目录
PCL双边滤波使用和结果显示
先用最简单的例子来说明PCL双边滤波如何使用
/*20201022 by 手口一斤*/
#include <pcl/point_types.h>
#include <pcl/io/pcd_io.h>
#include <pcl/kdtree/kdtree_flann.h>
#include <pcl/filters/bilateral.h>
#include <pcl/visualization/cloud_viewer.h>
typedef pcl::PointXYZI PointT;
int
main(int argc, char* argv[])
{
std::string incloudfile = "selfgen.pcd";
std::string outcloudfile ="testfilterssb.pcd";
float sigma_s = 1.0;
float sigma_r = 1.0;
// 读入点云文件
pcl::PointCloud<PointT>::Ptr cloud(new pcl::PointCloud<PointT>);
pcl::io::loadPCDFile(incloudfile.c_str(), *cloud);
pcl::PointCloud<PointT> outcloud;
// 建立kdtree
pcl::search::KdTree<PointT>::Ptr tree(new pcl::search::KdTree<PointT>);//用一个子类作为形参传入
pcl::BilateralFilter<PointT> bf;
bf.setInputCloud(cloud);
bf.setSearchMethod(tree);
bf.setHalfSize(sigma_s);
bf.setStdDev(sigma_r);
bf.filter(outcloud);
//结果点云显示
pcl::PointCloud<pcl::PointXYZI>::ConstPtr testpoint(&outcloud);
pcl::visualization::CloudViewer viewer("Cloud Viewer");
viewer.showCloud(cloud,"cloud");
viewer.showCloud(testpoint,"cloud filtered");
while (!viewer.wasStopped())
{
}
// 保存滤波输出点云文件
pcl::io::savePCDFile(outcloudfile.c_str(), outcloud);
return (0);
}
"selfgen.pcd"这个文件可以为自己的pcd文件,如果没有自己的点云文件,也可以看https://blog.csdn.net/h649070/article/details/111286598,中的genwrite函数得到一个正方体内的随机点生成的点云;
其中,需要注意,如果需要显示结果,需要对点云格式进行转换pcl::PointCloud<pcl::PointXYZI>::ConstPtr testpoint(&outcloud);
双边滤波实际计算过程(定义)
如果不会调参数,需要看双边滤波的原理,代码如下:
#include <pcl/point_types.h>
#include <pcl/io/pcd_io.h>
#include <pcl/kdtree/kdtree_flann.h>
typedef pcl::PointXYZI PointT;
float
G (float x, float sigma)
{
return std::exp (- (x*x)/(2*sigma*sigma));
}
int
main (int argc, char *argv[])
{
std::string incloudfile = argv[1];
std::string outcloudfile = argv[2];
float sigma_s = atof (argv[3]);
float sigma_r = atof (argv[4]);
// Load cloud
pcl::PointCloud<PointT>::Ptr cloud (new pcl::PointCloud<PointT>);
pcl::io::loadPCDFile (incloudfile.c_str (), *cloud);
int pnumber = (int)cloud->size ();
// Output Cloud = Input Cloud
pcl::PointCloud<PointT> outcloud = *cloud;
// Set up KDTree
pcl::KdTreeFLANN<PointT>::Ptr tree (new pcl::KdTreeFLANN<PointT>);
tree->setInputCloud (cloud);
// Neighbors containers
std::vector<int> k_indices;
std::vector<float> k_distances;
// Main Loop
for (int point_id = 0; point_id < pnumber; ++point_id)
{
float BF = 0;
float W = 0;
tree->radiusSearch (point_id, 2 * sigma_s, k_indices, k_distances);
// For each neighbor
for (std::size_t n_id = 0; n_id < k_indices.size (); ++n_id)
{
float id = k_indices.at (n_id);
float dist = sqrt (k_distances.at (n_id));
float intensity_dist = std::abs ((*cloud)[point_id].intensity - (*cloud)[id].intensity);
float w_a = G (dist, sigma_s);
float w_b = G (intensity_dist, sigma_r);
float weight = w_a * w_b;
BF += weight * (*cloud)[id].intensity;
W += weight;
}
outcloud[point_id].intensity = BF / W;
}
// Save filtered output
pcl::io::savePCDFile (outcloudfile.c_str (), outcloud);
return (0);
}
双边滤波适用场景和使用体验
一种非线性滤波,可以达到保持边缘的同时对图像进行平滑的效果;
经过实验,时间复杂度较高,大的点云不适合;
去噪效果效果需要根据实际点云情况,不是所有需要平滑和保留边缘的情况使用;
有两个参数需要调整,需要多次实验;
参考:https://pcl.readthedocs.io/projects/tutorials/en/latest/writing_new_classes.html#writing-new-classes