I want to now how I can optimize the point cloud I got from zed

I had gotten and saved detect target 's point cloud data with YOLOV3 and pyzed package’s retrive measure function and then I converted the data to .pcd file. But when I loaded point cloud by MATLAB, I found that the point cloud data’s accuracy was very bad. For example, I got a straight line’s point clouds which is on the pavement, but in MATLAB pcshow window, point cloud might be a curved line and might be not in the same plane(some points may be high and others may be low).
How can I make the point cloud show accurate information? Can you recommend some point cloud process method?
Thanks a lot for your help!

Hi @ujs1514lee
the quality of the depth map, and so the quality of the point cloud, is affected by many factors:

  • depth mode parameter
  • Confidence threshold parameter value
  • texturization level of the acquired surface
  • material of the acquired surface (is it reflective?)
  • environment light

I suggest you test QUALITY, ULTRA, and NEURAL depth modes, and you play with the confidence threshold parameter to filter out unrelieable depth values.