Classification and Segmentation of Terrestrial Laser Scanner Point Clouds Using Local Variance Information

With the use of terrestrial laser scanning, it is possible to capture thousands of 3-dimensional points on the surface of an object. The problem is that the vast quantity of data that needs to be manipulated sometimes hinders its application. There exist several techniques for the semi-automatic extraction of low-level features that help compensate for this problem. In addition to reviewing some of these techniques, some modifications will also be discussed, along with their use in developing a covariance based procedure to preform classification and feature extraction for terrestrial laser scanning point clouds using local neighbourhoods. Results from this procedure are then used to segment the surface features of the point cloud. The application of this is demonstrated on several captured data sets. Additional information such as methods for defining local surface intersections and direction of principle curvature will also be detailed based on using local information.