FFmpeg or gstreamer hardware acceleration on h264/h265 using USB camera

[dennis.zhang97]

Hi,

I was using Jetson Nano to do multiple livestreams, what I did is using gstreamer to split the CSI camera stream into 2 virtual devices, then one was used for opencv, another I was using ffmpeg directly pushing RTMP stream into my server, here’s the gstreamer command:

gst-launch-1.0 nvarguscamerasrc sensor-id=0 ! \
  'video/x-raw(memory:NVMM), width=1280, height=720, format=NV12, framerate=30/1' ! \
  nvvidconv flip-method=0 ! \
  'video/x-raw, width=1280, height=720, format=BGRx' ! \
  queue ! videoconvert ! \
  'video/x-raw, format=BGR' ! \
  tee name=t ! \
  queue ! v4l2sink device=/dev/video2 t. ! \
  queue ! v4l2sink device=/dev/video3

And for ffmpeg we used h264_nvmpi encoding, but CPU is still not enough to run our other programs, so we just updated to Jetson Orin Nano and we update camera from csi to USB.

Later I found out that If I directly use ffmpeg on Jetson Orin Nano, it costs too many CPU, and h264_nvmpi was based on jetson nano, how can I use the hardware acceleration to avoid massive CPU usage on spliting stream or pushing RMTP stream?

Thanks

[DaneLLL]

Hi,
Please use Jetpack 6.0GA and install the ffmpeg package:

Accelerated Decode with ffmpeg — NVIDIA Jetson Linux Developer Guide 1 documentation

And you can run the command to use hardware decoding:

$ ffmpeg -c:v h264_nvv4l2dec -i /home/nvidia/test.mp4 a.yuv

[dennis.zhang97]

Hi,

I am not decoding videos, I am useing ffmpeg encoding like ffmpeg -f v4l2 -i /dev/video2 -c:v libx264 -f flv rtmp://HOST_IP/live/livestream

[DaneLLL]

Hi,
Orin Nano does not have hardware encoders so you would need to use software encoder. It is expected CPU usage is high.

[dennis.zhang97]

Okay thank you for explaination, but 1920x1080 20 FPS resolution in h264 costs 400%(4 cores fully running) is too much…