Subpixel RMS error ~0.25 pixels after calibration (D415)

[THark]

Hi there,

According to the Best Known Methods for tuning the RealSense the subpixel RMS error should be lower than 0.1 (https://www.intel.com/content/dam/support/us/en/documents/emerging-technologies/intel-realsense-technology/BKMs_Tuning_RealSense_D4xx_Cam.pdf https://www.intel.com/content/dam/support/us/en/documents/emerging-technologies/intel-realsense-technology/BKMs_Tuning_R… ), measured with the Depth Quality Tool. If this subpixel RMS accuracy is not achieved a calibration is recommended with the Dynamic Calibrator.

However, my D415 has an RMS accuracy of roughly 0.25 pixels even after calibration, when recording the recommended target, mentioned in the BKM tuning document, on a flat surface at ~50 cm distance, in an office room with fluorescent lighting.

I've used the following software/hardware to get these values and I've tried calibrating multiple times, but the RMS accuracy stays around 0.25 pixels.

The dimensions of the printed calibration sheet have also been verified.

D415, firmware 05.09.14.00

Intel RealSense Dynamic Calibrator v2.6.4.0 used with default settings

Depth Quality Tool v2.13.0 used with default settings

Is there anything I can do/change to increase the RMS accuracy?

[MartyG]

Have you tried selecting the 'High Accuracy' visual preset in the Depth Quality Tool? You can do this by left-clicking on 'Custom' to open up the list of available camera configurations.

[THark]

Yes, I have tried this without a change in RMS accuracy. Only when enabling post processing the RMS drops to ~0.13 pixels.

[MartyG]

Real world values do not always match up to the values in the documentation. For example, image FOV can vary. The documentation values are best taken as an ideal aspiration to aim for, but various factors (including the environment the camera is in, the distance of the camera from an object, etc) mean that you may never get a perfect result.

[THark]

Ok thanks for the help. I just wanted to verify if this could be expected behavior.

I did find that I might have changed some settings after all in the Depth Quality Tool.

The region of interest was set to 40% and the emitter was enabled.

When changing the region to interest to 20% and the emitter was disabled and the camera to target distance was increased to ~60 cm the subpixel RMS did drop below 0.2 pixels.

Even the slightest changes in angle to the target could make a big difference in RMS accuracy.

In some cases the RMS even dropped below 0.1 pixels, without post processing. So it's good to see the RMS ended up being pretty close to the ones reported in the documentation.