Assuming that P 3D object points are observed from F views, we let
be 3-vector denoting the k-th object
point which projects onto a 2D image point represented by a 2-vector
in the i-th
frame. If each camera
accurately provides perspective projection, each object point is mapped
to an image point according to the following projection equation:
where and
are homogeneous coordinates of the
object and the image points,
and
are a
3-vector and a 3
3 orthogonal matrix that represent the
position and orientation of i-th camera, and
is a
3
3 upper triangular matrix giving intrinsic camera parameters
respectively[3]. The scaling factor
represents the depth of the k-th point from the
i-th camera. Equation (1) is called
a perspective camera model.