gicp-python

The implement is totally from https://github.com/avsegal/gicp.

I warp c++ code to a python. Suppose I am in a conda environment.

Installation

1. install pybind11, gsl, pkg-config, boost

   conda install pybind11
   conda install pkg-config
   conda install libboost
   conda install gsl
   

2. compile ANN

   Then run the command

   cd ann_1.1.2
   make linux-g++
   

3. compile the project

   cd ..
   vim Makefile
   

   Modify the first line in ./Makefile. Change CONDA_PREFIX in line 1 to your conda prefix path. and run

   make
   

   then it should generate warp_gicp.so

Usage

import warp_gicp
T = warp_gicp.gicp(source, target, max_distance, gicp_epsilon)
'''
Args:
    source: np.ndarray (N1,3)
        source point cloud

    target: np.ndarray (N2,4)
        target point cloud

    max_distance: float, default = 1.0
        filter corresponding pairs over max_distance
    
    gicp_epsilon: float, default = 1e-3
         epsilon constant for gicp paper; this is NOT the convergence tolerence

Returns:
    T: np.ndarray (4, 4)
        T @ source_homo = target_homo, *_homo is homogeneous coordinates with shape (4, N)

There are many optional parameters in gicp.cpp, like max_iteration, epsilon , but I just use it's default setting.

You can run python test_with_python.py to check if it's nicely compiled.