A robust O(n) solution to the perspective-n-point problem

Abstract

We propose a noniterative solution for the Perspective-n-Point (PnP) problem, which can robustly retrieve the optimum by solving a seventh order polynomial. The central idea consists of three steps: 1) to divide the reference points into 3-point subsets in order to achieve a series of fourth order polynomials, 2) to compute the sum of the square of the polynomials so as to form a cost function, and 3) to find the roots of the derivative of the cost function in order to determine the optimum. The advantages of the proposed method are as follows: First, it can stably deal with the planar case, ordinary 3D case, and quasi-singular case, and it is as accurate as the state-of-the-art iterative algorithms with much less computational time. Second, it is the first noniterative PnP solution that can achieve more accurate results than the iterative algorithms when no redundant reference points can be used (n≤ 5). Third, large-size point sets can be handled efficiently because its computational complexity is O(n).