[1e1a]

I read a paper a while ago (which I have failed to locate) which used around binary masks, each in front of a single photodiode, as input to a neural network, which estimated the number of people in front of this effectively ~9-pixel "camera". The binary masks and NN weights were trained at the same time. Presumably, something like this could be used to detect lack of driver focus in a far less invasive manner.