Iterative Pose Estimation Using Coplanar Feature Points

of the matrix in a singular value decomposition [9]. A general algorithm which assumes that these feature points This paper presents a new method for the computation of the position and orientation of a camera with respect to a known are noncoplanar and fails to detect that this case is degenerobject, using four or more coplanar feature points. Starting with ate would probably produce inaccurate camera pose estithe scaled orthographic projection approximation, this method mations. iteratively refines up to two different pose estimates, and proFor the case of coplanar feature points, researchers have vides an associated quality measure for each pose. When the formulated closed form solutions for configurations of camera distance is large compared with the object depth, or three feature points and four feature points. The P3P probwhen the accuracy of feature point extraction is low because lem (with three noncollinear points) can have as many as of image noise, the quality measures for the two poses are four possible solutions [6, 4, 11]. On the other hand, the similar, and the two pose estimates are plausible interpretations P4P problem has a single theoretical solution [3, 12, 1, 5] of the available information. In contrast, known methods using when the coplanar points are in an ordinary configuration a closed form pose solution for four coplanar points are not robust for distant objects in the presence of image noise because (no three collinear scene points, noncollinear image they provide only one of the two possible poses and may choose points). the wrong pose.  1996 Academic Press, Inc. Clearly, there is a problem with closed form calculations claiming a single solution for four coplanar points. The problem can be detected by the following reasoning: