BLOCK ADJUSTMENT OF LARGE-SCALE HIGH-RESOLUTION OPTICAL SATELLITE IMAGERY WITHOUT GCPS BASED ON THE GPU
Precise geo-positioning of high-resolution optical satellite imagery without ground control points (GCPs) has always been the goal pursued by photogrammetry scholars. This paper introduces the block adjustment (BA) without GCPs based on rational function model (RFM) model and its practical application in high-precision geo-positioning of optical satellite imagery. The mainly key technologies of BA model construction based on virtual control points (VCPs), gross error detection and elimination, and GPU parallel computing method of large-scale adjustment are studied. On this basis, experimental analysis and validation of 123 images of ZY-3 satellite in Taihu are carried out. The results show that the sparse matrix compression can reduce the memory requirement effectively. The GPU parallel computing can solve the problem of large-scale BA computational efficiency. In addition, after BA, the maximum residual is 3.79 pixel, the root mean square error (RMSE) is 0.37 pixel in the x (flight) direction, the maximum residual error is 7.18 pixel, and the RMSE is 0.66 pixel in the y (scan) direction. The proposed method has certain accuracy and stability in large-scale BA without GCPs. The relative positioning accuracy can reach sub-pixel level, which could meet the requirements of cartographic mosaicking.