I think this is for the depth sensor. I’d suggest a median 3X3 filter followed by an averaging 3×3 filter. You can also try a 5×5 filter, but I wouldn’t go beyond that given the resolution of the image output.
This is done to remove noise from the data and fill in ‘holes’ where dept information is missing.
Source: I am an image analyst specializing in recovering image features and deriving statistics.