*2.1. UAV Data*

To achieve the objective of the study, a rural area in Slovenia was selected as the number of visible (cadastral) boundaries in such areas is higher compared to dense urban ones. In addition, the selected rural area includes roads, agricultural field outlines, fences, hedges, and tree groups, which are assumed to indicate cadastral boundaries [22]. The UAV images of the case study area were indirectly geo-referenced, using an even distribution of ground control points (GCP) within the field as criteria. The GCPs were surveyed with real-time kinematic (RTK) by using Global Navigation Satellite System (GNSS) receiver, Leica Viva, connected in the Slovenian GNSS network, SIGNAL. The signals were received from satellite constellations of GPS and GLONASS. The total number of GCPs was 12. The Position dilution of precision (PDOP) values ranged from 1.2 to 1.7. The flight altitude was 80 m and 354 images were taken to cover the study area. The images were captured on October 19th, 2018 in the noon time (good weather conditions, clear sky) at solar zenith angle of approximately 35 degrees. The study site had a coverage area of 25 ha. The planimetric accuracy assessment of the UAV orthoimage was based on comparison between GCPs coordinates surveyed with the GNSS receiver and the coordinates of GCPs on the UAV orthoimage. The estimated root-mean-square-error (RMSE) was 2.5 cm. Table 1 shows the specifications of data capture and Figure 1 shows the UAV orthoimage of the study area.



**Figure 1.** (**a**) Cadastral map and ground control points (GCPs). (**b**) Manually delineated object visible boundaries used as reference data to determine the detection/extraction quality. (**a**,**b**) Overlaid on UAV orthoimage of Ponova vas, Slovenia (EPSG 3794).

#### *2.2. Reference data*

The current cadastral map for the selected area was retrieved from the e-portal of the Slovenian Surveying and Mapping Authority, which is an online platform for requesting official cadastral data [42]. The cadastral map was overlaid on the UAV orthoimage (Figure 1a). The visual interpretation of the combined dataset showed immediately that the cadastral map does not correspond with the visible objects that indicate land possession (land cover) boundaries (roads, agricultural field outlines) on the UAV orthoimage. From the initial analyses, it appeared that only 8% of cadastral boundaries correspond with the manually digitized visible boundaries (at 25 cm tolerance). This is because the current official cadastral map was created by digitizing previous analog cadastral maps whose origin goes back in the first half of 19th century. Due to the underestimated need for cadastral map updating as well as due to the land reforms in the 20th century (i.e., land nationalization and denationalization) the current possession land boundaries do not correspond with cadastral boundaries. Considering this, as reference data, manually digitized boundaries were used instead of the official cadastral data, as the aim of this research is to automatically delineate visible object boundaries from a UAV orthoimage and, at the same time, study the potential of the ENVI FX solution for the automatic detection of visible boundaries. Moreover, during the manual digitization of reference boundaries, some white stones considered as possession boundary signs were used as a guide for proper digitization (Figure 1b). The placement of white stones is a common practice in the selected study area, and for this reason, they were considered as reliable information during the manual digitization. In addition, the confidence in white stones as boundary signs is based on the authors' experiences in professional cadastral surveying.
