Three-dimensional point cloud reconstruction method based on deep learning

Abstract

The invention discloses a three-dimensional point cloud reconstruction method based on deep learning; the method comprises the steps that: the coordinate information of a three-dimensional point cloud is predicted through a point cloud predictor according to the inputted potential feature representation, wherein each branch takes potential feature representation output by the corresponding feature encoder as input, and learns complementary features combined with other branch information; by applying a cross-view interaction unit, each sparse point cloud reconstruction subnet captures cross-view complementary information and feeds back the information to the point cloud predictor to generate sparse point clouds; a global guidance dense point cloud reconstruction module composed of a plurality of point cloud feature extraction subnets, a global guidance feature learning subnet and a generation layer is constructed, each point cloud feature extraction subnet is composed of a series of multi-layer perceptron sharing weights, and the multi-layer perceptron extracts point cloud features from generated sparse point clouds; finally, chamfer distance loss is adopted as geometric consistency constraint, and semantic consistency constraint is constructed to optimize generation of dense point clouds.

Description

technical field [0001] The invention relates to the field of three-dimensional point cloud reconstruction, in particular to a three-dimensional point cloud reconstruction method based on deep learning. Background technique [0002] As one of the hot research tasks in the field of computer vision, the goal of 3D model reconstruction is to generate a real 3D model of the object contained in the 2D image through the information of the given 2D image. 3D models can be represented in a variety of ways, including: voxels, meshes, and 3D point clouds. As a typical representative of 3D models, 3D point cloud has been applied in many fields such as autonomous driving and virtual reality. Therefore, the task of 3D point cloud reconstruction has attracted extensive attention of researchers. Moreover, the quality of the generated point cloud model can significantly affect the performance of subsequent tasks, such as 3D model retrieval, classification, and segmentation, etc. However, ...

AI Technical Summary

Problems solved by technology

[0005] Although the existing 3D model reconstruction methods based on deep learning can predict a reasonable 3D shape from a limited number of input views, these methods usually directly generate relatively sparse and rough 3D point clouds.

Images