Revision history [back]

No you don't have to normalize the values for objectPoints. If you do it, the translation should be estimated up to a scale factor.

The common requirements are:

• the x, y, z components must be expressed in the same units
• the minimum number of points is 4 to estimate correctly the camera pose
• the 3D points should not be collinear

I think that some configurations could lead to singularities or camera pose badly esimated or completly wrong. For instance, if all the imagePoints are projected almost on a same line (with 4 coplanar 3D points for instance)?

No you don't have to normalize the values for objectPoints. If you do it, the translation should be estimated up to a scale factor.

The common requirements are:

• the x, y, z components must be expressed in the same units
• the minimum number of points is 4 to estimate correctly the camera pose
• the 3D points should not be collinear

I think that some configurations could lead to singularities or camera pose badly esimated or completly wrong. For instance, if all the imagePoints are projected almost on a same line (with 4 coplanar 3D points for instance)?

No you don't have to normalize the values for objectPoints. If you do it, the translation should be estimated up to a scale factor.

The common requirements are:

• the x, y, z components must be expressed in the same units
• the minimum number of points is 4 to estimate correctly the camera pose
• the 3D points should not be collinear

I think that some configurations could lead to singularities or camera pose badly esimated or completly wrong. For instance, if all the imagePoints are projected almost on a same line (with 4 coplanar 3D points for instance)?

Edit:

I found an interesting course, "The "The Perspective View of 3 Points" Points" by Bill Wolfe, which give the following information: