A 3D image registration method for AR system

A technology of 3D images and reference images, which is applied in image data processing, instruments, biological neural network models, etc., can solve the problems that the accuracy of 3D image registration cannot be guaranteed, so as to reduce training time, improve efficiency, ensure stability and The effect of precision

Inactive Publication Date: 2019-05-24
广州智能装备研究院有限公司
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved by the present invention is that in the existing AR system, the accuracy of three-dimensional image registration cannot be guaranteed

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A 3D image registration method for AR system
  • A 3D image registration method for AR system
  • A 3D image registration method for AR system

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment 1

[0045] Such as figure 1 As shown, the three-dimensional image registration method for the AR system provided in Embodiment 1 of the present invention includes the following steps:

[0046] Step S10, constructing a mathematical model T(t of three-dimensional image registration x ,t y ,t z ,θ x ,θ y ,θ z ):

[0047]

[0048] Where: parameter t x ,t y ,tz Indicates the translation amount of the 3D image along the x, y, z axes, θ x ,θ y ,θ z Indicates the rotation angle of the 3D image along the x, y, and z axes;

[0049] The foundation of the mathematical model is as follows:

[0050] Order I r Denotes a reference image or a target image, I f Indicates the need to register with the I r random image (registration image), then T g is a 4×4 homogeneous transformation matrix, representing the transformation matrix that needs to be estimated (that is, the direct means to realize 3D image registration), the purpose of this mathematical model is to obtain T g × I f ...

specific Embodiment 2

[0069] This specific embodiment 2 is a specific refinement of the four key parameter models constructed in step S20, thereby constructing the action estimation function model required for deep learning and the key parameter model required for training.

[0070] Such as figure 2 As shown, in this embodiment, step S20 specifically includes the following steps:

[0071] Step S211, constructing an optimal motion estimation function model;

[0072] Step S212, defining a minimum distance function model between transformations;

[0073] Step S213, obtain the timely feedback function model of the action;

[0074] Step S214, sort out the loss function model.

[0075] The core problem of 3D image registration is to find an optimal action execution strategy to command the intelligent agent to make decisions.

[0076] In the present invention, at first obtain the best action estimation function model according to the Bellman equation:

[0077] Q t (s t ,a t ) = max τ E[r t +γr ...

specific Embodiment 3

[0095] Such as image 3 As shown, in this embodiment, step S50 specifically includes the following steps:

[0096] Step S511, input image;

[0097] Step S512, performing coarse alignment processing;

[0098] Step S513, judging whether the coarse alignment is successful, if the coarse alignment is successful, go to step S514, otherwise go to step S512;

[0099] Step S514, perform fine alignment processing;

[0100] Step S515, judging whether the fine alignment is successful, if successful, go to step S516, otherwise go to step S514;

[0101] Step S516, completing the three-dimensional image registration.

[0102] Fine alignment is carried out starting from the last action position of the coarse alignment. Firstly, the coarse alignment and then the fine alignment can quickly converge to obtain the 3D registration result of the current image, which gradually improves the 3D registration accuracy and efficiency of the image in the present invention, which is different from the...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a three-dimensional image registration method for an AR system. The "estimation" of the best action of the reference image; the convolutional neural network training is performed on the sample image using a pyramid strategy, and two network models are generated for coarse alignment and fine alignment; the input scene image and the target image are used, and the key parameter model and The two network models perform coarse alignment and fine alignment respectively to complete 3D image registration. The present invention, based on the registration strategy of deep reinforcement learning, defines the three-dimensional image registration problem as a series of continuous actions to achieve image alignment and operation, and finds the best action in a limited solution to improve the alignment effect, which can ensure that The globally optimal alignment parameters ensure the accuracy of 3D image registration.

Description

technical field [0001] The invention relates to the field of augmented reality technology, in particular to a three-dimensional image registration method for an AR system. Background technique [0002] In recent years, AR (Augmented Reality, Augmented Reality) technology has gradually become a research hotspot and has a very broad application prospect. In order to achieve the perfect fusion of virtual and reality, high-precision 3D image registration is very important. [0003] Most of the traditional image registration algorithms are expressed as an optimization problem. A general matching strategy is used to measure the similarity between image pairs, and then the conversion parameters between images are calculated by the optimal criterion. This method faces two problems. Challenges, one is that the general matching strategy is usually non-convex in the registration parameter space, and the general optimal criterion does not perform well on this non-convex problem; the oth...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Patents(China)
IPC IPC(8): G06T17/00G06T19/00G06K9/00G06N3/04
CPCG06T17/00G06T19/006G06V40/50G06N3/045
Inventor 万磊赵常均李博肖定坤
Owner 广州智能装备研究院有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap