CN113052825A - Real-time three-dimensional deformation measurement method based on GPU parallel acceleration - Google Patents

Real-time three-dimensional deformation measurement method based on GPU parallel acceleration Download PDF

Info

Publication number
CN113052825A
CN113052825A CN202110333748.3A CN202110333748A CN113052825A CN 113052825 A CN113052825 A CN 113052825A CN 202110333748 A CN202110333748 A CN 202110333748A CN 113052825 A CN113052825 A CN 113052825A
Authority
CN
China
Prior art keywords
interest
deformation
dimensional
point
gpu
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110333748.3A
Other languages
Chinese (zh)
Other versions
CN113052825B (en
Inventor
董守斌
林傲宇
蒋震宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN202110333748.3A priority Critical patent/CN113052825B/en
Publication of CN113052825A publication Critical patent/CN113052825A/en
Application granted granted Critical
Publication of CN113052825B publication Critical patent/CN113052825B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
    • G01B11/00—Measuring arrangements characterised by the use of optical techniques
    • G01B11/16—Measuring arrangements characterised by the use of optical techniques for measuring the deformation in a solid, e.g. optical strain gauge
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G06T1/60—Memory management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
    • G06T7/85—Stereo camera calibration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • G06T2207/10012—Stereo images
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Abstract

The invention discloses a real-time three-dimensional deformation measurement method based on GPU parallel acceleration, which comprises the following steps: 1) obtaining a projection matrix through stereo calibration; 2) shooting images of an object before and after deformation; 3) selecting an interest area and an interest point; 4) transmitting the projection matrix, the image and the interest point to a GPU; 5) calculating a three-dimensional coordinate of the interest point before deformation; 6) calculating the three-dimensional deformation of the interest point at each moment in the deformation; 7) and transmitting the three-dimensional deformation data back to the CPU. According to the method, a left camera image before deformation is used as a reference image in all matching, so that IC-GN pre-calculated data such as Hessian matrixes corresponding to interest points are multiplexed; the GPU acceleration deformation measurement program developed based on the CUDA heterogeneous computing platform can exert the computing performance of GPU hardware equipment, the computing speed of three-dimensional deformation measurement is greatly improved aiming at the optimization technologies such as memory access of the GPU program, and the like, and the requirement of real-time three-dimensional deformation measurement is met.

Description

Real-time three-dimensional deformation measurement method based on GPU parallel acceleration
Technical Field
The invention relates to the technical field of optical measurement, in particular to a real-time three-dimensional deformation measurement method based on GPU parallel acceleration.
Background
In the fields of science and engineering, the three-dimensional digital image correlation method is widely applied to three-dimensional deformation measurement due to the advantages of simple device, non-contact type and the like. The three-dimensional digital image correlation method can measure the three-dimensional appearance and the full-field three-dimensional deformation of a curved surface object, and has very rich application scenes. However, since it directly processes a high-resolution digital image, the amount of calculation thereof is also considerable, with a problem that the calculation takes a long time. With the development of digital image acquisition technology, the image resolution and sampling rate are improved, and the problem is more prominent, so that the application of a three-dimensional digital image correlation method in some real-time monitoring scenes and the like is limited. Secondly, as a measuring method, it is also important to maintain its high accuracy.
In recent years, researchers have made great efforts to improve the computational efficiency of three-dimensional digital image correlation methods. However, the results of these studies are not satisfactory and always give a trade-off between accuracy and efficiency. There are a number of existing schemes that improve computational efficiency by optimizing correlation algorithms, reducing redundant computations, and using multi-threading techniques to speed up programs. However, because the computing capability of the multi-core processor is limited and the adopted computing strategy is simple, the highest computing speed reported at present is only 50000POI/s, and the requirement of real-time processing is far from being met; the speed is achieved on the premise of only using a simple first-order shape function, and the precision is not as good as that of a scheme using a second-order shape function.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a real-time three-dimensional deformation measuring method based on GPU parallel acceleration. In all matching, the left camera image L before deformation is used0As a reference image, the pre-calculation data of the IC-GN algorithm can be repeatedly used, so that a large amount of calculation is saved. The method has the advantages that the algorithm is accelerated by developing a GPU program based on CUDA, the computing capability of hardware is fully exerted, real-time three-dimensional deformation measurement is achieved, the computing speed can exceed 40 frames per second when the number of interest points is about 10000, and the problems that the computing speed of an existing three-dimensional deformation measurement method is low and real-time measurement requirements cannot be met are solved.
In order to achieve the purpose, the technical scheme provided by the invention is as follows: a real-time three-dimensional deformation measurement method based on GPU parallel acceleration comprises the following steps:
1) using Zhangyingyou scaling method to make stereo scaling on left and right fixed cameras to obtain projection matrix M of left and right camerasL、MR；
2) Synchronously shooting left and right images L of the surface of the target object before deformation by using the two cameras0、 R0And the left and right images L at the ith moment in the deformation processi、RiWherein i ═ 1,2,3, …, n;
3) in the image L0Selecting an interest area, and taking a batch of interest points P at equal intervals in the interest areaL0；
4) Copying a projection matrix of the camera, all shot images and interest points to a GPU;
5) on the GPU, for each interest point PL0Firstly, an IC-GN algorithm is used for searching the image R0Corresponding point P onR0Then using PL0、PR0The coordinates of the two points are calculated by a triangulation method to obtain the three-dimensional coordinates P of the interest point before deformationW0；
6) On the GPU, the following operations are carried out on each time i in the deformation process: first, for each point of interest PL0Use the IC-GN algorithm to find it in the image Li、RiCorresponding point P onLiAnd PRiThen using PLi、PRiThe coordinates of the two points are calculated by triangulation to obtain the three-dimensional coordinates P of the interest point at the moment iwiFinally with PWiMinus PW0Obtaining three-dimensional deformation data D of the interest point at the moment iWi；
7) And copying the three-dimensional deformation data of all the interest points at each moment back to the CPU, so as to obtain the three-dimensional deformation of the surface of the object at each moment.
In step 1), the projection matrix has a size of 3 × 4, which represents the relationship between the three-dimensional space coordinates and the two-dimensional coordinates on the camera image.
In step 3), the interest area refers to an area which needs to be measured and is designated by a user, and the interest point interval in the area is also selected according to the measurement requirement.
In step 4), the shot image is firstly stored in a page-locked memory allocated by using a runtime function cudamallocost in the CUDA, and then the data is copied to the GPU by using a runtime function cudaMemcpy in the CUDA; the camera parameters are copied to the constant memory of the GPU, enabling it to be accessed at high speed.
In step 5) and step 6), the IC-GN algorithm is used, wherein the input is a batch of interest points, reference images and target images, and a CUDA thread block is used to process a computation task corresponding to an interest point, including the following:
a. carrying out precalculation: for each interest point, calculating a corresponding Hessian matrix and storing data; using a CUDA thread block to complete the calculation task of each interest point, wherein the Hessian matrix is
Figure RE-GDA0003079967210000031
Figure RE-GDA0003079967210000032
Representing the accumulation of all pixel positions within a reference sub-area, which is a 33 x 33 sub-image centered at the point of interest; ψ is the coordinate of the point of interest, # is the local coordinate in the reference sub-area, # R (ψ + ζ) is the gradient of the reference image,
Figure RE-GDA0003079967210000033
is a Jacobian matrix, and T represents the transposition of the matrix;
b. and (c) for each interest point, setting the coordinate of the interest point as (x, y), and estimating the initial value p of the deformation vector of the interest point by using an image feature assisted method (u, u)x,uy,uxx,uxy,uyy,v,vx,vy,vxx,vxy,vyy) (ii) a Wherein u, v are translation amounts; u. ofx,uy,vx,vyIs the first order gradient component; u. ofxx,uxy,uyy,vxx,vxy,vyyIs the second order gradient component; here, a CUDA thread is used to accomplish a programCalculating the interest points;
c. for each interest point, iteratively updating the corresponding deformation vector p according to the following steps:
c1, calculating the deformation vector increment delta p:
Figure RE-GDA0003079967210000034
wherein H-1The inverse of the Hessian matrix is represented,
Figure RE-GDA0003079967210000035
and
Figure RE-GDA0003079967210000036
normalized coefficients for the reference and target sub-regions respectively,
Figure RE-GDA0003079967210000037
representing the target sub-area after subtraction of the mean value of the gray levels,
Figure RE-GDA0003079967210000038
representing the reference subarea after subtracting the gray mean value, and W (zeta; p) represents a transformation function from the local coordinate of the reference subarea to the local coordinate of the target subarea;
c2, calculating new transformation function W (ζ; p') as W (W) by using Δ p-1(ζ; Δ p); p) wherein W-1(ζ; Δ p) is the inverse of W (ζ; Δ p); then extracting a new deformation vector p 'from the new transformation function W (zeta; p');
c3, updating p, namely making p equal to p';
c4, repeating c1 to c3 until | | | Δ p | | < 0.001, and | | · | | | represents the modular length of the vector;
d. and (3) calculating the coordinates of the interest point at the corresponding point of the target map as (x ', y'), wherein x 'is x + u, and y' is y + v.
In step 6), the IC-GN algorithm used does not need to be pre-calculated, and the data stored after the IC-GN algorithm is pre-calculated in step 5) is directly used.
In step 5) and step 6), triangulation methods are used, including the following:
a. let the coordinate of the interest point on the left view be (x)L,yL) The point coordinate on the right view is (x)R,yR) The corresponding three-dimensional coordinate is (X)W,YW,ZW)；
b. (X) is calculated in the following mannerW,YW,ZW) Each CUDA thread calculates a three-dimensional coordinate;
Figure RE-GDA0003079967210000041
wherein A is a coefficient matrix, and
Figure RE-GDA0003079967210000042
b is a matrix of constant terms, an
Figure RE-GDA0003079967210000043
mL rcRepresenting a projection matrix MLM in the r-th row and c-th columnR rcRepresenting a projection matrix MRWhere r is 1,2,3, and c is 1,2,3, 4.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. in all matching of three-dimensional deformation measurement, the invention leads the left camera image L before deformation to be0As a reference image, data obtained by IC-GN pre-calculation such as Hessian matrix and the like corresponding to the interest points can be repeatedly utilized in the processing of images at all times in the three-dimensional deformation process, so that a large amount of calculation is saved, the processing speed is accelerated, and the calculation time is shortened.
2. The GPU program is developed based on the CUDA platform, so that development cost can be saved, development difficulty is reduced, meanwhile, the special CUDA program and corresponding program optimization can furthest exert the calculation performance of GPU hardware, and the requirement of measuring three-dimensional deformation in real time is met.
Detailed Description
The present invention will be described in further detail with reference to examples, but the embodiments of the present invention are not limited thereto.
The real-time three-dimensional deformation measurement method based on GPU parallel acceleration provided by the embodiment comprises the following steps:
1) using a Zhangyingyou calibration method to carry out three-dimensional calibration on the left and right fixed cameras to obtain the projection matrixes M of the left and right camerasL、MR。
2) Synchronously shooting left and right images L of the surface of the target object before deformation by using the two cameras0、 R0And the left and right images L at the ith moment in the deformation processi、RiWhere i is 1,2,3, …, n.
3) In the image L0Selecting an interest area, and taking a batch of interest points P at equal intervals in the interest areaL0(ii) a The interest area refers to an area which is specified by a user and needs to be measured, and the interest point interval in the area is also selected according to the measurement requirement.
4) Copying a projection matrix of the camera, all shot images and interest points to a GPU; the method comprises the steps that a shot image is stored in a page locking memory distributed by using a runtime function cudaMallocost in a CUDA (compute unified device architecture), and then data are copied to a GPU (graphics processing unit) by using a runtime function cudaMemcpy function in the CUDA; the projection matrix of the camera is copied to the constant memory of the GPU, allowing it to be accessed at high speed.
5) On the GPU, for each interest point PL0Firstly, an IC-GN algorithm is used for searching the image R0Corresponding point P onR0Then using PL0、PR0The coordinates of the two points are calculated by a triangulation method to obtain the three-dimensional coordinates P of the interest point before deformationW0。
6) On the GPU, the following operations are carried out on each time i in the deformation process: first, for each point of interest PL0Use the IC-GN algorithm to find it in the image Li、RiCorresponding point P onLiAnd PRiThen using PLi、PRiThe coordinates of the two points are calculated by triangulation to obtain the three-dimensional coordinate of the interest point at the time iMark PwiFinally with PWiMinus PW0Obtaining three-dimensional deformation data D of the interest point at the moment iWi(ii) a The IC-GN algorithm used does not need to be pre-calculated, and the data stored after the IC-GN algorithm is pre-calculated in the step 5) is directly utilized.
7) And copying the three-dimensional deformation data of all the interest points at each moment back to the CPU, so as to obtain the three-dimensional deformation of the surface of the object at each moment.
In step 5) and step 6), the IC-GN algorithm is used, wherein the input is a batch of interest points, reference images and target images, and a CUDA thread block is used to process a computation task corresponding to an interest point, including the following:
a. carrying out precalculation: for each interest point, calculating a corresponding Hessian matrix and storing data; using a CUDA thread block to complete the calculation task of each interest point, wherein the Hessian matrix is
Figure RE-GDA0003079967210000061
Figure RE-GDA0003079967210000062
Representing the accumulation of all pixel positions within a reference sub-area, which is a 33 x 33 sub-image centered at the point of interest; ψ is the coordinate of the point of interest, # is the local coordinate in the reference sub-area, # R (ψ + ζ) is the gradient of the reference image,
Figure RE-GDA0003079967210000063
is a Jacobian matrix, and T represents the transposition of the matrix;
b. and (c) for each interest point, setting the coordinate of the interest point as (x, y), and estimating the initial value p of the deformation vector of the interest point by using an image feature assisted method (u, u)x,uy,uxx,uxy,uyy,v,vx,vy,vxx,vxy,vyy) (ii) a Wherein u, v are translation amounts; u. ofx,uy,vx,vyIs the first order gradient component; u. ofxx,uxy,uyy,vxx,vxy,vyyIs twoAn order gradient component; here, a CUDA thread is used to perform a point of interest computing task;
c. for each interest point, iteratively updating the corresponding deformation vector p according to the following steps:
c1, calculating the deformation vector increment delta p:
Figure RE-GDA0003079967210000071
wherein H-1The inverse of the Hessian matrix is represented,
Figure RE-GDA0003079967210000072
and
Figure RE-GDA0003079967210000073
normalized coefficients for the reference and target sub-regions respectively,
Figure RE-GDA0003079967210000074
representing the target sub-area after subtraction of the mean value of the gray levels,
Figure RE-GDA0003079967210000075
representing the reference subarea after subtracting the gray mean value, and W (zeta; p) represents a transformation function from the local coordinate of the reference subarea to the local coordinate of the target subarea;
c2, calculating new transformation function W (ζ; p') as W (W) by using Δ p-1(ζ; Δ p); p) wherein W-1(ζ; Δ p) is the inverse of W (ζ; Δ p); then extracting a new deformation vector p 'from the new transformation function W (zeta; p');
c3, updating p, namely making p equal to p';
c4, repeating c1 to c3 until | | | Δ p | | < 0.001, and | | · | | | represents the modular length of the vector;
d. and (3) calculating the coordinates of the interest point at the corresponding point of the target map as (x ', y'), wherein x 'is x + u, and y' is y + v.
In step 5) and step 6), triangulation methods are used, including the following:
a. setting the point of interest atThe coordinates on the left view are (x)L,yL) The point coordinate on the right view is (x)R,yR) The corresponding three-dimensional coordinate is (X)W,YW,ZW)；
b. (X) is calculated in the following mannerW,YW,ZW) Each CUDA thread calculates a three-dimensional coordinate;
Figure RE-GDA0003079967210000076
wherein A is a coefficient matrix, and
Figure RE-GDA0003079967210000077
b is a matrix of constant terms, an
Figure RE-GDA0003079967210000081
mL rcRepresenting a projection matrix MLM in the r-th row and c-th columnR rcRepresenting a projection matrix MRWhere r is 1,2,3, and c is 1,2,3, 4.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (7)

1. A real-time three-dimensional deformation measurement method based on GPU parallel acceleration is characterized by comprising the following steps:
1) using Zhangyingyou scaling method to make stereo scaling on left and right fixed cameras to obtain projection matrix M of left and right camerasL、MR；
2) Synchronously shooting left and right images L of the surface of the target object before deformation by using the two cameras0、R0And the left and right images L at the ith moment in the deformation processi、RiWherein i ═ 1,2,3, …, n;
3) in the image L0Selecting an interest area, and taking a batch of interest points P at equal intervals in the interest areaL0；
4) Copying a projection matrix of the camera, all shot images and interest points to a GPU;
5) on the GPU, for each interest point PL0Firstly, an IC-GN algorithm is used for searching the image R0Corresponding point P onR0Then using PL0、PR0The coordinates of the two points are calculated by a triangulation method to obtain the three-dimensional coordinates P of the interest point before deformationW0；
6) On the GPU, the following operations are carried out on each time i in the deformation process: first, for each point of interest PL0Use the IC-GN algorithm to find it in the image Li、RiCorresponding point P onLiAnd PRiThen using PLi、PRiThe coordinates of the two points are calculated by triangulation to obtain the three-dimensional coordinates P of the interest point at the moment iwiFinally with PWiMinus PW0Obtaining three-dimensional deformation data D of the interest point at the moment iWi；
7) And copying the three-dimensional deformation data of all the interest points at each moment back to the CPU, so as to obtain the three-dimensional deformation of the surface of the object at each moment.
2. The real-time three-dimensional deformation measurement method based on GPU parallel acceleration as claimed in claim 1, characterized in that: in step 1), the projection matrix has a size of 3 × 4, which represents the relationship between the three-dimensional space coordinates and the two-dimensional coordinates on the camera image.
3. The real-time three-dimensional deformation measurement method based on GPU parallel acceleration as claimed in claim 1, characterized in that: in step 3), the interest area refers to an area which needs to be measured and is designated by a user, and the interest point interval in the area is also selected according to the measurement requirement.
4. The real-time three-dimensional deformation measurement method based on GPU parallel acceleration as claimed in claim 1, characterized in that: in step 4), the shot image is firstly stored in a page-locked memory allocated by using a runtime function cudamallocost in the CUDA, and then the data is copied to the GPU by using a runtime function cudaMemcpy in the CUDA; the projection matrix of the camera is copied to the constant memory of the GPU, enabling it to be accessed at high speed.
5. The real-time three-dimensional deformation measurement method based on GPU parallel acceleration as claimed in claim 1, characterized in that: in step 5) and step 6), the IC-GN algorithm is used, wherein the input is a batch of interest points, reference images and target images, and a CUDA thread block is used to process a computation task corresponding to an interest point, including the following:
a. carrying out precalculation: for each interest point, calculating a corresponding Hessian matrix and storing data; using a CUDA thread block to complete the calculation task of each interest point, wherein the Hessian matrix is
Figure FDA0002997371300000021
Figure FDA0002997371300000022
Representing the accumulation of all pixel positions within a reference sub-area, which is a 33 x 33 sub-image centered at the point of interest; ψ is the coordinate of the point of interest, ζ is the local coordinate in the reference sub-area,
Figure FDA0002997371300000023
is the gradient of the reference image or images,
Figure FDA0002997371300000024
is a Jacobian matrix, and T represents the transposition of the matrix;
b. and (c) for each interest point, setting the coordinate of the interest point as (x, y), and estimating the initial value p of the deformation vector of the interest point by using an image feature assisted method (u, u)x,uy,uxx,uxy,uyy,v,vx,vy,vxx,vxy,vyy) (ii) a Wherein u, v are translation amounts; u. ofx,uy,vx,vyIs the first order gradient component; u. ofxx,uxy,uyy,vxx,vxy,vyyIs the second order gradient component; here, a CUDA thread is used to perform a point of interest computing task;
c. for each interest point, iteratively updating the corresponding deformation vector p according to the following steps:
c1, calculating the deformation vector increment delta p:
Figure FDA0002997371300000025
wherein H-1The inverse of the Hessian matrix is represented,
Figure FDA0002997371300000031
and
Figure FDA0002997371300000032
normalized coefficients for the reference and target sub-regions respectively,
Figure FDA0002997371300000033
representing the target sub-area after subtraction of the mean value of the gray levels,
Figure FDA0002997371300000034
representing the reference subarea after subtracting the gray mean value, and W (zeta; p) represents a transformation function from the local coordinate of the reference subarea to the local coordinate of the target subarea;
c2, calculating new transformation function W (ζ; p') as W (W) by using Δ p-1(ζ; Δ p); p) wherein W-1(ζ; Δ p) is the inverse of W (ζ; Δ p); then extracting a new deformation vector p 'from the new transformation function W (zeta; p');
c3, updating p, namely making p equal to p';
c4, repeating c1 to c3 until | | | Δ p | | < 0.001, and | | · | | | represents the modular length of the vector;
d. and (3) calculating the coordinates of the interest point at the corresponding point of the target map as (x ', y'), wherein x 'is x + u, and y' is y + v.
6. The real-time three-dimensional deformation measurement method based on GPU parallel acceleration as claimed in claim 1, characterized in that: in step 6), the IC-GN algorithm used does not need to be pre-calculated, and the data stored after the IC-GN algorithm is pre-calculated in step 5) is directly used.
7. The real-time three-dimensional deformation measurement method based on GPU parallel acceleration as claimed in claim 1, characterized in that: in step 5) and step 6), triangulation methods are used, including the following:
a. let the coordinate of the interest point on the left view be (x)L,yL) The point coordinate on the right view is (x)R,yR) The corresponding three-dimensional coordinate is (X)W,YW,ZW)；
b. (X) is calculated in the following mannerW,YW,ZW) Each CUDA thread calculates a three-dimensional coordinate;
Figure FDA0002997371300000035
wherein A is a coefficient matrix, and
Figure FDA0002997371300000036
b is a matrix of constant terms, an
Figure FDA0002997371300000041
mL rcRepresenting a projection matrix MLM in the r-th row and c-th columnR rcRepresenting a projection matrix MRWhere r is 1,2,3, and c is 1,2,3, 4.
CN202110333748.3A 2021-03-29 2021-03-29 Real-time three-dimensional deformation measurement method based on GPU parallel acceleration Active CN113052825B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110333748.3A CN113052825B (en) 2021-03-29 2021-03-29 Real-time three-dimensional deformation measurement method based on GPU parallel acceleration

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110333748.3A CN113052825B (en) 2021-03-29 2021-03-29 Real-time three-dimensional deformation measurement method based on GPU parallel acceleration

Publications (2)

Publication Number Publication Date
CN113052825A true CN113052825A (en) 2021-06-29
CN113052825B CN113052825B (en) 2023-04-21

Family

ID=76516013

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110333748.3A Active CN113052825B (en) 2021-03-29 2021-03-29 Real-time three-dimensional deformation measurement method based on GPU parallel acceleration

Country Status (1)

Country Link
CN (1) CN113052825B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114526682A (en) * 2022-01-13 2022-05-24 华南理工大学 Deformation measurement method based on image feature enhanced digital volume image correlation method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2000007501A1 (en) * 1998-08-03 2000-02-17 Cardiac Pathways Corporation A dynamically alterable three-dimensional graphical model of a body region
WO2010005973A2 (en) * 2008-07-07 2010-01-14 The Johns Hopkins University Automated surface-based anatomical analysis based on atlas-based segmentation of medical imaging
CN102003946A (en) * 2010-09-02 2011-04-06 北京航空航天大学 High-temperature three-dimensional digital image related measurement system and measurement method
CN104634266A (en) * 2013-11-08 2015-05-20 南京化工职业技术学院 Mechanical sealing end surface deformation measurement system based on binocular vision DIC and measurement method thereof
CN110414573A (en) * 2019-07-09 2019-11-05 北京卫星制造厂有限公司 A kind of Route Dependence digital picture Image Matching accelerated based on GPU

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2000007501A1 (en) * 1998-08-03 2000-02-17 Cardiac Pathways Corporation A dynamically alterable three-dimensional graphical model of a body region
WO2010005973A2 (en) * 2008-07-07 2010-01-14 The Johns Hopkins University Automated surface-based anatomical analysis based on atlas-based segmentation of medical imaging
CN102003946A (en) * 2010-09-02 2011-04-06 北京航空航天大学 High-temperature three-dimensional digital image related measurement system and measurement method
CN104634266A (en) * 2013-11-08 2015-05-20 南京化工职业技术学院 Mechanical sealing end surface deformation measurement system based on binocular vision DIC and measurement method thereof
CN110414573A (en) * 2019-07-09 2019-11-05 北京卫星制造厂有限公司 A kind of Route Dependence digital picture Image Matching accelerated based on GPU

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
JIANG ZHENYU 等: "Path-independent digital image correlation with high accuracy，speed and robustness", <<网页>> *
LINGQI ZHANG等: "High accuracy digital image correlation powered by GPU-based parallel computing", <<网页>> *
黄健文: "高精度数字图像相关法的并行算法设计及异构计算实现" *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114526682A (en) * 2022-01-13 2022-05-24 华南理工大学 Deformation measurement method based on image feature enhanced digital volume image correlation method

Also Published As

Publication number Publication date
CN113052825B (en) 2023-04-21

Similar Documents

Publication Publication Date Title
US10628949B2 (en) Image processing with iterative closest point (ICP) technique
WO2018127007A1 (en) Depth image acquisition method and system
CN107230225B (en) Method and apparatus for three-dimensional reconstruction
JP3651590B2 (en) Method for restoring 3D scene structure and camera motion directly from points, lines and / or from image intensity
CN110070598B (en) Mobile terminal for 3D scanning reconstruction and 3D scanning reconstruction method thereof
WO2022206020A1 (en) Method and apparatus for estimating depth of field of image, and terminal device and storage medium
CN115690324A (en) Neural radiation field reconstruction optimization method and device based on point cloud
CN114399554A (en) Calibration method and system of multi-camera system
US20240046557A1 (en) Method, device, and non-transitory computer-readable storage medium for reconstructing a three-dimensional model
CN108022261B (en) Non-rigid image registration method based on improved optical flow field model
CN113111861A (en) Face texture feature extraction method, 3D face reconstruction method, device and storage medium
Eichhardt et al. Affine correspondences between central cameras for rapid relative pose estimation
CN114494150A (en) Design method of monocular vision odometer based on semi-direct method
CN113052825B (en) Real-time three-dimensional deformation measurement method based on GPU parallel acceleration
CN109300148B (en) Multi-source image registration method based on method cooperation
CN109215118B (en) Incremental motion structure recovery optimization method based on image sequence
CN112927251B (en) Morphology-based scene dense depth map acquisition method, system and device
WO2018133027A1 (en) Grayscale constraint-based method and apparatus for integer-pixel search for three-dimensional digital speckle pattern
CN115917597A (en) Promoting 2D representations to 3D using attention models
CN114399547B (en) Monocular SLAM robust initialization method based on multiframe
CN115690229A (en) Line structure light plane on-site calibration method based on two-dimensional target
CN115239559A (en) Depth map super-resolution method and system for fusion view synthesis
CN115631223A (en) Multi-view stereo reconstruction method based on self-adaptive learning and aggregation
CN112116640B (en) Binocular stereo matching method based on OpenCL
CN114549629A (en) Method for estimating three-dimensional pose of target by underwater monocular vision

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant