Video Support


The following dependencies are required to build PJSIP with video support (including H264)

  • ffmpeg (libavformat, libswscale, libavcodec, libavutil)
  • libx264

Versions I have tried:

  • ffmpeg 2.5.3
  • libx264 (snapshot-20141218-2245)

Installing dependencies (Debian / Ubuntu systems)

The situation here is a bit sad. Both Debian and Ubuntu ship with libav instead of FFmpeg, but libraries are called the same. PJSIP had to be patched in order to properly work with libav, and the patch as not yet been included upstream.

On Debian, when the Debian-Multimedia repositories are used (quite common) you get FFmpeg and not libav. Oh the joy!

Installing dependencies on Debian:

apt-get install libv4l-dev libavcodec-dev libavformat-dev libavutil-dev libswscale-dev libswresample-dev libx264-dev libavcodec-extra

If using the Debian-Multimedia repositories, do not install libavcodec-extra.

Installing dependencies on Ubuntu:

apt-get install libv4l-dev libavcodec-dev libavformat-dev libavutil-dev libswscale-dev libx264-dev libavcodec-extra

Note on H.264 support: In order to have H.264 support, FFmpeg (or libav) need to be compiled with support for it. The standard packages don't, hence the need for installing the libavcodec-extra packages.

Manually compiling dependencies (for OSX)

All dependencies will be compiled to a directory in the user's HOME directory:

export SIPSIMPLE_FFMPEG_PATH=$HOME/work/ag-projects/video/local

NOTE: yasm is required in order to enable asm optimizations. It does not come preinstalled on OSX, so it has to be manually installed. (brew install yasn will do)


./configure --enable-shared --disable-avs --disable-lavf --disable-ffms --disable-gpac --prefix=$SIPSIMPLE_FFMPEG_PATH
make install


# Some exports

./configure --enable-shared --disable-static --disable-lzma --enable-memalign-hack --enable-gpl --enable-libx264 --prefix=$SIPSIMPLE_FFMPEG_PATH --extra-cflags="`pkg-config --cflags x264`" --extra-ldflags="`pkg-config --libs x264`" 
make install


API for video components is based on 2 different types of video capable entities:

  • VideoProducer: a source for video data, for example a video camera or a remote video stream
  • VideoConsumer: a sink or destination for video data, for example a video rendering window

Data flow

Data flow works in pull fashion, that is, a producer doesn't start to produce data until there is a consumer which will consume it.


Produces video data.

Public API:

  • start: start producing video as soon as a consumer is attached
  • stop: immediately stop producing data
  • close: remove all consumers and stop producing video data (also deallocate all C structures)
  • producer_port: pointer to the pjmedia_port object


Consumes video data.

Public API:

  • producer: (r/w property) attach this consumer to a producer, in order to render the video data generated by the producer. If set to None, it's detached
  • consumer_port: pointer to the pjmedia_port object
  • close: detach from producer and free all resources (also deallocate all C structures)

Producer and consumer objects

  • VideoCamera: Producer, acquires video from a user camera.
  • FrameBufferVideoRenderer: Consumer, calls the user supplied callback with a video frame at a time.
  • LocalVideoStream: Consumer, takes video from a VideoCamera and sends it to the remote party.
  • RemoteVideoStream: Producer, produces video sent by the remote party.

These are just theoretical objects, they are not currently implemented.

  • VideoFileWriter: Consumer, saves incoming video data to a video file.
  • VideoFilePlayer: Producer, produces video data out of a video file.
  • VideoMixer: Producer/Consumer, consumes video from multiple sources and produces aggregated video data.

NOTE: pjsip does have a AVI file player, which also seems to support audio (this could be used to stream a movie, for example), it doesn't have a writer though.


Information about H264 profiles:

OpenH264 implementation

PJSIP has an initial version of a wrapper for Cisco's OpenH264 implementation (

OpenH264 seems to implement SVC, which is better than AVC. In practice it didn't outperform libx264 and it only implements the constrained baseline profile, so it was discarded.