Rotation Averaging Based PNP

In this paper, an extendable PNP method based on rotation averaging(RA) is proposed. Rotation matrices derived by P3P are averaged to suppress noise. By comparing the typical P3P methods, we find that the roots of the quartic polynomial consist of real ones and complex ones. Solutions based on the real roots are identical, while the solutions based on the complex ones are different. So the speed is the key parameter for the P3P implement and Kneip's method wins this contest. Next, four P4P methods are compared with each other, including the RA method proposed in this paper. For P4P problem, the RA method has the best performance. Finally, the comparison within four PNP methods shows that RA method based PNP is free from planar degeneration and can be extended easily.