- Lidar (LIght Detection And Ranging) is a remote sensing technology using light in the form of a pulsed laser, which enables efficient, accurate, 3-D data acquisition of a scene. Depending on the mounting platform, lidar data acquisition can be categorized into Airborne Laser Scanning (ALS, or airborne lidar), Terrestrial Laser Scanning (TLS, or terrestrial lidar), and Mobile Laser Scanning (MLS, or mobile lidar). The lidar technique has been widely used for a plethora of applications including topographic mapping, bathymetric mapping, utility mapping, engineering surveying, agriculture, forestry, geology, architecture, industrial facilities, cultural heritage, asset management, construction, and so forth. However, efficiently processing the dense datasets produced by lidar still remains challenging given the large data volume. In addition, because of the scan pattern, range, view angle, and other factors, the point density for terrestrial and mobile lidar data can vary dramatically across the scene, which raises different challenges in developing robust processing methods compared with an ALS point cloud, which tends to be more evenly distributed. To overcome the challenges in processing TLS and MLS data, in this research, the point cloud is structured into a 2-D grid structure called the scan pattern grid, which represents the way that a scanner collects data. This dissertation, comprising four manuscripts, explores the possibilities and performance improvements of exploiting this scan pattern grid to process point cloud data.
This first manuscript presents an efficient ground filtering method for TLS data via a Scanline Density Analysis. Ground filtering is a common procedure in lidar data processing, which separates the point cloud data into two classes: ground points and non-ground points. The proposed process first separates the ground candidates, density features, and unidentified points based on an analysis of point density within each scanline. Second, a region growth using the scan pattern clusters the ground candidates using the density features as boundaries and further refines the ground points. Both stages process and analyze the TLS data in each scan separately, exploiting the scan pattern grid for efficiency.
The next two manuscripts develop a novel point cloud segmentation with an approach that links the scan pattern grids from multiple scans during the analysis. Point cloud segmentation groups points with similar attributes with respect to geometric, colormetric radiometric, and/or other information to help with object extraction and interpreting the point cloud. The proposed segmentation method only requires the basic geometric information and consists of two main steps. First, a novel feature extraction approach, NORmal VAriation ANAlysis (Norvana), eliminates some noise points and extracts edge points without requiring a general (and error prone) normal estimation at each point. Second, region growing groups the points on a smooth surface using the edge points as boundaries to obtain the segmentation result.
Unlike TLS data that can be directly structured from a structured format (e.g., ASTM E57), Mobile lidar data is usually stored in an unorganized manner (e.g., ASPRS LAS). The final manuscript presents an efficient mobile lidar data processing framework including an approach to reconstruct the scanner trajectory such that an unorganized point cloud can be structured into the scan pattern grid based on the order of acquisition and revolutions of the scanner. Then the concept of Norvana for edge detection, normal estimation, feature extraction, and segmentation, is extended to be suitable for processing mobile lidar data and is named Mo norvana. Additionally, the proposed framework also introduces an efficient data visualization scheme exploiting the scan pattern grid.
All of the proposed methods implement parallel processing to obtain a higher computational performance. The effectiveness, efficiency, robustness, and versatility are demonstrated both qualitatively and quantitatively by testing multiple terrestrial and mobile lidar datasets collected by different scanners with different spatial scales, resolutions, and scene types. The key contribution of this research is a generalized point cloud processing framework that can efficiently support a wide range of refinements, processes, and analysis for a variety of applications.