Purpleish picture with custom camera (GMSL2 AR0821 Imager) on a custom Jetson TX2 NX platform

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

Hi,

I managed to connect a 8MP GMSL2 based camera (LI-AR0821) with AR0821 imager and to get the data into the nvidia camera stack and Iam able to stream it h264 encoded using gstreamer. However the picture is very purpleish (see attachment

Bild
Bild1920×1090 123 KB

From Leopard Imaging I got a camera_overrides.isp file and applied it to my Jetson rootfs (/var/nvidia/settings/camera) however according to the purple shades it had no effect on them.

The imager provides a 12 bit bayer pattern with a RCCB CFA, but according to source (sensor common.c) only rggb12 is supported by the nvidia kernel, thus I configured the format to be rggb12 in my device-tree source. Could this be the cause for the purpleish picture? If so, do you have an advise how to get a realistic picture then?

Thanks in advance for your support on this issue

Kind Regards,
Steve

2

P.S. My “badge” configuration is still the one from imx219, the configuration I’ve based my changes on:

/ {
	tegra-camera-platform {
		modules {
			cam_module0: module0 {
				badge = "lanai_front_RBP194";
				position = "topleft";

Do I need to change this badge string as well in order that the camera_overrides.isp tuning parameters are beeing used?

Thanks, Steve

3

Meanwhile I figured out by modifying the badge- parameter string to contain _ar0820 instead of RBP_194 the ISP override parameters seem to be active (please see attached screenshot). The picture is much clearer now. However there’s still a (more light) purple shade over the whole picture:

Screenshot 2024-04-17 175639
Screenshot 2024-04-17 1756391920×1034 95.7 KB

Could this have something to do with my rggb12 configuration, or would you say this is a normal picture coming from an RCCB-CFA based sensor?

Thanks, Steve

5

hello steve.kreyer,

that’s right, you should update badge property in device tree to match the unique string in ISP override file.

regarding to the latest capture results… may I know what’s the capture pipeline you’re using?
since this is RCCB format types, which using Clear (C) to replace the Green (G) channel. it looks normal results, or… you may target a color chart to evaluate the results.

6

Hi @JerryChang ,
thanks for your support. My pipeline is a very basic rtsp-server pipeline:

./test-launch "nvarguscamerasrc  ! video/x-raw(memory:NVMM), width=(int)3840, height=(int)2160, format=(string)NV12, framerate=(fraction)23/1 ! nvv4l2h264enc bitrate=28000000 ! h264parse ! rtph264pay name=pay0 pt=96"

Do you have some reference or some tool advise on how to perform this color chart analysis?

Thanks, Steve

7

hello steve.kreyer,

it usually check with a color checker, for instance, https://www.xrite.com/categories/calibration-profiling/colorchecker-classic.

besides,
please see-also Camera Architecture Stack. there’s a standard Linux V4L2 application that uses direct kernel IOCTL calls, which bypass ISP. you may capture the raw content with v4l IOCTL.
for instance, Applications Using V4L2 IOCTL Directly.
and… please use 3rdparty tool, such as… 7yuv to examine the content.

8

Hi @JerryChang
thanks for your sugggestion. 7Yuv seems to be a nice tool.

Meanwhile I got correct camera picture working by contacting Leopard Imaging once again. After applying their ISP file the pixel order “bayer_rggb12” was still incorrect for their sensor. Instead they need bayer_grbg12 in order to assign pixels and colors correctly

Thanks for your support on this issue.

Kind regards, Steve

9

thanks for status update, looks like the issue has resolved.

10

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.