Recherche avancée

Sur d’autres sites (9249)

• Revision 6522 : $GLOBALS[’liste_des_forums’] a disparu

  12 juin 2012, par kent1 — Log

  $GLOBALSliste_des_forums ? a disparu

• Retrieving RTP/RTSP Absolute Timestamps through FFmpeg using Opencv

  6 septembre 2019, par Fr0sty

  This is an update to my previous post. I am able to get the RTP timestamps by using ffmpeg and opencv currently, however, I am trying to actually get the timestamp at which the frame was captured. I did a work around to try find the time at which the frame was captured (code is in python).

  seconds_before_frame = cap.getRTPTimeStampSeconds()
  
  fractionofseconds_before_frame = cap.getRTPTimeStampFraction()
  
  ret, frame = cap.read()
  
  seconds_after_frame = cap.getRTPTimeStampSeconds()
  
  fractionofseconds_after_frame = cap.getRTPTimeStampFraction()

  By doing this I found that the time captured was off by 0.02359296 seconds and sometimes 0.2359296 seconds, that’s a lot more than I had expected.

  I found some work arounds with trying to get the timestamp of the frame by using AVFormatContext which I didn’t really understand through the priv_data.

  AVPacket* packet;
  
  av_read_frame(formatCtx, packet);
  
  RTSPState* rtspState = formatCtx->priv_data; 
  
  RTPDemuxContext *rtpdemux = rtspState->rtsp_streams[packet->stream_index]->transport_priv;

  as well as

  RTSPState *state = _formatCtx->priv_data; 
  
  RTSPStream *stream = state->rtsp_streams[0]; 
  
  RTPDemuxContext *demux = stream->transport_priv; 
  
  demux→timestamp

  I’d like to ask how can we call AVFormatContext from C++ side, is it through av_read_frame ? Is it possible to use python bindings or VideoCapture already wraps ffmpeg so there is no need to use av_read_frame but just call VideoCapture ?

• C++ ffmpeg sws_getCachedContext seg fault when reading opengl buffer

  3 juillet 2022, par user19068953

  I made a program that draws some stuff on opengl, then reads the frames on the screen as rgba and encodes it as a yuv420 video, but I get a segfault when I try to send the red opengl frame to sws_getCachedContext. I used the muxing.c as the encoder and my code is as follows (I used align 256 because i have another framebuffer that uses 128) :

  


      constexpr int ALIGNMENT = 256;
    if (posix_memalign((void**)&state.glBuffer, ALIGNMENT, gl_width * gl_height * 4) != 0) 
    {
        VI_ERROR("Couldn't allocate frame buffer ");
        return false;
    }
    REGISTER("done, returning");
    return true;


  


  Then i read the frame after drawing the stuff i want :

  


          glReadPixels(0, 0,
                 gl_width, gl_height, 
                 GL_RGBA, GL_UNSIGNED_BYTE, 
                 (GLvoid*) state.glBuffer);


  


  The window is the same size as the gl_buffer (I know the result will be flipped, thats ok)

  


  And finally i send it to the encoder(rgb_data is glBuffer) :

  


  void video_encoder::set_frame_yuv_from_rgb(AVFrame *frame, struct SwsContext *sws_context) {
    const int in_linesize[1] = { 4 * width };

width = gl_width;
height = gl_height;
sws_context = sws_getCachedContext(sws_context,
        width, height, AV_PIX_FMT_RGBA,
        width, height, AV_PIX_FMT_YUV420P,
        0, 0, 0, 0);
        
sws_scale(sws_context, (const uint8_t * const *)&rgb_data, in_linesize, 0,
        height, frame->data, frame->linesize);
}


  


  I get a seg fault on sws_getCachedContext any idea why this happens and how do i fix it ?

  


  Edit : width = 512, height = 256, valgrind says : Invalid read of size 16.