Analysis of algorithms for automatic classification of photogrammetric point clouds to build a digital terrain model in a shoreline area

Unmanned aerial vehicles (UAVs) allow relatively cheap and fast acquisition of high-resolution data for small areas, making it possible to produce not only an orthophoto, understood as a two-dimensional product, but also a three-dimensional point cloud, which is the basis for obtaining a digital terrain model (DTM). The use of high-resolution image and elevation data may allow accurate shoreline delineation in areas where such measurement is not possible with other methods and further use of these data, e.g. for the production of electronic navigation charts. The accuracy of the final product, the DTM, is significantly affected by the photogrammetric classification process of the point cloud and the correct separation of the ground class. The aim of this study was to assess the effectiveness of the algorithms used to classify ground in photogrammetric point clouds and obtain correct DTMs. Three algorithms were tested: Adaptive Triangulated Irregular Network, Progressive Triangulated Irregular Network, and Multiscale Curvature Classification. The study involved two test areas covering natural shorelines. Images acquired with a UAV on the X8 system and a Sony alpha camera with a mounted 15 mm wide-angle lens were used as data for the research experiment. Accuracy analysis of the developed models was performed using check points acquired by the GNSS-RTK method.