Video User’s Guide

Video is available on PJSIP version 2.0 and later (2.3 support video for iOS, 2.4 support video for Android). This document describes how to use the video feature, mostly with PJSUA-LIB.

Tip

For PJSUA2 video tutorial, please see Working with video media.

Building with Video Support

Follow Get Started for your platform on building pjsip with video support.

Building the GUI Sample Application

We have a GUI sample application with video support. The project is located under pjsip-apps/src/vidgui. It is not built by default, and you need Qt SDK to build it.

GNU Build System (Mac OS X, Linux, etc)

Follow these steps to build vidgui sample:

  1. Go to vidgui source directory:

    $ cd pjsip-apps/src/vidgui

  2. Generate Makefile. For Linux:

    $ qmake

    and for Mac OS X:

    $ qmake -spec macx-g++

  3. Build the app:

    $ make

Visual Studio

Follow these steps to build vidgui sample with Visual Studio:

  1. Open command prompt, and

    cd pjsip-apps\src\vidgui

  2. Generate project files:

    qmake -tp vc

  3. Open vidgui.vcproj project.

  4. Save the solution, and build the project

Using Video API (pjsua-lib)

This section provides several sample scenarios of using video in your application. Please see Video API Reference (pjsua-lib) section below for a more complete documentation about the Video API.

Enabling Video

By default, video is enabled in pjsua_call_setting::vid_cnt setting.

Incoming Video Call

Incoming video will be accepted/rejected depending on whether video is enabled in the call setting (see above). You can pass the call setting using the API pjsua_call_answer2() (so for example, to reject the video, set vid_cnt to 0 and call pjsua_call_answer2()). If video is enabled, incoming video will be accepted as long as we have matching codec for it. However, this does not necessarily mean that the video will be displayed automatically to the screen, nor that outgoing video will be transmitted automatically, as there will be separate settings for these. Outgoing video behavior will be explained in the next section.

Display Incoming Video Automatically

By default, incoming video is not displayed automatically, since the app may want to seek user approval first. Use the following code to change this behavior on per account basis:

pjsua_acc_config cfg;

pjsua_acc_config_default(&cfg);
cfg.vid_in_auto_show = PJ_TRUE;

Show or Hide Incoming Video

Regardless of the setting above, you can use the following steps to show or hide the display incoming video:

  1. Use pjsua_call_get_vid_stream_idx() or enumerate the call’s media stream to find the media index of the default video. If there are multiple video streams in a call, the default video is the first active video media in the call.

  2. Locate the media information of the specified stream index in the pjsua_call_info, and acquire the window ID associated with the remote video. Sample code:

int vid_idx; pjsua_vid_win_id wid;

vid_idx = pjsua_call_get_vid_stream_idx(call_id);
if (vid_idx >= 0) {
   pjsua_call_info ci;

   pjsua_call_get_info(call_id, &ci);
   wid = ci.media[vid_idx].stream.vid.win_in;

}

  1. Using the video window ID, you may retrieve the associated native video handle with pjsua_vid_win_get_info() and then show or hide the video window using native API, or use pjsua_vid_win_set_show() to show/hide the window using PJSUA API. See Working with Video Window section below for information on manipulating video windows.

Controlling Incoming Video Stream

Controlling the video window above will not cause any re-INVITE or UPDATE to be sent to remote, since the operation occurs locally. However, if you wish, you may alter the incoming video stream with pjsua_call_set_vid_strm() API, and this will cause re-INVITE or UPDATE to be sent to negotiate the new SDP. The relevant operation to control incoming video with pjsua_call_set_vid_strm() are:

Since pjsua_call_set_vid_strm() will result in renegotiation of the SDP in a re-INVITE or UPDATE transaction, the result of this operation will not be available immediately. Application can monitor the status by implementing pjsua_callback::on_call_media_state callback and enumerate the media stream status with pjsua_call_info.

Incoming Re-offer

If the re-offer contains video, incoming re-offer will be automatically answered with current video setting in the call setting. Currently there is no callback for this, however application can always watch for media update via pjsua_callback::on_call_media_state callback.

Outgoing Video Call

Outgoing video is enabled/disabled depending on the call setting. To initiate a call with video in the SDP as inactive, you can disable the video in the call setting and set pjsua_call_setting::flag with PJSUA_CALL_INCLUDE_DISABLED_MEDIA.

Outgoing Video Transmission

Outgoing video transmission is independent from the incoming video transmission; each can be operated separately. Note that outgoing video transmission is not started by default, not even when incoming offer contains video support. This behavior is controlled by pjsua_acc_config::vid_out_auto_transmit setting, which default to PJ_FALSE. Setting this to PJ_TRUE will cause video transmission to be started automatically on each outgoing calls and on incoming calls that indicates video support in its offer. However, it is more flexible and appropriate to leave this setting at PJ_FALSE, and add video later during the call by using pjsua_call_set_vid_strm() API, as will be explained shortly.

Default Capture Device

The default capture device that is used by an account is configured in pjsua_acc_config::vid_cap_dev setting. It is more convenient to set the “correct” device here rather than having to set it in every other API calls later.

Controlling Video Stream

Application uses pjsua_call_set_vid_strm() API to control video stream on a call.

Some of the video operations above require re-INVITE or UPDATE to be sent, hence the result will not be available immediately. In that case, application can implement pjsua_callback::on_call_media_state callback and inspect the resulting negotiation by looking at the pjsua_call_info. Please see Video Call Manipulation in the API reference section below for more information about the operations above.

Add or Remove Video

You can set pjsua_call_setting::vid_cnt to the desired video count to add/remove video, then send the reinvite/update. Alternatively, you can use pjsua_call_set_vid_strm() API to control the video stream on a call Controlling Incoming Video Stream or Controlling Video Stream above.

Working with Video Window

Video Window represents all window objects on the screen that the library creates. The video window can display incoming video, preview, and/or other video playbacks.

Application may retrieve video windows from the following places:

Application retrieves pjsua_vid_win_info with pjsua_vid_win_get_info(). The one window property that most applications will be interested with is the native window handle of the video. The native video handle is contained by pjmedia_vid_dev_hwnd structure inside pjsua_vid_win_info. Application can use the native handle to embed the video window into application’s GUI structure. Alternatively, the library also provides few simple and most commonly used API to operate the window, such as pjsua_vid_win_set_show(), pjsua_vid_win_set_size(), etc., however the availability of these APIs are not guaranteed since it depends on the underlying backend device.

Modifying video codec parameters for video call

Video codec parameters are specified in pjmedia_vid_codec_param. The codec parameters provide separate settings for each direction, encoding and decoding. Any modifications on video codec parameters can be applied using pjsua_vid_codec_set_param(), here is a sample code for reference:

const pj_str_t codec_id = {"H264", 4};
pjmedia_vid_codec_param param;

pjsua_vid_codec_get_param(&codec_id, &param);

/* Modify param here */
...

pjsua_vid_codec_set_param(&codec_id, &param);

Size or resolution

Specify video picture dimension.

  1. For encoding direction, configured via det.vid.size field of pjmedia_vid_codec_param::enc_fmt, e.g:

    /* Sending 1280 x 720 */
param.enc_fmt.det.vid.size.w = 1280;
param.enc_fmt.det.vid.size.h = 720;

    Note

    • Both width and height must be even numbers.

    • There is a possibility that the value will be adjusted to follow remote capability. For example, if remote signals that maximum resolution supported is 640 x 480 and locally the encoding direction size is set to 1280 x 720, then 640 x 480 will be used.

    • The library will find the closest size/ratio that the capture device supports. Application should choose the size ratio that the capture device supports, otherwise the video might get stretched. For example, if the device capture supports 640x480 and 1280x720 and the size is set to 500x500. The device camera will be opened at 640x480 and later converted to 500x500 and get the image stretched.

  2. For decoding direction, two steps are needed:

    1. The det.vid.size field of pjmedia_vid_codec_param::dec_fmt should be set to the highest value expected for incoming video size.

    2. If the resolution exceeds the supported maximum specified in the video codecs, you need to modify it (MAX_RX_WIDTH and MAX_RX_HEIGHT in openh264.cpp, vid_toolbox.m, or and_vid_mediacodec.cpp, or MAX_RES in vpx.c or ffmpeg_vid_codecs.c).

    3. signalling to remote, configured via codec specific SDP format parameter (fmtp): pjmedia_vid_codec_param::dec_fmtp.

      • H263-1998, e.g:

        /* 1st preference: 352 x 288 (CIF) */
param.dec_fmtp.param[n].name = pj_str("CIF");
/* The value actually specifies framerate, see framerate section below */
param.dec_fmtp.param[n].val = pj_str("1");
/* 2nd preference: 176 x 144 (QCIF) */
param.dec_fmtp.param[n+1].name = pj_str("QCIF");
/* The value actually specifies framerate, see framerate section below */
param.dec_fmtp.param[n+1].val = pj_str("1");

      • H264, the size is implicitly specified in H264 level (check the standard specification or this Wikipedia page) and on SDP, the H264 level is signalled via H264 SDP fmtp profile-level-id, e.g:

        /* Can receive up to 1280×720 @30fps */
param.dec_fmtp.param[n].name = pj_str("profile-level-id");
/* Set the profile level to "1f", which means level 3.1 */
param.dec_fmtp.param[n].val = pj_str("xxxx1f");

Framerate

Specify number of frames processed per second.

  1. For encoding direction, configured via det.vid.fps of pjmedia_vid_codec_param::enc_fmt, e.g:

    /* Sending @30fps */
param.enc_fmt.det.vid.fps.num   = 30;
param.enc_fmt.det.vid.fps.denum = 1;

    Note

    • that there is a possibility that the value will be adjusted to follow remote capability. For example, if remote signals that maximum framerate supported is 10fps and locally the encoding direction framerate is set to 30fps, then 10fps will be used.

    • limitation: if preview is enabled before call is established, capture device will opened using default framerate of the device, and subsequent calls that use that device will use this framerate regardless of the configured encoding framerate that is set above. Currently the only solution is to disable preview before establishing media and re-enable it once the video media is established.

  2. For decoding direction, two steps are needed:

    1. The det.vid.fps of pjmedia_vid_codec_param::dec_fmt should be set to the highest value expected for incoming video framerate.

    2. signalling to remote, configured via codec specific SDP format parameter (fmtp): pjmedia_vid_codec_param::dec_fmtp.

      • H263-1998, maximum framerate is specified per size/resolution basis, check RFC 4629 Section 8.1.1 for more info.

        /* 3000/(1.001*2) fps for CIF */
param.dec_fmtp.param[m].name = pj_str("CIF");
param.dec_fmtp.param[m].val = pj_str("2");
/* 3000/(1.001*1) fps for QCIF */
param.dec_fmtp.param[n].name = pj_str("QCIF");
param.dec_fmtp.param[n].val = pj_str("1");

      • H264, similar to size/resolution, the framerate is implicitly specified in H264 level (check the standard specification or MPEG-4 AVC levels) and the H264 level is signalled via H264 SDP fmtp profile-level-id, e.g:

        /* Can receive up to 1280×720 @30fps */
param.dec_fmtp.param[n].name = pj_str("profile-level-id");
param.dec_fmtp.param[n].val = pj_str("xxxx1f");

Bitrate

Specify bandwidth requirement for video payloads stream delivery.

This is configurable via det.vid.avg_bps and det.vid.max_bps fields of pjmedia_vid_codec_param::enc_fmt, e.g:

/* Bitrate range preferred: 512-1024kbps */
param.enc_fmt.det.vid.avg_bps = 512000;
param.enc_fmt.det.vid.max_bps = 1024000;

Note

  • This setting is applicable for encoding and decoding direction, currently there is no way to set asymmetric bitrate. By decoding direction, actually it just means that this setting will be queried when generating bandwidth info for local SDP (see next point).

  • The bitrate setting of all codecs will be enumerated and the highest value will be signalled in bandwidth info in local SDP (see ticket #1244).

  • There is a possibility that the encoding bitrate will be adjusted to follow remote bitrate setting, i.e: read from SDP bandwidth info (b=TIAS line) in remote SDP. For example, if remote signals that maximum bitrate is 128kbps and locally the bitrate is set to 512kbps, then 128kbps will be used.

  • If codec specific bitrate setting signalling (via SDP fmtp) is desired, e.g: MaxBR for H263, application should put the SDP fmtp manually, for example:

    /* H263 specific maximum bitrate 512kbps */
param.dec_fmtp.param[n].name = pj_str("MaxBR");
param.dec_fmtp.param[n].val = pj_str("5120"); /* = max_bps / 100 \*/

Setting video capture orientation

On mobile platforms, in order to send video in the proper orientation (i.e. head always up regardless of the device orientation), application needs to do the following:

  1. Setup the device to get orientation change notification.

  2. Inside the callback, call PJSUA API pjsua_vid_dev_set_setting(), e.g.:

    pjsua_vid_dev_set_setting(dev_id, PJMEDIA_VID_DEV_CAP_ORIENTATION,
                          &new_orientation, PJ_TRUE)

    or PJSUA2 API pj::VidDevManager::setCaptureOrient(), e.g.:

    Endpoint.instance().vidDevManager()
                   .setCaptureOrient(dev_id, new_orient, true)

    to tell the video device about the new orientation.

For sample usage, please refer to our sample apps, ipjsua for iOS, and pjsua2 for Android. Ticket #1861 explains this feature in detail.

When video orientation signaling is available

In case application has the capability to signal remote about video orientation (e.g: via SIP INFO or RTP header extension), instead of telling video device capturer (via pjsua_vid_dev_set_setting() or pj::VidDevManager::setCaptureOrient()), it may signal remote directly about the new orientation. This way the video sent to remote will always in full frame (no black bands in left+right sides due to forcing landscape video in portrait frame or vice versa), but it may not be in “proper” orientation, this should not be problem though as remote could get the orientation info from out of band signaling, so it should be able to render the incoming video frames in “proper” orientation.

However note that if portrait mode is prefered as the initial orientation in a video call session (default settings are set for landscape video orientation), the encoding part of video codec param should be configured as portrait too, i.e: width < height, e.g:

/* Sending 240 x 320 */
param.enc_fmt.det.vid.size.w = 240;
param.enc_fmt.det.vid.size.h = 320;

and the initial video device orientation should be set as portrait too, e.g:

/* After the capturer device is opened, e.g: using pjsua_vid_preview_start()
 * or opened automatically by video call, tell the capture device about
 * current orientation. Note this need to be done once only, so when orientation
 * is changed, never update the device about the new orientation.
 */

/* On Android, portrait mode is defined as PJMEDIA_ORIENT_ROTATE_270DEG */
current_orient = PJMEDIA_ORIENT_ROTATE_270DEG;

/* On iOS, portrait mode is defined as PJMEDIA_ORIENT_ROTATE_90DEG*/
current_orient = PJMEDIA_ORIENT_ROTATE_90DEG;

pjsua_vid_dev_set_setting(dev_id, PJMEDIA_VID_DEV_CAP_ORIENTATION,
                          &current_orient, PJ_TRUE);

...

then when device orientation is changed, application must not update the video device orientation, instead, it should just signal remote about device orientation. Updating orientation info to video capture device will cause device to rotate (and perhaps downsize the image) to make sure that the image always has “proper” orientation (head upside).

Video Conference

Available since 2.9.

Please check ticket #2181 for more info.

Additional Info

Using OpenGL with SDL

PJSIP supports OpenGL video rendering with SDL. Follow these steps to enable and use the OpenGL backend.

  1. Install OpenGL development libraries for your system. The instructions vary, and some platforms may have OpenGL development libraries installed by default.

    • For Ubuntu 12.04, you can run the following:

      $ sudo apt-get install freeglut3 freeglut3-dev
$ sudo apt-get install binutils-gold

    • Alternatively, you can use libgl-dev which is smaller. Please note that since Ubuntu 14.04 LTS, libsdl2-dev is available which comes with libgl-dev automatically, so it might not be needed anymore.

      $ sudo apt-get install libgl-dev

  2. Enable SDL OpenGL support in PJSIP, by declaring this in your config_site.h:

    #define PJMEDIA_VIDEO_DEV_SDL_HAS_OPENGL    1

  3. If you’re not using Visual Studio, add OpenGL library in your application’s input library list. If you’re using GNU tools, you can add this in user.mak file in root PJSIP directory:

    export LDFLAGS += -lGL

  4. Rebuild PJSIP

  5. Now “SDL openGL renderer” device should show up in video device list. Simply just use this device.

Mac OS X Video Threading Issue

On Mac OS X, our video implementation uses Cocoa frameworks, which require handling user events and drawing window content to be done in the main thread. Hence, to avoid deadlock, application should not call any PJSIP API which can potentially block from the main thread. We provide an API pj_run_app() to simplify creating a GUI app on Mac OS X, please refer to pjsua app located in pjsip-apps/src/pjsua for sample usage. Basically, pj_run_app() will setup an event loop management in the main thread and create a multi-threading environment, allowing PJSIP to be called from another thread.

int main_func(int argc, char *argv[])
{
    // This is your real main function
}

int main(int argc, char *argv[])
{
    // pj_run_app() will call your main function from another thread (if necessary)
    // this will free the main thread to handle GUI events and drawing
    return pj_run_app(&main_func, argc, argv, 0);
}

Video key frame transmission

Video API Reference (pjsua-lib)

This section explains and lists the Video API as it was available when this document is written. For a richer and more up to date list, please see Video API reference

The Video API is classified into the following categories.

Device enumeration API

In addition, the PJMEDIA videodev also provides this API to detect change in device availability:

Video preview API

The video preview API can be used to show the output of capture device to a video window:

Video Configuration

Video is enabled/disabled on pjsua_call_setting.

Video settings are mostly configured on the pjsua_acc_config with the following fields:

Video Call Manipulation

The default video behavior for a call is controlled by the account settings above. On top of that, the application can manipulate video of an already-going call by using pjsua_call_set_vid_strm() API.

Use pjsua_call_get_vid_stream_idx() to get the media stream index of the default video stream in the call.

Video Call Information

Video media information are available in pjsua_call_info.

Video Call Stream Information and Statistic

Use the following API to query call’s stream information and statistic.

Note

The pjsua_call_get_media_session has been deprecated since its use is unsafe.

Video Window API

A video window is a rectangular area in your monitor to display video content. The video content may come from remote stream, local camera (in case of preview), AVI playback, or any other video playback. Application mostly will be interested in the native handle of the video window so that it can embed it in its application window, however we also provide simple and commonly used API for manipulating the window.

See:

Video Codec API

API for managing video codecs: