Periodic Reporting for period 1 - TITANIUM (Software Components for Robust Geometry Processing)
Período documentado: 2017-01-01 hasta 2018-06-30
We developed a series of interoperable components for the Computational Geometry Algorithm Library (CGAL), specialized to 3D urban modeling. These components offer an algorithmic pipeline for converting physical measurements over urban environments (3D point clouds or meshes) into their simplified digital representations, that conform to the meaningful levels of detail of the CityGML standard. The pipeline is general and can handle different semantic data such as buildings, vegetation and ground.
The pipeline contains several algorithms that perform different geometry processing tasks, grouped into categories: semantic classification, shape detection, regularization, partitioning, visibility, and levels of detail. Classification is a set of machine learning algorithms that process general point clouds with no semantic information and label each point with one of the user-defined semantic labels. Shape detection combines points into groups that form canonical shapes such as planes. After detection the shapes are regularized by consolidating geometric regularities between shapes, such as collinearity or orthogonality. Partitioning creates a connectivity between the planar shapes, by subdividing the 2D/3D space into cells of an arrangement. Partitioning labels each cell of the said arrangements, and the last step generates three levels of details conforming to the LOD 0-1-2 of the CityGML format.
The software components have been devised from and tested on real-world use cases provided by industrial companies, the commercial potential being judged very positively. The classification and shape detection components are already released and accessible from the CGAL web site (see https://doc.cgal.org/latest/Classification/index.html#Chapter_Classification and https://doc.cgal.org/latest/Point_set_shape_detection_3/index.html#Chapter_Point_Set_Shape_Detection and online video https://www.youtube.com/watch?v=xLFm8Aw8vuY&t=325s) the other components being under submission.
For the SME GeometryFactory the outcomes of the project are likely to turn into an increased activity and competitive advantage.
The pipeline contains several algorithms that perform different geometry processing tasks, grouped into categories: semantic classification, shape detection, regularization, partitioning, visibility, and levels of detail. Classification is a set of machine learning algorithms that process general point clouds with no semantic information and label each point with one of the user-defined semantic labels. Shape detection combines points into groups that form canonical shapes such as planes. After detection the shapes are regularized by consolidating geometric regularities between shapes, such as collinearity or orthogonality. Partitioning creates a connectivity between the planar shapes, by subdividing the 2D/3D space into cells of an arrangement. Partitioning labels each cell of the said arrangements, and the last step generates three levels of details conforming to the LOD 0-1-2 of the CityGML format.
The software components have been devised from and tested on real-world use cases provided by industrial companies, the commercial potential being judged very positively. The classification and shape detection components are already released and accessible from the CGAL web site (see https://doc.cgal.org/latest/Classification/index.html#Chapter_Classification and https://doc.cgal.org/latest/Point_set_shape_detection_3/index.html#Chapter_Point_Set_Shape_Detection and online video https://www.youtube.com/watch?v=xLFm8Aw8vuY&t=325s) the other components being under submission.
For the SME GeometryFactory the outcomes of the project are likely to turn into an increased activity and competitive advantage.