Semantic Classification based Streaming of Large-Scale Building Information Models

Abstract With the rapid development of building intelligent design technology, the data volume of 3D building information models is also increasing, bringing aboutlenges for 3D model transmission, storage, and rendering. The streaming of large-scale 3D building information models is one of the key intelligent construction processing techniques. In this work, we present a semantic classification-based streaming approach for the large-scale 3D point-cloud models. We first compute the composition separation within the large data input by adapting the advanced deep learning-based labeling techniques. Then, optimal alignments among these components are calculated, and an Octree-based representation is constructed to organize and compress the data. Objects within each semantic group are processed independently, enabling data streaming and interactive visualization. The experimental results show the effectiveness of our streaming model, this is of great significance for the data transmission and visualization of large-scale 3D content.