How does IKEA boxes with random noise pattern compared with default calibration setting as to calibration accuracy?

VCL3D / VolumetricCapture

A multi-sensor capture system for free viewpoint video.

https://vcl3d.github.io/VolumetricCapture

487 stars 66 forks source link

How does IKEA boxes with random noise pattern compared with default calibration setting as to calibration accuracy? #57

Closed YinghaoHuang91 closed 3 years ago

YinghaoHuang91 commented 3 years ago

Hi I wonder how good it is to glue the random noise pattern on the IKEA boxes for calibration from your experience? Is it comparable with the QR based ones, or still worse?

Also if I'm glad to try other marker based calibration, can I do that with the current release of the system (VoCap)? I'm talking about the method used in this project: http://vcl.iti.gr/spacewars/.

zuru commented 3 years ago

The random noise pattern was used to improve their depth estimates of the Intel RealSense sensors on the untextured box sides. For the Kinect4Azure it is not really needed. In any case, the (median) average depth from a number of frames is used, which helps remove noise.

The marker-based calibration is not supported in this version. However, the raw color and depth data for each viewpoint are available in the ./Data/Calibrations folder, so you can use them to your liking.