Visual analytics of three-dimensional airborne LiDAR point clouds in urban regions

Abstract

Airborne LiDAR datasets, in the form of three-dimensional point clouds, provide geometric information, owing to their spatial nature. Their popularity as a geospatial data acquisition technique is owing to its characteristics of low noise and high point density. Our work is a step toward three-dimensional analysis of the point clouds, where we discuss the role of visual analytics in point cloud processing. We discuss two different scenarios/applications: (a) unsupervised classification of point clouds and (b) local geometry analysis of point clouds. For classification, we discuss both structural as well as semantic classification of the points. Structural classes are points, lines, and surfaces, and semantic classes are buildings, ground (asphalt), ground (natural), and vegetation. Owing to the nature of the object classes we focus on, the datasets of our interest pertain to urban regions, where structures belonging to these object classes are found in plenty. The local geometric descriptor is formulated by using tensor voting and gradient energy tensor, where it is comparable to the conventionally covariance matrix. Overall, our research demonstrates how adding elements of user interactivity and visualizations in a data science workflow enables users to perform first-cut exploration of large-scale point clouds from airborne LiDAR.