Tvheadend

I think better is mark over uniqe event in stream like change audio from 5.1 to stereo when in movie start show advertisement.

Best solution for this have MythTV (in 99% hit good in advertisement and mark)
Please check

http://www.mythtv.org/wiki/Commercial_Detection

Methods

Blank Frame Detection - Is used to determine when a programme fades to black (this invariably happens between show segments)
    Blank Frame & scene change detection - As above but tries to determine that a large amount of the picture has changed
    Scene change detection - Tries to determine that a large amount of the picture has changed
    Logo detection - Looks for a part of the picture that does not change during a recorded show - i.e. an onscreen logo. Logos are usually removed for the duration of commercial breaks, making them 'easier' to spot.
    All - Use all commercial detection methods at the same time. This detection method uses information from blank-frame, scene change, and logo detection and can easily be modified to take advantage of other detection information. It works in a fundamentally different way than the normal blank-frame and scene change methods. The code groups the frames into blocks separated by blank frames and rates each block based upon several factors such as the rate of scene change, what percentage of frames in the block have a logo present, how long/short the block is, etc.. Indications that a block is part of a commercial (such as very high rates of scene changes) lower the block's score, while indications that a block is part of the show (such as having a logo present on most frames) raise the block's score. In the end the blocks are looked at as a group and various bits of logic applied to make the final determination where commercials start and end. On most of the shows tested, this method seems to be doing a much better job than just blank-frame or blank-frame plus scene-change.

piotr galek wrote:

Best solution for this have MythTV (in 99% hit good in advertisement and mark)

Automatic detection systems are not reliable, but a nice preliminary stage. TV stations try to manipulate the video stream in a way to make automatic detection systems fail. It's just another hare and tortoise game. So you need the manual process anyway to adjust falsely detected marks.

Additionally, the system can fingerprint the frames and check three minutes before a opening mark and three minutes after a closing mark for overlapping video.

renne - wrote:

When the user presses the "Cut now" button, TVHeadEnd creates a cutted copy of the recording by:

• hardlinking complete or closing GOPs in the filesystem
• reencoding starting truncated GOPs
  OR
• reencoding the marked parts of the recording

Luckily, mkvmerge has a "--split" option. TVHE just has to pass the timecodes to mkvmerge to do frame exact splitting into several files at the timecodes.