Segmentation of heritage building by means of geometric and radiometric components from terrestrial laser scanning

Abstract

Nowadays, the terrestrial laser scanning represents an integral source of data for cultural heritage 3D storage and access through digital communication tools. The achievement of 3D models requires the implementation of several tasks such as segmentation. Segmentation is the key step during the point cloud processing where all homogeneous areas are identified, which describe a building facade. Usually, a large part of the segmentation approach focuses on the geometric information contained in the point cloud data by exploiting mathematical representation of a parametric surface. However, due to the complexity of the architecture, such segmentation does not suffice. Henceforth, other approaches turn to the use of color and laser intensity components. Although a variety of algorithms have been developed in this sense, problems of over-segmentation or under-segmentation are observed. In this context, we propose a new approach for point cloud segmentation aiming at a more accurate result. This approach relies on all the components of a colored point – both geometric and radiometric – combining the RGB values, laser intensity and geometric data. Our process begins with the extraction of homogeneous planar segments using the RANSAC algorithm. Next, the result is subjected to a radiometric-based segmentation, first through color similarity as one of the homogeneity criteria of a region growing algorithm, then through the use of intensity similarity for segment fusion. Experiments are performed on a facade presenting an example of Moroccan classical architecture located in Casablanca's Medina. Results show the importance of integrating all point cloud components, both geometric and radiometric.