There are several ways to project, generally for stress we use a very simple "lumped mass" type least squares projection. It is simple but not fully accurate near the boundary. A better way is a local least squares. Do least squares on each element to get values to element nodes, then average. See paper in CMAME, vol 213-216, pp 84, 2012. We have not test on vectors, but it should give reasonable performance for each component. If you have a constraint such as vectors being unit there will be an error.
To implement in feap you can use an element option not currently used (see attached files for simple example on how to loop over elements -- check that arguments on routine are same number as in your version). Note files are not complete but give an idea on where to do things in the element routine.