Camera Motion Estimation
In my SLAM video, I use the 2D-3D method.
From the SLAM textbook
- 2D-2D: When the camera is monocular, we only know the 2D pixel coordinates, so the problem is to estimate the motion according to two sets of 2D points. This problem is solved by Epipolar Geometry.
- 3D-3D: When the camera is binocular, RGB-D, or the distance is obtained by some method, then the problem is to estimate the motion according to two sets of 3D points. This problem is usually solved by ICP.
- 3D-2D: If one set is 3D and one set is 2D, we get some 3D points and their projection positions on the camera, and we can also estimate the camera’s pose. This problem is solved by PnP.
- This is the method I show in the video