Point cloud up-sampling method based on GAN network

Abstract

The invention discloses a point cloud up-sampling method based on a GAN network. A GAN network composed of a Generator and a Display discriminator is mainly used for carrying out the point cloud up-sampling. The Generator is composed of an annular arrangement module, a multi-frequency pooling module and a GRU network module, and can effectively learn the geometrical characteristics and global characteristics of each point in the point cloud, thereby better mastering the geometrical information of the input point cloud. The discriminator Discriminator is composed of two layers of feature extractors and a deconvolution network module, can discriminate whether the input point cloud is true or false, and helps to better optimize the GAN network. Through annular arrangement, disordered point neighborhoods can be arranged into an ordered annular structure through orthogonal projection and counterclockwise arrangement, and geometrical characteristics of different groups of neighborhoods of each point can be accurately extracted through multi-frequency pooling. The method can be applied to the preprocessing step of three-dimensional point cloud reconstruction.

Description

technical field [0001] The present invention belongs to the neighborhood of three-dimensional images of computer vision, specifically a point cloud upsampling framework based on a GAN network that includes 3D point cloud feature learning, point generation, and true / false judgment. Background technique [0002] A 3D image is an expression of 3D space, including geometric models, depth maps, and point clouds. Among them, point cloud is a kind of massive point data collection obtained by scanning devices such as 3D laser scanners. It can completely save the geometric information of the model without any discrete processing, and the representation is very simple, only including the basic attributes of points. , such as coordinates, normal vectors, etc., due to its ease of use, it has been increasingly applied to 3D reconstruction, reverse engineering, unmanned driving and other fields. [0003] In these fields, 3D reconstruction plays an important role. Three-dimensional recon...

AI Technical Summary

Problems solved by technology

In order to learn the features of different levels, the radius of the neighborhood is increased step by step, the local neighborhood features are obtained by the PointNet++ method, and finally the reverse interpolation is used to combine all the features, but it reuses the existing features to bring a lot of feature redundancy.

After that, Wang et al. proposed a Patch-based multi-step learning upsampling network 3PU, which can divide a 16-fold upsampling network into four 2-fold upsampling networks, thereby reducing the network size and better Learning the local feature information of the point cloud helps coordinate reconstruction, but it also has serious feature redundancy problems.

Images