Just an idea. Maybe some others will give their opinion on this:
I would set up a training scenario where you can control the distance from the dog to the other person, and show the dog that being calm is what works to gain access to that person. I know you have tried something like this, but I think a measure of greater control is in order to enable the dog to succeed.
Let's call it "unrestrained recall."
Place the dog in heel position with person A, in a sit, with person B standing on the other side of person A from the dog. 3 in a line. Person B, person A, dog; all facing front.
Wait for the dog to be calm. Once the dog is calm, person B takes one step forward. If the dog remains calm, it gets released to person B for quiet praise. If the dog breaks or becomes overly excited, it is called back to heel position and placed back in a sit.
After 10 consecutive successful repetitions of this, increase the distance that person B moves to 2 steps.
After 10 consecutive successful repetitions at 2 steps, increase the distance to 4 steps and so on, increasing the distance as the dog remains successful. If at any time the dog starts breaking the sit or becoming excited, go back to the previous distance and reinforce good behavior there until it is solid.
The desired end state would be playing hide and seek with the dog, where person B goes a great distance into the woods, observed by the dog, and the dog remains calm until released.
At no time should person B go to the dog if it fails. The only time the dog gains access to person B is if it remains calm throughout the exercise. Person B should not call the dog at any time or encourage excitement in the dog when the dog gets there, just calm praise if that. Getting to person B is the reward, so there shouldn't be a big party at the meeting point.
Any thoughts anyone?