CN110414573A - A kind of Route Dependence digital picture Image Matching accelerated based on GPU - Google Patents
A kind of Route Dependence digital picture Image Matching accelerated based on GPU Download PDFInfo
- Publication number
- CN110414573A CN110414573A CN201910615683.4A CN201910615683A CN110414573A CN 110414573 A CN110414573 A CN 110414573A CN 201910615683 A CN201910615683 A CN 201910615683A CN 110414573 A CN110414573 A CN 110414573A
- Authority
- CN
- China
- Prior art keywords
- point
- interest
- seed point
- matching
- pix
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/94—Hardware or software architectures specially adapted for image or video understanding
- G06V10/955—Hardware or software architectures specially adapted for image or video understanding using specific electronic processors
Abstract
A kind of Route Dependence digital picture Image Matching accelerated based on GPU, real-time DIC measurement is carried out using the Route Dependence digital picture related algorithm accelerated based on GPU, initial value pass through mechanism and IC-GN method are combined into core, it completes from whole pixel to the computational algorithm of sub-pix result, and realizes the parallel acceleration of algorithm based on GPU;In conjunction with unique seed point Diffusion Strategy boosting algorithm degree of parallelism.The present invention has the advantages that high calculating speed, high-precision, high robust, belongs to Digital-image correlation method algorithm field.
Description
Technical field
The present invention relates to a kind of Route Dependence digital picture Image Matchings accelerated based on GPU, belong to digital picture
Measurement of correlation algorithm field.
Background technique
Realize that real-time measurement needs pole using related (Digital Image Correlation, the DIC) method of digital picture
High image processing speed, this proposes challenge to the computational efficiency of DIC method.For being widely used at present with reversely combined
Gauss-newton method (Inverse Compositional Gauss-Newton, IC-GN) be core DIC algorithm, if only
The optimization in algorithm level is only carried out, when facing the calculating of close-packed lattice, due to the essence of its serial process, it may be difficult to reach
Satisfactory calculating speed.
The calculating speed of DIC algorithm will be substantially improved using the parallel computing based on GPU.By based on quick
The cross correlation algorithm (Fast Fourier Transform Cross Correlation, FFT-CC) of Fourier transformation, can be with
Whole pixel image matching result is obtained, which obtains Asia after series of iterations optimization for by the initial value as IC-GN method
Pixel matching result.Since the processing of point-of-interest each in image is independent from each other, which is that path is unrelated,
Have the basis of algorithm task parallelization, the tall and handsome unified calculation equipment framework (Compute provided up to company is provided
Unified Device Architecture, CUDA) realize GPU parallel computation, algorithm will reach satisfaction, and processing is wanted in real time
The calculating speed asked.
However, the whole pixel initial value of DIC algorithm dependence FFT-CC algorithm offer that path is unrelated, and FFT-CC algorithm is with template
It based on convolution, is limited by convolution window size, will be failed in the case where large deformation.In addition, FFT-CC algorithm is only
The zeroth order deformation of image is considered, however the deformation of high-order but has occurred between actual target image and reference picture, is used
FFT-CC algorithm, which carries out matching, will lead to biggish systematic error, and especially when torsional deformation is serious between image, this method is obtained
The matching result arrived can even will result directly in comprising errors more than five pixels, result inaccurate in this way as initial value
The slow convergence of IC-GN algorithm does not restrain even, has seriously affected the robustness of algorithm entirety.Therefore, new algorithm, In are developed
While having superelevation calculating speed, guarantee the accuracy and robustness of algorithm, high performance real-time DIC deformation and pattern are surveyed
Measure important in inhibiting.
Summary of the invention
The technical problem to be solved by the present invention is having overcome the deficiencies of the prior art and provide a kind of road accelerated based on GPU
The advantages of diameter relies on digital picture Image Matching, has both high calculating speed, high-precision, high robust.The method of the present invention is adopted
Real-time DIC measurement is carried out with the Route Dependence digital picture related algorithm accelerated based on GPU, i.e., by initial value pass through mechanism and IC-
GN method is combined into core, completes from whole pixel to the computational algorithm of sub-pix result, and realizes that the parallel of algorithm adds based on GPU
Speed;In conjunction with unique seed point Diffusion Strategy boosting algorithm degree of parallelism.
The object of the invention is achieved by the following technical programs:
A kind of Route Dependence digital picture Image Matching accelerated based on GPU, is included the following steps:
S1, region to be matched is chosen in reference to figure, point-of-interest and point-of-interest are determined in the region to be matched
Number N;
S2, n point-of-interest is chosen in point-of-interest as seed point, to seed point and seed point in target figure
Corresponding points carry out whole pixel matching;Using the whole pixel matching result as the matched initial value of sub-pix;
S3, counter counter is defined, the initial value of counter is 0;
S4, to having activated point-of-interest and described activated corresponding points of the point-of-interest in target figure using the side IC-GN
Method parallel computation sub-pix matching result;It is described activated point-of-interest be have initial value not carry out sub-pix matched interested
Point;
The number for passing through the point-of-interest that sub-pix matched and obtained convergence result in S4 is added on counter, and
S4 is matched by sub-pix and obtains the point-of-interest of convergence result as source seed point;
WhenAnd by sub-pix matching obtain convergence result point-of-interest number be 0 when, be transferred to
S7;
S5, the relative position by source seed point described in S4 with respect to the point-of-interest that source seed point is faced in domain notify to give source kind
Son point faces the point-of-interest in domain;
S6, scanning it is all do not carry out the matched point-of-interest of sub-pix, by it is all receive source seed point relative positions notice
Do not carry out the matched point-of-interest of sub-pix, as point to be activated, the notice of the relative position according to S5, according to by
It closely arrives remote sequence and finds the source seed point nearest apart from point to be activated, matched according to the sub-pix of the nearest source seed point
As a result the matched initial value of sub-pix of point to be activated is calculated with relative positional relationship;
Using the point to be activated for obtaining sub-pix matching initial value as point-of-interest has been activated, it is then transferred to S4;
S7, terminated based on the GPU Route Dependence digital picture Image Matching accelerated.
The above-mentioned Route Dependence digital picture Image Matching accelerated based on GPU, in S1, is used in region to be matched
Evenly spaced method chooses point-of-interest.
The above-mentioned Route Dependence digital picture Image Matching accelerated based on GPU, in S2, the n senses to be calculated
Point of interest is uniformly distributed in region to be matched;
The above-mentioned Route Dependence digital picture Image Matching accelerated based on GPU, in the case where Stereo matching, for the first time
Matching uses polar curve matching process, carries out whole pixel matching to the corresponding points of seed point and seed point in target figure, for the second time
And matching uses history deformation data prediction technique later, carries out whole picture to the corresponding points of seed point and seed point in target figure
Element matching;
In the matched situation of continuous modification, first fit uses Fast Fourier Transform (FFT) cross-correlation method, to seed point
Whole pixel matching is carried out with corresponding points of the seed point in target figure, second and matching later are predicted using history deformation data
Method carries out whole pixel matching to the corresponding points of seed point and seed point in target figure.
The above-mentioned Route Dependence digital picture Image Matching accelerated based on GPU, source seed point described in S5 faces domain and is
4 adjacent point-of-interests centered on the seed point of source.
The above-mentioned Route Dependence digital picture Image Matching accelerated based on GPU, source seed point described in S5 faces domain and is
Using source seed point 8 point-of-interests nearest as center distance sources seed point.
The above-mentioned Route Dependence digital picture Image Matching accelerated based on GPU, source seed point described in S5 faces domain and is
Using source seed point 8 point-of-interests nearest as center distance sources seed point, and, the source seed point and distance sources seed point are most
8 close point-of-interests form 4 point-of-interests that distance sources seed point is equal on cornerwise extended line of square;It is described
The equal point-of-interest of distance sources seed point is less than or equal to 50 pictures at a distance from the seed point of source on the cornerwise extended line of square
Element.
The above-mentioned Route Dependence digital picture Image Matching accelerated based on GPU, after S7, using all interested
The sub-pix matching result of point calculates the deformation with reference to region to be matched in figure or topographical information, then carries out visual in real time
Change processing.
A kind of computer readable storage medium, is stored thereon with computer program, when which is executed by processor, realizes
The step of above-mentioned Route Dependence digital picture Image Matching accelerated based on GPU.
The present invention has the following beneficial effects: compared with the prior art
(1) the method for the present invention calculating speed is high, can not only reach the requirement handled in real time, and better than most of path without
Close algorithm;
(2) the method for the present invention is accelerated using parallel computation, does not lose the high-precision of algorithm script;
(3) the method for the present invention uses initial value pass through mechanism, using the continuity of deformation as theoretical foundation, as long as providing seed
Point can provide accurate initial value for its neighborhood point, not only improve robustness, also improve convergence rate.
Detailed description of the invention
Fig. 1 is the step flow chart of the method for the present invention.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to implementation of the invention
Mode is described in further detail.
A kind of Route Dependence digital picture Image Matching accelerated based on GPU, is included the following steps, as shown in Figure 1:
S1, region to be matched is chosen in reference to figure, chosen in region to be matched using evenly spaced method and feel emerging
Interesting point (Point of Interest, POI), determines the number N of point-of-interest.
S2, n point-of-interest is chosen in point-of-interest as seed point, to seed point and seed point in target figure
Corresponding points carry out whole pixel matching;Corresponding points of the seed point in target figure are, with reference to the corresponding quilt of seed point in figure
Shoot the point in target figure corresponding to the point on body surface;It is matched just using the whole pixel matching result as sub-pix
Value;The n point-of-interests to be calculated are uniformly distributed in region to be matched;
S3, counter counter is defined, the initial value of counter is 0.
S4, to having activated point-of-interest and described activated corresponding points of the point-of-interest in target figure using the side IC-GN
Method parallel computation sub-pix matching result;It is described activated point-of-interest be have initial value not carry out sub-pix matched interested
Point;
By this recycle in convergence result is matched and obtained in S4 by sub-pix the number of point-of-interest be added to
On counter, and using this recycle in S4 by sub-pix match and obtain convergence result point-of-interest as source seed point;
WhenAnd by sub-pix matching obtain convergence result point-of-interest number be 0 when, be transferred to
S7, i.e., when the number of remaining point-of-interest is seldom and does not restrain still after cycle calculations, it is believed that the method for the present invention is
Terminate, the method for the present invention falls into endless loop caused by avoiding because of not the restraining of individual point-of-interests.
S5, the relative position by source seed point described in S4 with respect to the point-of-interest that source seed point is faced in domain notify to give source kind
Son point faces the point-of-interest in domain.
S6, scanning it is all do not carry out the matched point-of-interest of sub-pix, by it is all receive source seed point relative positions notice
Do not carry out the matched point-of-interest of sub-pix, as point to be activated, the notice of the relative position according to S5, according to by
It closely arrives remote sequence and finds the source seed point nearest apart from point to be activated, matched according to the sub-pix of the nearest source seed point
As a result the matched initial value of sub-pix of point to be activated is calculated with relative positional relationship;
Using the point to be activated for obtaining sub-pix matching initial value as point-of-interest has been activated, it is then transferred to S4.
S7, terminated based on the GPU Route Dependence digital picture Image Matching accelerated.
In the present embodiment, in the case where Stereo matching, first fit uses polar curve matching process, to seed point and seed
Corresponding points of the point in target figure carry out whole pixel matching, and second and matching later use history deformation data prediction technique,
Whole pixel matching is carried out to the corresponding points of seed point and seed point in target figure;
In the matched situation of continuous modification, first fit uses Fast Fourier Transform (FFT) cross-correlation method, to seed point
Whole pixel matching is carried out with corresponding points of the seed point in target figure, second and matching later are predicted using history deformation data
Method carries out whole pixel matching to the corresponding points of seed point and seed point in target figure.
In the present embodiment, domain is faced in source seed point described in S5 can be determined using one of following three kinds of methods:
A, it is 4 point-of-interests adjacent centered on the seed point of source, i.e. source seed point and phase that the source seed point, which faces domain,
4 adjacent point-of-interests form cross;
B, it is 8 point-of-interests nearest by center distance sources seed point of source seed point that the source seed point, which faces domain, i.e.,
8 point-of-interests of source seed point and distance sources seed point recently form square;
C, it is 8 point-of-interests nearest by center distance sources seed point of source seed point that the source seed point, which faces domain, and,
8 point-of-interests of the source seed point and distance sources seed point recently form distance sources on cornerwise extended line of square
4 equal point-of-interests of seed point;The equal point-of-interest of distance sources seed point on the cornerwise extended line of square
With at a distance from the seed point of source be less than or equal to 50 pixels.
Domain is faced using the source seed point that one of three of the above method determines, when the domain of facing is more than region to be matched
When, it only takes and described faces point-of-interest of the domain in region to be matched.
The present embodiment, using the sub-pix matching result of all point-of-interests, is calculated with reference to be matched in figure after S7
Then the deformation in region or topographical information carry out real-time visualization processing.
A kind of computer readable storage medium, is stored thereon with computer program, when which is executed by processor, realizes
The step of above-mentioned Route Dependence digital picture Image Matching accelerated based on GPU.
Claims (9)
1. a kind of Route Dependence digital picture Image Matching accelerated based on GPU, which comprises the steps of:
S1, region to be matched is chosen in reference to figure, the number of point-of-interest and point-of-interest is determined in the region to be matched
Mesh N;
S2, n point-of-interest is chosen in point-of-interest as seed point, to the correspondence of seed point and seed point in target figure
Point carries out whole pixel matching;Using the whole pixel matching result as the matched initial value of sub-pix;
S3, counter counter is defined, the initial value of counter is 0;
S4, to activated point-of-interest and it is described activated corresponding points of the point-of-interest in target figure using IC-GN method simultaneously
Row calculates sub-pix matching result;It is described activated point-of-interest be have an initial value do not carry out the matched point-of-interest of sub-pix;
It sub-pix will be passed through in S4 matches and obtains the number of point-of-interest of convergence result and be added on counter, and by S4
It is matched by sub-pix and obtains the point-of-interest of convergence result as source seed point;
WhenAnd by sub-pix matching obtain convergence result point-of-interest number be 0 when, be transferred to S7;
S5, the relative position by source seed point described in S4 with respect to the point-of-interest that source seed point is faced in domain notify to give source seed point
Face the point-of-interest in domain;
S6, scanning it is all do not carry out the matched point-of-interest of sub-pix, by it is all receive source seed point relative positions notice not
Carry out the matched point-of-interest of sub-pix, as point to be activated, the notice of the relative position according to S5, according to by closely to
Remote sequence finds the source seed point nearest apart from point to be activated, according to the sub-pix matching result of the nearest source seed point
The matched initial value of sub-pix of point to be activated is calculated with relative positional relationship;
Using the point to be activated for obtaining sub-pix matching initial value as point-of-interest has been activated, it is then transferred to S4;
S7, terminated based on the GPU Route Dependence digital picture Image Matching accelerated.
2. a kind of Route Dependence digital picture Image Matching accelerated based on GPU according to claim 1, feature
It is, in S1, point-of-interest is chosen using evenly spaced method in region to be matched.
3. a kind of Route Dependence digital picture Image Matching accelerated based on GPU according to claim 1, feature
It is, in S2, the n point-of-interests to be calculated are uniformly distributed in region to be matched;
4. a kind of Route Dependence digital picture Image Matching accelerated based on GPU according to claim 1, feature
It is, in the case where Stereo matching, first fit uses polar curve matching process, to seed point and seed point in target figure
Corresponding points carry out whole pixel matching, and second and matching later use history deformation data prediction technique, to seed point and seed
Corresponding points of the point in target figure carry out whole pixel matching;
In the matched situation of continuous modification, first fit uses Fast Fourier Transform (FFT) cross-correlation method, to seed point and kind
Corresponding points of the son point in target figure carry out whole pixel matching, and second and matching later use history deformation data prediction side
Method carries out whole pixel matching to the corresponding points of seed point and seed point in target figure.
5. a kind of Route Dependence digital picture Image Matching accelerated based on GPU according to claim 1, feature
It is, it is 4 point-of-interests adjacent centered on the seed point of source that domain is faced in source seed point described in S5.
6. a kind of Route Dependence digital picture Image Matching accelerated based on GPU according to claim 1, feature
It is, it is 8 point-of-interests nearest by center distance sources seed point of source seed point that domain is faced in source seed point described in S5.
7. a kind of Route Dependence digital picture Image Matching accelerated based on GPU according to claim 1, feature
It is, it is 8 point-of-interests nearest by center distance sources seed point of source seed point that domain is faced in source seed point described in S5, and,
8 point-of-interests of the source seed point and distance sources seed point recently form distance sources on cornerwise extended line of square
4 equal point-of-interests of seed point;The equal point-of-interest of distance sources seed point on the cornerwise extended line of square
With at a distance from the seed point of source be less than or equal to 50 pixels.
8. a kind of Route Dependence digital picture Image Matching accelerated based on GPU according to claim 1, feature
Be, after S7, using the sub-pix matching result of all point-of-interests, calculate with reference to region to be matched in figure deformation or
Then topographical information carries out real-time visualization processing.
9. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is held by processor
When row, the step of realizing one of claim 1~8 the method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910615683.4A CN110414573A (en) | 2019-07-09 | 2019-07-09 | A kind of Route Dependence digital picture Image Matching accelerated based on GPU |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910615683.4A CN110414573A (en) | 2019-07-09 | 2019-07-09 | A kind of Route Dependence digital picture Image Matching accelerated based on GPU |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110414573A true CN110414573A (en) | 2019-11-05 |
Family
ID=68360774
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910615683.4A Pending CN110414573A (en) | 2019-07-09 | 2019-07-09 | A kind of Route Dependence digital picture Image Matching accelerated based on GPU |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110414573A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113052825A (en) * | 2021-03-29 | 2021-06-29 | 华南理工大学 | Real-time three-dimensional deformation measurement method based on GPU parallel acceleration |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102221341A (en) * | 2011-03-16 | 2011-10-19 | 中国人民解放军国防科学技术大学 | Quick digital image correlation measurement method based on stochastic parallel gradient descent optimization technology |
CN103542815A (en) * | 2013-09-23 | 2014-01-29 | 西安新拓三维光测科技有限公司 | Large size speckle full-field strain measurement method |
CN103729849A (en) * | 2013-12-31 | 2014-04-16 | 南京航空航天大学 | Method for calculating digital image morphing initial value |
CN106485267A (en) * | 2016-09-24 | 2017-03-08 | 上海大学 | The measuring method of material low-cycle fatigue strain amplitude under hot environment |
-
2019
- 2019-07-09 CN CN201910615683.4A patent/CN110414573A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102221341A (en) * | 2011-03-16 | 2011-10-19 | 中国人民解放军国防科学技术大学 | Quick digital image correlation measurement method based on stochastic parallel gradient descent optimization technology |
CN103542815A (en) * | 2013-09-23 | 2014-01-29 | 西安新拓三维光测科技有限公司 | Large size speckle full-field strain measurement method |
CN103729849A (en) * | 2013-12-31 | 2014-04-16 | 南京航空航天大学 | Method for calculating digital image morphing initial value |
CN106485267A (en) * | 2016-09-24 | 2017-03-08 | 上海大学 | The measuring method of material low-cycle fatigue strain amplitude under hot environment |
Non-Patent Citations (3)
Title |
---|
席涛: "二维数字图像相关加速方法与GPU高速实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
肖颖: "基于双目视觉的DIC测量技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
邵新星 等: "数字图像相关方法若干关键问题研究进展", 《实验力学》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113052825A (en) * | 2021-03-29 | 2021-06-29 | 华南理工大学 | Real-time three-dimensional deformation measurement method based on GPU parallel acceleration |
CN113052825B (en) * | 2021-03-29 | 2023-04-21 | 华南理工大学 | Real-time three-dimensional deformation measurement method based on GPU parallel acceleration |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104200236B (en) | Fast target detection method based on DPM | |
TWI520078B (en) | Optical flow tracking method and device | |
CN111160298B (en) | Robot and pose estimation method and device thereof | |
CN109035327B (en) | Panoramic camera attitude estimation method based on deep learning | |
CN108540794B (en) | Method, device and equipment for detecting layout parameters of display equipment and storage medium | |
CN110570449A (en) | positioning and mapping method based on millimeter wave radar and visual SLAM | |
KR20130084849A (en) | Method and apparatus for camera tracking | |
CN105793892A (en) | Image processing method and apparatus and photographing device | |
CN105631939B (en) | A kind of three-dimensional point cloud distortion correction method and its system based on curvature filtering | |
CN103258328A (en) | Method for locating distorted center of wide-field lens | |
CN103366375A (en) | Method for registering image set based on dynamic digraphs | |
CN109584157A (en) | Object plane degree measurement method and device, storage medium, electronic metering equipment | |
CN105182378A (en) | LLL (Lenstra-Lenstra-LovaszLattice) ambiguity decorrelation algorithm | |
KR20150114950A (en) | Increasing frame rate of an image stream | |
CN106709870A (en) | Close-range image straight-line segment matching method | |
CN110414573A (en) | A kind of Route Dependence digital picture Image Matching accelerated based on GPU | |
CN107133986A (en) | A kind of camera calibration method based on two-dimensional calibrations thing | |
TWI462056B (en) | Image processing method, apparatus, and computer program product | |
CN110686610B (en) | Optical deformation measuring method based on self-adaptive grid and electronic equipment | |
CN108711415A (en) | Correct the method, apparatus and storage medium of the time delay between accompaniment and dry sound | |
CN103424087A (en) | Three-dimensional measurement splicing system and method for large-scale steel plate | |
CN109215118A (en) | A kind of increment type motion structure recovery and optimization method based on image sequence | |
CN107977980A (en) | A kind of method for tracking target, equipment and computer-readable recording medium | |
CN108495138A (en) | A kind of integer pixel motion estimation method based on GPU | |
CN109859313B (en) | 3D point cloud data acquisition method and device, and 3D data generation method and system |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191105 |