c++ - 从 RTP 流解码 h264 帧

Question

我正在使用 live555 和 ffmpeg 库从服务器获取和解码 RTP H264 流；视频流由 ffmpeg 编码，使用 Baseline profile 和

x264_param_default_preset(m_params, "veryfast", "zerolatency")

我读了this topic并在我从网络接收的每一帧中添加 SPS 和 PPS 数据；

void ClientSink::NewFrameHandler(unsigned frameSize, unsigned numTruncatedBytes,
    timeval presentationTime, unsigned durationInMicroseconds)
{
     ...
EncodedFrame tmp;
    tmp.m_frame = std::vector<unsigned char>(m_tempBuffer.data(), m_tempBuffer.data() + frameSize);
    tmp.m_duration = durationInMicroseconds;
    tmp.m_pts = presentationTime;

    //Add SPS and PPS data into the frame; TODO: some devices may send SPS and PPs data already into frame;
    tmp.m_frame.insert(tmp.m_frame.begin(), m_spsPpsData.cbegin(), m_spsPpsData.cend());

    emit newEncodedFrame( SharedEncodedFrame(tmp) );
    m_frameCounter++;

    this->continuePlaying();
}

我在解码器中收到的这些帧。

bool H264Decoder::decodeFrame(SharedEncodedFrame orig_frame)
{
...
while(m_packet.size > 0)
    {
        int got_picture;
        int len = avcodec_decode_video2(m_decoderContext, m_picture, &got_picture, &m_packet);
        if (len < 0)
        {
            emit criticalError(QString("Decoding error"));
            return false;
        }
        if (got_picture)
        {
            std::vector<unsigned char> result;
            this->storePicture(result);

            if ( m_picture->format == AVPixelFormat::AV_PIX_FMT_YUV420P )
            {
                //QImage img = QImage(result.data(), m_picture->width, m_picture->height, QImage::Format_RGB888);
                Frame_t result_rgb;
                if (!convert_yuv420p_to_rgb32(result, m_picture->width, m_picture->height, result_rgb))
                {
                    emit criticalError( QString("Failed to convert YUV420p image into rgb32; can't create QImage!"));
                    return false;
                }
                unsigned char* copy_img = new unsigned char[result_rgb.size()];
//this needed because QImage shared buffer, which used, and it will crash, if i use this qimage after result_rgb deleting
                std::copy(result_rgb.cbegin(), result_rgb.cend(), copy_img);
                QImage img = QImage(copy_img, m_picture->width, m_picture->height, QImage::Format_RGB32,
                [](void* array)
                {
                    delete[] array;
                }, copy_img);
                img.save(QString("123.bmp"));
                emit newDecodedFrame(img);
            }

avcodec_decode_video2 解码帧没有任何错误消息，但解码帧在转换后(从 yuv420p 到 rgb32)无效。 this link 上可用的图像示例

你知道我做错了什么吗？

Answer 1

我怀疑错误出在 convert_yuv420p_to_rgb32() 代码中。试试这个:

static SwsContext *m_swsCtx = NULL;
QImage frame =  QImage ( m_picture->width, m_picture->height,
                         QImage::Format_RGB32 );
m_swsCtx = sws_getCachedContext ( m_swsCtx, m_picture->width,
                                  m_picture->height, PIX_FMT_YUV420P,
                                  m_picture->width, m_picture->height,
                                  AV_PIX_FMT_RGB32, SWS_BICUBIC,
                                  NULL, NULL, NULL );
uint8_t *dstSlice[] = { frame.bits() };
int dstStride = frame.width() * 4;
sws_scale ( m_swsCtx, &m_picture.data, &m_picture.linesize,
            0, m_picture->height, dstSlice, &dstStride );

如果您还没有包含/链接 swscale，则需要这样做。

注意:您不需要每一帧都需要 SPS/PPS(在关键帧上就足够了)。但它也没有伤害。