Demand for Highly-Calibrated Cameras — More Accurate Than OpenCV

Community Groups Industrial robotics
1

Hi community,

We’re a team working on camera calibration for computer vision and robotics applications.
Our technology enables metrological-quality calibration, providing accuracy and reliability beyond standard OpenCV methods.

Some key features:

  • We use a method based on active targets and a machine-learning-inspired workflow, allowing us to achieve consistently lower calibration errors (up to 2–4 times lower forward/reprojection error than OpenCV, even for mass-produced or wide-angle cameras).
  • The process is flexible and suitable for various camera types (standard, wide-angle, fish-eye, and even exotic optics).
  • We generate a Forward Projection Error (FPE) chart (heatmap) for each camera.
  • Calibration can be tailored to your specific end task (e.g., robot localization, measurement, 3D reconstruction).

Questions:

  1. Would you be interested in purchasing pre-calibrated cameras with calibration accuracy proven to be better than OpenCV’s results?

  2. For which applications would this level of accuracy and certification be most valuable?

  3. What, in your opinion, would be a reasonable price premium for a camera with such calibration and quality analytics included?

Any feedback or comments on needs, pain points, or desired features are welcome.

Exaples:

Calibration examples
Calibration examples1920×2716 393 KB

Thanks in advance!

2

The answers probably depend on the target group. I assume that industry would prefer purchasing pre-calibrated cameras, whereas academia would prefer to have the methods available.

There is a multitude of cameras with different properties. Having only a few of those cameras available and pre-calibrated only to a certain working range naturally limits their application. At least for me, having the calibration tools available, integrated in the ROS ecosystem, using standard camera topics and services, would be much more valuable.

3

Yall are silly. If you’re using fit residuals as the ultimate metric of calibration accuracy, then you should simply throw out most of your calibration data, and resolve: that will get your residuals super low. And that means the calibration is correct… right?

You need better tools to do this correctly: mrcal. And if you’re trying to sell pre-calibrated cameras, then what you want to measure is how stable your calibration is. So you need to calibrate precisely (which you probably aren’t able to do today), expose the cameras to shaking, thermal stresses, whatever, and then to recalibrate and compare (which you definitely aren’t able to do today). I don’t know what the premium hardware cost would be, but the premium software cost is $0: mrcal is free and open.

4

Hello, Christian. Great points!

We’re actually considering developing a calibration suite based on our technology, possibly as a Community Edition for broader access.
I can’t promise full ROS integration right at the start, but we definitely see the value in making these tools available and are open to feedback from the community on desired features and integrations.

Thanks for sharing your perspective — it’s really helpful!