Application of Learning-based Multi-view Stereo in Digital Design

Abstract

Image-based 3D reconstruction technology has been widely used in automatic driving, medical image reconstruction, positioning and navigation. At present, the digital design is still dominated by drawing 3D models manually, which has some shortcomings such as time-consuming and limited accuracy. To solve the above problems and improve the efficiency of digital design, a learning-based multi-view stereo 3D reconstruction technique is introduced for dense reconstruction, and then a grid model is generated for parametric design. We experimented our method and the results show that it can greatly improve the efficiency of 3D digital design, obtain high precision, and retain the texture information of the object.