I'm not surprised that this is a research topic but compared to all the other applications, I still think this will take much longer to get running reliably. T

I'm not surprised that this is a research topic but compared to all the other applications, I still think this will take much longer to get running reliably. They are working on centered, pre-registered faces and stationary cameras. Once you add low-quality sensor that moves with the head (introducing huge amounts of motion blur to the image), and rich, real-world scenes full of other distractors, the technology will fall apart. I'm certain we'll get hacked up implementations that work half the time if you keep perfectly still for some of the very basic and easy to tell emotions such as happy and sad, but it will mostly be a disfunctional joke for a long time. Time will tell!