How to speed up video conversion in opencv's VideoCapture/Gstreamer

[LoveNvidia]

Hi guys,
I want to use multi-stream decoder using harware of jetson nano in opencv’s VideoCapture+Gstreamer1.0. I use the below codes for that processing.

gstream_elemets = (‘rtspsrc location={} latency=300!’ ‘queue !’
'rtph264depay ! h264parse ! omxh264dec enable-low-outbuffer=1 ! ’
'nvvidconv ! ’
‘video/x-raw , format=(string)RGBA !’
'videoconvert ! ’
‘appsink sync=false’)
cv2.VideoCapture(gstream_elemets, cv2.CAP_GSTREAMER)

My problem is that in the convertor step, opencv use CPU for this work, Is it possible to optimize this stage?

[fabian.solano]

Hi @LoveNvidia,

You may want to try with the ‘nvvidconv’ plugin which can be used as a replacement with hardware optimization for videoconvert. However, as you are using appsink you may want to check the following topic

NVMM memory

Hi Folks, Where can we know more about NVMM memory, in terms of its architectural location on Tegra SoC or external DDR ? Would like to understand where buffer are created when we do - Gstreamer CAPS like - filtercaps = gst_caps_from_string("video/x-raw(memory:NVMM), width=(int)1920, height=(int)1080, format=(string)I420, framerate=(fraction)30/1 "); I need to transfer contents of Gstreamer - GstBuffer to opencv CV::Mat. While doing so I run into problems/issues related to NVMM (I think). He…

For instructions on how to use the nvvidconv with appsink.

Regards,
Fabian
www.ridgerun.com