采用SAC-IA(采样一致性初始配准算法)进行粗匹配得到大概位置， 再结合ICP（迭代最近点算法(Iterative Cloest Point, ICP)）算法进行精确配准。 绿色是源点云，红色是目标点云，蓝色是配准之后的点云）   #include <pcl/registration/ia_ransac.h>#include <pcl/point_types.h>#include <pcl/point_cloud.h>#include <pcl/features/normal_3d.h>#include <pcl/features/fpfh.h>#include <pcl/search/kdtree.h>#include <pcl/io/pcd_io.h>#include <pcl/filters/voxel_grid.h>#include <pcl/filters/filter.h>#include <pcl/registration/icp.h>#include <pcl/visualization/pcl_visualizer.h>#include <time.h>using pcl::NormalEstimation;using pcl::search::KdTree;typedef pcl::PointXYZ PointT;typedef pcl::PointCloud<PointT> PointCloud;//点云可视化void visualize_pcd(PointCloud::Ptr pcd_src,    PointCloud::Ptr pcd_tgt,    PointCloud::Ptr pcd_final){       //int vp_1, vp_2;    // Create a PCLVisualizer object    pcl::visualization::PCLVisualizer viewer("registration Viewer");    //viewer.createViewPort (0.0, 0, 0.5, 1.0, vp_1);   // viewer.createViewPort (0.5, 0, 1.0, 1.0, vp_2);    pcl::visualization::PointCloudColorHandlerCustom<pcl::PointXYZ> src_h(pcd_src, 0, 255, 0);    pcl::visualization::PointCloudColorHandlerCustom<pcl::PointXYZ> tgt_h(pcd_tgt, 255, 0, 0);    pcl::visualization::PointCloudColorHandlerCustom<pcl::PointXYZ> final_h(pcd_final, 0, 0, 255);    viewer.addPointCloud(pcd_src, src_h, "source cloud");    viewer.addPointCloud(pcd_tgt, tgt_h, "tgt cloud");    viewer.addPointCloud(pcd_final, final_h, "final cloud");    //viewer.addCoordinateSystem(1.0);    while (!viewer.wasStopped())    {           viewer.spinOnce(100);        boost::this_thread::sleep(boost::posix_time::microseconds(100000));    }}//由旋转平移矩阵计算旋转角度void matrix2angle(Eigen::Matrix4f& result_trans, Eigen::Vector3f& result_angle){       double ax, ay, az;    if (result_trans(2, 0) == 1 || result_trans(2, 0) == -1)    {           az = 0;        double dlta;        dlta = atan2(result_trans(0, 1), result_trans(0, 2));        if (result_trans(2, 0) == -1)        {               ay = M_PI / 2;            ax = az + dlta;        }        else        {               ay = -M_PI / 2;            ax = -az + dlta;        }    }    else    {           ay = -asin(result_trans(2, 0));        ax = atan2(result_trans(2, 1) / cos(ay), result_trans(2, 2) / cos(ay));        az = atan2(result_trans(1, 0) / cos(ay), result_trans(0, 0) / cos(ay));    }    result_angle << ax, ay, az;}intmain(int argc, char** argv){       //加载点云文件    PointCloud::Ptr cloud_src_o(new PointCloud);//原点云，待配准    pcl::io::loadPCDFile("E:\\intern\\SAC-IA-master\\bun000_Structured.pcd", *cloud_src_o);    PointCloud::Ptr cloud_tgt_o(new PointCloud);//目标点云    pcl::io::loadPCDFile("E:\\intern\\SAC-IA-master\\bun045_Structured.pcd", *cloud_tgt_o);    clock_t start = clock();    //去除NAN点    std::vector<int> indices_src; //保存去除的点的索引    pcl::removeNaNFromPointCloud(*cloud_src_o, *cloud_src_o, indices_src);    std::cout << "remove *cloud_src_o nan" << endl;    //下采样滤波    pcl::VoxelGrid<pcl::PointXYZ> voxel_grid;    voxel_grid.setLeafSize(0.012, 0.012, 0.012);    voxel_grid.setInputCloud(cloud_src_o);    PointCloud::Ptr cloud_src(new PointCloud);    voxel_grid.filter(*cloud_src);    std::cout << "down size *cloud_src_o from " << cloud_src_o->size() << "to" << cloud_src->size() << endl;    //pcl::io::savePCDFileASCII("bunny_src_down.pcd", *cloud_src);    //计算表面法线    pcl::NormalEstimation<pcl::PointXYZ, pcl::Normal> ne_src;    ne_src.setInputCloud(cloud_src);    pcl::search::KdTree< pcl::PointXYZ>::Ptr tree_src(new pcl::search::KdTree< pcl::PointXYZ>());    ne_src.setSearchMethod(tree_src);    pcl::PointCloud<pcl::Normal>::Ptr cloud_src_normals(new pcl::PointCloud< pcl::Normal>);    ne_src.setRadiusSearch(0.02);    ne_src.compute(*cloud_src_normals);    std::vector<int> indices_tgt;    pcl::removeNaNFromPointCloud(*cloud_tgt_o, *cloud_tgt_o, indices_tgt);    std::cout << "remove *cloud_tgt_o nan" << endl;    pcl::VoxelGrid<pcl::PointXYZ> voxel_grid_2;    voxel_grid_2.setLeafSize(0.01, 0.01, 0.01);    voxel_grid_2.setInputCloud(cloud_tgt_o);    PointCloud::Ptr cloud_tgt(new PointCloud);    voxel_grid_2.filter(*cloud_tgt);    std::cout << "down size *cloud_tgt_o.pcd from " << cloud_tgt_o->size() << "to" << cloud_tgt->size() << endl;    pcl::io::savePCDFileASCII("bunny_tgt_down.pcd", *cloud_tgt);    pcl::NormalEstimation<pcl::PointXYZ, pcl::Normal> ne_tgt;    ne_tgt.setInputCloud(cloud_tgt);    pcl::search::KdTree< pcl::PointXYZ>::Ptr tree_tgt(new pcl::search::KdTree< pcl::PointXYZ>());    ne_tgt.setSearchMethod(tree_tgt);    pcl::PointCloud<pcl::Normal>::Ptr cloud_tgt_normals(new pcl::PointCloud< pcl::Normal>);    //ne_tgt.setKSearch(20);    ne_tgt.setRadiusSearch(0.02);    ne_tgt.compute(*cloud_tgt_normals);    //计算FPFH    pcl::FPFHEstimation<pcl::PointXYZ, pcl::Normal, pcl::FPFHSignature33> fpfh_src;    fpfh_src.setInputCloud(cloud_src);    fpfh_src.setInputNormals(cloud_src_normals);    pcl::search::KdTree<PointT>::Ptr tree_src_fpfh(new pcl::search::KdTree<PointT>);    fpfh_src.setSearchMethod(tree_src_fpfh);    pcl::PointCloud<pcl::FPFHSignature33>::Ptr fpfhs_src(new pcl::PointCloud<pcl::FPFHSignature33>());    fpfh_src.setRadiusSearch(0.05);    fpfh_src.compute(*fpfhs_src);    std::cout << "compute *cloud_src fpfh" << endl;    pcl::FPFHEstimation<pcl::PointXYZ, pcl::Normal, pcl::FPFHSignature33> fpfh_tgt;    fpfh_tgt.setInputCloud(cloud_tgt);    fpfh_tgt.setInputNormals(cloud_tgt_normals);    pcl::search::KdTree<PointT>::Ptr tree_tgt_fpfh(new pcl::search::KdTree<PointT>);    fpfh_tgt.setSearchMethod(tree_tgt_fpfh);    pcl::PointCloud<pcl::FPFHSignature33>::Ptr fpfhs_tgt(new pcl::PointCloud<pcl::FPFHSignature33>());    fpfh_tgt.setRadiusSearch(0.05);    fpfh_tgt.compute(*fpfhs_tgt);    std::cout << "compute *cloud_tgt fpfh" << endl;    //SAC配准    pcl::SampleConsensusInitialAlignment<pcl::PointXYZ, pcl::PointXYZ, pcl::FPFHSignature33> scia;    scia.setInputSource(cloud_src);    scia.setInputTarget(cloud_tgt);    scia.setSourceFeatures(fpfhs_src);    scia.setTargetFeatures(fpfhs_tgt);    //scia.setMinSampleDistance(1);    //scia.setNumberOfSamples(2);    //scia.setCorrespondenceRandomness(20);    PointCloud::Ptr sac_result(new PointCloud);    scia.align(*sac_result);    std::cout << "sac has converged:" << scia.hasConverged() << " score: " << scia.getFitnessScore() << endl;    Eigen::Matrix4f sac_trans;    sac_trans = scia.getFinalTransformation();    std::cout << sac_trans << endl;    //pcl::io::savePCDFileASCII("bunny_transformed_sac.pcd", *sac_result);    clock_t sac_time = clock();    //icp配准    PointCloud::Ptr icp_result(new PointCloud);    pcl::IterativeClosestPoint<pcl::PointXYZ, pcl::PointXYZ> icp;    icp.setInputSource(cloud_src);    icp.setInputTarget(cloud_tgt_o);    //Set the max correspondence distance to 4cm (e.g., correspondences with higher distances will be ignored)    icp.setMaxCorrespondenceDistance(0.04);    // 最大迭代次数    icp.setMaximumIterations(50);    // 两次变化矩阵之间的差值    icp.setTransformationEpsilon(1e-10);    // 均方误差    icp.setEuclideanFitnessEpsilon(0.2);    icp.align(*icp_result, sac_trans);    clock_t end = clock();    cout << "total time: " << (double)(end - start) / (double)CLOCKS_PER_SEC << " s" << endl;    //我把计算法线和点特征直方图的时间也算在SAC里面了    cout << "sac time: " << (double)(sac_time - start) / (double)CLOCKS_PER_SEC << " s" << endl;    cout << "icp time: " << (double)(end - sac_time) / (double)CLOCKS_PER_SEC << " s" << endl;    std::cout << "ICP has converged:" << icp.hasConverged()        << " score: " << icp.getFitnessScore() << std::endl;    Eigen::Matrix4f icp_trans;    icp_trans = icp.getFinalTransformation();    //cout<<"ransformationProbability"<<icp.getTransformationProbability()<<endl;    std::cout << icp_trans << endl;    //使用创建的变换对未过滤的输入点云进行变换    pcl::transformPointCloud(*cloud_src_o, *icp_result, icp_trans);    //保存转换的输入点云    //pcl::io::savePCDFileASCII("bunny_transformed_sac_ndt.pcd", *icp_result);    //计算误差    Eigen::Vector3f ANGLE_origin;    ANGLE_origin << 0, 0, M_PI / 5;    double error_x, error_y, error_z;    Eigen::Vector3f ANGLE_result;    matrix2angle(icp_trans, ANGLE_result);    error_x = fabs(ANGLE_result(0)) - fabs(ANGLE_origin(0));    error_y = fabs(ANGLE_result(1)) - fabs(ANGLE_origin(1));    error_z = fabs(ANGLE_result(2)) - fabs(ANGLE_origin(2));    cout << "original angle in x y z:\n" << ANGLE_origin << endl;    cout << "error in aixs_x: " << error_x << " error in aixs_y: " << error_y << " error in aixs_z: " << error_z << endl;    //可视化    visualize_pcd(cloud_src_o, cloud_tgt_o, icp_result);    return (0);}  报错解决： error C2079: “pcl::KdTreeFLANN::param_radius_”使用未定义的 struct“flann::SearchParams”  PCL和OpenCV库冲突，OpenCV的包含目录中opencv\build\include\opencv2\flann，解决方案是项目属性->VC++目录->包含目录中找到OpenCV的目录删除，或者在属性管理器中删除OpenCV的属性表，只留pcl的属性表。