Highly Efficient Deep Learning-Enabled Parameterization and 3D Reconstruction of Traditional Chinese Roof Structures

Area of Science:

• Digital Heritage
• Computer Vision
• Architectural Engineering

Background:

• Ancient Chinese architecture is culturally significant, characterized by unique symmetrical designs.
• Modern technologies like UAV photogrammetry and laser scanning aid in replicating ancient structures for Heritage Building Information Modeling (HBIM).
• Modeling traditional Chinese roofs is challenging due to complex curved surfaces and upturned eaves.

Purpose of the Study:

• To develop an efficient method for symmetry detection and parameter extraction in traditional Chinese roofs.
• To reduce the complexity and difficulty of creating accurate HBIM models for these structures.
• To leverage symmetry for rapid digital reconstruction using a mirror replication approach.

Main Methods:

• A deep learning network, TCRSym-Net, was developed for symmetry identification in roof point clouds.
• Point clouds were processed to obtain cross and longitudinal sections.
• Parametric modeling scripts were coded in Dynamo utilizing curve lofting and solid Boolean operations.

Main Results:

• TCRSym-Net demonstrated effectiveness in accurately detecting symmetry in point clouds of traditional Chinese roofs.
• Five distinct types of traditional Chinese roofs were successfully recreated using the proposed method.
• The study confirmed the dependability of the symmetry detection and modeling approach.

Conclusions:

• The proposed deep learning approach significantly enhances the accuracy and efficiency of symmetry detection for traditional Chinese roofs.
• This method facilitates the creation of detailed and dependable HBIM models of ancient Chinese architecture.
• The findings support the use of AI-driven techniques for preserving and reconstructing cultural heritage.