EYES_UP expression accuracy

[Mona_J_]

I need to know based on what sort of accuracy does the "EYES_UP" expression work in the RealSense SDK. Say if I have two regions and I want to set the EYES_UP in a way that if I look at the upper region it returns me 100 and if I am looking at the lower region it gives me 0, how should I define my regions? 

Also would that support more than two regions?

Any information is really appreciated.

Best regards,

Mona Jalal.

[Gal_G_Intel]

Hey Mona, great question. Eyes up / down / right / left expressions will only return the position of the eyes which is usually not good enough.

The feature you are looking for is gaze detection -> understanding where the user is looking at.

In the next version we will release a new sub-module inside of Face Module called Gaze Module. This module will require some calibration flow from the end-user but will try to solve your exact problem of finding out specific region the user is looking at

[Mona_J_]

Hi Gal,

Thank you for the answer. Can you please show in a simple example how I can define the bounding box that is used for the expressions_eyes_up and how that algorithm exactly work? Is it true if I look above the light in the camera it is considered as up and if I look down the camera light is considered as eyes not up? Also I want to know how do you define the bounding box (like how up, right, left , down we can look)? 

Thanks,

Mona Jalal.