Fast Ground Segmentation of 3D Point Clouds

Ground plane fitting algorithm for fast ground removal in LAS/LAZ Point Clouds. A preprocessing step, where ground points removal significantly reduces the number of points involved in proceeding computations.

Pseudocode

This project is based on the pseudocode of the ground plane fitting algorithm proposed in the work of Zermas et al. The pseudocode is described below.

P: point cloud
num_iter_: number of iterations
num_lpr_: number of points used to estimate the LPR
th_seeds_: threshold points to be considered initial seeds
th_dist_: threshold distance of the plane

mainLoop(P):
    P_ground = extractInitialSeeds(P)
    for i in 1 : num_iter_:
        model = estimatePlane(P_ground);
        clear(P_ground, P_notground)
        for point p in PointCloud P:
            if model(p) < th_dist_:
                add p to P_ground；
            else
                add p to P_notground;
    return P_ground, P_notground

extractInitialSeeds(P):
    P_sorted = sortOnHeight(P)
    LPR_height = Average(P_sorted(1 : num_lpr_));
    clear(P_seeds)
    for i in 1 : P_sorted.size:
        if P_sorted[i].z < LPR_height + th_seeds_:
            add p_sorted[i] to P_seeds
    return P_seeds

Prerequisites

• Install LAStools:

  wget  http://lastools.github.io/download/LAStools.zip
  unzip LAStools.zip
  cd LAStools; mv LASlib/src/LASlib-config.cmake LASlib/src/laslib-config.cmake
  mkdir build; cd build
  cmake ..
  sudo make install
  

• Install PCL:

  sudo apt install libpcl-dev
  

Installation instructions

• Download the project.

  git clone https://github.com/Amsterdam-AI-Team/3D_Ground_Segmentation.git
  

• Build the project as follows.

  mkdir build  
  cd build
  cmake ..
  make
  

• Run the project.

  cd build
  ./ground_plane_fit
  

Example