【摘要】 直接线性变换方法是数字摄影测量和计算机视觉领域最常用的解析处理方法,经典最小二乘方法令L12=1,且不考虑像点坐标和空间坐标的误差,总体最小二乘方法虽然考虑了设计矩阵的误差,但假设误差是独立等方差分布的,本文采用了考虑残差统计特性的近似最大似然估计方法,该方法能有效提高估计精度,并用模拟和实测数据将本文方法与常用方法比较。更多还原

【Abstract】 Direct Linear Transformation is a widely used analytical approach in photogrammetry and computer vision fields.Traditional method for DLT is least squares estimation with over-determined linear equations with L12=1,and without consideration of image points and corresponding 3D coordinates errors.Total least squares methods regard the design matrix error as the independent and identical Gaussian noise,which is always not true in application case.A new method named Approximated Maximum Likelihood Estimation is presented in this paper,it depends on the residual covariance through first order error propagation,also the simulated and real data experiments are made to verify the algorithm.更多还原