Adaptation of TIN-based Ortho-Mosaicking for DSM Error Mitigation
Sunghyeon Kim(a) , Taejung Kim(b*)

a) Program in Smart City Engineering, Inha University, 100 Inha-ro, Incheon 22212, Republic of Korea
b) Department of Geoinformatic Engineering, Inha University, 100 Inha-ro, Incheon 22212, Republic of Korea
*tezid[at]inha.ac.kr

Abstract

Ortho-mosaicking is a key process required to generate image maps over large areas from UAV imagery. For ortho-mosaicking, multiple UAV images are ortho-rectified by referencing a Digital Surface Model (DSM) and mosaicked to a common image plane. Traditionally, DSMs used for ortho-mosaicking are generated either from LiDAR surveys or dense matching of aerial/UAV imagery. While LiDAR-based DSMs offer high accuracy, they entail high costs and time due to expensive equipment and the very fine spatial resolution of UAV images. In contrast, stereo-based DSMs are more cost- and time-efficient, but frequently suffer from elevation errors in regions such as building boundaries or occlusion zones, leading to geometric inconsistencies in the mosaics. To address these limitations, this study extends the triangulated irregular network (TIN)-based ortho-mosaic approach developed in-house to handle DSM errors. The method divides the DSM space into grid patches, whose sizes and shapes are adaptively adjusted based on local terrain. Three-dimensional points are extracted at the corners of each patch and used to form a TIN for ortho-mosaicking. The TIN vertices are reprojected into the image coordinate system using the original image^s projection model. Image patches are then extracted from the reprojected points and mapped to the mosaicking plane. The method was experimentally validated using UAV datasets over urban areas with significant elevation variation and by incorporating DSMs and DEMs of various resolutions. The experiments evaluated mosaic quality using alignment error, color discontinuity, and mosaic coverage as indicators, under varying grid patch sizes and DSM accuracy levels. Since the proposed method operates on grid patches, it maintains global consistency even under DSM inaccuracies. By adjusting patch size and shape, TIN vertices can omit problematic regions, such as building edges or occlusion zones, thereby reducing distortion and positional errors.

Keywords: DSM, UAV Image mosaic, geometric correction, patch size adaptation, urban mapping

Topic: Topic B: Applications of Remote Sensing