You could assume that the non-thumb fingers of a hand span at most two rows and that one of these is the "home" row.
When you detect a tap visually, you could then infer which row that finger tapped by looking at where the other non-thumb fingers are.
i dont understand , are we still talking about a qwerty keyboard? or a special kind with just two rows, one of which is a "home" row that is placed wherever certain fingers are?
This would work with three rows, with the home row being the middle one.