The calculation results of the two versions PCL are inconsistent (PCL1.11.1 and PCL1.13.1)

[lzcchl]

Hi, There is a code with two results when using two PCL versions, please, is that something change?

• OS: WIN11
• Compiler: VS2019 and VS2022
• PCL Version: 1.11.1 and 1.13.1
• Environment: VS2019+PCL1.11.1 and VS2022+PCL1.13.1

code
#include

#include <Eigen/Core>

#include <opencv2/opencv.hpp>

#include <pcl/common/angles.h>
#include <pcl/common/impl/pca.hpp>
#include <pcl/common/common.h> // for getMinMax3D()
#include <pcl/common/eigen.h>
#include <pcl/common/centroid.h>
#include <pcl/common/geometry.h>
#include <pcl/point_types.h>
#include <pcl/point_cloud.h>
#include <pcl/io/ply_io.h>
#include <pcl/io/obj_io.h>
#include <pcl/segmentation/sac_segmentation.h> //基于采样一致性分割的类的头文件

int main()
{
using namespace std;
using namespace cv;
cout << "begin" << endl;

string obj_path = R"(E:\luozc\code\cpp\ovi\sstree\spine_fwsy\build\save\cylinder\AARJ33D0003_20230627150639940_99\cloudRgbPtr.ply)";
using T = pcl::PointXYZ;
pcl::PointCloud<T>::Ptr ptcld(new pcl::PointCloud<T>);
pcl::io::loadPLYFile(obj_path, *ptcld);

double dis = 4.0;
Eigen::Vector4f coeff_plane;

pcl::PointIndices::Ptr planeInd(new pcl::PointIndices);
pcl::ModelCoefficients::Ptr coefficients(new pcl::ModelCoefficients);
pcl::SACSegmentation<T> seg;
// Optional
seg.setOptimizeCoefficients(true);
// Mandatory
seg.setModelType(pcl::SACMODEL_PLANE);
seg.setMethodType(pcl::SAC_RANSAC);//pcl::SAC_RANSAC//pcl::LMedS
seg.setDistanceThreshold(dis);

seg.setInputCloud(ptcld);
seg.segment(*planeInd, *coefficients);
coeff_plane = { coefficients->values[0],coefficients->values[1],coefficients->values[2],coefficients->values[3] };
cout << coeff_plane << endl;

system("pause");
return 0;

}

this is the data.
cloudRgbPtr.zip

[mvieth]

@lzcchl It is possible that you get different results because something in PCL changed. It is also possible that you get different results because you used different compiler versions (VS2019 or VS2022).
What are the plane coefficients that are printed in each case? Are both a good fit for the plane in your point cloud?
Can you perhaps test with more PCL versions (PCL 1.12.0, PCL 1.12.1, PCL 1.13.0) and check with which PCL version the results changed?

[mvieth]

@lzcchl I investigated, pull request #4983 changed the result.
With PCL 1.11.1, I get these results:

0.00519627
  0.034637
  0.999386
  -671.016
330246 inliers

With PCL 1.13.1, I get these results:

0.00373465
 0.0342436
  0.999407
  -670.425
334283 inliers

Looking at the number of inliers, the results with PCL 1.13.1 are better. With PCL 1.11.1, the covariance matrix computation while refining/optimizing the plane coefficients was inaccurate because the points in your cloud are quite far away from the origin. This was fixed in the pull request I linked above. So use PCL 1.13.1, the results are more precise.