An online calibration method for three-dimensional point cloud and two-dimensional image based on edge features

By using graph neural networks with edge feature point extraction and attention mechanisms, efficient and accurate online calibration of 3D point clouds and 2D images is achieved, solving the problems of high computational cost and information loss in existing methods, and improving the real-time performance and accuracy of autonomous driving systems.

CN118298028BActive Publication Date: 2026-07-21SHANDONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SHANDONG UNIV
Filing Date
2024-04-18
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing deep learning methods suffer from high computational cost and information loss in 3D point cloud and 2D image calibration, especially due to insufficient calibration accuracy and efficiency caused by downsampling.

Method used

An online calibration method based on edge features for 3D point clouds and 2D images is adopted. The PointNet++ feature encoder and attention-based graph neural network are used to extract 2D and 3D feature points and their descriptors. The transformation matrix is ​​estimated by the EPnP algorithm for registration.

Benefits of technology

It improves calibration accuracy and computational efficiency, eliminates characteristic mode differences, expands the applicability of the calibration method, adapts to changes in sensor state, and reduces computational complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118298028B_ABST
    Figure CN118298028B_ABST
Patent Text Reader

Abstract

The present application relates to a kind of three-dimensional point cloud and two-dimensional image online calibration method based on edge feature, comprising: data preprocessing and edge extraction, generate the data that conforms to the input of neural network model and the data of task requirement;The data after data preprocessing is input into the neural network model built using Python programming, after outputting two-dimensional and three-dimensional feature points and their corresponding descriptors, the corresponding relationship of two-dimensional and three-dimensional feature points is found by the similarity between descriptors;Transform matrix is estimated from the given two-dimensional-three-dimensional point pair using EPnP algorithm.The error in rotation and translation and the success rate of calibration of the present application are superior to the mainstream method at present.The present application reduces the number of point pairs that need to be calculated by selecting representative feature points, reduces the computational complexity, and the significance of key points also guarantees accuracy, improves the calculation efficiency of calibration process on the basis of ensuring accuracy.
Need to check novelty before this filing date? Find Prior Art