An Intelligent Identification Method for Building Components Based on Synthetic Point Cloud and Deep Learning
This paper investigates the viability of using synthetic point clouds generated by 3D build-ing information model to train deep learning algorithms to realize intelligent identification of building components.To achieve this goal,firstly,this paper proposes an original method of converting build-ing information model into synthetic point clouds through three common commercial software.Then,these synthesized point clouds are used as datasets to train the deep learning model and the intelligent identification performance of the trained model is compared under different datasets(real dataset and synthetic dataset)to verify the effectiveness of synthetic point clouds.The experimental result proves the feasibility of using synthetic point cloud generated by building information models to achieve intel-ligent identification.The accuracy gap between model trained by synthetic dataset and real dataset is only 3％,which further indicates the possibility of using synthetic dataset instead of real dataset in in-telligent identification.This method also provides researchers with a new method to construct specific datasets for their own intelligent identification and semantic segmentation research and contributes to 3D reconstruction work.
building information modeldeep learning modelpoint cloudintelligent identification
万方数据
维普
