AS
Asa
Admin
Thanks guys.
An update! Recordings are no longer clipped thanks to a convoluted way of basically recording each programme twice - once as is currently with the 'accurate recording' and then a timed recording which covers the whole junction. I log the time of the accurate recording, knock a few seconds off and then trim the timed one, and upload that instead. It seems quite reliable now, although still getting the odd one where the flag is being detected well into the ident/titles. Still no idea if that's a software or broadcast issue.
This also now lets me -30secs off for the News Channel ident and similarly it'll do a lookup to see if a show was originally on BBC Three to account for the BBC Three ident on BBC One. The only other annoyance is the BBC Music one which crops up a lot on BBC Four but not always before a music programme.
Still a few little niggles to work on including identifying the idents - particularly BBC Two where the visuals can be quite different depending on which part of the ident is captured.
An update! Recordings are no longer clipped thanks to a convoluted way of basically recording each programme twice - once as is currently with the 'accurate recording' and then a timed recording which covers the whole junction. I log the time of the accurate recording, knock a few seconds off and then trim the timed one, and upload that instead. It seems quite reliable now, although still getting the odd one where the flag is being detected well into the ident/titles. Still no idea if that's a software or broadcast issue.
This also now lets me -30secs off for the News Channel ident and similarly it'll do a lookup to see if a show was originally on BBC Three to account for the BBC Three ident on BBC One. The only other annoyance is the BBC Music one which crops up a lot on BBC Four but not always before a music programme.
Still a few little niggles to work on including identifying the idents - particularly BBC Two where the visuals can be quite different depending on which part of the ident is captured.
AS
Asa
Admin
Absolutely, all the idents and recordings are saved into the database so slicing and dicing the data is pretty easy. Iāve started getting it to check against multiple thumbnails for BBC Two so it should better identify them now.
Iāll see if I can put a page together a page of some interesting stats.
Iāll see if I can put a page together a page of some interesting stats.
TE
The starting point fir a recording was set in the days of PDC ..
(Btw this is far better than DVB p/f)
The BBC decided that the recording should be "as the viewer should enjoy it as live"
So the last junction item before the programme - being network ident , is when the PRF flag was created and the VCR started to record ( having been threaded up by the PRF flag being set c 30 sec before orig start .. but none did !)
This meant that there was a BBC ident in the front of the recording ...this was In The days before "BBC blocks " in the programme , and any "health warning " would be recorded ... this was more important as the programme could be viewed at any time not just after watershed,.
Likewise the stop recording was sent after the junction item after the programme fir "if you were affected " messaging and also to guarantee that the signal was sent after the end if the programme thus avoiding it being cut off !!!
These basic simple rules had a few modifications like what it two programmes were abutted ....
like strart at start of programme and stop 10 sec after end of programme .. note in PDC there could be four "programmes" being controlled at any one time ..
(Btw this is far better than DVB p/f)
The BBC decided that the recording should be "as the viewer should enjoy it as live"
So the last junction item before the programme - being network ident , is when the PRF flag was created and the VCR started to record ( having been threaded up by the PRF flag being set c 30 sec before orig start .. but none did !)
This meant that there was a BBC ident in the front of the recording ...this was In The days before "BBC blocks " in the programme , and any "health warning " would be recorded ... this was more important as the programme could be viewed at any time not just after watershed,.
Likewise the stop recording was sent after the junction item after the programme fir "if you were affected " messaging and also to guarantee that the signal was sent after the end if the programme thus avoiding it being cut off !!!
These basic simple rules had a few modifications like what it two programmes were abutted ....
like strart at start of programme and stop 10 sec after end of programme .. note in PDC there could be four "programmes" being controlled at any one time ..
AS
I wouldnāt know where to start! a combination of image matching and Rekognition (which brings back some fairly distinct words) could be enough to more accurately identify them. Although somethingās gone wrong as it identified this morningās lifeguards as being the football kids.
Interesting about the PDC, good to see the logic carried through into the digital era. I think I remember being fascinated by it when my parents updated their VCR, only to then use it a couple of times!
Asa
Admin
Machine Learning Asa?
I wouldnāt know where to start! a combination of image matching and Rekognition (which brings back some fairly distinct words) could be enough to more accurately identify them. Although somethingās gone wrong as it identified this morningās lifeguards as being the football kids.
Interesting about the PDC, good to see the logic carried through into the digital era. I think I remember being fascinated by it when my parents updated their VCR, only to then use it a couple of times!
AS
Asa
Admin
Sounds like the Pi has crashed so itās going to miss todayās recordings. I wonāt be home until gone 10 to pull the plug out and back in again. Iāll need to dig into the error and maybe rebuild with a lightweight OS (it currently uses OSMC which isnāt necessary for what is effectively just a recorder).
Typically I only set up uptime monitoring the other day so was obviously tempting fate!
Accuracy is still a bit ropey - I have a way of quickly correcting them now but itās not ideal. I think I need to implement a checker using both image and label matching.
Typically I only set up uptime monitoring the other day so was obviously tempting fate!
Accuracy is still a bit ropey - I have a way of quickly correcting them now but itās not ideal. I think I need to implement a checker using both image and label matching.