Volume cloud new view angle synthesis and three-dimensional reconstruction method based on neural radiation field

Through simplified radiation transmission equations and mask loss optimization, the problem of density distribution of neural radiation fields in volumetric cloud reconstruction is solved, and more accurate cloud shape modeling and new perspective rendering effects are achieved.

CN119027582AActive Publication Date: 2024-11-26ZHEJIANG UNIV

Patent Information

Application Number
CN202411005190.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-26
Estimated Expiration
2044-07-25

AI Technical Summary

Technical Problem

The existing neural radiation fields are difficult to reconstruct the volumetric cloud model that meets the requirements, resulting in some viewing angles deviating from the target cloud image, and the density exists not only in the cloud, but also in the entire bounding box.

Method used

Using a new perspective synthesis and three-dimensional reconstruction method based on neural radiation field, including data preprocessing, network model forward propagation, rendering and backpropagation, variables and parameters are optimized through simplified radiation transmission equations and mask loss optimization, To reconstruct density that conforms to the shape of the cloud and achieve good new perspective rendering.

Benefits of technology

The problem of more correct modeling of volumetric clouds and new perspective synthesis is solved, avoids the problem of deviation of cloud images rendered by neural radiation fields at certain perspectives, and ensures that density is concentrated only in the cloud shape.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027582A_ABST
    Figure CN119027582A_ABST
Patent Text Reader

Abstract

The invention discloses a volume cloud new view angle synthesis and three-dimensional reconstruction method based on a neural radiation field. The method comprises a data preprocessing step, a network model forward propagation step, a rendering step and a back propagation step. In the data preprocessing step, a mask data set MASKS is obtained from an image data set IMAGES containing surround cloud; in the model forward propagation step, light stepping sampling is carried out on the camera attitude corresponding to each image, and density and phase are obtained through network model forward propagation calculation on coded sampling points and a single light source direction; in the rendering step, the density, the phase and other variables are rendered according to a simplified radiation transfer equation to obtain a predicted image and a cloud mask; in the back propagation step, error loss Loss is calculated by using a predicted image, a predicted cloud mask, IMAGES and MASKS so as to optimize variables and parameters; according to the method, the density conforming to the cloud shape can be reconstructed, and a good rendering result can be obtained for a new view angle and an input view angle.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Urban large-scale scene reconstruction method based on neural radiation field

    CN115841559A

  • Outdoor unbounded scene three-dimensional reconstruction method and system based on neural radiation field

    CN116051740A

  • Object-centric neural decomposition for image re-rendering

    US20230215085A1

Cited By

  • Neural radiation field DSM generation method based on two-dimensional triangulation network constraint

    CN120374898A