How to get the real coordinates from the image coordinates?

edit

As an example, I have a node that receives a point cloud, extracts an image and runs a vision algorithm on it, and then retrieves the XYZ coordinates of the detected features:

https://github.com/trainman419/target...

The key things to pick out are the conversion from a structured point cloud to an image:

sensor_msgs::ImagePtr image = boost::make_shared<sensor_msgs::Image>();
pcl::toROSMsg( *cloud, *image );

And the return conversion from image coordinates to a 3D point from the cloud:

pcl::PointXYZ p1 = depth.at(x, y);