It's possible that detecting crows vs anything else could allow for a ground version, but I'd suggest recognizing individual crows would be even more impactful. This way you could tell which birds were more willing to try things, who was newer to a particular feeder, if a specific bird was picking up training routines, etc. Some members of the group have already made some efforts here; anyone care to chime in?
In regard to feeding times, I've noticed that different birds will keep different schedules. For example, in a mated pair's territory they may start the morning in one location, then move to another, and always spend the afternoon in a third. While this "routine" can change with weather, predators, construction, food availability, etc., it's not a bad way of charting their movements and can create some regular appearances to time feedings to.
OTOH, if you got rid of the other animals impeding the machine could just serve them whenever worked best for the birds themselves, so that may be the most desirable solution. Let us know how your efforts go; this could be a game changer in creating more reliable means for the birds to absorb the training regime!
- J