Wide lens distortion, depth image artifacts

Dear Community,
I have 2 questions for ZED2 image, using SDK 4.1. :

  1. is there an easy way how to retrieve image without wide lens distortion ?
    As you can see in the image below, the image is stretched towards corners. I’d like to have a corrected image even if it would mean slightly narrower FOV.
    depth3

  2. In the depth image, there are artifacts when arms are close to body, but not touching it. Yet the camera shows them as connected object. Is there any way how to improve the detection? I am using the neural depth mode. See the image below where the space between arms is connected into one area together with the head.
    depth 1
    This has been improved with Neural+ mode, but some artifacts remain and rendering speed is much slower.
    depth4

Thank you for any ideas!
Tomas

Hi @Tomas,

Welcome to the Stereolabs forums :slight_smile:

  1. The images displayed in ZED_Depth_Viewer are already rectified, so there should not be any distortion. How are you measuring that you have lens distortion? To verify this, you can check that all straight lines in real life are indeed straight in the image (we usually use for e.g. a calibration checkerboard target in different areas of the image for this). If this is not the case, the camera may be not correctly calibrated, and you can use the ZED_Calibration tool to recalibrate the camera

  2. The issue of determining precisely the edges of objects is definitely a tricky one in stereo-vision. The performance of the depth algorithms can greatly vary depending on the texture and distance to the objects. In your case here, Neural is not precise enough to distinguish your arms from the rest of your body at that distance.

What I can suggest is to use the “Confidence” value in the Runtime parameters in ZED_Depth_Viewer, which can filter out values in which the algorithm is not confident in. This can help in most cases to remove the “unconfident” borders which most likely contain errors.