Accuracy optimisation and error detection in automatically generated elevation models derived using digital photogrammetry

Users of current Digital Photogrammetric Systems (DPS) can now rapidly generate dense Digital Elevation Models (DEMs) with a minimal amount of training. This procedure is controlled through a set of strategy parameters embedded in the software. Previous research into the effect of these parameters on the resulting DEMs produc'ed mixed results, with some researchers finding that significant changes to the DEM can be made through manipulation of the parameters whilst others suggested that they have little effect. This thesis builds upon this early work to develop two systems that provide assistance for novice users. The first technique optimises the parameters with respect to DEM accuracy and takes the form of an expert system and compares the output from the DEM with a knowledge base to prescribe an improved set of parameters. The results suggest that the system works and can produce improvements in the accuracy of a DEM. It was found that in certain circumstances, changes to the parameters can have a significant effect on the resulting DEM, but this change does not occur across the entire DEM. The second aspect of the thesis details the development of a completely new approach that automatically detects low accuracy areas of the DEM and presents this information graphically. This is an important development since, as documented in the current literature, few quality control procedures are offered to users. The user can use this information to assist in the manual checking and editing of the final DEM, thus speeding up the workflow and improving the accuracy of the output. The results of tests (using the ERDAS Imagine OrthoMAX software) on a wide variety of imagery are presented and show that the technique reliably detects areas of a DEM with high errors. More significantly, the technique has also been tested on two other DPSs (Zeiss Phodis TS and VirtuoZo) and it was found that it worked well for the Zeiss system but could not be applied to the VirtuoZo software. This demonstrates that the research is not limited to the users of one software package and is of interest to the wider photogrammetric community.