Hi everyone, I’m having trouble with an Arducam AR0234 (B0429, Jetvariety camera module) connected to a Jetson Orin Nano Dev Kit running JetPack 6.2. The images coming from the camera are extremely dark, almost unusable. It seems like the exposure control isn’t working at all — no matter what I do, the brightness doesn’t change. Has anyone else run into this? Any pointers on where to start debugging would be really appreciated.
That does sound like an exposure issue. Let’s start with the basics — can you check whether the exposure control is actually exposed via V4L2 on your system? Run this and share what you see:
v4l2-ctl -d 0 -l
This will list all the available controls for the camera device. Look for anything related to exposure in the output. If it’s listed, we can try setting it manually and see what happens.
Thanks for the quick reply. I ran that command and yes, the exposure setting does show up among the listed V4L2 controls. So the control is definitely there.
I then tried setting it manually:
v4l2-ctl -d 0 -c exposure=3000
The command completed without any errors, which made me think it worked. But when I capture images afterwards, they’re exactly the same — still dark, no change at all. It’s like the setting is being accepted but completely ignored. I’ve tried a few different exposure values and the result is always the same. Very frustrating.
Interesting — so the control is present and accepts values but doesn’t actually affect the output. That’s odd. Let’s dig a bit deeper into what’s happening at the hardware level.
Can you try reading some I2C registers directly from the camera sensor? If you have a demo app or a way to perform raw I2C reads, try reading a few different registers and let me know what values you get back. This will help us figure out whether the communication path to the sensor is working correctly when it’s being streamed via V4L2.
Alright, I did some more digging and the results are… strange. I used a demo application to do I2C register reads on the camera sensor, and no matter which register I read, the returned value is always 0xFFFE. Every single register. That can’t be right.
I also spent some time experimenting in OpenCV with C++. Here’s what I found:
- If I create the
VideoCaptureobject and try to set parameters (like exposure) before capturing the first frame, things break. The settings don’t stick and sometimes the stream gets weird. - But if I capture a single frame first, THEN adjust the settings afterwards, it seems to work better.
- I also tried several OpenCV debayering/conversion options to get proper color from the raw output, but none of them seem to match the AR0234’s Bayer pattern correctly.
On a slightly more positive note, I did test a Python script that Arducam support shared with me. After I tweaked the image resolution and pixel format in the script to match what the AR0234 expects, it ran correctly and I could actually get usable frames. So the hardware does work — but the whole V4L2 exposure control path seems broken somehow.
Thanks for the detailed update — this actually paints a pretty clear picture of what’s going on. Let me connect the dots:
The fact that I2C reads always return 0xFFFE regardless of which register you query is a big clue. It tells us that the standard I2C register reading approach simply doesn’t work reliably when the camera is being streamed through the V4L2 pipeline on the Jetson Orin Nano. The sensor isn’t responding to direct register reads the way you’d normally expect.
Now, this ties directly into your experience with v4l2-ctl -c exposure=3000. Here’s what I believe is happening: the Jetvariety camera module (AR0234, B0429) resets all of its settings — including exposure — back to default values the moment V4L2 streaming begins. So your v4l2-ctl command writes the value, the control reports success, but then as soon as the actual video stream starts, the module’s internal reset wipes it out. The setting never actually takes effect on the live stream.
This also explains your OpenCV C++ findings. Setting parameters before the first frame fails because streaming hasn’t really initialized yet. Setting them after the first frame works better because by then streaming is active — but on this particular module, even that approach is fighting against the reset behavior.
The Python script working after resolution and format adjustments confirms the hardware itself is fine. It’s purely a software sequencing issue with how this specific camera module initializes during V4L2 streaming.
So here’s the root cause and the workaround:
Root cause: The Jetvariety camera module (AR0234, B0429) performs a hard reset of all settings to factory defaults at the start of every V4L2 streaming session. Any exposure changes made before or during stream initialization get overwritten. This is a behavior specific to this module — it’s not a generic V4L2 or Jetson issue.
Workaround: The approach that works is to let V4L2 streaming start normally (so the module finishes its initialization and reset), and then use a subprocess to apply the exposure settings after the stream is already active. This way the settings land after the reset window has passed and they actually take effect on the live frames.
I have a Python demo that implements exactly this pattern — it streams via V4L2, then spawns a subprocess to apply exposure control once streaming is running, bypassing the reset issue. I’ve sent you the download link via private message — please check your inbox and give it a try.
Key takeaway: With this particular camera module, you cannot set exposure (or really any other sensor setting) before or independently of V4L2 streaming. Settings must always be applied after streaming has begun, and the subprocess method is the cleanest way to achieve that.