Abstract: Since the coefficient matrix of image point observation equations may contain random error, it is proposed that a bundle adjustment method for satellite linear array CCD imagery based on total least squares. Assuming that both point observation random error and the coefficient matrix random error are independent and identically distributed, a total least squares based bundle adjustment algorithm, which contains both virtual observation equations of exterior elements and error equations of ground control points, has been deduced using Lagrange conditional extremum. The variance of any type of virtual observation equation can be estimated using variance component estimation. Thus the proposed method can handle with adjustment problems with more than one type of virtual observation equation and chose the undetermined coefficient in proposed method using priori information or the ridge mark method, which overcomes the deficiency of existing virtual observation total least squares method and ensures that the adjustment problems can be solved correctly and effectively. Experiments have been taken on both simulative data and real satellite linear array images in two areas. Results indicate that the proposed method can get more accurate solutions than traditional least squares algorithm and recently proposed virtual observation total least squares method. The proposed method can also get more accurate bundle adjustment results when compared to conventional method.

Key words: linear array images, virtual observation equation, total least squares, bundle adjustment, ridge mark method