3D reconstruction of buildings based on transformer-MVSNet

Xiaofeng Zeng; Tang Jin

doi:10.1117/12.2686276

28 July 2023 3D reconstruction of buildings based on transformer-MVSNet

Xiaofeng Zeng, Tang Jin

Proceedings Volume 12756, 3rd International Conference on Applied Mathematics, Modelling, and Intelligent Computing (CAMMIC 2023); 127564S (2023) https://doi.org/10.1117/12.2686276
Event: 2023 3rd International Conference on Applied Mathematics, Modelling and Intelligent Computing (CAMMIC 2023), 2023, Tangshan, China

Abstract

Three-dimensional model is an important form of human cognition, and with the development of computer technology, 3D reconstruction technology has important applications in many fields, such as heritage conservation and virtual reality. The current mainstream 3D reconstruction techniques are based on deep learning methods. In this paper, based on the MVSNet model, the feature extraction module of grouped SDTA encoder and the attention module of 3D-SKNet are introduced. Among them, the feature extraction module of grouped SDTA encoder can establish long-range dependencies and provide the ability of the model to capture global features. The proposed attention module of the proposed 3D-SKNet is able to fuse multiple perceptual field features to obtain more robust features. This paper is validated on the DTU dataset, and compared with the original MSVNet model, the proposed model in this paper improves 0.195 mm in Acc metric, 0.09 mm in Comp metric, and 0.052 mm in Overall metric. experiments show that the proposed model in this paper can improve the quality of 3D reconstruction to a great extent.With the experimental conclusion, the algorithm in this paper can obtain applications in heritage conservation and building reconstruction.

Citation Download Citation

Xiaofeng Zeng and Tang Jin "3D reconstruction of buildings based on transformer-MVSNet", Proc. SPIE 12756, 3rd International Conference on Applied Mathematics, Modelling, and Intelligent Computing (CAMMIC 2023), 127564S (28 July 2023); https://doi.org/10.1117/12.2686276

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available