10.1. TV input

10.1. TV input

This section is about how to enable watching/grabbing from V4L compatible TV tuner. See the man page for a description of TV options and keyboard controls.

10.1.1. Compilation

1. First, you have to recompile. ./configure will autodetect kernel headers of v4l stuff and the existence of /dev/video* entries. If they exist, TV support will be built (see the output of ./configure).

2. Make sure your tuner works with another TV software in Linux, for example XawTV.

10.1.2. Usage tips

The full listing of the options is available on the manual page. Here are just a few tips:

• Use the channels option. An example:

  -tv channels=26-MTV1,23-TV2
  

  Explanation: Using this option, only the 26 and 23 channels will be usable, and there will be a nice OSD text upon channel switching, displaying the channel's name. Spaces in the channel name must be replaced by the "_" character.

• Choose some sane image dimensions. The dimensions of the resulting image should be divisible by 16.

• If you capture the video with the vertical resolution higher than half of the full resolution (i.e. 288 for PAL or 240 for NTSC), then the 'frames' you get will really be interleaved pairs of fields. Depending on what you want to do with the video you may leave it in this form, destructively deinterlace, or break the pairs apart into individual fields.

  Otherwise you'll get a movie which is distorted during fast-motion scenes and the bit-rate controller will be probably even unable to retain the specified bit-rate as the interlacing artifacts produce high amount of detail and thus consume lot of bandwidth. You can enable deinterlacing with -vf pp=DEINT_TYPE. Usually pp=lb does a good job, but it can be matter of personal preference. See other deinterlacing algorithms in the manual and give it a try.

• Crop out the dead space. When you capture the video, the areas at the edges are usually black or contain some noise. These again consume lots of unnecessary bandwidth. More precisely it's not the black areas themselves but the sharp transitions between the black and the brighter video image which do but that's not important for now. Before you start capturing, adjust the arguments of the crop option so that all the crap at the margins is cropped out. Again, don't forget to keep the resulting dimensions sane.

• Watch out for CPU load. It shouldn't cross the 90% boundary for most of the time. If you have a large capture buffer, MEncoder can survive an overload for few seconds but nothing more. It's better to turn off the 3D OpenGL screensavers and similar stuff.

• Don't mess with the system clock. MEncoder uses the system clock for doing A/V sync. If you adjust the system clock (especially backwards in time), MEncoder gets confused and you will lose frames. This is an important issue if you are hooked to a network and run some time synchronization software like NTP. You have to turn NTP off during the capture process if you want to capture reliably.

• Don't change the outfmt unless you know what you are doing or your card/driver really doesn't support the default (YV12 colorspace). In the older versions of MPlayer/ MEncoder it was necessary to specify the output format. This issue should be fixed in the current releases and outfmt isn't required anymore, and the default suits the most purposes. For example, if you are capturing into DivX using libavcodec and specify outfmt=RGB24 in order to increase the quality of the captured images, the captured image will be actually later converted back into YV12 so the only thing you achieve is a massive waste of CPU power.

• To specify the I420 colorspace (outfmt=i420), you have to add an option -vc rawi420 due to a fourcc conflict with an Intel Indeo video codec.

• There are several ways of capturing audio. You can grab the sound either using your sound card via an external cable connection between video card and line-in, or using the built-in ADC in the bt878 chip. In the latter case, you have to load the btaudio driver. Read the linux/Documentation/sound/btaudio file (in the kernel tree, not MPlayer's) for some instructions on using this driver.

• If MEncoder cannot open the audio device, make sure that it is really available. There can be some trouble with the sound servers like aRts (KDE) or ESD (GNOME). If you have a full duplex sound card (almost any decent card supports it today), and you are using KDE, try to check the "full duplex" option in the sound server preference menu.

10.1.3. Examples

Dummy output, to AAlib :)

mplayer -tv driver=dummy:width=640:height=480 -vo aa tv://

Input from standard V4L:

mplayer -tv driver=v4l:width=640:height=480:outfmt=i420 -vc rawi420 -vo xv tv://

A more sophisticated example. This makes MEncoder capture the full PAL image, crop the margins, and deinterlace the picture using a linear blend algorithm. audio is compressed with a constant bitrate of 64kbps, using LAME codec. This setup is suitable for capturing movies.

mencoder -tv driver=v4l:width=768:height=576 -oac mp3lame -lameopts cbr:br=64\
 -ovc lavc -lavcopts vcodec=mpeg4:vbitrate=900 \
 -vf crop=720:544:24:16,pp=lb -o output.avi tv://

This will additionally rescale the image to 384x288 and compresses the video with the bit-rate of 350kbps in high quality mode. The vqmax option looses the quantizer and allows the video compressor to actually reach so low bit-rate even at the expense of the quality. This can be used for capturing long TV series, where the video quality isn't so important.

mencoder -tv driver=v4l:width=768:height=576 \
 -ovc lavc -lavcopts vcodec=mpeg4:vbitrate=350:vhq:vqmax=31:keyint=300 \
 -oac mp3lame -lameopts cbr:br=48 -sws 8-n-1 -o output.avi\
 -vf crop=720:540:24:18,pp=lb,scale=384:288 tv://

It's also possible to specify smaller image dimensions in the -tv option and omit the software scaling but this approach uses the maximum available information and is a little more resistant to noise. The bt8x8 chips can do the pixel averaging only in the horizontal direction due to a hardware limitation.