Qt+GStreamer: How to take a snapshot while playing live video stream

Question

I've developed a video player based on Qt and QtGstreamer. It is used to play live streams (RTSP). I have to add the possibility for the user to take snapshots while he is playing a live stream without perturbing the video playback.

Here the graph of the pipeline I've made:

                                -->queue-->autovideosink
uridecodebin-->videoflip-->tee--|
            |                   -->queue->videoconvert-->pngenc-->filesink
            |
            |->audioconvert-->autoaudiosink

I use the pad-added signal from uridecodebin to add and link dynamically my elements to the pipeline, function of the received caps.

void Player::onPadAdded(const QGst::PadPtr &pad)
{
    QGst::CapsPtr caps = pad->currentCaps();
    if (caps->toString().startsWith("video/x-raw")) {
        qDebug("Received 'video/x-raw' caps");
        handleNewVideoPad(pad);
    }
    else if (caps->toString().startsWith("audio/x-raw")) {
        qDebug("Received 'audio/x-raw' caps");
        if (!m_audioEnabled) {
            qDebug("Audio is disabled in the player. Ignoring...");
            return;
        }
        handleNewAudioPad(pad);
    }
    else {
        qWarning("Unsuported caps, arborting ...!");
        return;
    }
}

[...]

void Player::handleNewVideoPad(QGst::PadPtr pad)
{
    m_player->videoTeeVideoSrcPad = m_player->videoTee->getRequestPad("src_%u");

    // Add video elements
    m_player->pipeline->add(m_player->videoFlip);
    m_player->pipeline->add(m_player->videoTee);
    m_player->pipeline->add(m_player->videoQueue);
    m_player->pipeline->add(m_player->videoSink);

    // Add snap elements
    m_player->pipeline->add(m_player->snapQueue);
    m_player->pipeline->add(m_player->snapConverter);
    m_player->pipeline->add(m_player->snapEncoder);
    m_player->pipeline->add(m_player->snapSink);

    // Link video elements
    m_player->videoFlip->link(m_player->videoTee);
    m_player->videoQueue->link(m_player->videoSink);

    // Link snap elements
    m_player->snapQueue->link(m_player->snapConverter);
    m_player->snapConverter->link(m_player->snapEncoder);
    m_player->snapEncoder->link(m_player->snapSink);

    // Lock snap elements
    m_player->snapQueue->setStateLocked(true);
    m_player->snapConverter->setStateLocked(true);
    m_player->snapEncoder->setStateLocked(true);
    m_player->snapSink->setStateLocked(true);

    m_player->videoFlip->setState(QGst::StatePlaying);
    m_player->videoTee->setState(QGst::StatePlaying);
    m_player->videoQueue->setState(QGst::StatePlaying);
    m_player->videoSink->setState(QGst::StatePlaying);

    // Link pads
    m_player->videoTeeVideoSrcPad->link(m_player->videoQueue->getStaticPad("sink"));
    pad->link(m_player->videoSinkPad);

    m_player->videoLinked = true;
}

The method to take a snapshot:

void Player::takeSnapshot()
{
    QDateTime dateTime = QDateTime::currentDateTime();
    QString snapLocation = QString("/%1/snap_%2.png").arg(m_snapDir).arg(dateTime.toString(Qt::ISODate));

    m_player->inSnapshotCaputre = true;

    if (m_player->videoTeeSnapSrcPad) {
        m_player->videoTee->releaseRequestPad(m_player->videoTeeSnapSrcPad);
        m_player->videoTeeSnapSrcPad.clear();
    }
    m_player->videoTeeSnapSrcPad = m_player->videoTee->getRequestPad("src_%u");

    // Stop the snapshot branch
    m_player->snapQueue->setState(QGst::StateNull);
    m_player->snapConverter->setState(QGst::StateNull);
    m_player->snapEncoder->setState(QGst::StateNull);
    m_player->snapSink->setState(QGst::StateNull);

    // Link Tee src pad to snap queue sink pad
    m_player->videoTeeSnapSrcPad->link(m_player->snapQueue->getStaticPad("sink"));

    // Set the snapshot location property
    m_player->snapSink->setProperty("location", snapLocation);

    // Unlock snapshot branch
    m_player->snapQueue->setStateLocked(false);
    m_player->snapConverter->setStateLocked(false);
    m_player->snapEncoder->setStateLocked(false);
    m_player->snapSink->setStateLocked(false);
    m_player->videoTeeSnapSrcPad->setActive(true);

    // Synch snapshot branch state with parent
    m_player->snapQueue->syncStateWithParent();
    m_player->snapConverter->syncStateWithParent();
    m_player->snapEncoder->syncStateWithParent();
    m_player->snapSink->syncStateWithParent();
}

The bus message callback:

void Player::onBusMessage(const QGst::MessagePtr & message)
{
    QGst::ElementPtr source = message->source().staticCast<QGst::Element>();
    switch (message->type()) {
    case QGst::MessageEos: { //End of stream. We reached the end of the file.
        qDebug("Message End Off Stream");
        if (m_player->inSnapshotCaputre) {
            blockSignals(true);
            pause();
            play();
            blockSignals(false);
            m_player->inSnapshotCaputre = false;
        }
        else {
            m_eos = true;
            stop();
        }
        break;
    }
    [...]
}

The problem is:

• When I set the snapshot property to true of the pngenc element, I receive the EOS event which stop my pipeline, so I need to restart it, which freeze the video playback for about half a second, which in not acceptable in my case.
• When I set the snapshot property to false of the pngenc element, I have no pipeline perturbations, but my png file keeps growing until I call again the Player::takeSnapshot() method.

Where am I wrong ? Is there a better way to do it ? I've tried unsuccessfully creating a QGst::Bin element for my snapshot branch. What about pad probe ?

Thanks by advance

Answer 1

Thanks to @sebastian-droge answer, I found the solution, using gst_video_convert_sample and the last-sample property of my video sink.

The solution I've implemented is:

void Player::takeSnapshot()
{
    QDateTime currentDate = QDateTime::currentDateTime();
    QString location = QString("%1/snap_%2.png").arg(QDir::homePath()).arg(currentDate.toString(Qt::ISODate));
    QImage snapShot;
    QImage::Format snapFormat;
    QGlib::Value val = m_videoSink->property("last-sample");
    GstSample *videoSample = (GstSample *)g_value_get_boxed(val);
    QGst::SamplePtr sample = QGst::SamplePtr::wrap(videoSample);
    QGst::SamplePtr convertedSample;
    QGst::BufferPtr buffer;
    QGst::CapsPtr caps = sample->caps();
    QGst::MapInfo mapInfo;
    GError *err = NULL;
    GstCaps * capsTo = NULL;
    const QGst::StructurePtr structure = caps->internalStructure(0);
    int width, height;

    width = structure.data()->value("width").get<int>();
    height = structure.data()->value("height").get<int>();

    qDebug() << "Sample caps:" << structure.data()->toString();

    /*
     * { QImage::Format_RGBX8888, GST_VIDEO_FORMAT_RGBx  },
     * { QImage::Format_RGBA8888, GST_VIDEO_FORMAT_RGBA  },
     * { QImage::Format_RGB888  , GST_VIDEO_FORMAT_RGB   },
     * { QImage::Format_RGB16   , GST_VIDEO_FORMAT_RGB16 }
     */
    snapFormat = QImage::Format_RGB888;
    capsTo = gst_caps_new_simple("video/x-raw",
                                 "format", G_TYPE_STRING, "RGB",
                                 "width", G_TYPE_INT, width,
                                 "height", G_TYPE_INT, height,
                                 NULL);

    convertedSample = QGst::SamplePtr::wrap(gst_video_convert_sample(videoSample, capsTo, GST_SECOND, &err));
    if (convertedSample.isNull()) {
        qWarning() << "gst_video_convert_sample Failed:" << err->message;
    }
    else {
        qDebug() << "Converted sample caps:" << convertedSample->caps()->toString();

        buffer = convertedSample->buffer();
        buffer->map(mapInfo, QGst::MapRead);

        snapShot = QImage((const uchar *)mapInfo.data(),
                          width,
                          height,
                          snapFormat);

        qDebug() << "Saving snap to" << location;
        snapShot.save(location);

        buffer->unmap(mapInfo);
    }

    val.clear();
    sample.clear();
    convertedSample.clear();
    buffer.clear();
    caps.clear();
    g_clear_error(&err);
    if (capsTo)
        gst_caps_unref(capsTo);
}

I've create a simple test application, which implement this solution. The code is available on my Github

Answer 2

You can take the last-sample property on any sink, e.g. your video sink. This contains a GstSample, which has a buffer with the very latest video frame in it. You can take that as a snapshot, and e.g. with gst_video_convert_sample() or the async variant of it, convert it to a PNG/JPG/whatever.

See https://gstreamer.freedesktop.org/data/doc/gstreamer/head/gstreamer-libs/html/GstBaseSink.html#GstBaseSink--last-sample and https://gstreamer.freedesktop.org/data/doc/gstreamer/head/gst-plugins-base-libs/html/gst-plugins-base-libs-gstvideo.html#gst-video-convert-sample

Alternatively, you would have to shut down the filesink snapshot pipeline after the first frame. For example by having a pad probe to know when the first frame happened, and then injecting an EOS event to prevent further PNG frames to be appended to the same file.