One of the greatest recent changes in the field of remote sensing is the addition of high-quality Light Detection and Ranging (LIDAR) instruments. In particular, the past few decades have been greatly beneficial to these systems because of increases in data collection speed and accuracy, as well as a reduction in the costs of components. These improvements allow modern airborne instruments to resolve sub-meter details, making them ideal for a wide variety of applications. Because LIDAR uses active illumination to capture 3D information, its output is fundamentally different from other modalities. Despite this difference, LIDAR datasets are often processed using methods appropriate for 2D images and that do not take advantage of its primary virtue of 3-dimensional data.
It is this problem we explore by using volumetric voxel modeling. Voxel-based analysis has been used in many applications, especially medical imaging, but rarely in traditional remote sensing. In part this is because the memory requirements are substantial when handling large areas, but with modern computing and storage this is no longer a significant impediment. Our reason for using voxels to model scenes from LIDAR data is that there are several advantages over standard triangle-based models, including better handling of overlapping surfaces and complex shapes. We show how incorporating system position information from early in the LIDAR point cloud generation process allows radiometrically-correct transmission and other novel voxel properties to be recovered. This voxelization technique is validated on simulated data using the Digital Imaging and Remote Sensing Image Generation (DIRSIG) software, a first-principles based ray-tracer developed at the Rochester Institute of Technology.
Voxel-based modeling of LIDAR can be useful on its own, but we believe its primary advantage is when applied to problems where simpler surface-based 3D models conflict with the requirement of realistic geometry. To show the voxel model's advantage, we apply it to several outstanding problems in remote sensing: LIDAR quality metrics, line-of-sight mapping, and multi-model fusion. Each of these applications is derived, validated, and examined in detail, and our results compared with other state-of-the-art methods. In most cases the voxel-based methods demonstrate superior results and are able to derive information not available to existing methods. Realizing these improvements requires only a shift away from traditional 3D model generation, and our results give a small indicator of what is possible. Many examples of possible areas for future improvement and expansion of algorithms beyond the scope of our work are also noted.
Library of Congress Subject Headings
Optical radar--Data processing; Remote sensing--Data processing; Image analysis
Imaging Science (Ph.D.)
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Jan van Aardt
Hagstrom, Shea T., "Voxel-Based LIDAR Analysis and Applications" (2014). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus