Development of 3D Point Cloud Classification Algorithm for Varied Data Collection Methods

Authors: Katrina Schweikert*, Blue Marble Geographics
Topics: Geographic Information Science and Systems, Spatial Analysis & Modeling, UAS / UAV
Keywords: GIS, Lidar, Drone, UAV, Spatial Analysis, Segmentation
Session Type: Virtual Paper
Day: 4/9/2021
Start / End Time: 4:40 PM / 5:55 PM
Room: Virtual 24
Presentation File: No File Uploaded

With the recent expansion of spatial data collection from drones, as well as the lower cost of lidar sensors, there is a growing proliferation of 3D point cloud data used in various surveying and mapping applications. Common point cloud collection methods include photogrammetry, terrestrial lidar, UAV and helicopter-mounted lidar, bathymetric sensors, and aerial lidar. These varied sources produce point clouds that share the same basic structure but can have major differences in the point cloud density and distribution. Traditional lidar classification algorithms for aerial lidar relied on differentiating vegetation from the built environment based on multiple returns and the vertical overlap of points exclusive to forested areas. A new approach is necessary that performs on data from a variety of sources. This approach clusters points in 3-dimensional space using various point characteristics. Principal Component Analysis (PCA) is performed to segment the point cloud. This analysis introduces the possibility of bringing further intelligence to point clouds by uniquely identifying objects within the point cloud based on the similarity of points.