Skip to content

Latest commit

 

History

History

PointCloudGenerator

README.md

Point Cloud Generator

Introduction

A set of Matlab scripts to generate point clouds with noise and outliers. The purpose of the scripts are:

  1. Apply a random rigid transformation
  2. Add noise
  3. Add outliers
  4. Save the point cloud (eventually the dataset)

Dependencies

  • Matlab R2015b
  • Computer Vision Toolbox

Usage

This mini-library aims to generate two or more point clouds to be aligned along with the ground truth; there are three example of usage

  1. generate two point clouds from a function describing a surface
  2. generate two point clouds starting from a pre-existing point cloud
  3. generate a dataset of point clouds starting from a pre-existing point cloud

Setup

To add all the required folder to the path, run setup_lib. If you want to save permanently the path run savepath.

Generate point cloud from a function

To generate a point cloud from a explicit 3D function, run the script MAIN_byFunction. Eventually, modify the first few lines (as below) according to your preferences.

%% Point Cloud Generation
% Sombrero function
f = @(X,Y) (10*sin(sqrt(X.^2 + Y.^2)) + 0.1) ./ sqrt(X.^2 + Y.^2);
x_space = -8:.25:8;
y_space = -8:.25:8;
sigma = 0.05; % noise
beta = 3/100; % outliers percentage

The result of the script generates two point clouds:

  • ptCloud_Q.pcd, the point cloud representing the surface
  • ptCloud_P.pcd, that is the point cloud Q transformed with a random rigid transformation T
  • T.txt, which contains the random transformation applied to point cloud Q.

Generate point cloud from an existing one

Sometimes it is necessary to take a pre-existing point cloud and add noise and outliers. The scriptMAIN_byPLY does precisely this. As before, modify the first few lines (as below) according to your preferences.

plyFile = 'bunny/reconstruction/bun_zipper_res3.ply';
sigma = 0.03; % noise
beta = 5/100; % outliers percentage

The result of this script is the same as the previous one, the only difference is that point cloud Q is the exact copy of the input point cloud.

Generate a dataset

In the root of this repository, you can find the Experiment python script; it is handy when you want do describe compactly and then execute on a batch of point clouds. The script MAIN_dataset generates a bunch of point clouds with noise, outliers and random transformation along with the relative dataset descriptor used from the python script.

To use it change the first lines of code according to your preferences:

plyFile_P = 'data/bunny/reconstruction/bun_zipper_res3.ply'; % moving
plyFile_Q = 'data/bunny/reconstruction/bun_zipper_res3.ply'; % fix
output_folder = 'bunny_noise'; % the output folder
output_basename = 'bunny'; %the basename of the folder containing
    % the point clouds associated with the current `value`level
    
dataset_multiplier = 20; % how many times the same noise level (or outlier
    % percentage) should be used?

variable = 'noise'; % possible values {'noise', 'outliers', ''}
% Note: outliers is expressed as percentage [0,1]
values = 0:0.0005:0.005; % noise
%values = 0:0.025:0.5; %outliers

% Random transformation parameters
max_rot = [ 2*pi, 2*pi, 2*pi ];
min_t = 'auto'; % a vector 3x1 or 'auto'
max_t = [ 5, 5, 5]; % if min_t is 'auto' thes are multipliers

This script will create (eventually overriding) the output folder as specified containing a sub-folder for each value specified for the variable. Inside each subfolder, you will find dataset_multiplier different pairs of point cloud P (a random rigid transformation of point cloud Q) and the transformation applied to it.

The output folder will contain the descriptor.json file to be used for the experiment descriptor and the fixed point cloud Q.

License

MIT Copyright (c) Pasquale Antonante