autowareDoN点云过滤三

一级目录

differenceNormalsSegmentation

void differenceNormalsSegmentation(const pcl::PointCloud<pcl::PointXYZ>::Ptr in_cloud_ptr,pcl::PointCloud<pcl::PointXYZ>::Ptr out_cloud_ptr)


DoN参数

事先定义两个不同范围的支持半径用于向量计算

float small_scale &＃61; 0.5;float large_scale &＃61; 2.0;float angle_threshold &＃61; 0.5;


KD-TREE

根据点云类型&＃xff08;无序点云、有序点云&＃xff09;建立搜索树

pcl::search::Search<pcl::PointXYZ>::Ptr tree;if (in_cloud_ptr->isOrganized())//有序点云{tree.reset(new pcl::search::OrganizedNeighbor<pcl::PointXYZ>());} else{tree.reset(new pcl::search::KdTree<pcl::PointXYZ>(false));}// Set the input pointcloud for the search treetree->setInputCloud(in_cloud_ptr);


求解法线向量信息

开始法向量求解器。OpenMP标准并行估计每个3D点的局部表面属性。加入搜索树。关于NormalEstimationOMP 使用&＃xff0c;

//法向量估计求解器pcl::NormalEstimationOMP<pcl::PointXYZ, pcl::PointNormal> normal_estimation;// pcl::gpu::NormalEstimation normal_estimation;normal_estimation.setInputCloud(in_cloud_ptr);normal_estimation.setSearchMethod(tree);// 设置视点源点&＃xff0c;用于调整点云法向&＃xff08;指向视点&＃xff09;&＃xff0c;默认&＃xff08;0&＃xff0c;0&＃xff0c;0&＃xff09;normal_estimation.setViewPoint(std::numeric_limits<float>::max(), std::numeric_limits<float>::max(),std::numeric_limits<float>::max());


计算法线数据 normals_small_scale/ normals_large_scale

pcl::PointCloud<pcl::PointNormal>::Ptr normals_small_scale(new pcl::PointCloud<pcl::PointNormal>);pcl::PointCloud<pcl::PointNormal>::Ptr normals_large_scale(new pcl::PointCloud<pcl::PointNormal>);normal_estimation.setRadiusSearch(small_scale);normal_estimation.compute(*normals_small_scale);normal_estimation.setRadiusSearch(large_scale);normal_estimation.compute(*normals_large_scale);


DoN估计

定义法向量并绑定点云 法线信息
创建DoN估计器。得到DoN特征向量diffnormals_cloud

//定义法向量pcl::PointCloud<pcl::PointNormal>::Ptr diffnormals_cloud(new pcl::PointCloud<pcl::PointNormal>);pcl::copyPointCloud<pcl::PointXYZ, pcl::PointNormal>(*in_cloud_ptr, *diffnormals_cloud);
//DoN// Create DoN operatorpcl::DifferenceOfNormalsEstimation<pcl::PointXYZ, pcl::PointNormal, pcl::PointNormal> diffnormals_estimator;diffnormals_estimator.setInputCloud(in_cloud_ptr);diffnormals_estimator.setNormalScaleLarge(normals_large_scale);diffnormals_estimator.setNormalScaleSmall(normals_small_scale);diffnormals_estimator.initCompute();diffnormals_estimator.computeFeature(*diffnormals_cloud);


条件滤波

对于 pcl::ConditionOr
解释为

/** \brief Determine if a point meets this condition.
* \return whether the point meets this condition.
* The ConditionOr evaluates to true when ANY
* comparisons or nested conditions evaluate to true
*/

曲率curvature 大于阀值angle_threshold 即认为满足条件。博客
最后加入ConditionalRemoval中。这里应该是保留满足上述条件的法向量。得到过滤结果diffnormals_cloud_filtered注意这里得到的数据类型&＃xff0c;需要转点云

//条件滤波pcl::ConditionOr<pcl::PointNormal>::Ptr range_cond(new pcl::ConditionOr<pcl::PointNormal>());//加入比较阀值 GT 大于, GE大于等于, LT 小于, LE小于等于, EQ等于range_cond->addComparison(pcl::FieldComparison<pcl::PointNormal>::ConstPtr(new pcl::FieldComparison<pcl::PointNormal>("curvature", pcl::ComparisonOps::GT, angle_threshold)));// Build the filterpcl::ConditionalRemoval<pcl::PointNormal> cond_removal;cond_removal.setCondition(range_cond);cond_removal.setInputCloud(diffnormals_cloud);pcl::PointCloud<pcl::PointNormal>::Ptr diffnormals_cloud_filtered(new pcl::PointCloud<pcl::PointNormal>);// Apply filtercond_removal.filter(*diffnormals_cloud_filtered);


转换点云结构…out_cloud_ptr传出

pcl::copyPointCloud<pcl::PointNormal, pcl::PointXYZ>(*diffnormals_cloud, *out_cloud_ptr);


对于上述条件滤波做了几组实验&＃xff1a;可以看出阀值越小保留的点月多。可以推断出条件滤波保留了满足条件的点。

angle_threshold&＃61;0.9
angle_threshold&＃61;0.5
angle_threshold&＃61;0.1
angle_threshold&＃61;0.05

其他相关博客
https://blog.csdn.net/zhan_zhan1/article/details/104755582