CN105678757B - A kind of ohject displacement measuring method - Google Patents

Abstract

The invention discloses a method for measuring object displacement, comprising the following steps: 1. Selecting the first frame image P1 and any frame image M1 after the first frame; 2. Performing grayscale processing on the image; 3. Obtaining P2 and M2; 4. Frame a specific area in P1 as the template T1; 5. Obtain a matching area similar to T1; 6. Realize pixel-level matching; 7. Calculate the correlation coefficient of each pixel in the n*n area, and select the fitting The coordinates of the maximum value of the obtained surface are used as matching points; 8. Obtain the corresponding coordinate position in M2; 9. Use the template T1 to select the corresponding template T2 in P2; 10. Calculate the correlation coefficient of T2 at each point ; 11. Comparing the coordinates corresponding to the best matching point with the coordinates of the template in P1; 12. Repeating steps 4-11 to calculate the statistical average value of the displacement, using the statistical average value as the final displacement result. It has the advantages of high computational efficiency.

Description

A method for measuring object displacement

technical field

The invention relates to a sub-pixel displacement measurement technology of a digital image, in particular to an object displacement measurement method, which can be widely used in the measurement of displacement, deformation and strain of an object, and can also perform vibration analysis on the object.

Background technique

Optical Mechanics, which uses experiments as a means and uses optical measurement methods to study displacement, stress and strain as its main task, is an interdisciplinary subject that combines multiple disciplines and technologies. It has been widely used in many testing and inspection fields. application and play a pivotal role.

Digital image correlation method (Digital Image Correlation Method, DICM), as a photomechanical technology, has the advantages of full-field measurement, non-destructive, and low requirements for the measurement environment. developed rapidly. In recent years, many scholars have done a lot of research work on the theory and engineering application of digital image related technology, and achieved certain research results. With continuous improvement, its application fields are getting wider and wider, and it will play an increasingly important role in the engineering field.

Correlation calculation is a key issue in digital image correlation technology, and improving the measurement accuracy of digital image correlation method is an urgent requirement for non-destructive testing of engineering quality. It is expensive and unrealistic to improve the measurement accuracy by upgrading the hardware, but from the perspective of optimization algorithm, the idea of improving the image sub-pixel positioning accuracy is economically feasible. The algorithm of integer pixel search is very mature and perfect. Relatively speaking, the calculation time is relatively less, and the sub-pixel level positioning is the key to calculation accuracy, and it is also a time-consuming link. It directly affects the efficiency and calculation accuracy of related searches. and stability.

Most of the correlation search algorithms have the same integer pixel positioning accuracy, but there are some differences in the amount of calculation, calculation efficiency, anti-noise performance, stability, etc. Therefore, the main factor determining the calculation accuracy of digital image correlation methods is sub-pixel positioning. Accuracy, common methods include sub-pixel grayscale interpolation method, surface fitting method, coordinate rotation method, Newton-Raphson method, quasi-Newton method, gradient method, frequency domain correlation method, genetic algorithm, neural network algorithm, etc. The positioning accuracy that the algorithm can achieve ranges from 0.005 to 0.1 pixels.

The grayscale interpolation method requires sub-pixel reconstruction of the discrete grayscale field through interpolation. The simplest grayscale interpolation method is the nearest neighbor interpolation method and the bilinear interpolation method. These two interpolation methods have very low precision. Interpolation reconstruction produces blur. Interpolation methods with relatively high precision include Lagrangian interpolation, cubic interpolation, bicubic spline interpolation, and quintic spline interpolation. By interpolating the discrete gray-scale field, the digital image becomes an approximate continuous image, and then performs a fine search to select the position with the largest correlation coefficient as the optimal matching position. This method has a huge amount of calculation and low efficiency.

Contents of the invention

The object of the present invention is to overcome the shortcomings and deficiencies of the prior art, and provide a method for measuring object displacement, which meets the requirements of high-precision measurement in actual engineering and ensures the rationality of calculation efficiency.

The object of the present invention can be achieved through the following technical solutions: a method for measuring object displacement mainly comprises the following steps:

S1: Fix the position of the image acquisition device, and then use the image acquisition device to collect continuous moving images of the target to be measured; the position of the image acquisition device cannot be changed to obtain an image that reflects the movement of the target; select the first frame image (P1) and the second frame Any frame of image after one frame (M1);

S2: If the image is a color image, first perform grayscale processing on it;

S3: Use an interpolation algorithm to interpolate P1 and M1 by k times, and obtain P2 and M2 respectively after interpolation;

S4: frame a specific area containing obvious features in P1 as a template (T1), and record the coordinates (x ₀ , y ₀ ) of the upper left corner of the template map T1 in P1;

S5: Use the hill-climbing method to perform matching in M1, and obtain a matching area roughly similar to T1;

S6: In the area obtained in the previous step, use the SDA\SSD algorithm to accurately match the coordinates (x ₁ , y ₁ ) of the template image to achieve pixel-level matching;

S7: With (x ₁ , y ₁ ) as the center, calculate the correlation coefficient of each pixel in the n*n area, use the obtained correlation coefficient to perform surface fitting, and select the coordinates of the maximum value of the fitted surface as the matching point (x ₂ ,y ₂ );

S8: Map the matching coordinate position (x ₂ , y ₂ ) in M1 obtained in step S7 to M2 to obtain the corresponding coordinate position (x ₃ , y ₃ ) in M2;

S9: Utilize the template image (T1) selected in P1 to select its corresponding template image (T2) in P2;

S10: In M2, with (x ₃ , y ₃ ) as the starting point, repeat step S6 to find the coordinates of the best matching point, take the coordinates as the center, select a rectangular area of m*m, and calculate T2 at each point The correlation coefficient on

S11: Use the correlation coefficient obtained in the previous step to perform surface fitting, select the coordinate (x ₄ , y ₄ ) of the fitted maximum value as the best matching point, and map the coordinate value back to the corresponding coordinate value in M1 (x ₅ , y ₅ ), and compare (x ₅ , y ₅ ) with the coordinates (x ₀ , y ₀ ) of the template in P1, so as to achieve accurate displacement measurement (Δx, Δy) of the target;

S12: Perform the above step S4 to step S11 cyclically p times, select a different template map each time, calculate the statistical average value of the displacement, and use this value as the final displacement result.

The calculation formula adopted by the NCC algorithm for calculating the correlation coefficient in the above steps S5, S6, and S10 is as follows:

In the above formula, r _xy is the correlation coefficient between the m*n sub-region with the point (x, y) as the origin and the template image, and S _{x, y} represents the point (x, y) in the image to be matched. The m*n sub-area intercepted at the origin, S _{x, y} (i, j) refers to the gray value of the coordinate (i, j) point on the sub-area, Refers to the average value of the gray level on the sub-region. T _x,y (i,j) refers to the gray value of the coordinate (i,j) point on the template, Refers to the average value of the grayscale on the template. m and n represent the number of columns and rows of the template, respectively.

In the above steps S7 and S11, methods for surface fitting include quadratic surface fitting, cubic surface fitting, Gaussian surface fitting, and two-dimensional Lagrangian surface fitting; generally, quadratic surface fitting is used That's it. The binary quadratic function used in the quadratic surface fitting method is as follows:

r(x _i ,y _i )=a ₀ +a ₁ x _i +a ₂ y _i +a ₃ x ² +a ₄ x _i y _i +a ₅ y _i ² ,

In the formula, r( _xi , y _i ) represents the correlation coefficient calculated at the coordinates ( _xi , y _i ) of the template graph, and the coefficients a ₀ to a ₅ are the coefficients of the quadric surface. The coordinate formula for the maximum value of a quadric surface is as follows:

In the formula, x and y respectively represent the abscissa and ordinate of the maximum value of the quadric surface, and the coefficients a ₀ to a ₅ are the coefficients of the quadric surface.

The formula of the coordinate mapping in the above step S8 is as follows:

x′ ₃ =(x ₂ -1)*n+1

y′ ₃ =(y ₂ -1)*n+1,

In the formula, the coordinate (x ₂ , y ₂ ) represents the matching coordinate position of the template obtained in step S7 in M1, the coordinate (x' ₃ , y' ₃ ) represents the result of the coordinate mapping, and the coordinate position (x ₃ , y ₃ ) represents the corresponding coordinate position of the template in M2.

The specific method of the above step S9 is as follows: first map the coordinates (x ₀ , y ₀ ) of the upper left corner point of the template map T1 in P1 to the coordinates in P2, denoted as (x ₀ ', y ₀ '), and at the same time select the template The coordinates of the lower right corner of the graph T1 in P1, and map it to the coordinates in P2, recorded as (x ₀ ", y ₀ "), then the template T2 is taken as the coordinates (x ₀ ', y ₀ ') and the coordinates A rectangular area between (x ₀ ", y ₀ "). The coordinate mapping formula is as follows:

x ₂ =(x ₁ -1)*k+1

y ₂ =(y ₁ -1)*k+1,

Wherein, the coordinates (x ₁ , y ₁ ) and (x ₂ , y ₂ ) are the corresponding coordinates in P1 and P2 respectively, and k represents the multiple of the interpolation in step S3.

In the above step S11, the coordinate mapping formula is as follows:

In the formula, the coordinates (x ₄ , y ₄ ) represent the overall best matching coordinates of M2, and the coordinates (x ₅ , y ₅ ) represent the best matching coordinates in M1, which are obtained from the above mapping formula, and k represents the Multiplier for interpolation.

In the above step S11, the displacement calculation formula is as follows:

In the formula, the coordinates (x ₅ , y ₅ ) represent the best matching coordinates in M1, the coordinates (x ₀ , y ₀ ) represent the coordinates of the upper left corner point of the template graph T1 in step S4 in P1, and Δx represents the target in horizontal The precise displacement in the coordinate direction, Δy represents the precise displacement of the target in the ordinate direction.

The values of k, n, m, and p in steps S3, S7, S10, and S12 can be any positive integers, and the larger the value, the greater the computational complexity, and the value range is preferably [1,10].

In steps S3, S7, S8, S9, S10 and S11, the sub-pixel interpolation algorithm of classic correlation search and the surface fitting method are combined to greatly improve the accuracy of sub-pixel level displacement measurement; at the same time, before interpolation and interpolation Matching in the final image effectively reduces the computational complexity.

The purpose of the present invention can also be achieved through the following technical solutions: a method for measuring object displacement, comprising the following steps: (1) utilizing image acquisition equipment to collect continuous displacement images of the target object to be measured; (2) utilizing an interpolation algorithm to collect (3) Select the template image in the first frame, and then use the hill-climbing method and SDA\SSD algorithm to find the rough coordinate position of the template in the subsequent image; (4) Fit it to obtain the similarity surface, Find the peak coordinates of the surface, compare with the coordinates in the original image, and calculate the displacement; (5) Repeat steps (3) to (4) several times, and calculate the average value of the displacement as the final result. When the method of the present invention is used in engineering, the template image is selected multiple times for measurement, and the calculation is performed in the image before and after interpolation at the same time, and the interpolation algorithm and the surface fitting algorithm are skillfully combined to improve the efficiency of calculation and realize accurate sub-pixel displacement measurement.

Compared with the prior art, the present invention has the following advantages and effects:

The surface fitting method assumes that the correlation coefficient matrix of the integer pixel displacement correlation search results and its adjacent points can be fitted into a continuous surface, and then the extreme point position of the surface is used as the center position of the deformed image sub-region. The types of surfaces used for fitting include quadratic surfaces, cubic surfaces, Gaussian surfaces, and two-dimensional Lagrangian surfaces. The surface fitting method has high calculation accuracy and strong anti-noise ability. The present invention cleverly combines the grayscale interpolation method and surface fitting method in the sub-pixel displacement measurement, so that the calculation accuracy has been greatly improved; at the same time, by first calculating the matching position of the template on the picture before interpolation, and then Then transfer to the interpolated image to calculate the precise position of the template image. The advantage of this method is that the image before moving is matched. The size of the template image and the original image are smaller, and the calculation efficiency is higher. Compared with directly after interpolation The template image is calculated on the image, and the calculation efficiency has been greatly improved; and multiple matching images are selected, and the statistical mean is finally calculated to obtain more accurate results.

Description of drawings

Figure 1 is a flow chart of the specific implementation of the hill-climbing algorithm.

Fig. 2 is a specific implementation flow chart of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

Example

As shown in Figure 2, an object displacement measurement method specifically includes the following steps:

S1: Fix the position of the image acquisition device, and then use the image acquisition device to collect continuous moving images of the target to be measured; the position of the image acquisition device cannot be changed to obtain an image that reflects the movement of the target; select the first frame image (P1) and the second frame Any frame of image (M1) after one frame; and in order to obtain high measurement accuracy, the acquired photos should be of sufficiently high quality.

S2: If the image is a color image, first perform grayscale processing on it;

S3: Use the interpolation algorithm to interpolate P1 and M1 by k times, and obtain P2 and M2 respectively after interpolation; there are many interpolation algorithms that can be selected. Generally, the cubic convolution interpolation algorithm can be used to obtain pictures with relatively high precision, and the interpolation multiple should be It depends on the actual needs of computing efficiency, generally 10 times is enough.

S4: Frame a specific area containing obvious features in P1 as a template (T1), and record the coordinates (x ₀ , y ₀ ) of the upper left corner of the template map T1 in P1; the size of the template can be determined according to the actual calculation efficiency It depends on the needs, generally it is 41*41 pixels or 51*51 pixels.

S5: Use the hill-climbing method to match in M1, and get a matching area roughly similar to T1;

S6: In the area obtained in the previous step, use the SDA\SSD algorithm to accurately match the coordinates (x ₁ , y ₁ ) of the template image to achieve pixel-level matching;

S7: With (x ₁ , y ₁ ) as the center, calculate the correlation coefficient of each pixel in the n*n area, use the obtained correlation coefficient to perform surface fitting, and select the coordinates of the maximum value of the fitted surface as the matching point (x ₂ , y ₂ ); the selection of the surface fitting method can be determined according to the specific calculation efficiency requirements. In general, the selection of the quadratic surface fitting method can meet the requirements.

S8: Map the matching coordinate position (x ₂ , y ₂ ) in M1 obtained in step S7 to M2 to obtain the corresponding coordinate position (x ₃ , y ₃ ) in M2;

S9: Utilize the template image (T1) selected in P1 to select its corresponding template image (T2) in P2;

S10: In M2, take T2 as the template map and (x ₃ , y ₃ ) as the starting point, repeat step S6 to find the coordinates of the best matching point, take the coordinates as the center, select a rectangular area of m*m, Calculate the correlation coefficient of T2 at each point; since the matching position (x ₂ , y ₂ ) obtained in step S8 is already very close to the actual matching position, this step uses step S6 to search for the optimal matching point of the integer pixel point The coordinate speed is very fast.

S11: Use the obtained correlation coefficient to perform surface fitting, select the coordinates (x ₄ , y ₄ ) of the fitted maximum value as the best matching point, and map the coordinates back to the corresponding coordinates in M1 (x ₅ ,y ₅ ), and compare (x ₅ ,y ₅ ) with the coordinates (x ₀ ,y ₀ ) of the template in P1 to obtain the precise displacement (Δx,Δy) of the target;

S12: Repeat steps S4 to S11 above p times, select a different template image each time, calculate the statistical average value of the displacement, and use this value as the final displacement result; how many times to repeat can be determined according to the actual calculation efficiency.

The performance of the hill-climbing search method in step S5 is affected by many factors such as the size of the template and the image to be matched, the gray distribution of the template image, and the algorithm for calculating the correlation coefficient. As shown in Figure 1, step S5 can be further divided into:

S5.1: According to the relative size relationship between the to-be-matched sub-image and the template sub-image, select the horizontal and vertical distances between all initial mountain-climbing starting points in the hill-climbing method, and generate all mountain-climbing starting points in the image to be matched. If the actually measured moving range of the target is relatively small, all initial climbing starting points can be set near the coordinates (x ₀ , y ₀ ) to increase the climbing speed. In order to more accurately reflect the correlation between the original image and the template sub-image, the selected starting points for climbing should be distributed as evenly as possible, and the distance between the starting points should be moderate.

S5.2: Calculate the correlation coefficients of all climbing starting points and arrange them in reverse order. After the reasonable selection of the starting point in S5.1, it can be considered that there is a certain positive correlation between the correlation coefficient of the starting point and the distance between it and the maximum value of the correlation coefficient. Checking these starting points in the reverse order of the correlation coefficient can find the target faster.

S5.3: According to the requirements for matching accuracy, select an appropriate upper limit of the correlation coefficient to terminate the search process. In addition, based on the reasonable selection of the climbing starting point in step S5.1, it is considered that the point with too small correlation coefficient is too far away from the target point, so it can be discarded directly. The selection of this lower limit is closely related to the selection of the starting point in S5.1, and 0.25 is taken in the embodiment.

S5.4: After determining the current point, calculate the correlation coefficient of each point in the surrounding 3*3 matrix in turn. If there is a point with a larger correlation coefficient, use it as a new starting point to continue searching, if not, go to the next step.

S5.5: Determine whether the maximum correlation coefficient found in S5.4 exceeds the upper limit set in step S5.3, if not, calculate and check the next starting point of mountain climbing, otherwise, consider that a point meeting the requirements has been found, and directly terminate the mountain climbing process.

The above-mentioned embodiment is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, Simplifications should be equivalent replacement methods, and all are included in the protection scope of the present invention.

Claims (4)

1. a kind of ohject displacement measuring method, it is characterised in that comprise the following steps：

The position of S1, fixed image capture device, then gather the continuous moving figure of target to be measured using image capture device Picture；Choose any one two field picture M1 after the first two field picture P1 and the first frame；

If S2, image are coloured image, gray processing processing first is carried out to it；

S3, using interpolation algorithm to P1 and M1 into k times of computing of row interpolation, respectively obtain P2 and M2 after interpolation；

S4, outline a specific region comprising obvious characteristic as template T1 in P1；

S5, matched using climbing method in M1, obtains the matching area substantially similar with T1；

S6, in region obtained in the previous step, using SDA SSD algorithms accurately matching obtain the coordinate (x of template image₁,y₁), Realize pixel matching；

S7, with (x₁,y₁) centered on, calculate the related coefficient of each pixel in n*n regions, using acquisition related coefficient into Row surface fitting, chooses the coordinate of maximum for the curved surface that fitting obtains as match point (x₂,y₂)；

Matching coordinate position (x in S8, the M1 for obtaining step S7₂,y₂) be mapped in M2, obtain the corresponding coordinate in M2 Position (x₃,y₃)；

S9, using the masterplate T1 chosen in P1, choose its corresponding template T2 in P2；

S10, in M2, using T2 as Prototype drawing, with (x₃,y₃) it is starting point, repeat step S6, finds the coordinate of optimal match point, Centered on the coordinate, the rectangular area of m*m is chosen, calculates related coefficients of the T2 on each point；

S11, carry out surface fitting, the coordinate (x of the maximum for the curved surface that selection fitting obtains using the related coefficient of acquisition₄,y₄) Corresponding coordinate value (x in M1 is mapped back as optimal match point, and by the coordinate value₅,y₅), and by (x₅,y₅) exist with template Coordinate in P1 is compared；

S12, perform the step S4~step S11 circulation p times, the assembly average of displacement is calculated, with assembly average As final mean annual increment movement result.

2. ohject displacement measuring method as claimed in claim 1, it is characterised in that k in step S3, S7, S10 and S12, N, the value of m and p is any positive integer.

3. ohject displacement measuring method as claimed in claim 2, it is characterised in that the value range of k, n, m and p for [1, 10]。

4. ohject displacement measuring method as claimed in claim 1, it is characterised in that in step S3, S7, S8, S9, S10 and S11 In, sub-pixel interpolation algorithm and Surface Fitting are combined, while is matched in the image before interpolation and after interpolation.