CN113095323A - SIFT-improvement-based digital image correlation method real-time detection method - Google Patents
SIFT-improvement-based digital image correlation method real-time detection method Download PDFInfo
- Publication number
- CN113095323A CN113095323A CN202110438916.5A CN202110438916A CN113095323A CN 113095323 A CN113095323 A CN 113095323A CN 202110438916 A CN202110438916 A CN 202110438916A CN 113095323 A CN113095323 A CN 113095323A
- Authority
- CN
- China
- Prior art keywords
- formula
- sub
- image
- gaussian
- scale
- 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.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 88
- 238000011897 real-time detection Methods 0.000 title claims abstract description 11
- 238000004364 calculation method Methods 0.000 claims abstract description 44
- 238000006073 displacement reaction Methods 0.000 claims abstract description 39
- 238000004458 analytical method Methods 0.000 claims abstract description 16
- 230000006872 improvement Effects 0.000 claims abstract description 13
- 239000013598 vector Substances 0.000 claims abstract description 7
- 238000012216 screening Methods 0.000 claims abstract description 6
- 230000008859 change Effects 0.000 claims description 29
- 230000000875 corresponding effect Effects 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 15
- 230000009466 transformation Effects 0.000 claims description 9
- 230000001276 controlling effect Effects 0.000 claims description 5
- 238000005314 correlation function Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000002035 prolonged effect Effects 0.000 claims description 3
- 230000002596 correlated effect Effects 0.000 claims description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims 1
- 238000001514 detection method Methods 0.000 abstract description 6
- 239000000463 material Substances 0.000 description 17
- 238000005259 measurement Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 238000005452 bending Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 239000004567 concrete Substances 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012886 linear function Methods 0.000 description 2
- 238000000691 measurement method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000010008 shearing Methods 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000007257 malfunction Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a digital image correlation method real-time detection method based on SIFT improvement, and relates to a deformation detection method; it comprises the following steps: firstly, acquiring gray level images before and after deformation; secondly, calculating sub-pixel surface fitting; thirdly, calculating sub-pixel interpolation; fourthly, establishing different Gaussian scale spaces to construct a continuous scale image pyramid; fifthly, searching and screening feature points from the continuous scale space; sixthly, constructing an analysis sub-area by combining the selected characteristic points; seventhly, calculating displacement of the deformation subarea; and eighthly, representing deformation by using the displacement feature vector. The method ensures the continuity of the displacement field, ensures that the adjustment of the calculation speed and the precision are simpler and more convenient to balance between two parameters, does not increase the calculation complexity of the improved algorithm, saves the pretreatment time, has better adaptability to complex boundaries, and can achieve higher calculation precision under the condition of the same calculation speed.
Description
Technical Field
The invention relates to a deformation detection method, in particular to a digital image correlation method real-time detection method based on SIFT improvement.
Background
Civil engineering materials such as sand, concrete, rock, and the like have been made by mixing a plurality of materials. These different materials are prone to form rough and uneven surfaces, and these surfaces have certain differences in light reflection effect, and these differences have the characteristics of randomness, non-uniformity and the like, so that capturing these differences can be used as a feature for describing the material surface in the digital image correlation method.
When the deformation monitoring method of the geotechnical material is used, a sensor method, an ultrasonic method and the like are generally used, and the sensor method is more common. A plurality of strain gauges are adhered or embedded on the surfaces or the interiors of stressed test pieces such as buildings and experimental materials, the voltage change of the strain gauges is monitored in real time and converted into the deformation condition, and the deformation detection of the buildings can be realized. However, this method has an effect on the strength or deformation state of the material itself, and is prone to malfunction and cumbersome to deploy.
Deformation detection of geotechnical materials or structures is widely used in laboratories or engineering, and resistance strain measurement is a traditional method in the field of material deformation monitoring. The strain gauge is adhered to the surface of a measured member, when the strain gauge deforms under stress, the strain gauge deforms along with the member, the resistance value of the strain gauge changes correspondingly, the resistance in the strain gauge is measured through the resistance strain gauge and converted into a strain value, and the strain measuring instrument mainly comprises a linear displacement sensor, an extensometer and the like. Because the strain gauge commonly used for measuring strain needs to be arranged complicatedly in advance, and the measured average strain in a certain area is usually, the strain gauge is not suitable for certain occasions which cannot be in direct contact.
The digital image correlation method is an optical deformation detection method, and the basic principle is that deformation images before and after deformation are collected and analyzed to obtain the deformation condition of the surface of a test piece, so that the defect of physical influence on an object to be detected by a resistance method, a sensor method and the like is overcome; meanwhile, the optical measurement method is used for deformation detection, so that the acquisition of the analysis material and the analysis process are more convenient and quicker. The SIFT algorithm is a feature matching algorithm based on scale transformation, and is used for constructing a plurality of continuous scale spaces after smoothing images through a Gaussian filter, extracting scale invariant feature points with extremely small variation in the continuous scales, wherein the features have rotation and scale invariance and are commonly used in computer vision feature matching. Meanwhile, some scholars at home and abroad improve the algorithm in the application direction of the algorithm applied to feature extraction.
Disclosure of Invention
The invention aims to provide a method for detecting the digital image correlation method in real time based on SIFT improvement by applying a computer vision correlation technology aiming at the defects and shortcomings of the prior art, and improves part of analysis methods and processes in the method on the basis of the technical scheme of analyzing material deformation by the traditional digital image correlation method, thereby greatly improving the analysis speed of the digital image correlation method while ensuring high precision and simultaneously not increasing the complexity of pretreatment and post-treatment during analysis.
In order to achieve the purpose, the invention adopts the technical scheme that: it comprises the following steps:
firstly, acquiring gray level images before and after deformation;
secondly, calculating sub-pixel surface fitting;
thirdly, calculating sub-pixel interpolation;
fourthly, establishing different Gaussian scale spaces to construct a continuous scale image pyramid;
fifthly, searching and screening feature points from the continuous scale space to control the calculation speed;
sixthly, constructing an analysis sub-area by combining the selected characteristic points;
seventhly, calculating displacement of the deformation subarea;
and eighthly, representing deformation by using the displacement feature vector.
Further, a specific method for acquiring the gray level images before and after the deformation in the first step is as follows: the CDD camera can directly capture a gray-scale image, or convert an input RGB color image through a computer to acquire a gray-scale image of a sub-area, wherein the gray-scale formula of the color image is shown as formula (1);
In the formula (1)
Is the intensity of the red color of the color image,
is the green color intensity of the color image,
the blue color of the color image is strong.
Further, the method for calculating the sub-pixel surface fitting in the step two is as follows: constructing a surface fitting function expression, and calculating coefficients in the surface fitting formula by using a Newton iteration method, wherein the surface fitting expression is shown as a formula (2):
In the formula (2), the first and second groups,
a surface function obtained by surface fitting is obtained,
respectively the distances from the center point of the curved surface;
respectively, the parameters are the generation parameters of the surface fitting.
Further, the method for calculating the sub-pixel interpolation in step three is as follows: interpolating according to the fitted shape function mapping relation, and inserting sub-pixel points between the positions of the integer pixel points; substituting the corresponding decimal position of the sub-pixel into the fitted shape function to approximately estimate the gray value corresponding to the pixel point of the sub-pixel precision difference; after interpolation, a smoother gray scale image with sub-pixels as index values can be obtained.
Further, the method for establishing the continuous scale image pyramid by using different gaussian scale spaces in the fourth step is as follows: carrying out Gaussian blur of different scales on the obtained multiple Gaussian spline curves in a sub-region; the calculation formula of a single gaussian scale template is as follows:
Wherein,
is a template of a gaussian convolution,
is the gray-scale image before the convolution,
for the convolved gray scale image, e is the base of the natural logarithm,
respectively the horizontal distance and the numerical distance from the point on the selected template to the center point,
for the spatial scale size of the Gaussian template, the size of the Gaussian convolution template is selected by the variance of the specified Gaussian convolution kernel
Determining;
constructing a top-down Gaussian according to the change of Gaussian convolution under different scales of the sub-regionVariance of convolution kernel
An ever-increasing digital pyramid, as shown in FIG. 3; and the number n of layers of the digital image pyramid is calculated by formula (5) and formula (6):
Wherein: m, N is the size of the length and width of the sub-region, i.e. how many multiples of pixels.
Further, the specific method for searching and screening the feature points from the continuous scale space in the step five to control the calculation speed is as follows: after a Gaussian image pyramid is constructed by establishing Gaussian continuous scales, differentiating Gaussian pyramids of adjacent scales to obtain a Gaussian differential matrix; the point with small gray value change under different scales is the scale-invariant feature point, the absolute value of the Gaussian difference matrix is calculated to search the minimum value, and the minimum value is used as a temporary feature point; performing surface fitting around the extreme points, and constructing a gradient change matrix, i.e. a hessian matrix, which represents the change of the gray values of the characteristic points in two directions, as shown in a formula (7):
In the formula (7), the first and second groups,
is the partial derivative of the gradient of the change of the gray scale in the x direction with respect to the x direction,
for gradient of variation of grey scale in x-direction versus y-directionA partial derivative; the Hessian matrix has two corresponding eigenvalues
Ratio of these two characteristic values
The main curvature is called, the distribution density of the selected characteristic points is controlled by controlling the size of the main curvature threshold, the distribution density directly determines the calculation speed, and meanwhile, higher calculation accuracy can be kept.
Further, the specific method for constructing the analysis sub-region by combining the selected feature points in the sixth step is as follows: constructing an image sub-region by taking each selected feature point to be calculated as a center and 18-40 pixels as side lengths; constructing a sub-region on a reference image, wherein the gray change characteristic of the sub-region is used as the speckle characteristic of the image; meanwhile, image sub-regions with the same size are constructed at corresponding positions in the deformed image, and finally the corresponding sub-regions in the two images are matched by calculating a correlation coefficient peak value; the gradient descent method is used for searching the sub-area with the maximum correlation coefficient, so that the searching time can be effectively prolonged;
calculating the similarity of speckles in the region of interest by using a correlation function in probability statistics, wherein a correlation coefficient solving formula is shown as a formula (8):
In the formula (8)
Represents a correlation coefficient;
representing the gray value size of the position;
representing the gray value of the next point position; wherein when
A value of 1 indicates that the speckle is linearly correlated.
Further, in the seventh step, the deformation subarea displacement calculation method is as follows:
assuming any two points on the sub-area before deformation
The coordinates are respectively
，
For the deformed subareas the same two points
The coordinates are respectively
，
(ii) a The correlation calculation formula is as follows:
Apoint:
formula (9)
BPoint:
formula (10)
And A, B there are two points in the following relationship:
According to the theory of continuous mechanical medium, when
When the magnitude transformation of (2) is close to 0, the solution relationship of the displacement of the point B can be approximately expressed by the relationship between the displacement of the point A and the first derivative of the displacement of the point A as follows:
Will be provided with
Substituting into the formula to obtain formula (13)
Wherein,
horizontal displacement and vertical displacement of the deformation field.
Further, in the step eight, an interpolation method of non-characteristic point displacement is adopted when the displacement characteristic vector is used for representing deformation, and the difference value of the non-characteristic point is interpolated through displacement change obtained by the calculated characteristic points.
After the scheme is adopted, the invention has the beneficial effects that: the invention relates to a method for detecting a digital image correlation method based on SIFT improvement in real time, which has the following advantages:
1. the method has the advantages of breakthrough and efficiency improvement: the stress of the test piece is analyzed, a large number of strain gauges are arranged, a large number of preprocessing and post-processing processes are performed, and meanwhile, the improved analysis process greatly reduces the overall calculated amount, so that the calculation precision can be greatly improved, and the purpose of real-time monitoring is achieved;
2. little external influence, stability is high: the SIFT feature is the basic feature of an image, which keeps invariance to rotation, scale scaling and brightness change and also keeps certain stability to view angle change, affine transformation and noise; therefore, the influence of external environments such as illumination on the surface characteristics of the test piece can be ignored within a certain range, and the whole analysis scheme has high environmental adaptability and calculation stability;
3. real-time detection, the influence to the object that awaits measuring is little: under the condition that the precision of the 400-300 pixel picture is 90%, the required time is only 1.5 seconds, and the calculation speed can be greatly increased to realize real-time measurement on the premise of ensuring higher precision; meanwhile, the surface of the test piece is not required to be changed in the whole analysis process, so that the physical influence on the test piece can be reduced;
4. the applicability is strong, the acceptance is big: the calculation speed and the precision are adjusted simply and conveniently, so that the relation between the speed and the precision can be balanced better; meanwhile, the method has strong expandability and can be conveniently combined with the characteristic vectors in other forms.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a graph showing the variation of the distortion coincidence rate and the accuracy of the present invention;
FIG. 2 is a flow chart of the algorithm improvement of the present invention;
FIG. 3 is a schematic block flow diagram of the present invention;
FIG. 4 is a digital pyramid constructed under different Gaussian templates;
FIG. 5 is an image before the embodiment is deformed;
FIG. 6 is an image after a modification of the embodiment;
FIG. 7 is a graph of program accuracy versus speed for the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings.
Example (b): a method for real-time detection based on SIFT improved digital image correlation method, as shown in fig. 2-4, comprising the following steps:
firstly, obtaining gray level images before and after deformation:
the CDD camera can directly capture a gray-scale image, or convert an input RGB color image through a computer to acquire a gray-scale image of a sub-area, wherein the gray-scale formula of the color image is shown as formula (1);
In the formula (1)
Is the intensity of the red color of the color image,
is the green color intensity of the color image,
is the blue color intensity of the color image;
secondly, calculating sub-pixel surface fitting:
constructing a surface fitting function expression, and calculating coefficients in the surface fitting formula by using a Newton iteration method, wherein the surface fitting expression is shown as a formula (2):
In the formula (2), the first and second groups,
a surface function obtained by surface fitting is obtained,
respectively the distances from the center point of the curved surface;
respectively calculating parameters for curved surface fitting;
thirdly, calculating sub-pixel interpolation:
interpolating according to the fitted shape function mapping relation, and inserting sub-pixel points between the positions of the integer pixel points; substituting the corresponding decimal position of the sub-pixel into the fitted shape function to approximately estimate the gray value corresponding to the pixel point of the sub-pixel precision difference; after interpolation, a smoother gray image with sub-pixels as index values can be obtained;
fourthly, establishing a continuous scale image pyramid with different Gaussian scale space structures:
carrying out Gaussian blur of different scales on the obtained multiple Gaussian spline curves in a sub-region; the calculation formula of a single gaussian scale template is as follows:
Wherein,
is a template of a gaussian convolution,
is the gray-scale image before the convolution,
for the convolved gray scale image, e is the base of the natural logarithm,
respectively the horizontal distance and the numerical distance from the point on the selected template to the center point,
of Gaussian formThe size of the spatial scale, the size of the Gaussian convolution template, the variance of the specified Gaussian convolution kernel
Determining;
the sub-region template size may be selected to be
；
Constructing a top-down Gaussian convolution kernel variance according to the Gaussian convolution change of the subareas under different scales
An ever-increasing digital pyramid, as shown in FIG. 3; and the number n of layers of the digital image pyramid is calculated by formula (5) and formula (6):
Wherein: m, N is the length and width of the sub-region, namely, how many pixel multiples;
according to the method, a continuous Gaussian scale space is constructed based on basic scales according to a Gaussian continuous scale transformation principle, and a Gaussian continuous scale image pyramid is established; according to the continuous scale space obtained by the Gaussian pyramid, carrying out image difference of adjacent scales, and finding out scale invariant feature transformation feature points under the continuous scale; the reference dimension of the continuous dimension is
The proportionality coefficients of successive gaussian scales are:
；
fifthly, searching and screening feature points from the continuous scale space:
after a Gaussian image pyramid is constructed by establishing Gaussian continuous scales, differentiating Gaussian pyramids of adjacent scales to obtain a Gaussian differential matrix; the point with small gray value change under different scales is the scale-invariant feature point, so the minimum value is searched for by calculating the absolute value of the Gaussian difference matrix, and the minimum value is used as the temporary feature point; generally, the method directly calculates and searches the obtained minimum value points, and can correspond to almost all contour features, so that a large number of pseudo feature points can be obtained; therefore, surface fitting is performed around the extreme points, and a gradient change matrix, i.e., a hessian matrix, which represents changes of gray values of the feature points in two directions is constructed, as shown in formula (7):
In the formula (7), the first and second groups,
is the partial derivative of the gradient of the change of the gray scale in the x direction with respect to the x direction,
is the partial derivative of the gradient of the change of the gray scale in the x direction to the y direction; the Hessian matrix has two corresponding eigenvalues
Ratio of these two characteristic values
The main curvature is called, so that the distribution density of the selected feature points can be controlled by controlling the size of a main curvature threshold; the larger the main curvature threshold value is selected, the smaller the number of the obtained characteristic points is, but the more obvious the characteristic position represented by the characteristic points is as a structure contour; if the density or number of the selected feature points is too small, random noise can be added to increase and fill the feature pointsThe density of the points to be calculated is used for ensuring the calculation precision;
sixthly, constructing an analysis sub-area by combining the selected characteristic points:
constructing an image sub-region by taking each selected feature point to be calculated as a center and 18-40 pixels as side lengths; constructing a sub-region on a reference image, wherein the gray change characteristic of the sub-region is used as the speckle characteristic of the image; meanwhile, image sub-regions with the same size are constructed at corresponding positions in the deformed image, and finally the corresponding sub-regions in the two images are matched by calculating a correlation coefficient peak value; the gradient descent method is used for searching the sub-area with the maximum correlation coefficient, so that the searching time can be effectively prolonged;
calculating the similarity of speckles in the region of interest by using a correlation function in probability statistics, wherein a correlation coefficient solving formula is shown as a formula (8):
In the formula (8)
Represents a correlation coefficient;
representing the gray value size of the position;
representing the gray value of the next point position; wherein when
When the number is 1, the speckles are linearly related;
seventhly, calculating displacement of the deformation subarea:
assuming any two points on the sub-area before deformation
The coordinates are respectively
，
For the deformed subareas the same two points
The coordinates are respectively
，
(ii) a The correlation calculation formula is as follows:
Apoint:
formula (9)
BPoint:
formula (10)
And A, B there are two points in the following relationship:
According to the theory of continuous mechanical medium, when
When the magnitude transformation of (2) is close to 0, the solution relationship of the displacement of the point B can be approximately expressed by the relationship between the displacement of the point A and the first derivative of the displacement of the point A as follows:
Will be provided with
Substituting into the formula to obtain formula (13)
Wherein,
horizontal displacement and vertical displacement of the deformation field;
the displacement change images before and after the experiment of this example are shown in FIGS. 5-6;
eighthly, representing deformation by using the displacement feature vector: interpolation method of non-characteristic point displacement:
the difference value of the non-feature points can be calculated through displacement change interpolation obtained by the calculated feature points; because concrete, soil body materials and the like can be approximately regarded as continuous mechanical media when micro deformation occurs; therefore, the calculated deformation state of the device is continuous within a certain range, and the deformation change trend in a section of area can be obtained through function fitting;
according to the relevant theory of material mechanics, when the bending, bending and shearing are in five stress states, only the deformation equation of the bending is a high-order function, and the other functions are linear functions, so that the deformation condition can be estimated uniformly by using the linear function; meanwhile, the distance between the characteristic points to be calculated cannot be too large and can be smaller than the side length of the selected calculation sub-area; therefore, linear interpolation can be carried out between the deformation values obtained by calculation in a small range to estimate the deformation of other points which are not calculated.
In the embodiment, a digital speckle correlation method is improved by applying a scale invariant feature transformation algorithm, and a pretreatment process for manufacturing speckles on the surface of a material can be omitted for civil engineering materials; meanwhile, the method of replacing full-field measurement by the characteristic points can greatly reduce the number of calculation so as to improve the calculation speed; because the deformation characteristic of the geotechnical material has continuity, the actual deformation of other non-characteristic positions can be accurately estimated by interpolating the calculation result of the characteristic points, so that higher precision can be kept; in addition, the invention realizes the sub-pixel measurement technology and greatly improves the measurement precision of the digital image correlation method.
Because the basic unit of the image is a pixel, a sub-pixel image processing method is needed to realize the measurement of sub-pixel precision; common methods include a surface fitting method, a correlation coefficient method, and the like; according to the characteristic that the gray scale of the test piece is continuous, the method applies and constructs a two-surface fitting function to perform surface fitting according to the distribution condition of the pixel gray scale and then performs interpolation calculation on the gray scale of the sub-pixel position; the accuracy choice of the sub-pixels can be determined in combination with the accuracy required for the actual deformation measurement to be measured; therefore, before the deformation calculation is executed, the image interpolation pretreatment can be realized by using a method of fitting interpolation by using a binary second-order curved surface function;
in the embodiment, a scale invariant feature transformation algorithm is adopted to search points, of which the gray values hardly change in different Gaussian scale spaces, as scale invariant feature points by constructing continuous Gaussian scale spaces, so as to replace the traditional measurement method of a digital image correlation method; the number and density of the feature points obtained after the scale invariant feature conversion can be controlled by controlling a curvature threshold, so that the speed and precision required by the algorithm are balanced and adjusted conveniently according to different requirements in the whole analysis method;
according to the principle of deformation continuity of geotechnical materials, for common deformation forms such as stretching, compression, bending, torsion and shearing, the displacement of a sub-region corresponding to an uncalculated pixel point between two adjacent key points is calculated. In the embodiment, the linear interpolation result of the pixel point between the two calculated adjacent key points can be used for estimating the deformation of other non-calculated points; the calculation results of the method and the full-field calculation method are compared to better estimate the full-field deformation condition, and the estimation method can simply achieve the precision of more than 90% through experimental analysis.
The embodiment achieves the effects of controllable precision and high speed by adjusting the proportion number of the characteristic points. The improved algorithm has the advantages that because the feature points are directly calculated, the calculated displacement of each pixel point cannot be completely the same as that of the traditional method; assuming that the deformation characteristics of each pixel point calculated by the original digital image correlation method are accurate, the deformation condition of each pixel point obtained by the improved algorithm can be compared with the deformation condition of each pixel point obtained by the original calculation method, wherein the deformation conditions are the same as or different from each other, so that the coincidence rate of the pixel deformation of the two sets of calculation methods is obtained.
FIG. 1 shows the variation of the deformed coincidence rate with the calculated density of the feature points; in the embodiment, the calculation accuracy of the improved digital image correlation method is evaluated by using the deformed coincidence rate, and the purpose of reducing the calculation amount is achieved by reducing the amount of calculation points, so that the calculation speed of a computer is improved; the proportion of the number of the characteristic points in the whole area is small, and the final precision is controlled according to the proportion of the regulating characteristic points.
Meanwhile, under different scale spaces, characteristic points are generally obtained by searching in an extreme point mode; the gradient change relation between the searched extreme point and the surrounding points is called curvature, the feature points can be classified into extreme points with different degrees according to the sizes of the curvature and the gyration radius, and therefore the grade and the density of the feature points required to be calculated can be controlled by controlling the size and the range of the selected curvature expression.
Compared with the prior art, the invention has the beneficial effects that: the method has the advantages of ensuring the continuity of the displacement field, along with high accuracy, simplicity and convenience in operation, saving of pretreatment time, better adaptability to complex boundaries and higher calculation accuracy.
The relationship between accuracy and speed is shown in FIG. 7; in fig. 7, N is a speed multiple, the abscissa is accuracy, the calculation speed can be increased by 4-6 times under the condition that the deformation coincidence rate of the calculation result and the conventional calculation result is 99.5%, the calculation speed can be increased by 12 times under the condition that the deformation coincidence rate is 90%, and the speed must be increased if the accuracy of the result obtained by the improved algorithm is to be increased.
The above description is only for the purpose of illustrating the technical solutions of the present invention and not for the purpose of limiting the same, and other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention should be covered within the scope of the claims of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims (9)
1. A digital image correlation method real-time detection method based on SIFT improvement is characterized by comprising the following steps:
firstly, acquiring gray level images before and after deformation;
secondly, calculating sub-pixel surface fitting;
thirdly, calculating sub-pixel interpolation;
fourthly, establishing different Gaussian scale spaces to construct a continuous scale image pyramid;
fifthly, searching and screening feature points from the continuous scale space to control the calculation speed;
sixthly, constructing an analysis sub-area by combining the selected characteristic points;
seventhly, calculating displacement of the deformation subarea;
and eighthly, representing deformation by using the displacement feature vector.
2. The method for detecting the digital image correlation method based on the SIFT improvement in real time as claimed in claim 1, wherein the specific method for acquiring the gray image before and after the deformation in the first step is as follows: the CDD camera can directly capture a gray-scale image, or convert an input RGB color image through a computer to acquire a gray-scale image of a sub-area, wherein the gray-scale formula of the color image is shown as formula (1);
In the formula (1)
Is the intensity of the red color of the color image,
is the green color intensity of the color image,
the blue color of the color image is strong.
3. The method for real-time detection of digital image correlation based on SIFT improvement as claimed in claim 1, wherein the method for calculating the sub-pixel surface fitting in the second step is as follows: constructing a surface fitting function expression, and calculating coefficients in the surface fitting formula by using a Newton iteration method, wherein the surface fitting expression is shown as a formula (2):
In the formula (2), the first and second groups,
a surface function obtained by surface fitting is obtained,
respectively the distances from the center point of the curved surface;
respectively, the parameters are the generation parameters of the surface fitting.
4. The method for real-time detection of digital image correlation based on SIFT improvement as claimed in claim 1, wherein the method for calculating sub-pixel interpolation in the third step is as follows: interpolating according to the fitted shape function mapping relation, and inserting sub-pixel points between the positions of the integer pixel points; substituting the corresponding decimal position of the sub-pixel into the fitted shape function to approximately estimate the gray value corresponding to the pixel point of the sub-pixel precision difference; after interpolation, a smoother gray scale image with sub-pixels as index values can be obtained.
5. The method for detecting the digital image correlation method based on the SIFT improvement in real time as claimed in claim 1, wherein the method for establishing the continuous scale image pyramid constructed by different Gaussian scale spaces in the fourth step is as follows: carrying out Gaussian blur of different scales on the obtained multiple Gaussian spline curves in a sub-region; the calculation formula of a single gaussian scale template is as follows:
Wherein,
is a template of a gaussian convolution,
is the gray-scale image before the convolution,
for the convolved gray scale image, e is the base of the natural logarithm,
respectively the horizontal distance and the numerical distance from the point on the selected template to the center point,
for the spatial scale size of the Gaussian template, the size of the Gaussian convolution template is selected by the variance of the specified Gaussian convolution kernel
Determining;
constructing a top-down Gaussian convolution kernel variance according to the Gaussian convolution change of the subareas under different scales
An ever-increasing digital pyramid, as shown in FIG. 3; and the number n of layers of the digital image pyramid is calculated by formula (5) and formula (6):
Wherein: m, N is the size of the length and width of the sub-region, i.e. how many multiples of pixels.
6. The method as claimed in claim 1, wherein the method for detecting the digital image correlation method based on the SIFT improvement in real time is characterized in that the specific method for searching and screening feature points from the continuous scale space in the step five to control the calculation speed is as follows: after a Gaussian image pyramid is constructed by establishing Gaussian continuous scales, differentiating Gaussian pyramids of adjacent scales to obtain a Gaussian differential matrix; the point with small gray value change under different scales is the scale-invariant feature point, the absolute value of the Gaussian difference matrix is calculated to search the minimum value, and the minimum value is used as a temporary feature point; performing surface fitting around the extreme points, and constructing a gradient change matrix, i.e. a hessian matrix, which represents the change of the gray values of the characteristic points in two directions, as shown in a formula (7):
In the formula (7), the first and second groups,
is the partial derivative of the gradient of the change of the gray scale in the x direction with respect to the x direction,
is the partial derivative of the gradient of the change of the gray scale in the x direction to the y direction; the Hessian matrix has two corresponding eigenvalues
Ratio of these two characteristic values
The main curvature is called, the distribution density of the selected characteristic points is controlled by controlling the size of the main curvature threshold, the distribution density directly determines the calculation speed, and meanwhile, higher calculation accuracy can be kept.
7. The method for detecting the relevance of the digital image based on the SIFT improvement in real time as claimed in claim 1, wherein the specific method for constructing the analysis sub-region by combining the selected feature points in the sixth step is as follows: constructing an image sub-region by taking each selected feature point to be calculated as a center and 18-40 pixels as side lengths; constructing a sub-region on a reference image, wherein the gray change characteristic of the sub-region is used as the speckle characteristic of the image; meanwhile, image sub-regions with the same size are constructed at corresponding positions in the deformed image, and finally the corresponding sub-regions in the two images are matched by calculating a correlation coefficient peak value; the gradient descent method is used for searching the sub-area with the maximum correlation coefficient, so that the searching time can be effectively prolonged;
calculating the similarity of speckles in the region of interest by using a correlation function in probability statistics, wherein a correlation coefficient solving formula is shown as a formula (8):
In the formula (8)
Represents a correlation coefficient;
representing the gray value size of the position;
representing the gray value of the next point position; wherein when
A value of 1 indicates that the speckle is linearly correlated.
8. The method for real-time detection of the SIFT-based improved digital image correlation method according to claim 1, wherein the method for calculating the displacement of the deformation subarea in the seventh step comprises the following steps:
assuming any two points on the sub-area before deformation
The coordinates are respectively
，
For the deformed subareas the same two points
The coordinates are respectively
，
(ii) a The correlation calculation formula is as follows:
Apoint:
formula (9)
BPoint:
formula (10)
And A, B there are two points in the following relationship:
According to the theory of continuous mechanical medium, when
When the magnitude transformation of (2) is close to 0, the solution relationship of the displacement of the point B can be approximately expressed by the relationship between the displacement of the point A and the first derivative of the displacement of the point A as follows:
Will be provided with
Substituting into the formula to obtain formula (13)
Wherein,
for horizontal displacement of deformation fieldAnd vertical displacement.
9. The method as claimed in claim 1, wherein the SIFT-based improved digital image correlation method is used for real-time detection, and the interpolation method using non-feature point displacement is used in the step eight when the displacement feature vector is used to represent the deformation, and the difference value of the non-feature point is interpolated by the displacement change obtained from the calculated feature points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110438916.5A CN113095323A (en) | 2021-04-23 | 2021-04-23 | SIFT-improvement-based digital image correlation method real-time detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110438916.5A CN113095323A (en) | 2021-04-23 | 2021-04-23 | SIFT-improvement-based digital image correlation method real-time detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113095323A true CN113095323A (en) | 2021-07-09 |
Family
ID=76679613
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110438916.5A Withdrawn CN113095323A (en) | 2021-04-23 | 2021-04-23 | SIFT-improvement-based digital image correlation method real-time detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113095323A (en) |
Cited By (2)
| 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 |
| CN117329977A (en) * | 2023-11-28 | 2024-01-02 | 中国飞机强度研究所 | Visual characteristic characterization and measurement processing method for structural fatigue crack under complex working condition |
-
2021
- 2021-04-23 CN CN202110438916.5A patent/CN113095323A/en not_active Withdrawn
Cited By (3)
| 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 |
| CN117329977A (en) * | 2023-11-28 | 2024-01-02 | 中国飞机强度研究所 | Visual characteristic characterization and measurement processing method for structural fatigue crack under complex working condition |
| CN117329977B (en) * | 2023-11-28 | 2024-02-13 | 中国飞机强度研究所 | Visual characteristic characterization and measurement processing method for structural fatigue crack under complex working condition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105139412B (en) | A kind of high spectrum image angular-point detection method and system | |
| CN109883533B (en) | Low-frequency vibration measurement method based on machine vision | |
| CN101539629B (en) | Remote sensing image change detection method based on multi-feature evidence integration and structure similarity | |
| CN105574534A (en) | Significant object detection method based on sparse subspace clustering and low-order expression | |
| CN107328502B (en) | Anchor rod tray load visualization digital imaging method | |
| CN104657955B (en) | The displacement field iteration smoothing method of Digital Image Correlation Method based on kernel function | |
| CN113095323A (en) | SIFT-improvement-based digital image correlation method real-time detection method | |
| CN104657999B (en) | A kind of Digital Image Correlation Method based on kernel function | |
| CN111354033B (en) | Digital image measuring method based on feature matching | |
| CN108765476A (en) | Polarized image registration method | |
| CN102930576A (en) | Feature flow-based method for generating abstract line drawing | |
| CN115096206B (en) | High-precision part size measurement method based on machine vision | |
| CN110378924A (en) | Level set image segmentation method based on local entropy | |
| CN112801141B (en) | Heterogeneous image matching method based on template matching and twin neural network optimization | |
| CN117745697A (en) | Steel surface defect detection algorithm based on improved YOLOv8 model | |
| CN102510437B (en) | Method for detecting background of video image based on distribution of red, green and blue (RGB) components | |
| CN113807238A (en) | Visual measurement method for area of river surface floater | |
| CN113642397A (en) | Object length measuring method based on mobile phone video | |
| CN117372480A (en) | Point cloud registration method utilizing self-adaptive downsampling and curvature enhancement point-to-point characteristics | |
| CN116664531A (en) | Deep learning-based large deformation measurement method and system | |
| CN111126508A (en) | Hopc-based improved heterogeneous image matching method | |
| CN116051808A (en) | YOLOv 5-based lightweight part identification and positioning method | |
| CN111899284B (en) | Planar target tracking method based on parameterized ESM network | |
| Deshpande et al. | Super resolution based low cost vision system | |
| Cai et al. | Model-informed deep learning-based compressive sensing for resampling and recovering 2D SHM vision data |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20210709 |