[mythtv] PATCH: Commercial detection/skip using blank frames

Vincent AE Scott mythtv at codex.dyndns.tv
Fri Feb 28 02:49:33 EST 2003


James McTavish(mctavish at shaw.ca)@Mon, Feb 24, 2003 at 10:41:58PM -0500:
> 
> Somebody has already mentioned how RCA vcrs have commercial skipping. 
> Their commercial skipping is flawless and their method could probably be
> used in MythTV recording.  They work by first recording the entire show
> including commercials.  Then it automatically rewinds the tape (after it
> has finished recording) and begins doing a seek to all black screens
> using a method similar to yours, and marks the spots on the tape with
> some sort of non-visible marker.  When playing it sees these markers and
> begins fast-forwarding when it sees one and resumes playing when it
> reaches the next.
> 
> MythTV could take this a step further by identifying the commercials and
> then editing them out in a background process.

I've been following this thread with much anticipation, altho i don't
have an actual fully working mthytv setup, this for me is certainly an
awesome feature.

I like this idea of tagging the ads on the fly, for later post recording
analysis, that got me thinking:

There's been lots of interesting discussions about mechanisms for
detecting adverts in the video stream.  it struck me that you could try
and take a slightly larger view of the problem.  If an entire program
has been completely recorded, the adverts within it will be bunched up
together, with larger gaps of program in between.  

If we've run some advert detecting algorithms on the stream, and
identified possible chronological points that may be adverts.  Then we
could apply some statistical models about the proximity of the markers
over the entire program recording.   i.e. looking for areas over the
time line that have a higher than average number of possible ads in
close proximity.

We could also apply the reverse theory.  looking for annomolies in the
time line where a small number of possible ads appear over a long period
of otherwise ad free recording.  This might better identify
false-postivies.  Where a program might have a black screen during the
actual content.

Now, taking all this yet another step further, we could build up a model
for advertising strategies for different stations.  for example, we
might spot a trend on one particular station, where the adverts are 5
minutes long in total, and never start within the first 17 minutes of
program.  Using this information, we might better be able to predict on
subsequent recordings for that channel, where the adverts are more
likely to appear.  By associating the data with a channel, we can
hopefully better predict ad placing.  



As i said at the start, this is a fascinating topic, especially given
what the eventual outcomes might be.  In my enthusiasm i hope i haven't
over looked something, or that this has already been better explained by
someone else previously.  Better to have ideas re-iterated than
forgotten.


ttfn,
-v

-- 
keys:  http://codex.net/pgp/gpg.asc http://codex.net/pgp/pgp.asc

 The bug is mightier than the fix



More information about the mythtv-dev mailing list