Multi-view three-dimensional reconstruction method

Abstract

The invention discloses a multi-view three-dimensional reconstruction method, and the method comprises the following steps: based on an end-to-end deep learning framework, taking a plurality of colorimages at different visual angles as network input, and taking a reconstructed three-dimensional grid model as output; according to the network, firstly, extracting multi-view fusion features of different scales by using a multi-level fusion feature extraction module; and based on the fusion features of different scales, utilizing a grid deformation module to deform the input initialized ellipsoidal grid model, and reconstructing a three-dimensional grid model corresponding to the input image. According to the method, multi-view information is fully utilized, and higher reconstruction precision and a better reconstruction effect are achieved.

Description

technical field [0001] The present invention relates to the technical fields of image processing and stereo vision, in particular to a multi-view Figure three Dimensional reconstruction method. Background technique [0002] As a hot issue in the field of computer graphics and computer vision, 3D reconstruction has received more and more attention. The task of 3D reconstruction is to construct the 3D structural model of the scene and the objects in the scene according to the mapping relationship between 2D and 3D space on the premise of using the visual inspection equipment to obtain the image. This technology has been widely used in urban planning, medical diagnosis and unmanned driving, and has important research significance and application value. [0003] Traditional 3D reconstruction methods use the principle of binocular vision geometry to reconstruct complex 3D scenes based on disparity information, so as to achieve a certain degree of realistic scene modeling. Howe...

AI Technical Summary

Problems solved by technology

[0007] When using multi-view for 3D reconstruction in existing technologies, although information from multiple perspectives can be better used to obtain a 3D reconstruction model, most algorithms have strict requirements on the number of input multi-view images, making it difficult to Reconstruction of 3D mesh models in the case of perspective images; the existing training data used to study 3D reconstruction methods is relatively rich, but compared with the rich diversity of objects and scenes in nature, the types of reconstruction objects in the training set are relatively few, limiting The applicability of 3D reconstruction technology in real scenes

Images