0

I can get the depth map from depth estimation which uses supervised or unsupervised learning, but here is an obstacle that I want to get 3D coordinates from depth maps. If anyone can provide some suggestions.

Feona
  • 67
  • 2
  • 8
  • Do you have access to the camera's intrinsics? – Ivan Aug 20 '21 at 08:38
  • I can get the intrinsics – Feona Sep 01 '21 at 01:12
  • Given `D` the 2D depth map , `K^-1` the inverse of the camera intrinsics matrix, and `p` the 2D pixel coordinates of the image. The transformation pipeline is `D*K^-1*p`. Have you heard about such transformation before? – Ivan Sep 01 '21 at 08:27
  • Yes, I have heard about this transformation, change the camera world to the true world – Feona Sep 03 '21 at 07:26
  • This transformation unprojects the pixel elements to 3D view using the camera intrinsics matrix `K`, and your depth map `D` which removes ambiguity in the problem. I've you had a go at it? I suggest you try on paper and see how you can unproject a 2D point from image coordinate to camera view using the transform above. Then I can help with the implementation details. – Ivan Sep 03 '21 at 07:49
  • 1
    Any solution to this? – Kasra Jan 07 '22 at 22:20

0 Answers0