The capabilities of modern three-dimensional (3D) capture technology such as laser scanning and image-based 3D reconstruction are well suited to enhance the practice and research of civil engineering. However, given the often-overwhelming focus placed on the incredible capabilities of these tools and techniques, it is important to investigate the limitations of these technologies to ensure they are not misused. Currently, limited resources are available to assist in the evaluation of 3D geospatial data quality, which renders it difficult to efficiently quantify, and communicate the limitations of such data.
To this end, this research investigates the occurrence of data gaps in terrestrial laser scanning (TLS)-derived digital elevation models (DEMs), the quality and accuracy of Structure from Motion (SfM) image-based 3D reconstructions, and the inherent positional uncertainty of individual points in a TLS point cloud. Novel approaches for the detection and classification of data gaps, the evaluation of data suitability, and the efficient computation and visualization of per-point TLS point cloud uncertainty will be discussed.
With regards to TLS data gaps, a novel data gap classification methodology for TLS-derived DEMs was developed, which automatically detects and differentiates between occlusion and dropout-based data gaps (Chapter 2). This methodology facilitates the assessment of TLS survey quality and can be used to determine the location and surface area of pooled water for scientific research. The data gap classification methodology can also be used for efficient quality evaluation of DEMs and subsequent optimization of TLS acquisition strategies (Chapter 3).
In situations where laser scanning results in significant occlusion-based data gaps that are difficult to mitigate, image-based 3D reconstruction (i.e., SfM) is a possible alternative. In support of investigating the capabilities and limitations of SfM-based 3D reconstructions, a suitability evaluation of unmanned aircraft systems (UAS) and handheld camera-based SfM was performed in the context of automated unstable rock-slope assessment (Chapter 4). The evaluation includes both a rigorous accuracy assessment and quality evaluation of SfM-derived 3D geometry. TLS-derived 3D geometry was found to be more accurate; however, using both UAS and handheld camera-based imagery is a viable option for unstable rock slope characterization when tied to rigorous survey control. Nevertheless, concerns such as over-smoothing and inconsistencies question the suitability of SfM reconstruction for reliably detecting small rock-slope changes over time.
Lastly, a TLS point cloud uncertainty visualization framework was developed to intuitively communicate per point uncertainty during interactive 3D visualization (Chapter 5). Uncertainty propagation amongst the points is performed out-of-core using vertex computations in the OpenGL pipeline made possible by OpenGL Shader Language (GLSL) programming. The flexibility of this visualization solution provides the ability to efficiently adjust error parameters and perform visual-sensitivity analyses. The proposed framework was tested on four unique datasets and aided in the development of a new beamwidth-derived range error equation that incorporates laser beam exit diameter.