基于点对应的相机姿态估计算法缺少科学的性能评价方法，增加了工程应用中算法选择的难度。针对该问题，提出了在特定代价函数下的相机姿态估计算法性能评价方法，主要包括3个性能评价参数：精度、效率和存在区域最优解成功率。其中，区域最优解不同于局部最优解，若给定区域为代价函数定义域，则区域最优解等价于全局最优解。着重阐述了存在区域最优解的判断方法，以角度残差为基础建立代价函数，利用姿态矩阵计算代价函数的Hessian矩阵下界，若Hessian矩阵下界半正定，则代价函数在以该姿态矩阵为中心、大小由像点噪声模型确定的邻域内为凸函数，即存在区域最优解。借助仿真实验平台，对9种经典相机姿态估计算法进行了性能评价。结果表明，RPnP+LHM算法的综合性能最优。

Camera pose estimation algorithm based on point correspondence is lack of scientific performance evaluation method, which increases the selection difficulty for algorithms in the engineering applications. Aimed at the problem, the evaluation strategy for camera pose estimation algorithm under specific cost function is put forward, mainly including three performance evaluation parameters: precision, efficiency and success rate of domain optimal solution. Domain optimal solution is different from local optimal solution. If a given area is at the definition domain of cost function, the optimal solution of this area is equivalent to the global optimal solution. The judgment method of domain optimal solution is described emphatically, the cost function is set up based on angle residual, and the lower bound Hessian matrix of cost function is calculated using attitude matrix. If lower bound of Hessian matrix is positive semi-definite, the cost function is convex function in the neighborhood that centers on pose matrix and the size is confirmed by the image point noise model. There exists optimal solution in given domain. In virtue of simulation experimental platform, nine classical camera pose estimation algorithms performance are evaluated. Results indicate that the comprehensive performance of RPnP+LHM method is the best.