Jump to Main Content
Precise georeferencing using the rigorous sensor model and rational function model for ZiYuan-3 strip scenes with minimum control
- Pan, Hongbo, Tao, Chao, Zou, Zhengrong
- ISPRS journal of photogrammetry and remote sensing 2016 v.119 pp. 259-266
- cameras, geometry, georeferencing, image analysis, models, satellites
- The rigorous sensor model (RSM) and the rational function model (RFM) are the most widely used geometric models for georeferencing. Even though geometric calibration and bundle adjustment with the RFM has been carried out for the ZiYuan-3 (ZY-3) earth observation satellite, few studies determined the major error sources affecting the three line cameras (TLCs). In this work, we propose a new set of compensation parameters, the shift and drift of both pitch and roll angle, for the RSM, since the yaw angle error is not as significant as the pitch angle for very narrow field of view images. Corresponding bias compensation methods are also validated for the RFM. Seven continuous strip scenes from the ZY-3 TLCs are used for the experiments, for which the root mean square error (RMSE) in the image space and object space are calculated. The experimental results demonstrate that the proposed method can model the major errors and achieve the same accuracy as the use of redundant parameters. With this model, the RMSEs of the checkpoints are 2.048m in planimetry and 1.256m in height. The RMSEs would increase to 2.522m in planimetry and 2.635m in height if the drift parameters were ignored. However, subpixel georeferencing accuracy is not as sensitive as the RMSE in the object space, since the RMSE of the height increases to 2.6m compared to 1.3m, while the change of the RMSE in the image space is within 0.1 pixels. In addition, the relationships among the TLCs are dynamic during imaging. Compensation for the TLCs as a unit introduces a height error of about 1m, while maintaining subpixel georeferencing accuracy. Two ground control points (GCPs) placed at the beginning and the end of a strip are preferred to reduce oscillation and point picking errors. Compared with the RSM, the RFM can achieve similar accuracy when the drift compensation model and shift compensation model are applied.