Contemporary measurement techniques facilitate the rapid and highly precise development of three-dimensional models of any spatial object. Terrestrial laser scanning (TLS) stands as one of the most precise methodologies. Nevertheless, instances arise wherein restrictions imposed by the terrain configuration or infrastructure design impede the acquisition of comprehensive information regarding its geometry. In such scenarios, the optimal resolution lies in the integration of data sourced from diverse measurement instruments. In the context of working with large objects, the optimal approach to capturing comprehensive data, particularly pertaining to the upper parts, involves utilising an unmanned aerial vehicle (UAV). The high resolution of images acquired at a low altitude enables the generation of a point cloud with remarkable accuracy, delivering a satisfactory outcome. When it comes to the modelling of special objects, such as brine graduation towers, the selection of suitable software that facilitates the creation of realistic three-dimensional models is of paramount significance. The study utilised the integration of data acquired from a low altitude using the DJI Air 2S Fly More Combo unmanned aerial vehicle. Diverse mission types were employed, and the data was subsequently recorded using a terrestrial Leica ScanStation P40 laser scanner. The research was conducted on a brine graduation tower situated above the Nowa Huta reservoir in Kraków. The tower’s dimensions necessitated the incorporation of TLS and UAV data. This study analyses three 3D models of the brine graduation tower in Nowa Huta. The models were generated using various computer programmes, namely MeshLab, Agisoft Metashape, and Cyclone 3DR, each of which demonstrated specific capabilities and suitability for modelling a special object like a brine graduation tower. The accuracy of the constructed three-dimensional model of the tower was determined by comparing sections that were measured in the field on the structure between photographic points marked by discs and the equivalent points on the model. Eighteen sections were measured, yielding a mean error of 0.039 m.