Method and apparatus for fiducial point image recognition

Abstract

The present application relates to a method and device for identifying a marker point image. The method comprises: acquiring multiple images of a marker point collected continuously to form an image sequence of the marker point, selecting some images in the image sequence of the marker point as reference images; selecting at least one group of short image sequences in the image sequence of the marker point; converting the short image sequence into an equivalent image; acquiring the displacement of the equivalent image relative to the reference image according to a grayscale invariant model, solving the problem of low marker point image recognition accuracy, thereby reducing the influence of the temporal and spatial inconsistency of the image grayscale on the marker point image recognition accuracy, so that the marker point image recognition accuracy can meet the technical requirements of the relevant application field.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular, to a method and device for recognizing fiducial point images. Background Art

[0002] With the development of image recognition technology, digital image recognition technology based on fiducial points has been widely applied in fields such as medical feature localization, artificial intelligence (AI) control, and engineering structure deformation measurement, promoting the development of various application fields and bringing many conveniences to life.

[0003] In related technologies, the recognition accuracy of fiducial point images is easily affected by the temporal and spatial inconsistencies of image grayscale, resulting in the existing fiducial point image recognition technology being unable to meet the technical requirements for recognition accuracy in related application fields.

[0004] Aiming at the problem of low recognition accuracy of fiducial point images in related technologies, no effective solution has been proposed yet. Summary of the Invention

[0005] Aiming at the problem of low recognition accuracy of fiducial point images in related technologies, the present invention provides a method and device for recognizing fiducial point images to at least solve the above problems.

[0006] According to one aspect of the present invention, a method for recognizing fiducial point images is provided, including the following steps:

[0007] Obtain multiple continuously acquired images of fiducial points to form an image sequence of the fiducial points, and select some images from the image sequence of the fiducial points as reference images;

[0008] Select at least one group of short image sequences from the image sequence of the fiducial points;

[0009] Equivalent the short image sequence into an equivalent image;

[0010] Obtain the displacement of the equivalent image relative to the reference image according to the gray-scale invariant model.

[0011] In one embodiment, the step of selecting at least one group of short image sequences from the image sequence of the fiducial points includes:

[0012] Select multiple groups of short image sequences of the fiducial points from the image sequence of the fiducial points, and there is a preset number of image lengths difference between adjacent two groups of short image sequences.

[0013] In one embodiment, the step of equivalenting the short image sequence into an equivalent image includes:

[0014] Equivalently convert multiple groups of the short image sequences into multiple equivalent images respectively, and use the multiple equivalent images as an equivalent image sequence.

[0015] In one embodiment, obtaining the displacement of the equivalent image relative to the reference image according to the gray-scale invariant model includes:

[0016] Match the reference image with the equivalent image sequence according to the gray-scale invariant model, and obtain the deformation field data of the displacement between the equivalent image sequence and the reference image.

[0017] In one embodiment, selecting at least one group of short image sequences from the image sequences of the fiducial points includes:

[0018] Select one group of short image sequences with a preset sequence length from the image sequences of the fiducial points.

[0019] In one embodiment, obtaining the displacement of the equivalent image relative to the reference image according to the gray-scale invariant model includes:

[0020] Obtain the displacement of the equivalent image of the short image sequence that has been selected relative to the reference image according to the gray-scale invariant model;

[0021] Take the region where the short image sequence is located as the sequence selection region, translate the sequence selection region by a preset number of image lengths in the time direction of the image sequences of the fiducial points, update the short image sequence, and calculate the corresponding displacement until all the image sequences of the fiducial points are selected;

[0022] Obtain the deformation field data according to all the calculated displacements.

[0023] In one embodiment, the according to the gray-scale invariant model includes:

[0024]

[0025] Wherein, X is the displacement value of the marked point, U is the displacement of the marked point image, α is a related parameter, Δt is the time interval between two images, F i represents the gray-scale value of the reference image, G i represents the gray-scale value of the deformed image, and m represents m frames of images before and after the i-th frame.

[0026] In one embodiment, obtaining the images of multiple continuously collected fiducial points includes:

[0027] Select the image sub-region including the fiducial point, and obtain the images of the multiple continuously collected image sub-regions.

[0028] In one embodiment, selecting partial images from the image sequence of the fiducial points as reference images includes:

[0029] Continuously selecting multiple images from the image sequence of the fiducial points;

[0030] Equating the multiple selected images to reference images.

[0031] According to another aspect of the present invention, there is also provided a device for fiducial point image recognition, the device including:

[0032] An image acquisition module, configured to acquire multiple continuously acquired images of fiducial points to form the image sequence of the fiducial points;

[0033] An image selection module, configured to select partial images from the image sequence of the fiducial points as reference images, and select at least one group of short image sequences;

[0034] An image equivalence module, configured to equate the short image sequences to equivalent images;

[0035] A displacement solving module, configured to obtain the displacement of the equivalent image relative to the reference image according to the gray-scale invariant model.

[0036] The above-mentioned method and device for fiducial point image recognition acquire multiple continuously acquired images of fiducial points to form the image sequence of the fiducial points, select partial images from the image sequence of the fiducial points as reference images; select at least one group of short image sequences from the image sequence of the fiducial points; equate the short image sequences to equivalent images; obtain the displacement of the equivalent image relative to the reference image according to the gray-scale invariant model, solving the problem of low recognition accuracy of fiducial point images, thereby reducing the influence of the temporal and spatial inconsistency of image gray-scale on the recognition accuracy of fiducial point images, and enabling the recognition accuracy of fiducial point images to meet the technical requirements of relevant application fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described herein are used to provide a further understanding of the present invention, form a part of this application, and the schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0038] Figure 1 is a flowchart of the method for fiducial point image recognition according to an embodiment of the present invention;

[0039] Figure 2 is a schematic diagram of an image sequence simulating the vibration process of a fiducial point according to an embodiment of the present invention;

[0040] Figure 3 is a schematic diagram of the displacement-time curve of a fiducial point measured by different methodsFigure 1 ;

[0041] Figure 4 is a schematic diagram of the random error and systematic error of displacement measurement according to different methods;

[0042] Figure 5 is a schematic diagram of a uniaxial tensile experiment with marked points in an embodiment of the present invention;

[0043] Figure 6 is a schematic of the displacement-time curve of the marked points measured by different methods Figure 2 ;

[0044] Figure 7 is a schematic of the strain-time curve of the marked points measured by different methods Figure 3 ;

[0045] Figure 8 is a schematic diagram of an experimental device for measuring the truss movement of a deployable antenna in an embodiment of the present invention;

[0046] Figure 9 is a schematic diagram of the distance between two marked points measured by different methods;

[0047] Figure 10 is a structural block diagram of a device for identifying marked point images in an embodiment of the present invention. Detailed implementation manners

[0048] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0049] It should be noted that the terms "first", "second", and "third" involved in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific order for the objects. "First", "second", and "third" can be interchanged with a specific order or sequence when permitted. It can be understood that the objects distinguished by "first", "second", and "third" can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein.

[0050] The method for identifying marked point images provided by the present application can be applied to fields such as medical feature localization, artificial intelligence AI control, and engineering structure deformation measurement, such as real-time monitoring of the movement and deformation of large bridges, geotechnical structures, and large deployable structures in aerospace.

[0051] In one embodiment, Figure 1is a flowchart of a method for fiducial point image recognition according to an embodiment of the present invention. As Figure 1 shown, a method for fiducial point image recognition is provided, including the following steps:

[0052] Step S110, obtain multiple continuously acquired images of fiducial points to form an image sequence of fiducial points, and select some images from the image sequence of fiducial points as reference images;

[0053] Among them, from time t i-m to time t i+m the image acquisition device obtains 2m + 1 consecutive images of fiducial points, forming an image sequence of fiducial points.

[0054] It should be noted that the fiducial points can be the positions to be monitored in medical feature localization, artificial intelligence AI control, and engineering structure deformation measurement. Select the monitoring area as the fiducial points according to the requirements of actual applications, and collect multiple images of the fiducial points within the time period to be monitored through the image acquisition device. The number of collected fiducial point images can be determined according to the needs of actual research and calculation costs.

[0055] Step S120, select at least one short image sequence from the image sequence of fiducial points.

[0056] It should be noted that the short image sequence refers to an image sequence of fiducial points composed of multiple fiducial point images.

[0057] Step S130, equivalent the short image sequence into an equivalent image.

[0058] It should be noted that an image sequence of fiducial points composed of multiple fiducial point images is equivalent to a high-quality equivalent image according to the following formula:

[0059]

[0060] where X is the displacement value of the fiducial point, U is the displacement of the fiducial point image, f(X) represents the gray value of the equivalent image, g t represents the gray value of the deformed image, and m represents m frames of images before and after the i-th frame.

[0061] Step S140, obtain the displacement of the equivalent image relative to the reference image according to the gray invariant model.

[0062] It should be noted that according to the gray-scale invariant model, the reference image and the equivalent image are matched and analyzed to obtain the displacement function of the equivalent image. The gray-scale invariant model is simplified. By performing a Taylor expansion on the displacement function and simplifying the displacement function through digital image correlation (DIC) of the fiducial points, a displacement function related to time, displacement, and relevant parameters is obtained. Then, according to the displacement solved from the simplified gray-scale invariant model and using the Newton-Raphson iteration method, the relevant parameters of the displacement function are solved to obtain the displacement of the equivalent image relative to the reference image.

[0063] Among them, the displacement of the displacement function refers to the lateral displacement value and the longitudinal displacement value of the fiducial points. Considering that solid deformation is continuous in time, the displacement function of the fiducial points is a continuous function of time. For the digital image correlation (DIC) of the fiducial points, the deformation of the fiducial points themselves is not considered, and only the rigid body displacement of the fiducial points is considered.

[0064] It should be further noted that digital image correlation introducing time continuity can establish 2m + 1 equations with gray-scale invariant assumptions:

[0065]

[0066] Among them, f(x) represents the gray value of the reference image, and g t represents the gray value of the deformed image;

[0067] According to the gray-scale invariant model, the reference image and the equivalent image are matched and analyzed to obtain the displacement data of each point on the surface of the equivalent image, generate the deformation field data of the displacement between the equivalent image sequence and the reference image, and thus obtain the displacement function of the equivalent image sequence.

[0068] Considering that solid deformation is continuous in time, the displacement function U(X, t) of the marked points is a continuous function of time. Performing a first-order Taylor expansion on the function U(X, t), it can be expressed as:

[0069]

[0070] Among them, Δt is the time interval between two images.

[0071] For the digital image correlation of the marked points, the deformation of the marked points themselves is not considered, and only the rigid body displacement of the marked points is considered. Therefore, its displacement can be simplified as:

[0072]

[0073] Among them,

[0074]

[0075] Finally, the digital correlation matching algorithm for marker points considering temporal continuity expressed by the gray-scale invariance assumption can be represented as:

[0076]

[0077] The α in this equation can be solved using the Newton-Raphson method, and μ and v can be obtained through optimization, that is, the lateral and longitudinal displacement values of the marker points.

[0078] Through the above method for marker point image recognition, by selecting at least one set of short image sequences in the image sequence of marker points, equivalent images are obtained by equating the short image sequences. According to the gray-scale invariance model, the displacement of the equivalent image relative to the reference image is obtained, reducing the temporal noise of the marker point images, solving the problem of low recognition accuracy of marker point images, and thus reducing the impact of temporal and spatial inconsistencies in the gray-scale of marker point images on the recognition accuracy of marker point images. Finally, the recognition accuracy of marker point images is improved, enabling the recognition accuracy of marker point images to meet the technical requirements of relevant application fields.

[0079] In one embodiment, a method for marker point image recognition is provided, and step S120 includes step S220:

[0080] Step S220: In the image sequence of marker points, multiple sets of short image sequences of marker points are selected, and the length of the preset number of images differs between adjacent two sets of short image sequences.

[0081] Taking the continuous acquisition of 9 images as an example: From time t 0 to time t 8 the image acquisition device acquires 9 consecutive images of marker points, forming an image sequence of marker points. The preset sequence length of the short image sequence is 3. In this image sequence, the first set of short image sequences is selected, and the images of this first set of short image sequences are P1, P2, and P3 respectively. In this image sequence, the second set of short image sequences is selected, and the images of this second set of short image sequences are P2, P3, and P4 respectively. The preset number of image lengths is 1; in this image sequence, the first set of short image sequences is selected, and the images of this first set of short image sequences are P1, P2, and P3 respectively. In this image sequence, the second set of short image sequences is selected, and the images of this second set of short image sequences are P3, P4, and P5 respectively. The preset number of image lengths is 2.

[0082] It should be noted that a short image sequence refers to a sequence of landmark images composed of multiple landmark images; the landmarks can be positions to be monitored in medical feature localization, artificial intelligence (AI) control, and engineering structure deformation measurement. The monitoring area is selected as the landmark according to the requirements of actual applications, and multiple landmark images within the time period to be monitored are collected by an image acquisition device. The number of collected landmark images can be determined according to the needs of actual research and calculation costs.

[0083] Through the above method for identifying landmark images, by selecting multiple short image sequences of landmarks in the image sequence of landmarks, and according to the gray-scale invariant model, the displacement of the equivalent image relative to the reference image is obtained, reducing the time-domain noise of the landmark images, solving the problem of low recognition accuracy of landmark images, thereby reducing the influence of the temporal and spatial inconsistency of the gray scale of landmark images on the recognition accuracy of landmark images, and finally improving the recognition accuracy of landmark images, enabling the recognition accuracy of landmark images to meet the technical requirements of relevant application fields.

[0084] In one embodiment, a method for identifying landmark images is provided, and step S130 includes step S330:

[0085] Step S330: Equivalent multiple groups of short image sequences into multiple equivalent images respectively, and use the multiple equivalent images as an equivalent image sequence.

[0086] It should be noted that the sequence of landmark images composed of multiple landmark images is equivalent to a high-quality equivalent image.

[0087] Through the above method for identifying landmark images, by respectively equivalent multiple groups of short image sequences into multiple equivalent images and using the multiple equivalent images as an equivalent image sequence, and according to the gray-scale invariant model, the displacement of the equivalent image relative to the reference image is obtained, reducing the time-domain noise of the landmark images, solving the problem of low recognition accuracy of landmark images, thereby reducing the influence of the temporal and spatial inconsistency of the gray scale of landmark images on the recognition accuracy of landmark images, and finally improving the recognition accuracy of landmark images, enabling the recognition accuracy of landmark images to meet the technical requirements of relevant application fields.

[0088] In one embodiment, a method for identifying landmark images is provided, and step S140 includes step S440:

[0089] Step S440: Match the reference image with the equivalent image sequence according to the gray-scale invariant model to obtain the deformation field data of the displacement of the equivalent image sequence relative to the reference image.

[0090] Among them, according to the gray-scale invariant model, the reference image is matched with the equivalent image to obtain the displacement data of each point on the surface of the equivalent image, and the deformation field data of the displacement between the equivalent image sequence and the reference image is generated.

[0091] In one embodiment, a method for identifying fiducial point images is provided, and step S120 includes step S520:

[0092] Step S520, in the image sequence of the fiducial points, a set of short image sequences is selected with a preset sequence length.

[0093] Among them, by changing the preset sequence length, short image sequences with different sequence lengths can be selected. The longer the length of the short image sequence, the higher the image quality of the equivalent image obtained equivalently. Therefore, increasing the preset sequence length can improve the image quality of the equivalent image, and further reduce the error of the finally calculated displacement and improve the accuracy of the calculated displacement.

[0094] Through the above method for identifying fiducial point images, in the image sequence of the fiducial points, a set of short image sequences is selected with a preset sequence length, the short image sequences are equivalently converted into equivalent images, according to the gray-scale invariant model, the reference image is matched and analyzed with the equivalent image sequence, the displacement function of the equivalent image is obtained, and by improving the image quality of the equivalent image, the calculation accuracy is improved, thereby improving the recognition accuracy of the fiducial point images.

[0095] In one embodiment, a method for identifying fiducial point images is provided, and step S140 includes step S640:

[0096] Step S640, according to the gray-scale invariant model, obtain the displacement of the equivalent image of the short image sequence that has been selected relative to the reference image; take the area where the short image sequence is located as the sequence selection area, in the time direction of the image sequence of the fiducial points, translate the sequence selection area by a preset number of image lengths, update the short image sequence, and calculate the corresponding displacement until all the image sequences of the fiducial points are selected; obtain the deformation field data according to all the calculated displacements.

[0097] Among them, according to the gray-scale invariant model, the reference image is matched with the equivalent image to obtain the displacement data of each point on the surface of the equivalent image, and the deformation field data of the displacement between the equivalent image sequence and the reference image is generated.

[0098] It should be further noted that the first group of short image sequences is equivalent to an equivalent image, and the displacement is calculated. The area where the short image sequences are located is used as the sequence selection area. In the time direction of the image sequence of the fiducial points, the sequence selection area is translated by the length of a preset number of images, images of the length of the preset number of images are loaded, the images of the length of the preset number of images at the front in the first group of short image sequences are removed, and the loaded images are arranged behind the images of the previous group of short image sequences to form the second group of short image sequences. The second group of short image sequences is equivalent to an equivalent image, and the displacement is calculated, and so on until all the images in the image sequence of the fiducial points are selected, and the deformation field data is obtained according to all the calculated displacements.

[0099] For example, from time t0 to time t8, the image acquisition device acquires 9 consecutive images of the fiducial points, forming an image sequence of the fiducial points. The sequence length of the preset short image sequence is 3. In this image sequence, the first group of short image sequences is selected. The images of the first group of short image sequences are P1, P2, and P3 respectively. The first group of short image sequences is equivalent to an equivalent image, and the displacement is calculated. The area where the short image sequences are located is used as the sequence selection area. In the time direction of the image sequence of the fiducial points, the sequence selection area is translated by the length of a preset number of images. The preset number of images is set to 1, and one image is loaded. The previous image P1 in the first group of short image sequences is removed, and the loaded image P4 is arranged behind the images of the first group of short image sequences to form the second group of short image sequences. The images of the second group of short image sequences are P2, P3, and P4. The second group of short image sequences is equivalent to an equivalent image, and the displacement is calculated, and so on until all the images in the image sequence of the fiducial points are selected, and the deformation field data is obtained according to all the calculated displacements.

[0100] Through the above method for fiducial point image recognition, according to the gray-scale invariant model, the displacement of the equivalent image relative to the reference image is obtained. In the time direction of the image sequence of the fiducial points, images of the length of the preset number of images are loaded, the short image sequence is updated, and the corresponding displacement is calculated until all the images in the image sequence of the fiducial points are selected. The deformation field data is obtained according to all the calculated displacements, and the displacement of the equivalent image is calculated in real time, reducing the time-domain noise of the fiducial point images, solving the problem of low recognition accuracy of the fiducial point images, thereby reducing the influence of the time and space inconsistency of the gray scale of the fiducial point images on the recognition accuracy of the fiducial point images, and realizing high-precision and real-time recognition and calculation of the marked point images.

[0101] In one embodiment, a method for fiducial point image recognition is provided. Step S110 includes step S710:

[0102] Step S710: Select an image sub-region including the landmark points, obtain images of multiple continuously acquired image sub-regions to form an image sequence of the landmark points, and select some images from the image sequence of the landmark points as reference images.

[0103] It should be noted that the marked points can be positions to be monitored in medical feature localization, artificial intelligence AI control, and engineering structure deformation measurement. Select the monitoring area as the marked points according to the requirements of actual applications. Collect multiple images of the marked points within the required monitoring time period through an image acquisition device. The number of collected marked point images can be determined according to the needs of actual research and calculation costs. Further, select the image sub-region including the marked points as the area for marked point image recognition and monitoring, and ensure that the marked points are within the image sub-region throughout the monitoring period.

[0104] Through the above method for landmark point image recognition, select an image sub-region including the landmark points, obtain images of multiple continuously acquired image sub-regions. On the premise of ensuring that the landmark points are always within the image sub-region throughout the deformation process, accurately locate the image sub-region for landmark point recognition, improve the efficiency of image recognition, and reduce the costs of image recognition and calculation.

[0105] In one embodiment, a method for landmark point image recognition is provided. Step S110 includes step S810:

[0106] Step S810: Obtain images of multiple continuously acquired landmark points to form an image sequence of the landmark points. In the image sequence of the landmark points, continuously select multiple images and equivalent the selected multiple images into a reference image.

[0107] It should be noted that the marked points can be positions to be monitored in medical feature localization, artificial intelligence AI control, and engineering structure deformation measurement. Select the monitoring area as the marked points according to the requirements of actual applications. Collect multiple images of the marked points within the required monitoring time period through an image acquisition device. The number of collected marked point images can be determined according to the needs of actual research and calculation costs.

[0108] Through the above method for landmark point image recognition, by equating multiple reference images of the landmark points into one reference image of the landmark points, the image quality of the reference image of the landmark points is improved, and further the recognition accuracy of the landmark point images is improved.

[0109] This application also provides the following three specific embodiments to further elaborate on the method for landmark point image recognition, including Embodiment 1, Embodiment 2, and Embodiment 3:

[0110] In Embodiment 1, the above method for landmark point image recognition includes the following steps:

[0111] Step S910,Figure 2 is a schematic diagram of an image sequence simulating the vibration process of fiducial points in an embodiment of the present invention. As Figure 2 shown, a reference image is selected. The reference image contains a fiducial point with a size of 6×6 pixels. The size of the image sub-region is selected as 15×15 pixels to ensure that the fiducial point is always within the image sub-region during the entire deformation process. Through simulation, a sine function y = 1.5sinx is applied to the reference image to obtain a deformed image, and Gaussian noise with a standard deviation σ = 2 (255 gray levels) and a mean of 0 is introduced into the deformed image to generate a deformed image sequence of the fiducial point.

[0112] Step S920: In the deformed image sequence of the fiducial point, select a deformed image sequence of the fiducial point with a preset sequence length to form a short image sequence of the fiducial point.

[0113] Step S930: Equivalent the short image sequence into an equivalent image, and generate multiple equivalent images to form an equivalent image sequence of the fiducial point.

[0114] Step S940: According to the gray-scale invariant model, perform matching analysis on the reference image and the equivalent image sequence to obtain the deformation field data of the displacement of the equivalent image sequence and the reference image.

[0115] In this embodiment, Figure 3 is a schematic diagram of the displacement-time curve of the fiducial point measured by different methods Figure 1 , as Figure 3 shown, by comparing the displacement (displacement / pixel)-time (time / s) curves of the fiducial point measured by the gray centrobaric arithmetic method, the traditional DIC (traditional DIC) method, and the method of fiducial point image recognition proposed by the present invention with the theoretical curve (theoretical curve), it can be seen that the displacement obtained by using the gray centrobaric arithmetic method fluctuates greatly. This is because compared with other methods, the gray centrobaric arithmetic method is more sensitive to the change of the spatial gray scale of the fiducial point image. From Figure 3 the enlarged part, it can be seen that the displacement measurement results of the traditional DIC method deviate from the theoretical value at some moments. The above-mentioned method of fiducial point image recognition with short image sequences having preset sequence lengths of 3 and 5 is respectively called STS-DIC-3 and STS-DIC-5, and their measurement results are in good agreement with the theoretical values.

[0116] In addition, the measurement effect can also be quantitatively evaluated by calculating the displacement measurement errors of different methods. Calculate the systematic error and random error using the formula to obtain the displacement measurement random error and systematic error of different analysis methods. Figure 4It is a schematic diagram of the random error and systematic error of displacement measurement according to different methods. As shown in Table 4, according to Figure 4 the results, for the above method of landmark image recognition, when the preset sequence length of the short image sequence is 5, the measurement error is the smallest, and the measurement error of the gray centroid method is the largest. Compared with the traditional DIC, the measurement error of the above method of landmark image recognition is reduced by two-thirds. It is found by comparison that the computational cost of the method proposed by the present invention increases with the increase of the preset sequence length of the short image sequence. However, compared with the improvement of the measurement accuracy by the method proposed by the present invention, the increase in computational cost is not a major issue. According to the above analysis, although the computational cost has increased, the recognition accuracy of the method of landmark image recognition proposed by the present invention is superior to that of the traditional DIC and the gray centroid method.

[0117] In Example 2, the method of landmark image recognition described above includes the following steps:

[0118] Step S1010, in a uniaxial tensile experiment, spray a dog-bone-shaped polymer specimen with black paint, and then make two white landmarks at both ends. Figure 5 It is a schematic diagram of the uniaxial tensile experiment with landmarks according to an embodiment of the present invention. As Figure 5 shown, perform tensile testing on a tensile testing machine at a loading rate of 1 mm / min, and use a high-resolution camera to collect images of the landmarks on the surface of the specimen at a capture rate of 1 frame / second. In the initial stage of loading, the deformation of the specimen can be regarded as linear elastic, and use the camera to record the deformation images at this stage, obtain multiple consecutive images of the image sub-regions where the landmarks are located, and form an image sequence of the landmarks. According to the current imaging settings, each white landmark approximately occupies 105×105 pixels, and the size of the image sub-region is set to 121×121 pixels, which can ensure that the landmarks are within the image sub-region during the entire deformation process.

[0119] Step S1020, in the deformation image sequence of the landmarks, select multiple short image sequences of the landmarks, and there is a difference of a preset number of images between each group of short image sequences.

[0120] Step S1030, respectively equivalent multiple groups of short image sequences into multiple equivalent images, and use the multiple equivalent images as an equivalent image sequence.

[0121] Step S1040, according to the gray-scale invariant model, match the reference image with the equivalent image sequence to obtain the deformation field data of the displacement of the equivalent image sequence and the reference image.

[0122] In this embodiment, considering the computational cost, when using the method proposed by the present invention, the preset sequence length of the selected short image sequence is 3, which is called STS-DIC-3. Figure 6It is a schematic diagram of the displacement-time curve of the marked points measured by different methods Figure 2 , such as Figure 6 shown Figure 7 It is a schematic diagram of the strain-time curve of the marked points measured by different methods Figure 3 , such as Figure 7 shown. According to the displacement-time curve (displacement / time) and strain-time (strain / time) curve of the marked points, it can be seen that the proposed method of the present invention has the strongest smoothing effect, followed by the traditional DIC method, and the robustness of the displacement measurement result using the gray centroid method is the worst, which confirms the superiority of the method for identifying the marked point images proposed by the present invention.

[0123] In Embodiment 3, the method for identifying the marked point images includes the following steps:

[0124] Step S1110, Figure 8 It is a schematic diagram of the experimental device for measuring the truss movement of the deployable antenna according to the embodiment of the present invention, such as Figure 8 shown. There are two marked points on the measured target, and the distance between the two marked points remains unchanged during the measurement. The distance between the two points can be obtained by an image measuring instrument, and the distance is used as the reference value for measurement. During the deployment process, a 3D motion measurement system is used to photograph the deployment process of the truss. Before starting the measurement, the camera is first calibrated to obtain the internal parameters and external parameters of the camera. The image sub-region is selected as 31×31 pixels. During the measurement, the image acquisition frequency of the camera is 30 frames per second, and multiple consecutive images of the two marked points are collected respectively to form an image sequence of the two marked points.

[0125] Step S1120, obtain multiple consecutive images of the marked points to form an image sequence of the marked points, select some images in the image sequence of the marked points as reference images, and select at least one group of short image sequences in the image sequence of the marked points.

[0126] Step S1130, equivalent the short image sequence into an equivalent image, and generate multiple equivalent images to form an equivalent image sequence of the marked points.

[0127] Step S1140, according to the gray-scale invariant model, obtain the displacement of the equivalent image relative to the reference image. In the time direction of the image sequence of the marked points, translate the short image sequence by the length of the preset number of images, update the short image sequence, and calculate the corresponding displacement until all the image sequences of the marked points are selected, and obtain the deformation field data according to all the calculated displacements.

[0128] In this embodiment, the traditional DIC method, the grey barycentre method, and the method for identifying fiducial point images proposed by the present invention are used for analysis. During the analysis, short image sequences with preset sequence lengths of 3 and 5 are respectively selected. First, the image sequences collected by two cameras are processed using different methods to obtain the 2D motions of two fiducial points. Then, these are used for spatial position calibration and 3D reconstruction of multiple cameras. Finally, the distance between point A and point B is calculated. Figure 9 is a schematic diagram of the distances between two fiducial points measured by different methods, as Figure 9 shown. According to the measurement results, it can be seen that the results measured by the grey barycentre method fluctuate the most, and the results measured by the method for identifying fiducial point images proposed by the present invention are relatively reliable and in good agreement with the reference value, among which the results measured by STS-DIC-5 are closest to the reference value.

[0129] The measurement effects of the grey barycentre method, the traditional DIC method, and the method for identifying fiducial point images proposed by the present invention are compared through simulation and experimental means. It is verified that the method for identifying fiducial point images proposed by the present invention improves the recognition accuracy of fiducial point images at the cost of slightly increasing the calculation cost.

[0130] It should be understood that although Figure 1 the steps in the flowchart of Figure 1 are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this article, the execution of these steps has no strict order limitation, and these steps can be executed in other orders. Moreover, Figure 1 at least a part of the steps in

[0131] correspond to the manufacturing method of the above flexible electronic device. In this embodiment, a device for identifying fiducial point images is further provided. The device is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated here. As used hereinafter, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0132] According to another aspect of the present invention, a device for identifying fiducial point images is further provided, Figure 10 is a structural block diagram of the device for identifying fiducial point images according to an embodiment of the present invention, asFigure 10 As shown in the figure, the device includes:

[0133] An image acquisition module 101, configured to acquire images of a plurality of continuously acquired fiducial points to form an image sequence of the fiducial points;

[0134] An image selection module 102, configured to select some images as reference images and at least one set of short image sequences from the image sequence of the fiducial points;

[0135] An image equivalence module 103, configured to equivalent the short image sequence into an equivalent image;

[0136] A displacement solving module 104, configured to obtain the displacement of the equivalent image relative to the reference image according to the gray invariant model.

[0137] In the above-mentioned device for fiducial point image recognition, the image acquisition module 101 is connected to the image selection module 102. The image selection module 102 selects reference images and short image sequences from the image sequence acquired by the image acquisition module 101. The image equivalence module 103 is connected to the image selection module 102 to equivalent the short image sequence into an equivalent image, reducing the temporal noise of the fiducial point image, solving the problem of low recognition accuracy of the fiducial point image, thereby reducing the influence of the temporal and spatial inconsistency of the gray level of the fiducial point image on the recognition accuracy of the fiducial point image, and finally achieving an improvement in the recognition accuracy of the fiducial point image, such that the recognition accuracy of the fiducial point image can meet the technical requirements of relevant application fields.

[0138] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0139] The above embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for identifying fiducial point images, characterized in that, the method includes: Obtaining images of fiducial points collected continuously in multiple numbers to form an image sequence of the fiducial points, and selecting some images from the image sequence of the fiducial points as reference images; Selecting at least one set of short image sequences from the image sequence of the fiducial points; Equivalent the short image sequence into an equivalent image; the equivalent of the short image sequence into an equivalent image includes: equivalent a marker point image sequence composed of multiple marker point images into a high-quality equivalent image according to the following formula: where X is the displacement value of the marker point, U is the displacement of the marker point image, f(X) represents the gray value of the equivalent image, and g t represents the gray value of the deformed image, and m represents m frames of images before and after the i-th frame; According to the gray-scale invariant model, obtain the displacement of the equivalent image relative to the reference image; the said according to the gray-scale invariant model includes: where X is the displacement value of the marked point, U is the image displacement of the marked point, α is a related parameter, Δt is the time interval between two images, F i represents the gray-scale value of the reference image, G i represents the gray-scale value of the deformed image, and m represents m frames of images before and after the i-th frame.

2. The method according to claim 1, characterized in that, the selecting at least one set of short image sequences from the image sequence of the fiducial points includes: Selecting multiple sets of short image sequences of the fiducial points from the image sequence of the fiducial points, with a preset number of image lengths difference between adjacent two sets of short image sequences.

3. The method according to claim 2, characterized in that, the equivalenting the short image sequences into equivalent images includes: Equivalenting multiple sets of the short image sequences into multiple equivalent images respectively, and using the multiple equivalent images as an equivalent image sequence.

4. The method according to claim 3, characterized in that, the obtaining the displacement of the equivalent image relative to the reference image according to the gray-scale invariant model includes: According to the gray-scale invariant model, matching the reference image with the equivalent image sequence, and obtaining the deformation field data of the displacement between the equivalent image sequence and the reference image.

5. The method according to claim 1, characterized in that, the selecting at least one set of short image sequences from the image sequence of the fiducial points includes: Selecting one set of short image sequences with a preset sequence length from the image sequence of the fiducial points.

6. The method according to claim 5, characterized in that, the obtaining the displacement of the equivalent image relative to the reference image according to the gray-scale invariant model includes: According to the gray-scale invariant model, obtaining the displacement of the equivalent image of the short image sequence that has been selected relative to the reference image; Taking the area where the short image sequence is located as a sequence selection area, translating the sequence selection area by a preset number of image lengths in the time direction of the image sequence of the fiducial points, updating the short image sequence, and calculating the corresponding displacement until all of the image sequence of the fiducial points is selected; Obtaining deformation field data according to all the calculated displacements.

7. The method according to claim 1, characterized in that, the obtaining images of fiducial points collected continuously in multiple numbers includes: Selecting an image sub-region including the fiducial point, and obtaining images of the multiple continuously collected image sub-regions.

8. The method according to claim 1, characterized in that, the selecting some images from the image sequence of the fiducial points as reference images includes: Continuously selecting multiple images from the image sequence of the fiducial points; Equivalenting the multiple selected images into a reference image.

9. A device for identifying fiducial point images, characterized in that, the device includes: An image acquisition module, configured to obtain images of fiducial points collected continuously in multiple numbers to form an image sequence of the fiducial points; An image selection module, configured to select some images from the image sequence of the fiducial points as reference images, and select at least one set of short image sequences; An image equivalence module for equating the short image sequence into an equivalent image; the equating of the short image sequence into an equivalent image includes: equating a marker point image sequence composed of multiple marker point images into a high-quality equivalent image according to the following formula: where X is the displacement value of the marker point, U is the displacement of the marker point image, f(X) represents the gray value of the equivalent image, and g t represents the gray value of the deformed image, and m represents m frames of images before and after the i-th frame; A displacement solving module, configured to obtain the displacement of the equivalent image relative to the reference image according to the gray-invariant model; the gray-invariant model includes: where X is the displacement value of the marked point, U is the displacement of the marked point image, α is a related parameter, Δt is the time interval between two images, F i represents the gray value of the reference image, G i represents the gray value of the deformed image, and m represents m frames of images before and after the i-th frame.

Images