Abstract:
The research involves study of two schemes of quality assessments for photo control points namely: Hold Out Validation (HOV) and Leave One Out Cross Validation (LOOCV). Quality photo control point is crucial for sensor model refinement and further orthorectification processing. Since the sensor model for the used high-resolution satellite image WorldView-1 is not specified by the operator. At the firstly the appropriated sensor models were extensively carried out. The opted sensor model for exterior orientation parameters were two biases for the three positions ( ) and two biases for Euler angles namely. However for the angle around Z-axis, polynomials of second degree were adopted. For this kind of sensor model refinement, it required 8 unknowns; therefore it demanded at least 4 GCPs. In the research the quality assessment engaged 78 photo control points. The first HOV method could discriminate good quality photo control points from the poor and obtained 72 points or 92%. These 72 qualified photo controls were deployed as 13 GCPs and 59 independent check points. The HOV could identify 6 outlier GCPs. In contrast to the LOOCV, it identified 52 qualified photo control points. These photo control points were deployed as 8 GCPs and 44 independent check points. Apparently HOV methods could identify more qualified photo control points than the LOOCV. Nevertheless LOOCV behaved more strict than HOV did. After the sensor model refinement were finished, then orthorectifications were followed. The chosen GCPs located along the edges of satellite image and they were well parallel to the orbit track of the scene. The RMSE after sensor model refinement was 0.17 meter. The orthophoto resulting from GCPs qualified from different schemes were assessed according to the US NSSDA standard. The orthophoto involved 13 GCPs and 59 ICPs from HOV, the horizontal accuracy was 1.31 meters (2.62 pixels), where the one with 8 GCPs and 44 ICPs from LOOCV, the accuracy was 0.70 meters (1.40 pixels). Apparently the LOOCV scheme was superior than the HOV in this research.