pcl的kdtree与nanoflann速度比较

最新推荐文章于 2024-10-10 07:36:44 发布

Rain-jay

最新推荐文章于 2024-10-10 07:36:44 发布

阅读量2.4k

点赞数 4

分类专栏：三维重建

本文链接：https://blog.csdn.net/weixin_43236428/article/details/118522018

版权

三维重建专栏收录该内容

12 篇文章

订阅专栏

这篇博客对比了PCL库和nanoflann库在进行10万点云数据的KdTree搜索时的效率。结果显示，nanoflann在执行相同任务时比PCL的KdTree快约33毫秒，表明nanoflann在搜索性能上有优势。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

typedef KDTreeSingleIndexAdaptor<L2_Simple_Adaptor<double, PointCloud_flann<double> >, PointCloud_flann<double>, 3> my_kd_tree;
pcl::PointCloud<pcl::PointXYZ>::Ptr p_cloud(new pcl::PointCloud<pcl::PointXYZ>);
// pcl kdtree
pcl::KdTreeFLANN<pcl::PointXYZ>::Ptr pcltree(new pcl::KdTreeFLANN<pcl::PointXYZ>);
pcltree->setInputCloud(p_cloud);
double SearchRadius = 0.3;
float ppp = clock();
for (int i = 0; i < 100000; i++)
{
	std::vector<int> k_indices; std::vector<float> distance;
	pcltree->radiusSearch(p_cloud->at(i), SearchRadius, k_indices, distance);
}	

// nanoflann tree
PointCloud_flann<double> cloud_flann;
cloud_flann.pts.resize(p_cloud->size());
for (int i = 0; i < p_cloud->size(); i++)
{
	cloud_flann.pts[i].x = p_cloud->points[i].x;
	cloud_flann.pts[i].y = p_cloud->points[i].y;
	cloud_flann.pts[i].z = p_cloud->points[i].z;
}
my_kd_tree kd_tree(3, cloud_flann, KDTreeSingleIndexAdaptorParams(10));
kd_tree.buildIndex();
const double query_radius_sq = 0.3*0.3;
for (int i = 0; i < 100000; i++)
{
	double query_p[3] = { cloud_flann.pts[i].x, cloud_flann.pts[i].y, cloud_flann.pts[i].z };
	std::vector<std::pair<size_t, double>> matches;
	nanoflann::SearchParams searchparams;
	kd_tree.radiusSearch(query_p, query_radius_sq, matches, searchparams);
}