In the documentation, it says the output contains the 3D coordinates of the point (x,y), but I don't know if this means the output is (x,y,z) or (x,y,d). This is further confused by the input ddepth, which says -1 means the output will have CV_32F depth, but in this case, the output is still a 3 channel array.
So, what is the output of this function?