Neural Radiation Field Enhancement Method Based on Joint Pose Optimization

Abstract

The invention belongs to the field of neural rendering under the field of computer graphics and the field of camera pose optimization under the field of computer vision, and more specifically relates to a neural radiation field enhancement method based on joint pose optimization. By combining the neural rendering method and the expressive ability of the neural network, the camera pose can be further effectively optimized. With a more accurate camera pose, the trained neural radiation field is closer to the real scene, and the rendered image quality is higher , the details are clearer.

Description

technical field [0001] The invention belongs to the field of neural rendering under the field of computer graphics and the field of camera pose optimization under the field of computer vision, and more specifically relates to a neural radiation field enhancement method based on joint pose optimization. Background technique [0002] New view synthesis is a very challenging task in the field of computer vision as well as in the field of computer graphics. It is defined as synthesizing images from a new perspective from a series of captured images of a certain scene. Neural Radiative Fields (NeRF) have achieved remarkable results in the task of novel view synthesis. In practical applications, the camera pose is often obtained through the SfM algorithm. Since the SfM algorithm relies on the detection and matching of key points, in some scenes with many repeated textures, many outliers will often be generated to cause incorrect matching. It can have a large adverse effect on th...

AI Technical Summary

Problems solved by technology

[0003] In some existing methods, the implicit expression focuses on geometric properties, but cannot simulate more complex properties, such as color and density

Most neural rendering methods require 3D display information as input, and accurate geometry is difficult to obtain, which is not conducive to pose calculation

The camera pose depends on the matching points between the images. If a wrong match is taken into account in the algorithm process, it will introduce different degrees of error.

Images