Recherche avancée

Sur d’autres sites (5828)

• very low latency streaminig with ffmpeg using a webcam

  5 avril, par userDtrm

  I'm trying to configure ffmpeg to do a real-time video streaming using a webcam. The ffmpeg encoder command I use is as follows.

  



  ffmpeg -f v4l2 -input_format yuyv422 -s 640x480 -i /dev/video0 -c:v libx264 -profile:v baseline -trellis 0 -subq 1 -level 32 -preset superfast -tune zerolatency -me_method epzs -crf 30 -threads 0 -bufsize 1 -refs 4 -coder 0 -b_strategy 0 -bf 0 -sc_threshold 0 -x264-params vbv-maxrate=2000:slice-max-size=1500:keyint=30:min-keyint=10: -pix_fmt yuv420p -an -f mpegts udp://192.168.1.8:5001


  



  The ffplay command used to display the video feed is,

  



  ffplay -analyzeduration 1 -fflags -nobuffer -i udp://192.168.1.8:5001


  



  However, I'm experiencing a latency of 0.5 - 1.0s latency in the video stream. Is there a way to reduce this to a number less than 100ms. Also, when I replace the v4l2 camera capture with a screen capture using x11grab, the stream is almost real-time and I experience no noticeable delays. Moreover, changing the encoder from x264 to mpeg2 had no effect on the latency. In addition, the statistics from the ffmpeg shows that the encoder is performing at a 30fps rate, which I believe indicates that the encoding is real-time. This leaves me with only one reason for the experienced delay.

  



  

  • Is there a significant delay in buffers when using v4l2 during video capturing in a webcam ?

  


  • I don't think the transmission delay is in effect in this case as I see no latencies when screen capture is used under the same conditions.

  


  • Can this latency be further reduced ?. Can someone think of a different encoder configuration to be used instead of the one that I've been using ?

  


  


• Broadcast Screen using ffmpeg to encode H.264

  29 avril 2018, par Erick Weil

  Background History :

  So. doing a program, I need a way to broadcast my desktop screen, trought local network( maybe wi-fi ), to aproximately 40 clients, primarily in windows, but if it also works in linux, will be great.

  Started doing my own video compression protocol, in Java, taking the hilbert curve as a inspiration ( it transmit 10 fps, 1024x720, horrible 8-bit colors, using 5 MB of bandwith) but the color quality was very low, and fps wasn’t that good, cpu usage blowing up to 50%, memory 800M. 

  So i realized i’ll never be able to do a proper encoding algorithm and tried to use ffmpeg to encode the screenshots in a video and transmit this.

  I tried :

  • Xuggler : not supported anymore, examples worked, but it does lack tutorials and things that explain how it works
  • VLCJ : VLC wraper for java, requires the client of the transmission to have VLC installed.
  • JavaCV : java wrapper of openCV, to wrap ffmpeg. my program size blows up if i include this
  • FFMPEG : in c++ i created a project, communication with java trought JNI, the remuxing example worked,
    but i don’t now how to use it to do what I want

  The actual question :

  How can i make a c++ program able to encode image frames into a H264 stream and send it trought UDP ?

  i mean, is there any tutorial on ffmpeg, any place to start learning how to use it.

• Create H.264 packet trace from video

  3 avril 2018, par Mello

  I’m using ns-3 for real time video transmission. The peculiarity of my scenario is that, using JSVM software, I obtain a text file corresponding to the trace video packets encoded using H.264/SVC. After a little bit of post processing, I get an input file suitable for network simulation ; then, I try to reconstruct the trace with the packets at the receiver, and after that of course I elaborate some metrics to compare different transmission strategy.
  However, last update on JSVM seems to quite old. In fact, if some (not fundamental) packets are missing from the received trace, the bitstream is reconstructed but video decoding procedure won’t finish (which to me seems counter-intuitive given H.264 decoding policies).
  This means that I can potentially obtain inconsistent metrics.

  I tried also ffmpeg for the bitstream processing procedure, but in this case it doesn’t recognize the enhancement quality layer which I encoded in my video-stream.

  Given all the listed problems, I would like to know if there’s any software that, given a video with any extension as input, it returns as output a text file corresponding to the packet trace. Moreover, it is important that it can do also the reverse process : from a packet trace text file, it must reconstruct the bit stream.