EuclideanClusterExtraction这个名字起的很奇怪，欧式距离聚类这个该如何理解？欧式距离只是一种距离测度的方法呀！有了一个Cluster在里面，我以为是某一种聚类算法，层次聚类？k-NN聚类？K-Means？还是模糊聚类？感觉很奇怪，看下代码吧。
找一个实例cluster_extraction.cpp的main入口函数。
找到computer函数，该方法中定义了一个pcl::EuclideanClusterExtraction<pcl::PointXYZ> ec;对象，接着就是参数赋值。
具体的算法执行在ec.extract (cluster_indices)中。因此进入EuclideanClusterExtraction查看具体的方法。
void compute (const pcl::PCLPointCloud2::ConstPtr &input, std::vector<pcl::PCLPointCloud2::Ptr> &output,
int min, int max, double tolerance)
{
// Convert data to PointCloud<T>
PointCloud<pcl::PointXYZ>::Ptr xyz (new PointCloud<pcl::PointXYZ>);
fromPCLPointCloud2 (*input, *xyz); // Estimate
TicToc tt;
tt.tic (); print_highlight (stderr, "Computing "); // Creating the KdTree object for the search method of the extraction
pcl::search::KdTree<pcl::PointXYZ>::Ptr tree (new pcl::search::KdTree<pcl::PointXYZ>);
tree->setInputCloud (xyz); std::vector<pcl::PointIndices> cluster_indices;
pcl::EuclideanClusterExtraction<pcl::PointXYZ> ec;
ec.setClusterTolerance (tolerance);
ec.setMinClusterSize (min);
ec.setMaxClusterSize (max);
ec.setSearchMethod (tree);
ec.setInputCloud (xyz);
ec.extract (cluster_indices); print_info ("[done, "); print_value ("%g", tt.toc ()); print_info (" ms : "); print_value ("%d", cluster_indices.size ()); print_info (" clusters]\n"); output.reserve (cluster_indices.size ());
for (std::vector<pcl::PointIndices>::const_iterator it = cluster_indices.begin (); it != cluster_indices.end (); ++it)
{
pcl::ExtractIndices<pcl::PCLPointCloud2> extract;
extract.setInputCloud (input);
extract.setIndices (boost::make_shared<const pcl::PointIndices> (*it));
pcl::PCLPointCloud2::Ptr out (new pcl::PCLPointCloud2);
extract.filter (*out);
output.push_back (out);
}
}
可以在pcl_segmentation项目下的extract_clusters.h文件中查看EuclideanClusterExtraction的定义，可知这是一个模板类。
template <typename PointT> class EuclideanClusterExtraction: public PCLBase<PointT>
实现文件在项目下的extract_clusters.hpp中，（还有一个extract_clusters.cpp文件），查看其中的extract方法：
template <typename PointT> void pcl::EuclideanClusterExtraction<PointT>::extract (std::vector<PointIndices> &clusters)
{
if (!initCompute () ||
(input_ != && input_->points.empty ()) ||
(indices_ != && indices_->empty ()))
{
clusters.clear ();
return;
} // Initialize the spatial locator
if (!tree_)
{
if (input_->isOrganized ())
tree_.reset (new pcl::search::OrganizedNeighbor<PointT> ());
else
tree_.reset (new pcl::search::KdTree<PointT> (false));
} // Send the input dataset to the spatial locator
tree_->setInputCloud (input_, indices_);
extractEuclideanClusters (*input_, *indices_, tree_, static_cast<float> (cluster_tolerance_), clusters, min_pts_per_cluster_, max_pts_per_cluster_);
//tree_->setInputCloud (input_);
//extractEuclideanClusters (*input_, tree_, cluster_tolerance_, clusters, min_pts_per_cluster_, max_pts_per_cluster_); // Sort the clusters based on their size (largest one first)
std::sort (clusters.rbegin (), clusters.rend (), comparePointClusters); deinitCompute ();
}
注意在extract_clusters.h中定义了四个
template <typename PointT> void
extractEuclideanClusters (
const PointCloud<PointT> &cloud, const boost::shared_ptr<search::Search<PointT> > &tree,
float tolerance, std::vector<PointIndices> &clusters,
unsigned int min_pts_per_cluster = , unsigned int max_pts_per_cluster = (std::numeric_limits<int>::max) ());
template <typename PointT> void
extractEuclideanClusters (
const PointCloud<PointT> &cloud, const std::vector<int> &indices,
const boost::shared_ptr<search::Search<PointT> > &tree, float tolerance, std::vector<PointIndices> &clusters,
unsigned int min_pts_per_cluster = , unsigned int max_pts_per_cluster = (std::numeric_limits<int>::max) ());
template <typename PointT, typename Normal> void
extractEuclideanClusters (
const PointCloud<PointT> &cloud, const PointCloud<Normal> &normals,
float tolerance, const boost::shared_ptr<KdTree<PointT> > &tree,
std::vector<PointIndices> &clusters, double eps_angle,
unsigned int min_pts_per_cluster = ,
unsigned int max_pts_per_cluster = (std::numeric_limits<int>::max) ())
template <typename PointT, typename Normal>
void extractEuclideanClusters (
const PointCloud<PointT> &cloud, const PointCloud<Normal> &normals,
const std::vector<int> &indices, const boost::shared_ptr<KdTree<PointT> > &tree,
float tolerance, std::vector<PointIndices> &clusters, double eps_angle,
unsigned int min_pts_per_cluster = ,
unsigned int max_pts_per_cluster = (std::numeric_limits<int>::max) ())
前两个的方法实现在文件extract_clusters.hpp中，后两个直接在头文件中就以内联函数的形式实现了，两个大同小异。择其中第一个加点注释，发现其实是采用的区域生长算法实现的分割。理解错误了，区域生长需要种子点，这里应该叫层次聚类方法。
template <typename PointT, typename Normal> void
extractEuclideanClusters (
const PointCloud<PointT> &cloud, const PointCloud<Normal> &normals,
float tolerance, const boost::shared_ptr<KdTree<PointT> > &tree,
std::vector<PointIndices> &clusters, double eps_angle,
unsigned int min_pts_per_cluster = ,
unsigned int max_pts_per_cluster = (std::numeric_limits<int>::max) ())
{
if (tree->getInputCloud ()->points.size () != cloud.points.size ())
{
PCL_ERROR ("[pcl::extractEuclideanClusters] Tree built for a different point cloud dataset (%lu) than the input cloud (%lu)!\n", tree->getInputCloud ()->points.size (), cloud.points.size ());
return;
}
if (cloud.points.size () != normals.points.size ())
{
PCL_ERROR ("[pcl::extractEuclideanClusters] Number of points in the input point cloud (%lu) different than normals (%lu)!\n", cloud.points.size (), normals.points.size ());
return;
} // Create a bool vector of processed point indices, and initialize it to false
std::vector<bool> processed (cloud.points.size (), false); std::vector<int> nn_indices;
std::vector<float> nn_distances;
// Process all points in the indices vector
for (size_t i = ; i < cloud.points.size (); ++i)//遍历点云中的每一个点
{
if (processed[i])//如果该点已经处理则跳过
continue; std::vector<unsigned int> seed_queue;//定义一个种子队列
int sq_idx = ;
seed_queue.push_back (static_cast<int> (i));//加入一个种子 processed[i] = true; while (sq_idx < static_cast<int> (seed_queue.size ()))//遍历每一个种子
{
// Search for sq_idx
if (!tree->radiusSearch (seed_queue[sq_idx], tolerance, nn_indices, nn_distances))//没找到近邻点就继续
{
sq_idx++;
continue;
} for (size_t j = ; j < nn_indices.size (); ++j) // nn_indices[0] should be sq_idx
{
if (processed[nn_indices[j]]) // Has this point been processed before ?种子点的近邻点中如果已经处理就跳出此次循环继续
continue; //processed[nn_indices[j]] = true;
// [-1;1]
double dot_p = normals.points[i].normal[] * normals.points[nn_indices[j]].normal[] +
normals.points[i].normal[] * normals.points[nn_indices[j]].normal[] +
normals.points[i].normal[] * normals.points[nn_indices[j]].normal[];
if ( fabs (acos (dot_p)) < eps_angle ) //根据内积求夹角，法向量都是单位向量，种子点和近邻点的法向量夹角小于阈值，
{
processed[nn_indices[j]] = true;
seed_queue.push_back (nn_indices[j]);//将此种子点的临近点作为新的种子点。
}
} sq_idx++;
} // If this queue is satisfactory, add to the clusters
if (seed_queue.size () >= min_pts_per_cluster && seed_queue.size () <= max_pts_per_cluster)
{
pcl::PointIndices r;
r.indices.resize (seed_queue.size ());
for (size_t j = ; j < seed_queue.size (); ++j)
r.indices[j] = seed_queue[j]; // These two lines should not be needed: (can anyone confirm?) -FF
std::sort (r.indices.begin (), r.indices.end ());
r.indices.erase (std::unique (r.indices.begin (), r.indices.end ()), r.indices.end ()); r.header = cloud.header;
clusters.push_back (r); // We could avoid a copy by working directly in the vector
}
}
}