Three-dimensional reconstruction of computer vision (1) (camera geometry)

pinhole camera
  • Add a barrier: Use pinhole ( o=aperture=pinhole=camera center) to achieve one-to-one mapping of reality and imaging to reduce blur. The distance between the pinhole and the image plane is f(focal length); the virtual image plane is located between the pinhole and the real object, and has an inverted relationship with the image plane.
  • Position mapping: Use similar triangles to obtain the mapping coordinates of real coordinates on the image plane.
  • Aperture: As the aperture decreases, the image becomes clearer and darker.
lens
  • Concentrate multiple rays of light onto the film to increase the brightness of the image.
  • PS: All rays parallel to the optical axis will converge to the focus , and the distance from the focus to the center of the lens becomes the focal length; the direction of the rays passing through the center does not change.
  • Out of focus: The lens focuses light onto the film for a specific distance.
  • Radial distortion: pincushion distortion, barrel distortion. Light rays farther away from the center of the pinhole are more bent.
camera geometry
  • Points in space (with the camera aperture as the original point) are mapped to image coordinates: linear transformation is achieved using a projection matrix (fixed parameters for each camera).
  • Projection matrix: represents the mapping relationship.
  • PS: Camera deflection: When the image coordinate system is non-vertical, the angle parameters are put into the projection matrix.
    Insert image description here
    Insert image description here
    ( x , y , z ) (x,y,z)(x,y,z ) Realistic coordinates,MMM is the projection matrix, whereMMThe first three columns of M are the intrinsic parameter matrix of the camera (completely determining the mapping of spatial points to image points).

Normalized projection transformation: the image coordinates are consistent with the real coordinates.

world coordinate system

A conversion that does not take the camera aperture as the starting point is nothing more than adding an extra conversion matrix.

Guess you like

Origin blog.csdn.net/qq_44116998/article/details/132334489