A hardware accelerator for three-dimensional reconstruction based on gaussian sputtering

CN122115695APending Publication Date: 2026-05-29INST OF COMPUTING TECH CHINESE ACAD OF SCI
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
INST OF COMPUTING TECH CHINESE ACAD OF SCI
Filing Date
2026-04-16
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

When existing Gaussian sputtering 3D reconstruction methods are deployed on the edge of unmanned robots, the reusable data information between frames is not systematically utilized, resulting in high overhead for memory access and reordering. Redundant calculations in the rendering and compositing stage consume a lot of computing power and energy, making it difficult to meet the requirements for real-time performance and energy efficiency.

Method used

Design a hardware accelerator for 3D reconstruction based on Gaussian sputtering. Through a frame reuse-oriented memory access and sorting optimization mechanism, retain the image information of similar regions to the next frame to reduce memory access. Perform reordering optimization on dissimilar regions and introduce an opacity critical point stopping strategy to terminate the calculation early and reduce redundant computation.

Benefits of technology

It enables data reuse, reduces memory access bandwidth overhead, improves throughput, significantly reduces redundant computing load, and improves overall energy efficiency density, meeting the performance and energy efficiency requirements of unmanned robots in real-time 3D reconstruction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122115695A_ABST
    Figure CN122115695A_ABST
Patent Text Reader

Abstract

The application provides a three-dimensional reconstruction hardware accelerator based on Gaussian sputtering, comprising: a sorting optimization subsystem, which is used for determining similar areas and dissimilar areas of a next frame and a current frame according to camera poses of the current frame, the current frame and the next frame, reserving picture information of the similar areas, and inputting Gaussian body parameters intersecting with pixels in the dissimilar areas without occlusion relationship into a calculation array subsystem; for multiple Gaussian bodies intersecting with pixels in the dissimilar areas with occlusion relationship, Gaussian bodies with picture contribution degrees greater than or equal to a preset threshold are divided into a high contribution group for sorting, and Gaussian bodies with picture contribution degrees less than the preset threshold are divided into a low contribution group for sorting, so as to be input into the calculation array subsystem; the calculation array subsystem is used for accumulating and calculating pixel states according to input Gaussian body parameters for each pixel in the dissimilar areas; and an operation clipping subsystem is used for outputting the pixel states when the opacity of each pixel with the occlusion relationship reaches an opacity threshold.
Need to check novelty before this filing date? Find Prior Art