中文摘要：

根据有限数目的参考点,通过二维单目观测图像估计三维目标的姿态参数（又称PnP问题,perspective-n-point）是计算机视觉研究中的一个经典难题。当参考点的数目n〈6时,PnP问题为高度非线性问题并可能存在多个可行解。目前求解PnP问题的方法主要分为迭代解法和闭式解法两类。迭代解法数值精度高,但是只能收敛到多解中的一个解,无法同时得到全部可行解;闭式解法的优点是可以一次得到全部可行解,但是现有算法在数值精度和数值稳定性上要逊于迭代解法。针对以上问题,以P3P问题为研究对象,提出一种可以同时得到全部可行解并具有高数值精度的半闭式解法,并通过详细的实验验证该方法的有效性。

英文摘要：

Estimating 3D object＇s pose parameters from 2D monocular image with limited number of reference points（known as PnP problem） is a classical difficult problem in computer vision research.When the number of reference points n6,PnP problem is highly nonlinear and multiple feasible solutions may exist.State-of-art methods for solving PnP problem mainly divide into two categories： iterative methods and closed-form methods.Iterative methods are highly precise,but can converge to only one of all the feasible solutions at a time;whereas closed-form methods can retrieve all the feasible solutions at a time,but are inferior to iterative methods in numerical precision and stability.This article proposed a new semi-closed method for P3P problem,which had high numerical precision.And proved the effectiveness of proposed method with thorough experiments.