One of the problems with the current method of getting access to media
is that it occurs as a side effect of a call done for other purposes
(GetUserMedia).

The parameters for this aren't ideal to determine the full range of
permissions that a page would want - for instance, a page might first
open a video device, and later open an audio device separately; in our
current models, that would give 2 permissions prompts.

Instead of doing it this way, we could make an explicit call:

void GetMediaPermissions(Permissions permissions,
                            successCallback, errorCallback)

Permissions = enum(
  "videoInputDevices",
  "audioInputDevices",
  // and maybe extend this to
  "deviceEnumeration",
  "screenCapture",
  "windowCapture",
  ....
)

The UA could then use the list of permissions requested to construct an
appropriate UI element for asking permission from the user (or use a
stored permissions model to grant access immediately, if that's the
Right Thing).
In any case, all programs that know what class of permissions they want
can get those permissions with one call, one prompt, no matter what they
do later.

For backwards compatibility, getUserMedia would be documented to have an
implicit call to GetMediaPermissions "behind the curtains".

An earlier attempt was http://dev.w3.org/2009/dap/perms/FeaturePermissions.html - an important difference between that and this proposal is that this proposal explicitly states that one can ask for a set of features, not just one at a time, and that user prompts (and therefore reasonably long delays) are to be expected.