pcl点云匹配ICP、NICP、NDT用法实例C++

提示：文章写完后，目录可以自动生成，如何生成可参考右边的帮助文档

---

前言

pcl常见点云匹配方法使用方式，包括ICP、NICP、NDT。若想了解大致原理，可参考该地址内容：https://blog.csdn.net/zzr1024/article/details/118567649

---

一、ICP

#include <pcl/registration/icp.h>

	int iterations = 200;
	//time.tic();
	pcl::IterativeClosestPoint<pcl::PointXYZ, pcl::PointXYZ> icp;
	icp.setMaximumIterations(iterations);
	icp.setInputSource(cloud_src);
	icp.setInputTarget(cloud_tar);
	icp.setTransformationEpsilon(1e-8);
	icp.setEuclideanFitnessEpsilon(1e-5);
	icp.setRANSACOutlierRejectionThreshold(2);
	icp.setMaxCorrespondenceDistance(2); //添加一个最大距离的阈值，超过最大距离的点不作为对应点。

	// 输出配准后点云
	icp.align(*cloud_icp);

	if (icp.hasConverged()) {
		std::cout << "\nICP has converged, score is " << icp.getFitnessScore() << std::endl;
		std::cout << "\nICP transformation " << iterations << " : cloud_icp -> cloud_in" << std::endl;
		transformation_matrix = icp.getFinalTransformation().cast<double>();
		//print4x4Matrix(transformation_matrix);
		std::cout << std::endl;
		std::cout << "transformation_matrix\n" << transformation_matrix << std::endl;
	}
	else {
		PCL_ERROR("\nICP has not converged.\n");
	}

二、NICP

#include <pcl/registration/icp.h>

	int iterations = 200;
	// ICP 参数设置
	pcl::IterativeClosestPointWithNormals<pcl::PointNormal, pcl::PointNormal> nicp;
	nicp.setInputSource(cloud_src);
	nicp.setInputTarget(cloud_tar);
	nicp.setMaxCorrespondenceDistance(5);
	nicp.setTransformationEpsilon(1e-10);
	nicp.setEuclideanFitnessEpsilon(1e-5);
	nicp.setMaximumIterations(iterations);

	// ICP 迭代
	nicp.align(*cloud_nicp);

	if (icp.hasConverged()) {
		std::cout << "\nNICP has converged, score is " << nicp.getFitnessScore() << std::endl;
		std::cout << "\NICP transformation " << iterations << " : cloud_nicp -> cloud_in" << std::endl;
		transformation_matrix = nicp.getFinalTransformation().cast<double>();
		//print4x4Matrix(transformation_matrix);
		std::cout << std::endl;
		std::cout << "transformation_matrix\n" << transformation_matrix << std::endl;
	}
	else {
		PCL_ERROR("\nNICP has not converged.\n");
	}

---

三、NDT

#include <pcl/registration/ndt.h>

	int iterations = 200;
	//初始化正态分布变换（NDT）
	pcl::NormalDistributionsTransform<pcl::PointXYZ, pcl::PointXYZ> ndt;
	//设置依赖尺度NDT参数
	//为终止条件设置最小转换差异
	ndt.setTransformationEpsilon(1e-8);
	//为More-Thuente线搜索设置最大步长
	ndt.setStepSize(4);
	//设置NDT网格结构的分辨率（VoxelGridCovariance）
	ndt.setResolution(3);
	//设置匹配迭代的最大次数
	ndt.setMaximumIterations(iterations);
	// 设置要配准的点云
	ndt.setInputCloud(cloud_src);
	//设置点云配准目标
	ndt.setInputTarget(cloud_tar);

	ndt.align(*cloud_ndt);

	if (ndt.hasConverged()) {
		std::cout << "\nNDT has converged, score is " << ndt.getFitnessScore() << std::endl;
		std::cout << "\nNDT transformation " << iterations << " : cloud_ndt -> cloud_in" << std::endl;
		transformation_matrix = ndt.getFinalTransformation().cast<double>();
		//print4x4Matrix(transformation_matrix);
		std::cout << std::endl;
		std::cout << "transformation_matrix\n" << transformation_matrix << std::endl;
	}
	else {
		PCL_ERROR("\nNDT has not converged.\n");
	}

---

总结

总结了下pcl常用的点云匹配方法，后续遇到新的再继续更新。
总体上NDT耗时会长很多，效果上三者孰优孰劣不好说，听说在开源领域，NICP是效果最好的ICP匹配方法。