Single-frame real-time three-dimensional imaging method based on deep learning

By constructing a lightweight phase retrieval network and a multi-vision structured light projection system, combined with GPU acceleration technology, real-time single-frame 3D imaging was achieved. This solved the problems of high computational resource consumption and difficulty in imaging dynamic scenes in existing methods, and achieved efficient and real-time 3D imaging results.

WO2026108570A1PCT designated stage Publication Date: 2026-05-28NANJING UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2025/131477
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2024-11-19
Filing Date
2025-10-31
Publication Date
2026-05-28

AI Technical Summary

Technical Problem

Existing single-frame deep learning methods struggle to achieve real-time, high-precision 3D imaging in complex and dynamic scenes. Furthermore, traditional structured light 3D imaging requires multiple shots and consumes significant computational resources, making it unsuitable for real-time applications.

Method used

A deep learning-based real-time single-frame 3D imaging method was designed. By constructing a lightweight phase retrieval network model and building a multi-view structured light stripe projection system, a real-time single-frame 3D imaging method was achieved by utilizing multi-view geometric constraints and stereo phase matching algorithms, combined with GPU acceleration technology.

Benefits of technology

It achieves real-time, high-precision 3D imaging in complex and dynamic scenes, overcomes the computational resource limitations of traditional methods, provides efficient 3D reconstruction capabilities, and is suitable for real-time application scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2025131477_28052026_PF_FP_ABST
    Figure CN2025131477_28052026_PF_FP_ABST
Patent Text Reader

Abstract

Disclosed in the present invention is a single-frame real-time three-dimensional imaging method based on deep learning, comprising: constructing a lightweight phase retrieval network model based on a physical model; building a multi-vision structured light fringe projection system; constructing a training data set to complete training of the lightweight phase retrieval network model; and using TensorRT to deploy the lightweight phase retrieval network model. Three wrapped phase images are obtained in real time by means of parallel reasoning of three monochromatic view angles; wrapped phases are unwrapped into absolute phases by using a multi-view geometric constraint and three-dimensional phase matching algorithm; and a three-dimensional point cloud having color texture information is generated by means of phase-height mapping and a color image. In the present invention, deep learning technology is integrated with a conventional fringe projection physical model, such that rapid phase acquisition can be achieved using only a single image frame, thereby overcoming motion interference and enabling online, real-time, high-precision three-dimensional imaging in complex dynamic scenes.
Need to check novelty before this filing date? Find Prior Art