K-means clustering on point cloud data - c++

Question

I am trying to implement K-means clustering algorithm on a point cloud. I am not sure, however, how to import the data as input for the k-means member of pcl. The documentation has proven to be a little confusing.
So far I have imported the pcd into a point cloud and transferred it into a vector but I dont know how to proceed from here and initialize Kmeans directly.
int main (int argc, char** argv)
{
pcl::PointCloud<pcl::PointXYZ>::Ptr cloud_in(new pcl::PointCloud<pcl::PointXYZ>);
std::vector<pcl::PointXYZ> cloud;
pcl::io::loadPCDFile ("Scene02 - Cloud.pcd", *cloud_in);
for (int i = 0; i < cloud_in->size(); i++)
{
cloud[i] = cloud_in->points[i];
}
pcl::Kmeans real(300000, 3);
real.setInputData(cloud);
}
I realize that the syntax is wrong but I am not sure what the right one is either.

Answer 1

This function is very odd as compared to how pcl generally does things (centering around custom point types). Basically, the oddity is that you have to enter points via a specified dimension vector rather than a custom point type. Here is tested and functional sample code: (obviously you need to provide your own file name, and you will likely want to adjust cluster size)
int main(int argc, char** argv) {
std::string filePath = "../PointCloudFiles/beaconJR.pcd";
pcl::PointCloud<pcl::PointXYZ>::Ptr tempCloud(new pcl::PointCloud<pcl::PointXYZ>);
if (pcl::io::loadPCDFile(filePath, *tempCloud) == -1) //* load the file
{printf("failed file load!\n");}
else
{
pcl::Kmeans real(static_cast<int> (tempCloud->points.size()), 3);
real.setClusterSize(3); //it is important that you set this term appropriately for your application
for (size_t i = 0; i < tempCloud->points.size(); i++)
{
std::vector<float> data(3);
data[0] = tempCloud->points[i].x;
data[1] = tempCloud->points[i].y;
data[2] = tempCloud->points[i].z;
real.addDataPoint(data);
}
real.kMeans();
// get the cluster centroids
pcl::Kmeans::Centroids centroids = real.get_centroids();
std::cout << "points in total Cloud : " << tempCloud->points.size() << std::endl;
std::cout << "centroid count: " << centroids.size() << std::endl;
for (int i = 0; i<centroids.size(); i++)
{
std::cout << i << "_cent output: x: " << centroids[i][0] << " ,";
std::cout << "y: " << centroids[i][1] << " ,";
std::cout << "z: " << centroids[i][2] << std::endl;
}
}
std::cin.get();
std::cin.get();
}
Cheers!
--edit
As far as visualizing the clusters. I think (untested) that "pcl::Kmeans::PointsToClusters" is going to give you a vector with custer labels per point which you can use to index through the original cloud and separate them.