YPbPr Protocol for Senior DSP Project (University of Michigan) - Page 2

Do you have a question? Post it now! No Registration Necessary

Translate This Thread From English to

Threaded View
Re: YPbPr Protocol for Senior DSP Project (University of Michigan)
Hi Clifford,

Quoted text here. Click to load it

I.e., the OP is looking for REAL TIME commercial detection.
If he was looking to provide this functionality in a PVR
environment, there would be no need to "play music until the
show comes back on" -- he could just skip *past* it.

Quoted text here. Click to load it

That influence doesn't extend to smaller firms who can make a device,
sell it and *not* care if the "system changes" later and makes its
use ineffective.

You will note that there have been several devices marketed
purporting to achieve this goal.  I.e., PVR manufacturers
don't seem too bothered by all that "influence".

Quoted text here. Click to load it

Most MythTV users are geeks.  And, more accommodating of
problems with technology.  Everyone that I know who has built
a MythTV setup *played* with it for a while and then silently
stopped using it.  I.e., if it was *that* good, one would
expect it to be "all the rage".

Quoted text here. Click to load it

Detecting commercials RELIABLY in real-time (i.e., a false positive
deprives the viewer of the content of the broadcast) is a different
problem than ex post facto.

Commercials (USA) *tend* to have durations that are integer multiples
of 15 seconds -- with most falling in the 15, 30 and 60 second
categories.

Commercials *tend* to come in groups of 2 - 6 (or 8).

Commercials *tend* to be delineated by a few black frames (defining
"black" can be a problem as dark scenes can appear "black").

Commercials *tend* to occur at certain points *in* the program.  Note
that this has changed, historically, as more broadcasters try to slip
an extra "commercial block" into each half-hour.  Note, also, that
materials presented "with limited/no commercial interruptions"
still have built-in "commercial insertion points".

Commercials *tend* to inhibit closed captioning information (though
detecting it as "missing" vs. "not present" is ambiguous).

Commercials *tend* to have louder audio than the programming
that they interrupt (though there is no guarantee that the
programming's audio won't have a similar dynamic range).

Commercials tend to be more "frenetic" (for want of a better
word  :< ) than most programming (i.e., the scene changes
occur at much higher frequency as the camera flits back and
forth between "shots").

Commercials tend to have more "flashing"/beating in the image
(in addition to the scene changes) -- flashing text, etc.

No single/few of these criteria are enough to *reliably*
(i.e., no false positives; false negatives are more tolerable)
detect a commercial as it happens.  If you have knowledge of the
future (i.e., have *seen* the future and can now go back and
"edit the past"), you can increase your accuracy -- as long as you
have lots of processing power at your disposal (i.e., aren't
also busy displaying stored video).

The MANUAL "commercial skipper" appears to be the 80/20 solution
requiring a minimum of user interaction for maximum results.
And, of course, the old faithful:  getting up for a 2 minute trip
to the bathroom!  :>

As such, I think most PVR users are happiest with the "skip 30
seconds" means of "commercial avoidance" as they don't have to
worry that the machine has elided some portion of the program
without their knowledge.

Re: YPbPr Protocol for Senior DSP Project (University of Michigan)
Quoted text here. Click to load it

I'd think the best way to do this is with some kind of spam filter
logic, and a time-delay so you're watching the stream delayed say 30
seconds. Of course, training the system would be required (e.g. if you
see Billy Mays anywhere in the frame, assume it's commercial).

Re: YPbPr Protocol for Senior DSP Project (University of Michigan)


Quoted text here. Click to load it

Sure. Offer a subscription service which would send a special signal
(via the Internet or some other way) when commercials are on for every
channel. This signal could be generated by a computer or even by a human
operator.

Quoted text here. Click to load it

There is no need to process video to do that; it is enough to process
sound. A common MCU is sufficient to do basic feature detection and
pattern matching.

But I found the simpler solution: there is no TV in our house. None at
all. Since 99.9% of TV is shit anyway, there is no point watching it.

http://s.lurkmore.ru/images/4/42/Tvass.jpg



Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

Re: YPbPr Protocol for Senior DSP Project (University of Michigan)

Quoted text here. Click to load it

I knew I liked you! :) This is of course the only sensible solution.
We have a couple of TV sets but no cable/satellite/etc. And the TVs
are never switched on. In fact, yesterday I thought I might turn on
the TV downstairs to play some movies off the server, and I just about
choked on the dust.

Re: YPbPr Protocol for Senior DSP Project (University of Michigan)
Hi Vladimir,

Quoted text here. Click to load it

This won't work "as is".  Each "local affilliate" can time shift
(small amounts) the programming material.  I'm not sure what the
contractual arrangement is with the national broadcasters
(i.e., how many of "their" commercials are they OBLIGATED to
pass through).  Obviously, local (commercial) content is also
generated -- is the time for this set-aside "universally"?
Or, does each affiliate inject them willy-nilly?

Quoted text here. Click to load it

I.e., "knowledge of the future".

Note, however, the parallel:  if your spam filter *unilaterally*
decided to delete anything that it considered spam, you run the risk
of a false positive causing you to lose a valid email message.

As such, most spam filters flag/divert the
suspect message and give *you* the final say over whether or not
it is "spam".  This is a tacit admission by the filter designer that
it is not "100% perfect".

A "commercial detector" would need the same ability -- i.e., at
the very least, "memory" so you could reinject the erroneously
silenced audio/video into the actual viewing stream in the event
that it mistakenly declares a portion of "content" as "commercial".

Quoted text here. Click to load it

The problem with that approach is you can't watch *movies*.
Granted, many of *those* are "shit".  But, some are actually
entertaining.  (watching video -- or listening to music -- on a
PC is a crippling experience... as is "reading a book" with such
a device)

Quoted text here. Click to load it

Re: YPbPr Protocol for Senior DSP Project (University of Michigan)
Quoted text here. Click to load it

Call me crippled then - I haven't watched a movie except on a PC
(usually a laptop) for quite a few years now. And most of the reading
I do now is ebooks on my phone (which I put in a ziploc bag to make it
safe-ish for my favorite reading location, which is the bath).

Re: YPbPr Protocol for Senior DSP Project (University of Michigan)
Hi Lewin,

Quoted text here. Click to load it

I find doing anything at a PC other than work to be
uncomfortable (physically, since the seated position
most favorable to working is geared toward alert
interaction -- not "pleasurable relaxation").  When
flying, I will sometimes watch a movie on a PSP, iPod
or whatever PMP I happen to have with me at the time.

We are currently arguing (negotiating?) the merits of
installing a motorized projection screen "in the ceiling"
to move to a DLP instead of a traditional "TV" (I don't
like having electronic devices cluttering up the house
and have been systematically re-engineering them out of
the environment).

Quoted text here. Click to load it

Unfortunately, I have not found any of the reading devices
to yield a comparable experience to a paper book.  Perhaps
a generational thing but I suspect the harshness of most displays
plays an important role.  The same applies for PC's -- if I
am faced with a 1,000 page datasheet, I'll run it off on one
of the laser printers, read it and then discard (recycle) the
pages thus "consumed".  Anything more than a few page technical
paper *or* a quick "consult" of something I have already read
goes this route.

This is unfortunate as I am a voracious reader (30,000+pp annually,
not counting "technical materials") and could benefit from keeping
a library on a portable device (books take up a LOT of space and
have a *lot* of mass!) -- though audio books (or, audio renderings
of text) would be even better (reading while walking is next to
impossible, for me).

Re: YPbPr Protocol for Senior DSP Project (University of Michigan)

Quoted text here. Click to load it

Ah, but I should add that I never - at least not for ~8 years - sit
down for an immersive "all I'm doing is watching a movie" experience.
I am working on one computer (typically a laptop), and have another
monitor behind or to one side of it playing the movie - I glance at it
occasionally. For me, movies are background noise/side entertainment.
I get much more absorbed in books than movies.

Quoted text here. Click to load it

I find that as soon as I boot up the laptop and start playing movies,
I fall asleep :) Still, nothing beats working on a boring PowerPoint
presentation at 25,000 feet with a whisky in my hand. Truly I feel
that I've joined the marketing club.

Quoted text here. Click to load it

Ah, I would not claim that they do. The paper-book experience for me
is most enjoyable with an old book, where the smell and texture of the
paper are part of the reading experience. I have some very interesting
late-1800s blackletter books printed around Philadelphia, for
instance, that would not be fun to read electronically, because most
of the enjoyment is in handling the artifact, not just understanding
the text.

However, I have all my old favorites - Dickens, Verne, Wells, Thoreau,
Forester, ... - electronically on my phone, where they're accessible
whenever I need to escape for a while.

Quoted text here. Click to load it

Heh. My boss is similar - send her a 1,000 page document
electronically and she'll print it. I am the exact opposite - give me
a 1,000 page document and I'll drop it into one of the office
scanners, PDF it to myself, and throw away the original (our facility
does not recycle paper).

Quoted text here. Click to load it

Algorithmic TTS isn't very enjoyable for me. But I do have some
audiobooks for the car. If you get a moment, go to
http://www.archive.org/details/Mindwebs_230 and download the Mindwebs
series - very good readings of short SF stories, produced in the 70s
(but spanning writings from the 30s to 70s). Free.

Re: YPbPr Protocol for Senior DSP Project (University of Michigan)
Hi Lewin,

Quoted text here. Click to load it

I treat movies similarly -- lying down on the floor with a laptop
while glancing up at the screen from time to time.

Quoted text here. Click to load it

Agreed.  But having two or three hours "undisturbed" to consume
a novel is a luxury I rarely have.  Notable exceptions are
long flights -- which makes my reliance on *paper* books even
more annoying (there are only so many things you can cram into
your pockets -- I never have "luggage" -- when travelling)

Quoted text here. Click to load it

I find a book ideal in these situations.  Usually I am traveling
alone and gambling on having a good conversationalist in the seat
nearby is a losing proposition.  OTOH, immersing oneself in a
book is easy to do (lighting is thankfully sufficient) and the
constant drone of the engines helps tune out the activity around me.

Quoted text here. Click to load it

Exactly!  It is a far more intimate experience than the "confrontation"
between a screen and reader/viewer.

Quoted text here. Click to load it

I just can't deal with the screens.  It tires my eyes too quickly
and makes what should be an enjoyable experience into a tedious
one.  I consider reading to be one of the most delightfully
*selfish* experiences in which a person can indulge.  So, I
take pains to avoid anything that can "cheapen" that for me
E.g., I have learned *not* to read while "on the can" (if you
get caught up in the story, you end up with a *really* sore
ass!) or "in the tub" (the "shriveled up" sensation of prolonged
soaking can become painful after a couple of hours!  :-/

Quoted text here. Click to load it

I will preserve the electronic copy of everything that I use/reference.
However, to "digest" something, I want paper to touch and mark up
and dog-ear, etc.  I suspect it has historical throwbacks to how
I "learned" to study...  This would be consistent with other
research I've seen regarding study/recall mechanisms.

Quoted text here. Click to load it

I am not averse to "mechanical" TTS -- I've considerable exposure
to speech synthesizers over the years (at one point, I could do a
remarkable impersonation of a Votrax VS6.3!).  Rather, my brain
is not wired to process information "aurally".  When digesting
something in this manner, I have to close my eyes in order to
process it -- this is not a good idea when walking along the roadside!

So, I stick with music -- selecting material with a cadence appropriate
for my walking speed -- and let my brain wander on architectural
issues (where you don't have to track/remember "detail").  I console
myself that I'm only "wasting" an hour -- and, technically, the
walking *is* beneficial  :-/

Re: YPbPr Protocol for Senior DSP Project (University of Michigan)
Quoted text here. Click to load it

The A/D converters that you need are called "video capture cards".
They come in both PCI and USB varieties for around $100. If you're
planning on building your own video capture card, start with a
reference design from one of the manufacturers that produces video
capture chips. If you're planning on building your own video capture
chip (and sending the signal to where?) you probably won't finish your
project before the end of the semester.

There is already a fairly mature commercial detector here:
http://www.kaashoek.com/comskip /
It's been in development for several years and the results are
surprisingly good. Not all the algorithms used here are applicable in
real-time though.

Quoted text here. Click to load it

I have no idea what you're talking about here. Sadly, I suspect
neither do you.

Site Timeline