Comparative Analysis of Different UAV-Based Photogrammetric Processes to Improve Product Accuracies

Producing geographic information has become increasingly widespread over the last few years thanks to the development of new sensors, tools and algorithms, easily to be implemented and user friendly. Nevertheless, the accuracy of photogrammetric outcomes, obtained by automated software instructions and the use of un-calibrated cheap sensors, is often unsatisfying. As a consequence, the results accuracy and the potentialities and repeatability of those procedures have to be improved, defining the best workflow to be adopted. In literature, different approaches related to the correct and efficient reduction of geometric errors are available, although, to date, comprehensive method has not yet been defined. This research work is aimed to detect an optimized workflow on the base of the comparison the accuracies achieved by applying different 3D reconstruction procedures of the cultural heritage. The process was subdivided in two steps: the former is related to a processing test, executed by analyzing four image-datasets acquired with a prosumer UAV equipped with a non-metric camera and a low-accuracy GNSS/INS receiver; the latter is based on an high-accuracy ground-truth survey, performed to evaluate the lever-arm and the camera self-calibration parameters, which are fundamental for reducing the errors propagation in the final accuracy. Thus, different 3D models were generated, modifying the processed image datasets and the spatial and numerical distribution of the Ground Control Points (GCPs). The root-mean-square error (RMSE) values on the Control Points (CPs), compensated differently in the Bundle Block Adjustment process, were assessed for each processing. Promising results were achieved in order to validate an optimal photogrammetric workflow.