Selecting specific areas in a pointcloud

I have a point cloud of this object

And I want to measure the distance between the plane where the piece is sitting on and the points marked as red, and as blue. I already have the plane equation, allowing me to remove the points corresponding to the plane and to calculate the distance between it and the points. But I can't figure out how to select the red and blue points, any methods that could help solve this problem?

Solution

If your cloud has high quality normals, you can try to leverage the discontinuities between the sub-parts using pcl::RegionGrowing with a strict smoothness criteria (If the deviation between points normals is less than the smoothness threshold then they are suggested to be in the same cluster).
If you know the size of all sub-parts, you can calculate an oriented bounding-box, and then (roughly) cut-out the two round parts.
If you don't know the size, then after you calculate an oriented bounding-box (and rotate so that x is the long axis - for example), you bin the x values, and calculate for each x, the min/max y values. This should give you a sense of it's shape.

Either way, once you have a sub-part, you can calculate it's center of mass, and then calculate for each point, the distance to the center. This will give you a distribution of radius.

If you have finer sub-parts (from method #1 above), you can determine which is what by considering the mean and min radius.
If you have a rough sub-part (from method #2/#3 above), you should segment points as "red"/"blue" based on the radius. You might have to do this iteratively (refining the center of mass each time).