An Automated Framework for Cadastral Parcel Adjustment Using UAV Orthophotos, SAM, and ICP

Abstract. Accurate cadastral boundary data is essential for land administration, yet inconsistencies remain in many registered parcels in Indonesia due to legacy georeferencing systems, fragmented survey methods, and non-uniform base maps. This study presents an automated pipeline for improving the spatial quality of cadastral polygons by leveraging UAV photogrammetry, artificial intelligence, and geometric adjustment techniques. High-resolution orthophotos (5 cm GSD) were acquired over Cimahi City, Indonesia, using a VTOL UAV platform at 300 meters altitude. A zero-shot segmentation approach based on the Segment Anything Model (SAM) was employed to extract boundary indication segments, which were then classified into five land cover types using a Random Forest model. The SAM-derived segments were preprocessed using geometric rules (simplification, orthogonalization, and centerline extraction) and used as reference layers for alignment with existing cadastral polygons. Parcel grouping into blocks was performed automatically based on spatial proximity. Block-level alignment was conducted using the Iterative Closest Point (ICP) algorithm, followed by hierarchical least squares adjustment at three levels (LS1, LS2, LS3), depending on residual magnitude. Outliers were detected through statistical analysis of displacement vectors and flagged for shape and attribute matching. The methodology was tested on two urban villages—Karangmekar and Baros—comprising 177 blocks and 6,198 parcels. Results showed significant improvement in spatial accuracy, with mean displacement reduced from 0,8 meters to 0.4 - 0.5 meters. Geometric quality metrics (compactness, rectangularity, elongation, etc.) confirmed increased parcel regularity post-adjustment. The integrated workflow supports scalable cadastral data enhancement and is suitable for deployment in national land registration initiatives.