点云欧式聚类分割

代码实现：

void euclideanclusters(pcl::PointCloud<pcl::PointXYZ>::Ptr& cloud, std::vector<pcl::PointCloud<pcl::PointXYZ>::Ptr>& clusters,
	float threshold, int min_cluster_size, int max_cluster_size)
{
	pcl::KdTreeFLANN<pcl::PointXYZ> tree;
	tree.setInputCloud(cloud);

	std::vector<pcl::PointIndices> cluster_indices;				//定义储存聚类点云的索引
	std::vector<bool> processed(cloud->points.size(), false);	//定义点云的点是否被处理过的标记
	std::vector<int> nn_indices;								//kdtree搜索到的点的索引
	std::vector<float> nn_distances;							//kdtree搜索到的点的距离

	for (int i = 0; i < cloud->points.size(); ++i)   //遍历点云中的每一个点
	{
		if (processed[i])   continue; //如果该点已经处理则跳过
			
		std::vector<int> seed_queue;  //定义一个种子点队列

		int sq_idx = 0;				//记录已经处理种子点队列中的种子点的个数
		seed_queue.push_back(i);	//加入一个种子点
		processed[i] = true;		//标记这个种子点已经被搜索过了

		while (sq_idx < seed_queue.size())   //遍历每一个种子点
		{
			//kdtree近邻搜索
			if (!tree.radiusSearch(seed_queue[sq_idx], threshold, nn_indices, nn_distances))
			{
				++sq_idx;
				continue;   //没找到近邻点就继续
			}

			for (size_t j = 0; j < nn_indices.size(); ++j)     //遍历搜索到的种子点的邻近点       
			{
				if (processed[nn_indices[j]])	continue;   //种子点的近邻点中如果已经处理则跳出循环

				seed_queue.push_back(nn_indices[j]);	//将此种子点的临近点作为新的种子点，入队操作
				processed[nn_indices[j]] = true;		//该点已经处理，打标签
			}

			++sq_idx;
		}

		//过滤满足最大点数和最小点数的类
		if (seed_queue.size() >= min_cluster_size && seed_queue.size() <= max_cluster_size)
		{
			pcl::PointIndices r;
			r.indices = seed_queue;
			cluster_indices.push_back(r);   
		}
	}

	//存储每个聚类
	for (int i = 0; i < cluster_indices.size(); ++i)
	{
		pcl::PointCloud<pcl::PointXYZ>::Ptr cluster_temp(new pcl::PointCloud<pcl::PointXYZ>);
		pcl::copyPointCloud(*cloud, cluster_indices[i].indices, *cluster_temp);
		clusters.push_back(cluster_temp);
	}
}

参考：PCL 的欧式距离聚类

代码传送门：https://github.com/taifyang/PCL-algorithm