I am trying to capture audio and video from a blackmagic decklink capture card using Windows 7 @ 720p, but I cant seem to get the ffmpeg command line settings right.
Finally got it working. My setup has this all running on a single machine.
For taking the video and serving it via UDP I use the following command :
ffmpeg -f dshow -video_size 1280x720 -rtbufsize 702000k -framerate 60 -i video="Decklink Video Capture":audio="Decklink Audio Capture" -r 30 -threads 4 -vcodec libx264 -crf 0 -preset ultrafast -f mpegts "udp://239.255.12.42:6666"
-f dshow
tells ffmpeg we need to use direct show.-video_size 1280x720
sets the source size, since I am using a 720p60 source, this is it.702000k
is really important since without it the realtime buffer would be full in a matter of seconds.-framerate 60
tells ffmpeg the source is using 60fps.video="Decklink Video Capture":audio="Decklink Audio Capture"
tells ffmpeg to use those device as input, but by specifying them in this fashion, the lag between audio and video will be substantially less (and/or gone).-r 30
forces the output to be 30fps instead of the 60fps in source.-threads 4
does what you think, use 4 threads.-vcodec libx264
encodes the source stream to h264 while broadcasting.-crf 0
sets the "constant rate factor" (quantizer scale) to 0, meaning lossless.-preset ultrafast
means we dont have any patience, so use as little compression as possible. This causes a high bitrate, but that's fine for my purpose.-f mpegts
option tells ffmpeg to use MPEG-TS packets, this will "force" ffmpeg to use a constant bitrate mpeg format, since mpeg itself is normally variable bitrate.udp://239.255.12.42:6666
specifies that we want to broadcast this stream to the multicast address 239.255.12.42 using port 6666 over udp. The reason I chose to use a multicast address here is simply because I need to display the stream (preview) and record at the same time, with as least processing as possible. This prevents me from having to copy the audio and video stream to two different network addresses.For capturing this video using VLC player, I open the following network streaming address :
udp://@239.255.12.42:6666
Finally for recording the stream I spawn a new process and issue the following command :
ffmpeg -y -threads 4 -i udp://239.255.12.42:6666 -map 0 -acodec copy -vcodec copy output.mkv
-y
option is for always overwriting the file if it exists without questions.-threads 4
option does what you think, it uses 4 threads.-i udp://239.255.12.42:6666
connects to the stream we are broadcasting.-map 0
tells ffmpeg that we need all streams (both video and audio).-acodec copy
and -vcodec copy
are there to ensure that the streams are taken as is, instead of doing any compression/transcoding.The only thing left to do (which is a work in progress atm) is creating a c# gui for this. Basic workflow will be to spawn the stream process when the form loads. Use the vlc com+ control to display the video in the application.
Then when the record button is pushed, spawn another process to record and stop that process to end the recording.
I do however stop the stream when I start to record, this makes the recording/detection go much more smoothly. If the stream stays on and I start recording it will take some time before the recording process can "tune in" to the stream. By stopping the stream, starting the recording (which will do nothing until the stream is back on) and starting the stream again, the recording will pickup from the first frame without any problems.
This small delay/flicker is totally acceptable for my purposes.