A digital twin method and system based on 3D model matching

Abstract

The invention discloses a digital twin method and system based on 3D model matching. The method includes four steps: model and texture training, model matching, texture fusion, and scene placement; wherein, model and texture training is an offline preprocessing step, and model matching, texture Fusion and scene placement are real-time processing steps. The training of the model uses multi-viewpoint silhouette images to learn the 3D structure, and realizes the 3D reconstruction of the polygonal mesh; the training of the texture obtains the texture flow through a fixed UV mapping, and realizes the reasoning from the 2D image to the 3D texture. Model matching uses the reconstructed model and the database model to perform IOU iterative calculation, and obtains the model whose IOU value is closest to 1 as the specified model after matching. Texture fusion is to fuse the specified model after matching with the 3D texture stream after texture inference to form a standard 3D model. Scene placement is to place the 3D model accurately in the 3D scene. This digital twin method automatically generates 3D model scenes and improves the efficiency of digital twins.

Description

technical field [0001] The invention relates to the field of digital twins, in particular to a digital twin method and system based on 3D model matching. Background technique [0002] At present, most methods for automatic generation of 3D models only match simple data such as coordinates or model types with existing models, and do not use the latest deep neural network machine learning methods. For example, patent CN202010701243.3 discloses an automatic generation of tunnel 3D models The method includes: S1, establishing a graphic database, storing objects including two-dimensional design drawings and entity objects of three-dimensional models; S2, creating an external database for storing multiple design parameters input by users; S3, based on user input parameters to generate 2D design drawings and 3D models of tunnels. Patent CN202011386970.1 discloses a method for automatically generating a 3D model of a road intersection, including: obtaining the basic parameters of t...

AI Technical Summary

Problems solved by technology

[0004] Patent CN 201310343131.5 discloses a realization method of automatic generation and online interaction of a three-dimensional digital model of a plane space structure. It is relatively rare to use the method of server and client, but the model generation is relatively rough, and only a simple background model such as a wall is generated. Body, windows, and furniture models are selected by the user through the operation of the client.

[0005] In addition, under the background of digital twin technology as the core technology of the next generation Internet, existing patents only focus on the similarity of shape, but do not take into account the similarity of texture between the original object and the virtual model

Images