Digital surface model generation for drifting Arctic sea ice with low-textured surfaces based on drone images

Abstract

Arctic sea ice is constantly moving and covered with low-textured surfaces, making it difficult to generate reliable digital surface models (DSMs) from drone images. The movement of sea ice makes georeferencing of DSMs difficult, and the low-textured surfaces of sea ice cause the uncertainty of image matching. This paper proposes a robust method to generate high-quality DSMs for drifting sea ice. To overcome the challenges, the proposed method introduces four improvements to the object-space-based image-matching pipeline: relative georeferencing to recover the horizontality and scale of sea-ice DSMs using a terrestrial light detection and ranging (LiDAR) dataset, match inspection to verify the matched points using several matching constraints, adaptive search-window adjustment to ensure distinct texture information through simple texture analysis, and robust vertical positioning to reduce the matching uncertainty via matching-indicator modeling. Performance evaluations were conducted with drone and LiDAR datasets obtained from a sea-ice campaign using the Korean Icebreaker Research Vessel (IBRV) Araon in the summer of 2017. The experimental results indicated that the proposed method can achieve significant quality enhancements compared with the existing matching method and that all the considerations contributed significantly to the enhancements.