Correspondance issue with ZED X

Software Python
, ,
1

  • We are using ZED X camera to capture the pointcloud of a fetal phantom,
    similar to this one Fetus Ultrasound Examination Phantom "SPACE FAN-ST" | KYOTO KAGAKU

  • The pointcloud is flat in the abdomen region. It fails to capture the curvature of the abdomen region.

  • We tried to capture the pointcloud with some black dots on the abdomen region, and this time the curvature was visible.

We understand that the ZED X has IR filters and so, we cannot use IR similar to the one used in realsense cameras to illuminate/enhance the textures.

We would like to know if this issue can be solved in any way such as using a random dot projector (light based) or through any computational means?

Regarding random dot projectors, we would like to know which kind of pattern will give the best results with ZED X camera.

2

Hi @Gopi-htic

According to the environment illumination the skin can have not too much visual features to match and provide a precise detection.
I recommend using the NEURAL depth mode. A random dot projection can be helpful, but it must project visible light points, no infrared that the IR filter of the optics would filter out.

Any kind of pattern is good because it adds visual features to match.

3

Hi @Myzhar,
We have tried with NEURAL depth mode as well.
It provides similar results.

The abdomen region is completely flat.

Is there any other solutions for this issue apart from the random dot projectors?
like using any algorithms such as illuminant invariant algorithms or any interpolation algorithms?

Any other way to solve this issue?

4

It’s not easy to reply without data to test.
Can you record an SVO in the same condition by using ZED Explorer and share it with us at support@stereolabs.com?