Virtual channels work correctly with V4L2-utils but NVarguscamerasrc fails sometimes on Xavier JetPa...

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Xavier
1

Hi everyone,

RidgeRun made enhanced the FPD Link III drivers to support Virtual Channel IDs, currently we are testing with 2 cameras at the same time and it is working. However, we are facing some issues to enable it with nvarguscamersrc. Details:

  • A video device is created per link enabled, now testing with 2 cameras.
  • Using V4L2, we got output from 2 cameras at 60 FPS.
  • The cameras are able to work at 60 FPS simultaneously with V4L2-utils.
  • Currently due to available hardware we have tested with two cameras, but 16 cameras should be possible, more testing and develop will be reported soon.
  • Using nvarguscamerasrc, the cameras work 1 time each 6 attempts. We found a timeout when a camera does not work so we have to run nvargus-daemon with param enableCamInfiniteTimeout=1 in order to get output in some cases.
  • When 2 cameras are working together on NVargusCamerasrc, the framerate decrease from 60 FPS to ~7 FPS on both cameras.
  • The framerate seems to be more stable using NVarguscamerasrc with 2 cameras at 30 FPS.
  • The camera indexes are inverted, in other words, the video output from sensor-id 0 is /dev/video1 and sensor-id 1 is /dev/video0

The hardware setup is as follows:

We are interested in any missing property in Device-Tree to correctly enable video output from NVarguscamerasrc.
Since we can get simultaneous video output at 60 FPS with V4L2-utils, we believe that there is something missing at Device-tree level to correctly enable the ISP.
As mentioned before, sometimes we get simultaneous output from NVarguscamerasrc but we want to understand what is going on when the cameras fail.

Do you have any information that might help us to enable the virtual channels through ISP using NVarguscamerasrc?

For more information feel free to contact: https://www.ridgerun.com/

Thanks.

2

Please have a try to boost the clock to avoid the bandwidth concern.
https://elinux.org/Jetson_TX2_Camera_BringUp

BTW, the sensor order fixed on next release.

3

Hi Greivin,

Just a thought: could max_lane_speed be too low in tegra-camera-platform now that you’re pushing more data through a single CSI port?

Regards,
Greg

4

hi all,

JetPack SDK | NVIDIA Developer, JetPack-4.2.1 is now live.

FYI,
we had some performance improvements for multi-session use-case with JetPack-4.2.1/ l4t-r32.2
suggest moving to the latest JetPack public release for verification.
thanks