Lightweight, rapid, and accurate self-supervised depth estimation method and system

Through the combination of weighted coefficient matrix and sparse coding module, a lightweight depth estimation network is built, which solves the problems of slow inference speed and large parameters caused by complex models, and achieves fast and accurate depth estimation.

WO2024227286A1PCT designated stage expired Publication Date: 2024-11-07SHENZHEN INST OF ADVANCED TECH CHINESE ACAD OF SCI

Patent Information

Application Number
PCT/CN2023/092055
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-04
Publication Date
2024-11-07

AI Technical Summary

Technical Problem

The prior art restores scene depth information from two-dimensional images through neural networks requires complex depth estimation network models, resulting in slower algorithm inference speed and increased model parameters.

Method used

By constructing a weighted coefficient matrix calculation module and a sparse encoding module, the contribution degree of feature maps is calculated and pixel-level fusion is performed, a small-scale depth estimation network model is constructed, and the small-scale model is used to approximate the feature map distribution characteristics of the large-scale depth estimation network model using function approximation of the feature map distribution characteristics of the large-scale depth estimation network model .

Benefits of technology

It realizes lightweight, fast and accurate depth estimation, improves model inference speed and accuracy, and reduces the amount of parameters.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2023092055_07112024_PF_FP_ABST
    Figure CN2023092055_07112024_PF_FP_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of computer vision and image processing, in particular to a lightweight, rapid, and accurate self-supervised depth estimation method and system, which can solve the problem that a complex depth estimation network model is required when scene depth is recovered from a two-dimensional image to a certain extent. The method comprises: acquiring an initial image data set; constructing a weighting coefficient matrix calculation module and a sparse coding module; constructing, on the basis of the weighting coefficient matrix calculation module and the sparse coding module, a small depth estimation network model to increase the algorithm reasoning speed and reduce the number of parameters in the model; constructing and training a large depth estimation network model by taking a photometric loss as a supervision signal at the same time; training the small depth estimation network model by using the photometric loss and a function approximation loss as supervision signals to optimize the depth estimation precision of the small depth estimation network model; and finally, inputting input data into the small depth estimation network model to obtain an estimation result.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Image depth information monocular estimation method and device and readable storage medium

    CN113192149A

  • Self-supervised monocular depth estimation method and device

    CN114022799A

  • Monocular three-dimensional reconstruction method fusing attention mechanism

    CN115375844A

  • Monocular depth estimation method based on convolution sparse coding

    CN115861723A

  • Systems and methods for weakly supervised training of a model for monocular depth estimation

    US20210004646A1

Cited By

  • Logistics management system based on artificial intelligence

    CN119904158A

  • Density measurement method and system fusing sparse measurement value and ground penetrating radar data

    CN121936308A

  • Long-time-interval electrocardiosignal analysis method and system based on sparse self-attention

    CN122413337A

  • Lightweight monocular depth estimation method and system based on space-channel cooperation

    CN122574051A