Dear developers,
We use the appsrc with udp sink to deliver live video from camera to client
on TX2 with the following signal processing pipeline.
------------------------------------- begin signal processing pipeline
-------------------------------------
Server side:
SceneTime(t0) ->camera → (t1) Capture(at 5Hz) → BayerImgQueue →
SignalProcessing–> (t2) outputYuvImgQueue →
(t3)GstreamerPipeLineSendVideoToClient
GStreamer pipeline:
pipelineStr = “appsrc format=3 is-live=true !
video/x-raw,format=(string)I420, width=(int)” + std::to_string(w) +
“,height=(int)” + std::to_string(h) +
" , framerate=" + std::to_string(frameRate)+ “/ 1” +
" ! identity check-imperfect-timestamp=true" +
" ! nvvidconv" +
" ! omxh264enc control-rate=2 target-bitrate=4000000 tune=zerolatency" +
" ! video/x-h264,stream-format=(string)byte-stream ! h264parse ! rtph264pay"
+
" ! udpsink host=" + clientIp +" port=50000";
Client side:
UdpSrc → display (t4)
------------------------------- end signal processing pipeline
-------------------------------------------
Let’s define the corresponding timestamps and time intervals as follows.
t0: time stamp on a computer screen which is captured by camera
t1: time at which a frame y is just captured
t2: time at which the frame y is processed and written into the
ouputYuvImgQueue
t3: time at which the frame y is read from ouputYuvImgQueue by
need_data_callback()
t4:time stamp on receiver side screen
dt1 = t2-t1
dt2 = t3-t2
dt3 = t4-t0 (glass-to-glass latency)
Our experiments show that the glass-to-glass latency strongly depends on the
frameRate parameter in video/x-raw cap. For example
frameRate dt1(ms) dt2 range(ms) max dt3(ms)
5 25 [700,850] ~1290
6 25 [270,300] ~740
7 25 [200,260] ~690
10 25 [150,200] ~630
15 25 [110,120] ~550
20 25 [92,100] ~530
30 25 [75,79] ~500
Note: when there is no new image in outputYuvImgQueue and
need_data_callback() is called, we just repeat the previous frame.
My Question is why the frameRate parameter affects dt2 so much? To void
this, what kind of parameter shall we tune?
Thanks,
--
Sent from:
http://gstreamer-devel.966125.n4.nabble.com/_______________________________________________
gstreamer-devel mailing list
[hidden email]
https://lists.freedesktop.org/mailman/listinfo/gstreamer-devel
Locked
