A Grating High Dynamic Range Measurement Method Based on Double Exposure Branch Fusion Network

Abstract

The present invention belongs to the technical field of oil and gas device detection, and in particular, relates to a grating high dynamic range measurement method based on a double exposure branch fusion network. The grating high dynamic range measurement method reconstructs a high dynamic range fringe pattern of an object to be measured through a double exposure branch fusion network, and realizes fast and accurate high dynamic range three-dimensional measurement of the object to be measured. A grating high dynamic range measurement method specifically includes the following steps: constructing a measurement system required for a grating high dynamic range measurement method, collecting fringe patterns with different exposure times; using an image fusion algorithm to synthesize a grating high dynamic range fringe pattern; constructing and training a double exposure branch fusion network; calibrating the constructed measurement system; obtaining a high dynamic range fringe pattern of the object to be measured, and calculating the absolute phase of the object to be measured; traversing all coordinate points of the object to be measured, and reconstructing the three-dimensional morphology of the object to be measured.

Description

Technical Field

[0001] The invention belongs to the technical field of oil and gas device detection, and in particular relates to a grating high dynamic range measurement method based on a double-exposure branch fusion network. Background Art

[0002] Oil pipelines and storage tanks are important equipment for transporting and storing oil and natural gas. Their working environment is harsh and they are easily affected by mechanical shock, medium corrosion and other factors, which may cause defects such as deformation and collapse. There is a risk of oil and gas leakage. Technical personnel are required to conduct regular inspections and maintenance on them to ensure that effective intervention measures are taken before oil and gas leakage occurs to avoid major economic losses and environmental pollution.

[0003] Among them, grating high dynamic range (HDR) 3D measurement technology is widely used in the field of oil and gas equipment inspection such as oil pipelines and storage tanks as a high-precision, efficient and low-cost detection method. However, in the actual measurement process, due to the fact that the surface of the equipment to be inspected is often coated with high-brightness or light-absorbing materials, and is restricted by factors such as the limited dynamic range of the camera and uneven ambient light, the dynamic range of the captured fringe pattern is low, which affects the phase demodulation accuracy and leads to the loss of target 3D information.

[0004] After further research, it can be found that the existing grating HDR measurement technology mainly includes the following two types of measurement methods: one is to adjust the camera parameters to collect fringe patterns with different exposure times; and synthesize a single HDR image through an image fusion algorithm. The other is to add additional hardware (such as polarizers) and adjust the angle of the polarizers to filter out highlights, thereby improving the dynamic range of the fringe pattern. For the former, the patent number CN114440794 A "A high-precision three-dimensional measurement method for high-dynamic surfaces" invented by Jiang Gedong et al. of Xi'an Jiaotong University proposed a grating HDR measurement method based on multiple exposures. The camera collects the multi-exposure sequence fringe pattern of the object, generates the corresponding mask sequence according to the degree of pixel overbrightness, and obtains a set of HDR fringe patterns through image fusion algorithm processing. Then, the phase measurement profilometry (PMP) is used to reconstruct the target three-dimensional morphology. However, the choice of camera exposure is usually determined by experience, with poor universality, and the camera needs to be imaged multiple times, which is time-consuming and has low measurement efficiency. For the latter, Yang Bin and others from Deep Computing (Changsha) Information Technology Co., Ltd. invented patent number CN110146036 A "A three-dimensional measurement method and system based on grating projection and binocular polarization camera", which proposed a binocular polarization camera grating HDR three-dimensional measurement method. The angle between the polarizer transmission axis and the object is adjusted to filter out highlights to improve the dynamic range of the system. However, external auxiliary equipment is required, which not only increases the hardware cost, but also requires multiple adjustments to the polarizer angle, seriously affecting the measurement efficiency.

[0005] In summary, the existing grating high dynamic range measurement methods have many limitations such as long time consumption and high cost, and the imaging system (characteristic parameters) are usually adjusted by experience, and the measurement accuracy is difficult to guarantee. Therefore, technical personnel in this field are urgently needed to provide a high-precision, efficient and low-cost grating high dynamic range measurement method. Summary of the invention

[0006] The present invention provides a grating high dynamic range measurement method based on a double-exposure branch fusion network. The grating high dynamic range measurement method reconstructs a high dynamic range fringe pattern of an object to be measured through a double-exposure branch fusion network, thereby realizing fast and accurate high dynamic range three-dimensional measurement of the object to be measured.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] A grating high dynamic range measurement method based on a double-exposure branch fusion network comprises the following steps:

[0009] Step 1: construct a measurement system required for the grating high dynamic range measurement method, and collect fringe patterns with different exposure times; wherein the measurement system is composed of a camera and a projector;

[0010] Using the image fusion algorithm, the grating high dynamic range fringe image is synthesized and used as the training truth value of the double exposure branch fusion network;

[0011] Step 2: Build and train a double-exposure branch fusion network;

[0012] Step 3: Calibrate the measurement system constructed in step 1;

[0013] Step 4: Based on the double-exposure branch fusion network trained in step 2, a high dynamic range fringe image of the object to be measured is obtained, and the absolute phase of the object to be measured is calculated;

[0014] According to the absolute phase of the object to be measured and the calibration parameters of the measurement system, all coordinate points of the object to be measured are traversed to reconstruct the three-dimensional shape of the object to be measured.

[0015] Preferably, the training truth value of the double-exposure branch fusion network in step 1 satisfies: Formula (1);

[0016] In formula (1), Indicates the pixel coordinates in the double exposure branch fusion network training as The true gray value of is the number of fringe images to be fused, , and Respectively The fusion weight factors of brightness, contrast and saturation in the fringe image, For the The pixel coordinates in the fringe image are The measured gray value.

[0017] Preferably, the double-exposure branch fusion network constructed in step 2 consists of a coding layer, a fusion layer and a decoding layer;

[0018] The encoding layer includes a convolutional layer, a normalization layer, a residual connection module DBF-Block and a channel attention mechanism module ECA, which is used to extract fringe patterns with different exposure times from the training truth value of the double exposure branch fusion network. , As a multi-channel feature map input, it is processed by the Relu activation function to obtain the fusion layer input image , ;

[0019] The fusion layer is used to input the image to the fusion layer , Perform adaptive fusion processing to obtain the fusion layer output image ;

[0020] The decoding layer is used to output the image of the fusion layer through the loss function Perform iterative solution to obtain the final image .

[0021] Preferably, the fusion layer outputs the image to the encoding layer , Perform adaptive fusion processing to obtain the fusion layer output image The process can be specifically described as:

[0022] Output image to the encoding layer , After normalization, mask calculation is performed;

[0023] Among them, the mask calculation process satisfies: Formula (2);

[0024] In formula (2), Indicates the network The pixel coordinates in the mask image are The gray value of For the The pixel coordinates in the output image of the coding layer are The gray value of is the exposure factor, is the image standard deviation, is the natural exponential function;

[0025] right Perform Gaussian filtering;

[0026] Among them, the Gaussian filtering process satisfies: Formula (3);

[0027] In formula (3), The pixel coordinates after Gaussian filtering are The gray value of is the standard deviation of the Gaussian function;

[0028] Feature fusion to obtain the fusion layer output image ; Among them, the fusion layer outputs the image ,satisfy: Formula (4).

[0029] Preferably, the decoding layer uses the loss function to output the image of the fusion layer Perform iterative solution to obtain the final image The process also includes the following steps:

[0030] Select a weighted combination loss function; the weighted combination loss function is used to improve the final image Dynamic range;

[0031] Among them, the weighted combination loss function ,satisfy: Formula (5);

[0032] In formula (5), Represents the input image With the output image The gray value difference is is the weight factor of the loss function;

[0033] Represents the structural similarity loss function, satisfying:

[0034] Formula (6);

[0035] In formula (6), and are the input and output image grayscale values, and , and They are and The mean and standard deviation of and is a constant;

[0036] Represents the pixel mean square error loss function, which is used to enhance the texture features of the fused image and satisfies: Formula (7);

[0037] In formula (7), and are the input and output image grayscale values, Represents the mean square error between the calculated input and output values.

[0038] Preferably, the process of calibrating the measurement system in step 3 can be specifically described as follows:

[0039] The inverse camera method is used to calibrate the extrinsic parameter matrix and distortion parameters of the camera and projector;

[0040] Place a black and white checkerboard calibration plate in the camera's field of view, project longitudinal and axial fringe patterns onto the calibration plate with a projector, and collect images synchronously with the camera;

[0041] Place the calibration plate at different spatial locations and perform multiple calibrations to complete the calibration process of the measurement system.

[0042] Preferably, the process of calibrating the measurement system in step 3 satisfies:

[0043] Formula (8);

[0044] In formula (8), is the scale factor, is the coordinate of the principal point of the image taken by the camera, is the equivalent focal length of the camera in the horizontal and vertical directions, is the rotation matrix, is the translation matrix; is the three-dimensional coordinates of the corner points of the chessboard calibration plate in the world coordinate system, The pixel coordinates of the corner points of the chessboard calibration plate on the camera;

[0045] During the calibration process, the coordinates of the corner points of the chessboard on the projector satisfy:

[0046] Formula (9);

[0047] In formula (9), Indicates the camera pixel coordinates are The absolute phase value of is the stripe period.

[0048] Preferably, the high dynamic range fringe pattern of the object to be measured obtained in step 4 satisfies:

[0049] Formula (10);

[0050] In formula (10), is the light intensity information, is the background light intensity, is the modulation amplitude, represents the wrapping phase, Represents the phase shift magnitude;

[0051] Among them, the wrapping phase ,satisfy:

[0052] Formula (11);

[0053] In formula (11), Indicates The light intensity information of an image.

[0054] Preferably, the process of calculating the absolute phase of the object to be measured in step 4 can be specifically described as:

[0055] The absolute phase of the object to be measured ,satisfy:

[0056] Formula (14);

[0057] in, satisfy:

[0058] Formula (13);

[0059] In formula (13), The fringe frequency is The fringe order at is the rounding function, and The fringe frequency is and The stripe period of time;

[0060] Package Phase Difference ,satisfy:

[0061] Formula (12);

[0062] In formula (12), The representative frequency is and Wrapping phase difference of fringe pattern.

[0063] The invention provides a grating high dynamic range measurement method based on a double-exposure branch fusion network. The grating high dynamic range measurement method specifically includes the following steps: step one: constructing a measurement system required by the grating high dynamic range measurement method, and collecting fringe patterns with different exposure times; using an image fusion algorithm to synthesize a grating high dynamic range fringe pattern, and using it as a training true value of the double-exposure branch fusion network; step two: constructing and training the double-exposure branch fusion network; step three: calibrating the measurement system constructed in step one; step four: based on the double-exposure branch fusion network trained in step two, obtaining a high dynamic range fringe pattern of an object to be measured, and calculating the absolute phase of the object to be measured; according to the absolute phase of the object to be measured and the calibration parameters of the measurement system, traversing all coordinate points of the object to be measured, and reconstructing the three-dimensional morphology of the object to be measured.

[0064] Compared with the prior art, the grating high dynamic range measurement method based on a double exposure branch fusion network with the above-mentioned step characteristics constructs a double exposure branch fusion network and takes the collected fringe patterns with different exposure times as input; then, the high dynamic range fringe pattern of the object to be measured is predicted by the double exposure branch fusion network, thereby realizing high dynamic range three-dimensional measurement of the object to be measured and significantly improving the efficiency and accuracy of the measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the following drawings:

[0066] Figure 1 A schematic flow chart of a grating high dynamic range measurement method based on a double-exposure branch fusion network provided by the present invention;

[0067] Figure 2 Schematic diagram of the structural architecture of the double-exposure branch fusion network constructed in step 2;

[0068] Figure 3 Schematic diagram of the training process of the double-exposure branch fusion network;

[0069] Figure 4a is a schematic diagram of the over-dark fringe pattern;

[0070] Figure 4b is a schematic diagram of an overexposed fringe pattern;

[0071] Figure 4c A schematic diagram of a high dynamic range fringe image predicted by a double-exposure branch fusion network provided by the present invention;

[0072] Figure 5a for Figure 4aMeasurement results in the dark fringe pattern scene;

[0073] Figure 5b for Figure 4b Measurement results in an overexposed fringe pattern scene;

[0074] Figure 5c for Figure 4c The measurement results in the scenario of the grating high dynamic range measurement method based on the double exposure branch fusion network provided by the present invention. DETAILED DESCRIPTION

[0075] The present invention provides a grating high dynamic range measurement method based on a double-exposure branch fusion network. The grating high dynamic range measurement method reconstructs a high dynamic range fringe pattern of an object to be measured through a double-exposure branch fusion network, thereby realizing fast and accurate high dynamic range three-dimensional measurement of the object to be measured.

[0076] The present invention provides a grating high dynamic range measurement method based on a double exposure branch fusion network, such as Figure 1 As shown, the following steps are included:

[0077] Step 1: Construct the measurement system required for the grating high dynamic range measurement method and collect fringe patterns with different exposure times; the measurement system consists of a camera and a projector.

[0078] The image fusion algorithm is used to synthesize the grating high dynamic range fringe image, which is used as the training truth value of the double exposure branch fusion network.

[0079] It should be noted that the camera in the measurement system is specifically used to collect images of the fringe pattern, and the projector is specifically used to project the fringes in the fringe pattern. Here, in order to facilitate the technical personnel in the field to understand this solution, a grating high dynamic range fringe pattern measurement system built under laboratory conditions is selected, which includes a camera and a projector. The frame rate of the camera is selected as 227fps, the resolution is 1440×1080 pixels, the focal length of the lens is 12mm, the resolution of the projector is selected as 1280×720 pixels, the wavelength of the projection light is 450nm, and the objects to be measured are selected as three objects with different reflectivity: plaster portraits, standard balls and welds.

[0080] The specific implementation process is further described in detail below. In the process of collecting fringe patterns with different exposure times, the camera collects fringe patterns at different exposure times of 2000μs, 5000μs, 10000μs, 15000μs, 25000μs, and 40000μs respectively; then, through the image fusion algorithm, the grating high dynamic range fringe pattern is synthesized. Finally, the technicians select images with exposure times of 2000μs and 15000μs as the input of the training truth value of the double exposure branch fusion network. , , a total of 200 sets of different scenes were shot for network training.

[0081] Further, as a more preferred implementation of the present invention, the training truth value of the double-exposure branch fusion network in step 1 satisfies: Formula (1);

[0082] In formula (1), Indicates the pixel coordinates in the double exposure branch fusion network training as The true gray value of is the number of fringe images to be fused, , and Respectively The fusion weight factors of brightness, contrast and saturation in the fringe image, For the The pixel coordinates in the fringe image are The measured gray value.

[0083] Step 2: Build and train the double-exposure branch fusion network.

[0084] After completing step 1, further implement step 2. It is worth noting that, refer to Figure 2 As shown in FIG. 1 , the double-exposure branch fusion network constructed in step 2 consists of three parts: an encoding layer, a fusion layer, and a decoding layer.

[0085] The encoding layer includes a convolutional layer, a normalization layer, a residual connection module DBF-Block and a channel attention mechanism module ECA, which is used to extract fringe patterns with different exposure times from the training truth value of the double exposure branch fusion network. , As a multi-channel feature map input, it is processed by the Relu activation function to obtain the fusion layer input image , .

[0086] The fusion layer is used to input the image to the fusion layer , Perform adaptive fusion processing to obtain the fusion layer output image Among them, as a more preferred embodiment of the present invention, the fusion layer outputs the image to the coding layer , Perform adaptive fusion processing to obtain the fusion layer output image The process can be specifically described as:

[0087] Output image to the encoding layer , After normalization, mask calculation is performed;

[0088] Among them, the mask calculation process satisfies: Formula (2);

[0089] In formula (2), Indicates the network The pixel coordinates in the mask image are The gray value of For the The pixel coordinates in the output image of the coding layer are The gray value of is the exposure factor, is the image standard deviation, is a natural exponential function.

[0090] In addition, further Gaussian filtering is performed. The purpose of performing Gaussian filtering is to enhance the sinusoidal property in the fringe pattern fusion process, thereby improving the quality of phase solution.

[0091] Specifically, the Gaussian filtering process satisfies: Formula (3);

[0092] In formula (3), The pixel coordinates after Gaussian filtering are The gray value of is the standard deviation of the Gaussian function;

[0093] Finally, feature fusion is performed to obtain the fusion layer output image ; Among them, the fusion layer outputs the image ,satisfy: Formula (4).

[0094] The decoding layer is used to output the image of the fusion layer through the loss function Perform iterative solution to obtain the final image One point that needs to be added is that in order to ensure that the final image obtained by iterative solution of the decoding layer This is the optimal result of the double exposure branch fusion network convergence. Here, the following parameters are selected: Using the Adam optimizer, the double exposure branch fusion network training rounds are 300, and the learning rate is set to automatic update. The learning rate is multiplied by 0.1 every 100 training rounds, the initial learning rate is set to 0.01, and the training batch size is 8. Figure 3 , Attachment Figure 4a , Figure 4b , Figure 4c As shown. Among them, Figure 3 Schematic diagram of the training process of the double-exposure branch fusion network. Figure 4ais a schematic diagram of the dark fringe diagram. Figure 4b is a schematic diagram of an overexposed fringe pattern. Figure 4c It is a schematic diagram of a high dynamic range fringe image predicted by the double-exposure branch fusion network provided by the present invention.

[0095] In addition, as a more preferred embodiment of the present invention, in order to improve the final image The dynamic range of the fusion layer output image is Perform iterative solution to obtain the final image The process also includes the following steps:

[0096] Select a weighted combination loss function; the weighted combination loss function is used to improve the final image Dynamic range;

[0097] Among them, the weighted combination loss function ,satisfy: Formula (5);

[0098] In formula (5), Represents the input image With the output image The gray value difference is is the weight factor of the loss function;

[0099] Represents the structural similarity loss function, satisfying:

[0100] Formula (6);

[0101] In formula (6), and are the input and output image grayscale values, and , and They are and The mean and standard deviation of and is a constant;

[0102] Represents the pixel mean square error loss function, which is used to enhance the texture features of the fused image and satisfies: Formula (7);

[0103] In formula (7), and are the input and output image grayscale values, Represents the mean square error between the calculated input and output values.

[0104] Step 3: Calibrate the measurement system constructed in step 1.

[0105] On the basis of completing step 2, further implement step 3. As a preferred implementation of the present invention, the process of calibrating the measurement system in step 3 can be specifically described as:

[0106] The inverse camera method is used to calibrate the extrinsic parameter matrix and distortion parameters of the camera and projector;

[0107] Place a black and white checkerboard calibration plate in the camera's field of view, project longitudinal and axial fringe patterns onto the calibration plate with a projector, and collect images synchronously with the camera;

[0108] Place the calibration plate at different spatial locations and perform multiple calibrations to complete the calibration process of the measurement system.

[0109] It is worth noting that the above process of calibrating the measurement system satisfies:

[0110] Formula (8);

[0111] In formula (8), is the scale factor, is the coordinate of the principal point of the image taken by the camera, is the equivalent focal length of the camera in the horizontal and vertical directions, is the rotation matrix, is the translation matrix; is the three-dimensional coordinates of the corner points of the chessboard calibration plate in the world coordinate system, The pixel coordinates of the corner points of the chessboard calibration plate on the camera;

[0112] During the calibration process, the coordinates of the corner points of the chessboard on the projector satisfy:

[0113] Formula (9);

[0114] In formula (9), Indicates the camera pixel coordinates are The absolute phase value of is the stripe period.

[0115] The following specific embodiment is further provided to explain the calibration process of the measurement system in step 3. Specifically, the calculation method of 12-step phase shift and complementary Gray code is used to solve the phase in the horizontal and vertical directions of the calibration plate. Among them, the stripes and frequency are selected as 80 and 45 respectively; the external parameter matrix and distortion parameters of the camera and projector obtained through this calibration process can be referred to the following table.

[0116]

[0117] Step 4: Based on the double-exposure branch fusion network trained in step 2, obtain the high dynamic range fringe pattern of the object to be measured, and calculate the absolute phase of the object to be measured. According to the absolute phase of the object to be measured and the calibration parameters of the measurement system, traverse all the coordinate points of the object to be measured to reconstruct the three-dimensional shape of the object to be measured.

[0118] On the basis of completing step 3, further implement step 4. Specifically, combined with the calibration parameters of the measurement system obtained in step 3, the three-dimensional shape of the object to be measured is measured and reconstructed. Set to 127, the modulation amplitude is 128, Take 0.25 to generate three different frequencies of sinusoidal fringes: 70, 64, and 59. The camera captures the fringe pattern with exposure time of 2000μs and 15000μs and inputs it into the double exposure branch fusion network to predict the fused image. , process different frequency images in turn, and finally get 12 grating high dynamic range fringe images, and calculate the absolute phase map of the fused image. Based on the calibration parameters of the measurement system, the PMP measurement principle is used to traverse all pixel points in turn to obtain the three-dimensional morphology (point cloud) of the object to be measured.

[0119] As a preferred embodiment of the present invention, the high dynamic range fringe pattern of the object to be measured obtained in step 4 satisfies:

[0120] Formula (10);

[0121] In formula (10), is the light intensity information, is the background light intensity, is the modulation amplitude, represents the wrapping phase, Represents the phase shift magnitude;

[0122] Among them, the wrapping phase ,satisfy:

[0123] Formula (11);

[0124] In formula (11), Indicates The light intensity information of an image.

[0125] In addition, as another preferred embodiment of the present invention, the process of calculating the absolute phase of the object to be measured in step 4 can be specifically described as:

[0126] The absolute phase of the object to be measured ,satisfy:

[0127] Formula (14);

[0128] in, satisfy:

[0129] Formula (13);

[0130] In formula (13), The fringe frequency is The fringe order at is the rounding function, and The fringe frequency is and The stripe period of time;

[0131] Package Phase Difference ,satisfy:

[0132] Formula (12);

[0133] In formula (12), The representative frequency is and Wrapping phase difference of fringe pattern.

[0134] Finally, in order to verify that the grating high dynamic range measurement method based on the double-exposure branch fusion network provided by the present invention is authentic and effective, the same camera is used to measure three objects with different reflectivity, namely, a plaster figure, a standard ball and a weld, as a control. Figure 4a Schematic diagram of the dark fringe diagram. Figure 4b Schematic diagram of overexposed fringe pattern. Figure 4c The schematic diagram of the high dynamic range fringe pattern predicted by the double exposure branch fusion network provided by the present invention is used as a reference, and the measurement results of the dark fringe pattern scene are statistically compared (refer to Figure 5a ), measurement results in the overexposed fringe pattern scene (refer to Figure 5b ) and the measurement results of the grating high dynamic range measurement method based on the double exposure branch fusion network provided by the present invention (refer to Figure 5c ). Experimental verification shows that the present invention provides a grating high dynamic range measurement method based on a double-exposure branch fusion network, and the predicted phase root mean square error (RMS) is 3.13 and the mean absolute error (PMAE) is 0.28.

[0135] At this point, technicians can find that the present invention provides a grating high dynamic range measurement method based on a double-exposure branch fusion network, and its reconstruction result is better.

[0136] The invention provides a grating high dynamic range measurement method based on a double-exposure branch fusion network. The grating high dynamic range measurement method specifically includes the following steps: step one: constructing a measurement system required by the grating high dynamic range measurement method, and collecting fringe patterns with different exposure times; using an image fusion algorithm to synthesize a grating high dynamic range fringe pattern, and using it as a training true value of the double-exposure branch fusion network; step two: constructing and training the double-exposure branch fusion network; step three: calibrating the measurement system constructed in step one; step four: based on the double-exposure branch fusion network trained in step two, obtaining a high dynamic range fringe pattern of an object to be measured, and calculating the absolute phase of the object to be measured; according to the absolute phase of the object to be measured and the calibration parameters of the measurement system, traversing all coordinate points of the object to be measured, and reconstructing the three-dimensional morphology of the object to be measured.

[0137] Compared with the prior art, the grating high dynamic range measurement method based on a double exposure branch fusion network with the above-mentioned step characteristics constructs a double exposure branch fusion network and takes the collected fringe patterns with different exposure times as input; then, the high dynamic range fringe pattern of the object to be measured is predicted by the double exposure branch fusion network, thereby realizing high dynamic range three-dimensional measurement of the object to be measured and significantly improving the efficiency and accuracy of the measurement.

[0138] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A grating high dynamic range measurement method based on a double-exposure branch fusion network, characterized in that: The steps include: Step 1: construct a measurement system required for the grating high dynamic range measurement method, and collect fringe patterns with different exposure times; wherein the measurement system is composed of a camera and a projector; Using the image fusion algorithm, the grating high dynamic range fringe image is synthesized and used as the training truth value of the double exposure branch fusion network; Step 2: Build and train a double-exposure branch fusion network; Step 3: Calibrate the measurement system constructed in step 1; Step 4: Based on the double-exposure branch fusion network trained in step 2, a high dynamic range fringe image of the object to be measured is obtained, and the absolute phase of the object to be measured is calculated; According to the absolute phase of the object to be measured and the calibration parameters of the measurement system, all coordinate points of the object to be measured are traversed to reconstruct the three-dimensional shape of the object to be measured; The double-exposure branch fusion network constructed in step 2 consists of a coding layer, a fusion layer and a decoding layer; The encoding layer includes a convolutional layer, a normalization layer, a residual connection module DBF-Block and a channel attention mechanism module ECA, which is used to extract fringe patterns with different exposure times from the training truth value of the double exposure branch fusion network. , As a multi-channel feature map input, it is processed by the Relu activation function to obtain the fusion layer input image , ; The fusion layer is used to input the image to the fusion layer , Perform adaptive fusion processing to obtain the fusion layer output image ; The decoding layer is used to output the image of the fusion layer through the loss function Perform iterative solution to obtain the final image .

2. The grating high dynamic range measurement method based on a double-exposure branch fusion network according to claim 1, characterized in that: The training truth value of the double-exposure branch fusion network in step 1 satisfies: Formula (1); In formula (1), Indicates the pixel coordinates in the double exposure branch fusion network training as The true gray value of is the number of fringe images to be fused, , and Respectively The fusion weight factors of brightness, contrast and saturation in the fringe image, For the The pixel coordinates in the fringe image are The measured gray value.

3. The grating high dynamic range measurement method based on double exposure branch fusion network according to claim 1, characterized in that: The fusion layer outputs the image to the encoding layer , Perform adaptive fusion processing to obtain the fusion layer output image The process can be specifically described as: Output image to the encoding layer , After normalization, mask calculation is performed; Among them, the mask calculation process satisfies: Formula (2); In formula (2), Indicates the network The pixel coordinates in the mask image are The gray value of For the The pixel coordinates in the output image of the coding layer are The gray value of is the exposure factor, is the image standard deviation, is the natural exponential function; right Perform Gaussian filtering; Among them, the Gaussian filtering process satisfies: Formula (3); In formula (3), The pixel coordinates after Gaussian filtering are The gray value of is the standard deviation of the Gaussian function; Feature fusion to obtain the fusion layer output image ; Among them, the fusion layer outputs the image ,satisfy: Formula (4).

4. The method for measuring grating high dynamic range based on double exposure branch fusion network according to claim 1, characterized in that: The decoding layer outputs the image of the fusion layer through the loss function Perform iterative solution to obtain the final image The process also includes the following steps: Select a weighted combination loss function; the weighted combination loss function is used to improve the final image Dynamic range; Among them, the weighted combination loss function ,satisfy: Formula (5); In formula (5), Represents the input image With the output image The gray value difference is is the weight factor of the loss function; Represents the structural similarity loss function, satisfying: Formula (6); In formula (6), and are the input and output image grayscale values, and , and They are and The mean and standard deviation of and is a constant; Represents the pixel mean square error loss function, which is used to enhance the texture features of the fused image and satisfies: Formula (7); In its formula (7), and are the input and output image grayscale values, Represents the mean square error between the calculated input and output values.

5. The method for measuring grating high dynamic range based on double exposure branch fusion network according to claim 1, characterized in that: The process of calibrating the measurement system in step 3 can be specifically described as: The inverse camera method is used to calibrate the extrinsic parameter matrix and distortion parameters of the camera and projector; Place a black and white checkerboard calibration plate in the camera's field of view, project longitudinal and axial fringe patterns onto the calibration plate with a projector, and collect images synchronously with the camera; Place the calibration plate at different spatial locations and perform multiple calibrations to complete the calibration process of the measurement system.

6. The method for measuring grating high dynamic range based on double exposure branch fusion network according to claim 5, characterized in that: The process of calibrating the measurement system in step 3 satisfies: Formula (8); In formula (8), is the scale factor, is the coordinate of the principal point of the image taken by the camera, is the equivalent focal length of the camera in the horizontal and vertical directions, is the rotation matrix, is the translation matrix; is the three-dimensional coordinates of the corner points of the chessboard calibration plate in the world coordinate system, The pixel coordinates of the corner points of the chessboard calibration plate on the camera; During the calibration process, the coordinates of the corner points of the chessboard on the projector satisfy: Formula (9); In formula (9), Indicates the camera pixel coordinates are The absolute phase value of is the stripe period.

7. The method for measuring grating high dynamic range based on double exposure branch fusion network according to claim 1, characterized in that: The high dynamic range fringe pattern of the object to be measured obtained in step 4 satisfies: Formula (10); In formula (10), is the light intensity information, is the background light intensity, is the modulation amplitude, represents the wrapping phase, Represents the phase shift magnitude; Among them, the wrapping phase ,satisfy: Formula (11); In formula (11), Indicates The light intensity information of an image.

8. The method for measuring grating high dynamic range based on double exposure branch fusion network according to claim 1, characterized in that: The process of calculating the absolute phase of the object to be measured in step 4 can be specifically described as: The absolute phase of the object to be measured ,satisfy: Formula (14); in, satisfy: Formula (13); In formula (13), The fringe frequency is The fringe order at is the rounding function, and The fringe frequency is and The stripe period of time; Package Phase Difference ,satisfy: Formula (12); In formula (12), The representative frequency is and Wrapping phase difference of fringe pattern.

Images