Reconstruction of multiple image-based high-resolution digital elevation model using planetary reflectance

Abstract

This dissertation discusses restoration of lunar surface topography with shadows over time. The digital elevation model (DEM) is generated by a gradient of a single image and a stereo image pair from imaging sensor or laser altimeter sensor on the orbiter. These DEMs are affected by intrinsic problems such as stereo noise in the triangulation process or the low spatial resolution of the sparse measurement point. Many image-based techniques are used to enhance the resolution of DEM which can be very challenging as the calculation of each surface point direction vector from one or partially overlapped images requires many assumptions. In this paper, an approach to enhance the resolution of coarse DEM using the information of surface normal extracted from ortho-rectified planetary image dataset is proposed. The surface normal extracted from ortho-rectified planetary image dataset and coarse DEM are well-aligned and the surface normal also offers direction vectors of uniform quality in every surface point. The key to this improvement is the separation of the surface normal as a map to overcome the ambiguity of the role of surface direction information. Using the high-resolution surface normal information, the coarse DEM is optimized to show the characteristic of terrain in the surface normal map. For performance analysis, the result of the proposed method was compared with a DEM generated using Socet Set Next-Generation Automatic Terrain Extraction (NGATE) of BAE Systems. It was demonstrated that the detail of terrain is well estimated in the high-resolution surface normal image and also in the output DEM. Restoration of surface shape information from planetary images requires planetary reflectance that are affected by many parameters including incidence, emission, and phase angles at each pixel. Complicated relationship between these angles and the terrain shape leads to difficulty in this restoration process. This paper proposes a method to effectively estimate the terrain surface shape using ortho-rectified image dataset taking into account the effect of planetary reflectance. The method extracts the Lambertian reflectance information, which depends only on the incidence angle, from the planetary reflectance by posing a sparse regression problem. This allows for more accurate estimation of the surface normal information. Moreover, this work visualizes the surface normal vector as an image for more intuitive assessment of the quality of the estimation. The effectiveness of the proposed method is demonstrated in the lunar and Mars surface image dataset.