All I can think of is this dropcam video except it is not a human screaming "Oh, Henry! Bad dog "
On the positive side, a lot of this image recognition and pattern recognition can probably happen without sending any information over the network and without storing anything beyond what's required for the task.
