Point Cloud Simplification and Civil Building Layout Extraction from Large Data-Set Involving Multiple Objects

K.K. Sareen, G.K. Knopf, and R. Canas (Canada)


3D scanning, data decimation, layout extraction, shape reconstruction, geometric modeling.


Geometric shape modeling from 3D scanned data of existing objects is a common task encountered in many engineering and scientific applications such as anatomical reconstruction, cartography, artefact modeling, digital archaeology, and infrastructure renewal. The range scanned data captures the object’s geometry in the form of spatial points, which are often huge and partially spurious. Accurate identification of desired features and geometric information extraction from this data is a complex task, especially when the data set is cohesive, and represents multiple objects. Most of the existing simplification and information extraction methods work on the reconstructed geometric model, which may not be directly available and is not always required. Thus, an effective, direct point-based method is proposed in this paper that simplifies the point cloud data involving multiple objects and extracts the shape layout. The process retains the boundary points based on their importance, computed in terms of average angular deviation value. This method is quite effective in identifying points representing different objects or features in the scanned scene and can extract accurate layouts with a fraction of data set (5-12%). Accuracy of the layout extraction process can be further improved by using fewer neighbourhood points and smaller angular deviation values.

Important Links:

Go Back