A new method for automatic gross error detection in remote sensing image geometric correction

Conventional gross error detection methods are mainly based on Gauss-Markov model and Least Squares Estimation, and are not adapted to gross error detection for control points of satellite remote sensing images, due to the serious ill-condition of satellite remote sensing imaging model and many iterations in the solving process. This paper proposed a method automatically detecting gross error of control points for geometric correction of satellite remote sensing images. This method substitutes the simple Least Squares Estimation with Levenberg-Marquardt (LM) algorithm, and removes one control point with the maximum standardized residual before updating the imaging model each time, until all the gross errors are eliminated. The disadvantages of traditional data detecting methods based on hypothesis testing were analyzed first, then a new approach determining the stopping point of gross error detection was put forward, that is clustering the absolute values of standardized residual differences. Experimental results and comparisons with other methods confirmed the validity of the proposed method.