看文献里也没具体说要几个点的 都说用严格物理模型正射校正对GCP点数目要求少 然后就没了···囧

以下内容是在erdas的field guide pdf里找到的

（p.s. 往笔记里记的时候 那2张图真是让我郁闷良久啊··· 唉···怎么也画不好 =_=||）

GCP Requirements

The minimum GCP requirements for an accurate mapping project vary with respect to the size of the project. With respect to establishing a relationship between image space and ground space, the theoretical minimum number of GCPs is two GCPs having X, Y, and Z coordinates and one GCP having a Z coordinate associated with it. This is a total of seven observations.

In establishing the mathematical relationship between image space and object space, seven parameters defining the relationship must be determined. The seven parameters include a scale factor (describing the scale difference between image space and ground space); X, Y, Z (defining the positional differences between image space and object space); and three rotation angles (omega, phi, and kappa) that define the rotational relationship between image space and ground space.
In order to compute a unique solution, at least seven known parameters must be available. In using the two X, Y, Z GCPs and one vertical (Z) GCP, the relationship can be defined. However, to increase the accuracy of a mapping project, using more GCPs is highly recommended.
The following descriptions are provided for various projects:

• Processing One Image:

If processing one image for the purpose of orthorectification (i.e., a single frame orthorectification), the minimum number of GCPs required is three. Each GCP must have an X, Y, and Z coordinate associated with it. The GCPs should be evenly distributed to ensure that the camera/sensor is accurately modeled.

• Processing a Strip of Images

If processing a strip of adjacent images, two GCPs for every third image is recommended. To increase the quality of orthorectification, measuring three GCPs at the corner edges of a strip is advantageous. Thus, during block triangulation a stronger geometry can be enforced in areas where there is less redundancy such as the corner edges of a strip or a block.

下图 illustrates the GCP configuration for a strip of images having 60% overlap. The triangles represent the GCPs. Thus, the
image positions of the GCPs are measured on the overlap areas of the imagery.

• Processing Multiple Strips of Imagery

下图depicts the standard GCP configuration for a block of images, comprising four strips of images, each containing eight
overlapping images.

In this case, the GCPs form a strong geometric network of observations. As a general rule, it is advantageous to have at least one GCP on every third image of a block. Additionally, whenever possible, locate GCPs that lie on multiple images, around the outside edges of a block, and at certain distances from one another within the block.