Sparse Aerial Ground Height Recovery Using Normalized Weighted Nonlocal Laplacian

Accurate distance information is essential for remote sensing mapping and 3D information reconstruction. Dense depth reconstruction from sparse measurements in indoor and outdoor natural scenes has been extensively researched. However, in aerial scenes, dense height image reconstruction from sparse height measurements has received relatively limited attention. This paper investigates the challenge of reconstructing aerial ground height from sparse height measurements. In this paper, we comprehensively evaluate the feasibility and performance of various sparse reconstruction algorithms in aerial scenes for the first time, and explore the optimal sampling rate and pattern for preserving the image structure and details. Additionally, we propose two novel algorithms based on the normalized weighted non-local Laplacian, which can achieve state-of-the-art results without relying on color image guidance. Moreover, our algorithm is inherently robust to different sampling configurations. The solver tools relied upon also have proven hardware and software optimisation and acceleration schemes in the industry, making them well suited for deployment in real-world scenarios. We validate the effectiveness and robustness of the proposed algorithms on three datasets.