THE CONSTRUCTION AND APPLICATION OF RESIDENTIAL BUILDING INFORMATION MODEL BASED ON DEEP LEARNING ALGORITHMS

In order to explore the construction and application of building BIM models, the construction industry is actively exploring a method that can quickly reshape the 3D information model of existing buildings in the wave of digital twins and smart cities. Starting from the perspective of deep learning 3D object detection algorithms, the author starts with the generation of large-scale building datasets and the theory of point cloud deep learning, analyzes the input data types required for point cloud deep learning frameworks, and focuses on the creation process of 3D bounding boxes and 3D point clouds for various building components. The author compares different point cloud datasets with the same data structure and implements an object detection algorithm based on the ScanNet dataset, furthermore, a feasible technology route for automatic generation of BIM models from 3D point clouds based on deep learning is integrated. Through this technology route, the trained neural network can input unknown building 3D point clouds and output BIM model parameters.