What is your current go to USB Camera

Hey Everyone,

Looking to for recommendations on USB cameras to use for AprilTag processing.

In the past we’ve had great success with PS3 Eye Cameras, as they are very high fps and cheap…but are very large.

Obviously there are tons and tons of cameras on amazon, but its hard to tell what info on there is accurate, what camera has a good fps, which one isn’t too fish eye’ed. ETC.

Also, obviously this post has been made many times before, I’m just hoping to find more recent info.

Excited to hear your thoughts!

1 Like

This is the one that I’ve seen recommended like crazy over the past year:
Arducam OV9281

4 Likes

I’ve seen that as well! I know the 6328 crew runs 4 of those on their 2023 bot ( shoutout @jonahb55 ).

$100 for their 50 FPS one feels a little pricey. Not sure why their 120 FPS one is cheaper, must be some drawbacks of that high FPS

We just got these color global shutter cameras from Arducam:

The 50 FPS version is higher resolution (~1080p vs ~720p). In our case we were trying to maximize resolution before frame rate.

2 Likes

Maybe I should take some time and read data sheets! That makes sense!

I recommend the Arducam brand OV9281, or a generic AR0144 cam if you want to save a bit of money. Otherwise, most 60fpa webcams will work fine, though it’s hard to find one with a good mount.

I have a case design and camera test details in this folder: Coprocessor - Google Drive

3 Likes

We’ve used several of the Arducam OV9281. One came out-of-focus, but it has a threaded lens mount and it was easy to fix the focus. We also had one fail with no apparent contact or damage. Overall, we still like them and have no plans to change in 2024, unless the Next Great Thing ™ is significantly better.

The defacto easy USB camera is Microsoft life cam. It’s not great, it’s easy to find and doesn’t require special drivers. At one point it was a first choice item for like 20 credits.

That being said we have used other fisheye cameras from Amazon for driver POV, but that’s not at all helpful for AprilTags.

I’d also second the arducams or cameras for the rpi/Jetson nano.

econ_ecam30cunano

If you are going to do image processing you might want a coprocessor anyways and typically you get much better direct control over those ribbon cable types vs generic USB.

Arducams makes a really cool optical zoom adjustable ribbon cable camera that would really make finding long distance tags easier without losing resolution.

Livonia Warriors actually had a really great demo at their pit of their own limelight-esque solution with Jetson nano and rpi with the ribbon style cameras

1 Like

Lifecams are great,they work with literally anything. The only downside is low FPS.

I forgot to mention that the USB cable attachment on Arducam and other camera modules sucks. I strongly recommend using hot glue to reinforce the bond between the came and connector. The mount I linked before has a cable strain relief as well which is a HUGE help.

2 Likes

MS Lifecams are popular but I dislike using them, mostly for the fact that they’re rather annoying to work with from a hardware POV.
I tend to use built in cameras at this point (Apple’s stuff is so much better than a cheap Toshiba from 2008), but I know Logitech makes some pretty solid stuff.

1 Like

Interesting. 70 degree lense instead of 150

Tempted to go with the OV9281, just not cheap when you factor in buying multiple of them!

Fisheye lenses have too much distortion, and PhotonVision isn’t designed to calibrate for them correctly.

Maybe it’s time? @bankst
I have no idea what it actually takes but it seems like it would have great benefits.

ah. fair. Yeah whatever USB camera we used last year was fish eye’d like crazy. Ended up just heavily favoring odometrey

can’t imagine thats a simple calibration haha

It’s not like the current calibration algorithm was written by the PV folks
Maybe there’s already something out there

fair, i haven’t looked at the code base at all

For Duckietown and also ROS we use calibration grids to figure out the cameras properties.

Essentially by moving the grid around, skewing the grid, adjusting as it directed it can figure out the properties intrinsicly.

We need something like that for getting the properties of common FRC cameras and a way to test/cal unknowns. Maybe those existing tools are fine but limited to Linux afaik

2 Likes