A high-precision linear method for camera pose determination

This paper concerns a high-precision linear method for camera pose determination. The key idea is to replace the algebraic error in the classical linear method with the weighted algebraic error as an approximation of the geometric error. The proposed method provides a linear solution with accuracy closed to that of maximum likelihood (ML) estimation, but more efficient than ML estimation. Based on the EPnP algorithm [15], we obtain the weighted EPnP (WEPnP) algorithm by using the weighted linear technique. Experiments on simulative data and real images show that the WEPnP algorithm significantly outperforms the EPnP algorithm, what's more, it even outperforms the Lu's nonlinear algorithm [13] in the case of small depth ratio.