目的
想使用一下CloudCompare自带的八叉树,替换掉nanoflann。
nanoflann返回值出现异常
用于最邻近搜索,在radiussearch下,如果能找到最近点,那么确实找到的是最近点,如果找不到最近点,那么就会出现问题,如下:
nanoflann::KdTreeFLANN<PB::pmPointXYZ>::Ptr kdtree = std::make_shared<nanoflann::KdTreeFLANN<PB::pmPointXYZ>>();
if (!kdtree) {
std::cerr << "KD tree build failed!" << std::endl;
return;
}
kdtree->setInputCloud(cloud);
for (int i = 0; i < checkcorrs.size(); ++i) {
std::vector<int> searchIndexes;
std::vector<SCALAR_TYPE> distances;
SCALAR_TYPE *search_pt = new SCALAR_TYPE[3];
search_pt[0] = checkcorrs[i].x;
search_pt[1] = checkcorrs[i].y;
search_pt[2] = checkcorrs[i].z;
kdtree->radiusSearch(search_pt, 2.0, searchIndexes, distances);
}
在for循环中,每次更新query的点坐标,然后试图从点云中返回其最近的2.0m范围内的点,如果query点周围没有符合条件的点,那么就会返回上一次for循环的值。详见issue。
使用CloudCompare自带的Octree
- 核心为直接接入cc内部的octree实现最近点的索引,位于
#include <DgmOctree.h>
,这样检索速度大大加快。过程如下:
- 首先找到当前选中哪些entity,此时需要引用主窗口的头文件并获取全局静态变量
#include <mainwindow.h>
;
const ccHObject::Container &selectedEntities = MainWindow::TheInstance()->getSelectedEntities();
- 遍历每个选中的entity,并将其转换为点云类型,需要
#include <ccHObjectCaster.h>
头文件;
for (ccHObject* ent : selectedEntities)
{
ccPointCloud* pc = ccHObjectCaster::ToPointCloud(ent);
}
- 获取ccPointCloud点云类中的八叉树索引,如果未构建,则重新构建;
ccOctree::Shared octree = pc->getOctree();
if (!octree)
{
octree = pc->computeOctree();
if (!octree)
{
ccLog::LogMessage(QString::fromLocal8Bit("索引构建失败ÿ