Point Cloud Data Segmentation Using RANSAC and Localization

Abstract

In this paper, we present 3D point cloud data segmentation from a Terrestrial Laser Scanner (TLS) using RANSAC and localization approaches. Our work is using the real world data acquired from outdoor and indoor scene of Tawau Bell Tower (also known as the Belfry), one of the heritage building located in Sabah. We adapt the methods to segment ground and nonground data for data reduction. Based on the non-ground information and geometrical constraint, the data managed to show the building structure. The aim is to use and handle a smaller set of data rather than using the entire point cloud data in processing the region of interest. This will also help to reduce processing time as point cloud data from a TLS is usually large in size. The ground plane was fitted with RANSAC algorithm restricted by distance to describe the geometrical of the plane. As a result, the point cloud data was reduced to the environment around Bell Tower and able to highlight the region of interest. Building segmentation is then retrieved from the reference centre location to form a bounding box and data were organized using kd-tree to denote the Bell Tower structure. The framework is tested for outdoor and indoor scan of the Bell Tower and able to segment the data appropriately.