubuntu 18.04 LTS)
$> sudo apt install git cmake libeigen3-dev libsuitesparse-dev qtdeclarative5-dev qt5-qmake libqglviewer-dev-qt5 libqglviewer2-qt5 libqglviewer-headers libflann-dev libopencv-dev freeglut3-dev
$> git clone https://github.com/yorsh87/nicp.git
$> cd nicp
$> mkdir build
$> cd build
$> cmake ..
$> make
Branches
- master : current stable branch
- develop: current development branch
- iros2015_experiments: code snapshot used for IROS 2015 paper publication [pdf]
- ras2016_experiments: code snapshot used for RAS 2017 paper publication [pdf]
Examples
Once you compiled the code you will have the following exmaple binaries:
nicp_simple_aligneris a binary that, given a set of depth images and a .txt file containing the list of depth images to align, perform the point cloud registrations