Big Data and Machine Learning
ABSTRACT3D modeling of the built environment has become common practice in the AEC/FM industry. Practitioners take advantage of the geometric and semantic information embedded in the 3D model to perform engineering analysis. Despite the benefits provide by the 3D model, the process is time-consuming, labor-intensive, and error-prone. In this paper, we propose a new neural network-based method for 3D point cloud semantic segmentation of building scenes using a hierarchical approach: first, we reason on the local and global contents of raw point cloud data to extract geometrical features. Second, the features are used as input to an artificial neural network that performs semantic segmentation on the points. These points are classified into: beam, ceiling, clutter, column, door, floor, pipe, wall, and window. We evaluated our approach on a dataset of several buildings and we obtained an accuracy of 73%. Our experiments produce robust results readily useful for practical Scan2BIM applications.