Single drone-based 3D reconstruction approach to improve public engagement in conservation of heritage buildings: A case of Hakka Tulou

Abstract

Public engagement in protecting architectural heritage is a critical component of sustainable development. This study has developed an innovative single drone-based 3D reconstruction (SD-3DR) approach for public-involved architectural heritage conservation. A new software tool named after CU-Recon is also developed to enable easy access and efficient 3D reconstruction using drone photography. Unlike traditional photogrammetry-based reconstruction, CU-Recon adopts our unique deep learning-based multi-view stereo network named LCM-MVSNet, enabling the public to employ only one drone for image capture and 3D reconstruction. LCM-MVSNet applies a learnable cost metric (LCM) to adaptively aggregate multi-view matching similarity into the 3D cost volume by leveraging sparse point hints. Its outstanding reconstruction performance for building scale applications is proved by the extensive experiments on the DTU training dataset and BlendedMVS dataset. A remarkable architectural heritage Hakka Tulou is selected to verify the effectiveness of the SD-3DR on large-scale heritage buildings. The results show our approach outperformed the other four 3D reconstruction tools. Moreover, the reconstruction quality in CU-Recon is evaluated from a perspective of heritage conservation using the concept of satisfaction. Practitioners expressed an overall satisfaction score of 4.175 (out of 5) for the reconstruction quality of the method. Survey results reveal a higher level of satisfaction with single drone photography for architectural heritage conservation compared to LiDAR-based scanning in terms of portability, operability, and cost. The research outcome changes the current situation of government-led top-down architectural heritage conservation by providing valuable insights for individual practitioners in creating bottom-up heritage conservation routes.