A method for imaging with high dynamic range

Abstract

The present invention proposes an imaging method with a high dynamic range, which relates to the technical field of high dynamic range imaging. Under high dynamic lighting, a target uses a dedicated optical path system and obtains high dynamic images at the same time based on parallel arranged image acquisition devices, thereby ensuring the consistency of light path difference and improving the frame rate of imaging. The acquired images are then scaled, registered and fused in parallel to achieve high dynamic imaging, form photos with rich features and clear imaging, locate the position of the target within the field of view, and provide a basis for distinguishing the welding direction trajectory at the next moment of industrial welding.

Description

Technical Field

[0001] The present invention relates to the technical field of high dynamic range imaging, and more particularly to an imaging method with a high dynamic range. Background Art

[0002] Dynamic Range is the ratio of the maximum value to the minimum value of a variable signal (such as a light signal or a sound signal). In computer graphics and related industries, especially those involving movies, games, simulations, virtual reality, and military applications, it is required to display more realistic images, that is, synthetic images that more accurately match the real scene to be represented. In the industrial world, such as the welding trajectory of large components, higher requirements (more than 100db) are also put forward for the traditional visual dynamic imaging range (40-60db). Therefore, the application of visual imaging technology to the industrial world requires a larger imaging dynamic range threshold to adapt to different imaging scenarios.

[0003] However, under normal lighting conditions, when conventional imaging equipment is used for imaging, there are two situations: (a) when the bright target is imaged clearly, the target in the dark area cannot be acquired; (b) when the dark target is imaged clearly, the image of the bright target is relatively blurry. The above two situations make it impossible for high-dynamic imaging technology to obtain relatively complete target imaging under high-dynamic lighting conditions, especially in industrial welding applications, and the imaging quality is poor and inaccurate.

[0004] The prior art discloses an imaging method with a high dynamic range. Generally, a low dynamic range image taken by an ordinary camera and an image taken by a neuromorphic camera (later reconstructed into a high dynamic range grayscale image) are input into a trained neural network to complete a fusion imaging operation. However, this method of first acquiring multiple low dynamic range images with different exposure parameters of the same scene and then generating a high dynamic range image according to a certain fusion rule from the acquired multiple exposure images is relatively time-consuming and difficult to achieve real-time high dynamic imaging. In addition, the parameters are variable during registration and fusion, the amount of calculation is large, and the accuracy is not high. It is difficult to accurately measure the spatial and temporal connections between different images, which is not conducive to further guidance and application in the actual welding field in the industry after imaging. Summary of the invention

[0005] In order to solve the problems that traditional high dynamic range imaging methods are time-consuming and the spatial and temporal connections between images are difficult to measure, the present invention proposes an imaging method with a high dynamic range, which has high imaging quality and is easy to obtain effective image information. It provides a full-information stable imaging solution for high dynamic imaging fields such as welding under strong arc light, surface targets under strong light, and backlight imaging.

[0006] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0007] A method for imaging with a high dynamic range comprises the following steps:

[0008] S1. Build an optical path system for high dynamic illumination light propagation, and arrange α image acquisition devices in parallel along the propagation of light in the optical path system;

[0009] S2. Place a target calibration plate on the plane to be measured, change the position of the target calibration plate, use high dynamic light to illuminate the target calibration plate on the plane to be measured, and the optical path system divides the calibration plate under high dynamic light into α imaging light paths, which are then incident on α image acquisition devices;

[0010] S3. Using α image acquisition devices to respectively acquire images of the target calibration plate, obtain α target calibration plate images, calibrate the intrinsic and extrinsic parameters of the α image acquisition devices based on the α target calibration plate images, and determine the intrinsic and extrinsic parameters of the α image acquisition devices;

[0011] S4. Perform feature extraction on α target calibration plate images, extract the image region of interest, and output the feature point set of the image region of interest;

[0012] S5. Based on the intrinsic and extrinsic parameters of α image acquisition devices and the feature point set of the image region of interest, the scale transformation of α target calibration plate images at the same time is performed using bilinear interpolation method;

[0013] S6. registering the scale-transformed α target calibration plate images;

[0014] S7. Perform parallel fusion on the two-dimensional arrays corresponding to the α registered target calibration plate images to obtain an array corresponding to the fused full-frame image, and then locate the target within the field of view.

[0015] Preferably, in step S1, the constructed optical path system includes a filter, an α prism arranged directly above the filter and having a bottom area matching the area of ​​the filter, and a plane mirror, where α is an integer not less than 3; the plane mirrors have α planes in common, one end of each plane mirror stands on the plane where the filter is located, and each plane mirror is tilted toward the other end along one end located on the plane where the filter is located, and the α plane mirror surrounds the α prism, and the α mirror surfaces respectively face one facet of the α prism. The high dynamic illumination light is incident from the filter and then incident on the α prism through the filter. The α prism splits the incident light into α paths of light on the same plane, which respectively enter the α plane mirrors, and the α plane mirrors respectively reflect the α paths of light.

[0016] Here, when high-dynamic light enters the filter of the optical path system from the plane to be measured, the light beam will be separated and enter the α prism with a certain wavelength band. The α prism divides the incident light into α paths on the same plane. The α path light enters the plane mirror respectively, and the plane mirror uses the light as the incident light of the image acquisition device. At this time, the light encoder on the image acquisition device senses the propagation of the light and transmits the generated signal to the processor end for control. After the processing end analyzes the signal, it issues a capture command to the image acquisition device to perform image capture. The image acquisition device can obtain images in the same field of view at the same time, reducing image errors caused by timing. The optical path system ensures that the optical path difference of the light is consistent, and the α image acquisition devices are arranged in parallel, with little motion interference, reducing subsequent calculation workload, and improving the frame rate of imaging.

[0017] Preferably, in step S2, the target calibration plate is a checkerboard calibration plate with a size of (2n+1)*(2n+1), wherein n represents the number of checkerboard grids, and the position of the target calibration plate can be changed 5 to 6 times within the six degrees of freedom of the field of view of α image acquisition devices, and each time the position of the target calibration plate is changed, the α image acquisition devices simultaneously acquire images of the target calibration plate 1 to 2 times. By adjusting different positions of the target calibration plate, the image acquisition device can acquire images of the calibration target.

[0018] Preferably, based on α target calibration plate images, the Zhang Zhengyou calibration method is used to calibrate the intrinsic parameters and extrinsic parameters of the α image acquisition devices, and the intrinsic parameters of the α image acquisition devices determined after calibration are respectively expressed as {K1, K2, ..., K α}, the external parameters are represented as {C1, C2, ..., C α}.

[0019] Here, since the external parameters of the image acquisition device will change when the position of the target calibration plate changes, the external parameters of the image acquisition device need to be calibrated separately, but the change of the external parameters does not affect the determination of the properties of the image acquisition device itself.

[0020] Preferably, in step S4, the image regions of interest extracted from the α target calibration plate images are represented as {I1, I2, ..., I α}, in the process of extracting the image region of interest from each target calibration plate image, for each target calibration plate image, when the gray value of a pixel is less than the gray mean square error of the target calibration plate image, the gray value of the pixel defaults to 0; otherwise, the gray value of the pixel remains unchanged, and the process satisfies the expression:

[0021]

[0022] Where f(x,y) represents the grayscale value of a pixel, E(∑(f(x,y)-I)) 2 Represents the grayscale mean square error of the target calibration plate image, represents the average gray value of the target calibration plate image, and u*v represents the pixel of the target calibration plate image;

[0023] The process of outputting the feature point set of the image area of ​​interest is:

[0024] The SIFT method is used to find the extreme points in the region of interest of each target calibration plate image. Each pixel in the region of interest of a target calibration plate image is traversed, and the grayscale value of the pixel is compared with all its adjacent points. When its grayscale value is greater than or less than the grayscale values ​​of all adjacent points in the image domain and scale domain, it is regarded as a maximum point; and each pixel is traversed and searched in the scale space and two-dimensional image space to ensure that the extreme points are fully detected. Then the feature point sets of the region of interest of each target calibration plate image are {P1, P2, ..., P α}.

[0025] Here, before extracting the image region of interest, the image is also preprocessed to remove redundant information.

[0026] Preferably, let the internal parameters {K1, K2, ..., K α} is represented by {(u1, v1), (u2, v2), ..., (u α , v α )}, the image size is represented as {(U1, V1), (U2, V2), ..., (U α , V α )},U1≠U2,...,≠U α , V1≠V2,...,≠V α ;

[0027] Assume that the target image is based on the least common multiple of the resolution of the α-frame target calibration plate images at the same time, that is:

[0028] k u =〔u1,u2,...,u α 〕, k v =〔v1,v2,...,v α 〕;

[0029] Among them, k u represents the least common multiple of the long sides of the α-frame target calibration plate images, k v Represents the least common multiple of the wide edges in the α-width target calibration plate image;

[0030] Assume that the pixels occupied by the target calibration plate image after size transformation are represented as {(u1', v1'), (u2', v2'), ..., (u α ',v α ')}, the image size is expressed as {(U 1-1 , V 1-1 )、(U 2-1 , V 2-1 ),...,(U α-1 , V α-1 )}, the target calibration plate image pixels after scale transformation and the target calibration plate image pixels before scale transformation have the following relationship:

[0031] u1'=u2'=....=u α '=k u

[0032] v1'=v2'=...=v α '=k v

[0033] For any target calibration plate image λ in the α target calibration plate images, λ=1,2,...,α; the size of the target calibration plate image after the scale transformation and the size of the target calibration plate image before the scale transformation have the following relationship:

[0034] U λ-1 =U λ *(k u / u λ ), V λ-1 =V λ *(k v / v λ ).

[0035] Here, since the resolution of each image acquisition device is different, the size of the image captured by each image acquisition device is inconsistent, and the image pixels occupied by the target calibration plate are inconsistent. In order to ensure the accuracy of subsequent image registration, the α-frame target calibration plate images are uniformly resized at the same resolution.

[0036] Preferably, the pixel size occupied by each checkerboard grid in the α-frame target calibration plate image is expressed as {(u1 / n,v1 / n), (u2 / n,v2 / n), ..., (u α / n,v α / n)}, and use bilinear interpolation to scale the α target calibration images at the same time, so that the number of pixels occupied by each chessboard grid in the α target calibration images after size transformation is consistent. The number of pixels occupied by a grid on the target image is (k u / (2n+1),k v / (2n + 1)). After image scale transformation, the pixel change of a single grid corresponding to α target calibration plate images is (k u / (2n + 1), k v / (2n + 1)). For any target calibration plate image λ in the α target calibration plate images, where λ = 1, 2,..., α, the difference between the pixels of a single grid and the pixels occupied by a grid on the target image is |u λ -u λ '|. Then, |u λ -u λ '| times of linear interpolation need to be performed on the target calibration plate image λ to obtain the α target calibration plate images after scale transformation. The process is as follows:

[0037] S51. Define the upper left corner of the target calibration plate image λ and the target image as the coordinate origin, and take the coordinate Q of the point on the target calibration plate image λ 11 (x1, y1), Q 12 (x2, y2), Q 21 (x3, y3), Q 22 (x4, y4) of the gray value function f, and Q 11 , Q 12 , Q 21 , Q 22 , R1, R2, Z ∈ P λ , P λ represents the set of feature points output from the region of interest of the target calibration plate image λ;

[0038] S52. Perform linear interpolation in the x direction, and the relationship is as follows:

[0039]

[0040]

[0041] Then perform interpolation in the y direction, and the relationship is as follows:

[0042]

[0043] S53. For a point T(x, y) on the target calibration plate image λ, where x1 < x < x2, y1 < y < y2, its gray value f(x, y) is obtained by the following formula:

[0044]

[0045] S54. Align the geometric centers between the interpolated target calibration plate image λ and the target image:

[0046] It is known that the side length ratios between the target calibration plate image λ and the target image are respectively Assume that the pixel of a certain point on the target calibration plate image λ is (i, j), and the pixel point with the coordinates of the point (i', j') on the target image is mapped to the pixel point on the target image through the side length ratio, that is, the relationship between the pixel point on the image to be converted and the pixel point on the original image after the scale transformation is as follows:

[0047]

[0048]

[0049] At this point, the size transformation between the target calibration plate image λ and the target image is unified. The bilinear interpolation method is used to transform the image scale of the heterogeneous image to ensure that the pixel accuracy of the target calibration plate image is consistent at the same time.

[0050] Preferably, in step S6, when registering the scale-transformed α target calibration plate images, a registration model between the scale-transformed α target calibration plate images is established based on the characteristic base point identification and space transformation, so as to transform them into the same coordinate system;

[0051] Assume that the image plane coordinate system of any image acquisition device λ among the α image acquisition devices is O λ -X λ Y λ , the image plane coordinate system of any image acquisition device η is O η -X η Y η , the coordinate transformation relationship between the image acquisition device λ and the image acquisition device η is given by (R λη , T λη ) indicates that R λη represents the rotation matrix between the image plane coordinate system of the image acquisition device λ and the image plane coordinate system of the image acquisition device η, T λη represents the translation matrix between the image plane coordinate system of the image acquisition device λ and the image plane coordinate system of the image acquisition device η; T λ , T η Respectively represent the translation matrices of the image acquisition device λ and the image acquisition device η, R λ , R η denote the rotation matrices of the image acquisition device λ and the image acquisition device η respectively;

[0052] Among them, the mapping relationship between space points and plane coordinates and the transformation relationship between different imaging coordinate systems satisfy:

[0053] For a point Q on the surface of the actual object to be measured, its world coordinate is Q w , the coordinates of point Q in the image plane coordinate system of the image acquisition device λ and the image acquisition device η are expressed as:

[0054] Qλ =R λ Q w +T λ

[0055] Q η =R η Q w +T η

[0056] The image acquisition device λ is set as the target coordinate system, then:

[0057] Q λ =R λη Q η +T λη

[0058] in,

[0059] The feature point sets of the regions of interest of the target calibration plate image λ and the target calibration plate image η acquired by the image acquisition device λ and the image acquisition device η are P respectively. λ With P η , and its transformation relationship is expressed as:

[0060] P η (x,y)=kR λη Pη(x,y)+T λη .

[0061] Here, the purpose of registration is to transform the target plate image to the same coordinate system. The matching points on the image coordinate system can be transformed to overcome the error caused by the installation of the imaging device and ensure the imaging quality.

[0062] Preferably, registering the scale-transformed α target calibration plate images further comprises:

[0063] Establish chessboard coordinates, obtain the coordinates of the four outer vertices of the target calibration plate under the transformed posture, and after obtaining the coordinates of the four outer vertices, establish the local coordinate system of the outer vertices, so as to establish the correspondence between the outer vertices of the target calibration plate image and the target calibration plate encoding, and according to the correspondence between the world coordinate number of the inner corner point of the chessboard and the matrix coordinate system number of the inner corner point, achieve the matching of the target calibration plate image landmark points, and obtain the degree of registration between the α target calibration plate images after scale transformation, the corresponding discrete signal of the image to be tested is expanded, and the registration reference point is in the same position in the corresponding two-dimensional array of the expanded full-frame image. This process can improve the accuracy of feature point matching between images, and coordinates can be adaptively obtained for simple multi-deformation vertices.

[0064] Preferably, the target calibration plate image λ acquired by the image acquisition device λ in the α image acquisition devices is transformed into image a after size conversion and registration, and the target calibration plate image η acquired by the image acquisition device η is transformed into image b after size conversion and registration, and the discrete feature two-dimensional point sets of image a and image b are P1, P2, f respectively. A , f B are the grayscale values ​​of image a and image b respectively, then:

[0065]

[0066]

[0067] Among them, i = 1, 2, 3, 4, 5..., n, m are pixel coordinate values

[0068] Assume that the effective feature space of image a is F A , the effective feature space of image b is F B , the feature space of the target image is Ω, for any sample ξ∈Ω, and the two feature vectors corresponding to ξ are g∈F A ,β∈F B , the parallel fusion formula is defined as:

[0069] ξ=g+β

[0070] Based on the above formula, image a and image b are fused to obtain a two-dimensional array image corresponding to the full frame of the target image; the array calculation is as follows:

[0071]

[0072] In this way, the α sub-images at the same moment can be spliced, and then effective image information can be obtained, providing a basis for determining the welding direction trajectory at the next moment.

[0073] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0074] The present invention proposes an imaging method with a high dynamic range. Under high dynamic lighting, the target uses a dedicated optical path system and obtains high dynamic images at the same time based on parallel arranged image acquisition devices, which ensures the consistency of light path difference and improves the frame rate of imaging. The acquired images are then scaled, registered and fused in parallel to achieve high dynamic imaging, form photos with rich features and clear imaging, and locate the position of the target within the field of view, thereby providing a basis for determining the direction of the travel trajectory of industrial welding at the next moment. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 A flow chart showing a target imaging and positioning method with a high dynamic range proposed in Embodiment 1 of the present invention;

[0076] Figure 2 A structural diagram showing an optical path system constructed in Example 1 of the present invention;

[0077] Figure 3 A schematic diagram showing a process of scaling the target calibration plate image at the same time proposed in Embodiment 2 of the present invention;

[0078] Figure 4 represents the difference transformation diagram proposed in embodiment 2 of the present invention;

[0079] Figure 5 A pixel scale transformation diagram of a single grid of the target calibration plate proposed in Example 2 of the present invention;

[0080] Figure 6 A diagram showing the spatial variation of corner points proposed in Embodiment 3 of the present invention;

[0081] Figure 7 A schematic diagram showing the principle of landmark point matching proposed in Embodiment 3 of the present invention;

[0082] Figure 8 A schematic diagram showing image a proposed in embodiment 4 of the present invention;

[0083] Fig. 9 A schematic diagram showing image b proposed in embodiment 4 of the present invention;

[0084] Fig.10 An image schematic diagram showing the parallel fusion of image a and image b proposed in embodiment 4 of the present invention. DETAILED DESCRIPTION

[0085] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0086] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0087] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0088] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0089] Example 1

[0090] This embodiment proposes a target imaging positioning method with a high dynamic range. The flowchart of the method is shown in FIG. Figure 1 , specifically comprising the following steps: S1. constructing an optical path system for propagation of high dynamic illumination light, and arranging α image acquisition devices in parallel along the propagation of light in the optical path system;

[0091] S2. Place a target calibration plate on the plane to be measured, change the position of the target calibration plate, use high dynamic light to illuminate the target calibration plate on the plane to be measured, and the optical path system divides the calibration plate under high dynamic light into α imaging light paths, which are then incident on α image acquisition devices;

[0092] S3. Using α image acquisition devices to respectively acquire images of the target calibration plate, obtain α target calibration plate images, calibrate the intrinsic and extrinsic parameters of the α image acquisition devices based on the α target calibration plate images, and determine the intrinsic and extrinsic parameters of the α image acquisition devices;

[0093] S4. Perform feature extraction on α target calibration plate images, extract the image region of interest, and output the feature point set of the image region of interest;

[0094] S5. Based on the intrinsic and extrinsic parameters of α image acquisition devices and the feature point set of the image region of interest, the scale transformation of α target calibration plate images at the same time is performed using bilinear interpolation method;

[0095] S6. registering the scale-transformed α target calibration plate images;

[0096] S7. Perform parallel fusion on the two-dimensional arrays corresponding to the α registered target calibration plate images to obtain an array corresponding to the fused full-frame image, and then locate the target within the field of view.

[0097] In specific implementation, the target under high dynamic lighting uses a dedicated optical path system, and based on parallel arranged image acquisition devices, high dynamic images at the same time are obtained, which ensures the consistency of light path difference and improves the frame rate of imaging. The acquired images are then scaled, registered and fused in parallel to achieve high dynamic imaging, forming feature-rich and clear images, thereby locating the position of the target within the field of view. In the industrial field of welding, for welding images, this method provides a basis for determining the direction of the welding trajectory at the next moment.

[0098] In this embodiment, the optical path system is introduced. Figure 2As shown, the optical path system includes a filter 1, an α prism 2 which is arranged directly above the filter 1 and whose bottom area matches the area of ​​the filter 1, and a plane mirror 3, where α is an integer not less than 3; the plane mirrors 3 have α planes in common, one end of each plane mirror 3 stands on the plane where the filter 1 is located, and each plane mirror 3 is tilted toward the other end along one end located on the plane where the filter 1 is located, and the α plane mirror 3 surrounds the α prism 2, and the α mirror surfaces respectively face one prism facet of the α prism 2. The high dynamic illumination light is incident from the filter 1, and then incident on the α prism 2 through the filter 1. The α prism 2 splits the incident light into α paths of light on the same plane, which respectively enter the α plane mirror 3, and the α plane mirror 3 reflects the α paths of light respectively.

[0099] Here, the α prism 2 is a quadrangular prism. When the high dynamic light enters the filter 1 of the optical path system from the plane to be measured, the light beam will be separated and the light of a certain wavelength band will enter the quadrangular prism. The quadrangular prism will divide the incident light into four paths on the same plane, such as Figure 2 As shown, the four rays of light are ①②③④ indicated by arrows, and enter the plane mirror 3 respectively, and the plane mirror 3 uses the rays of light as incident rays of the image acquisition device. Figure 2 Reference numeral 4 indicates an image acquisition device. There are four image acquisition devices, namely an infrared camera, a low-light camera, a large-exposure color camera, and a small-exposure color camera. At this time, the light-sensitive encoder on the image acquisition device senses the propagation of light and transmits the generated signal to the processor for control. After the processor analyzes the signal, it issues an acquisition command to the image acquisition device to acquire images and obtain images under different exposure parameters at the same time, thereby avoiding the influence of timing differences. The optical path system ensures that the optical path difference of the light is consistent, and the four image acquisition devices are arranged in parallel, with small motion interference, reducing the subsequent computing workload and improving the frame rate of imaging. Then, the images taken by all cameras are transmitted to the Gigabit switch, and then transmitted to the processor by the Gigabit switch, that is, transmitted to the image preprocessing subsystem to analyze the images taken by different cameras at a certain time. The four images obtained are scaled, registered, and fused to form photos with rich features and clear imaging, and finally the position of the target within the field of view of the image acquisition device is located.

[0100] Example 2

[0101] In step S2, the target calibration plate is a checkerboard calibration plate with a size of (2n+1)*(2n+1), where n represents the number of checkerboard grids. The position and posture of the target calibration plate can be changed 5 to 6 times within the six degrees of freedom of the field of view of the α image acquisition devices. Each time the position and posture of the target calibration plate is changed, the α image acquisition devices simultaneously acquire images of the target calibration plate 1 to 2 times. By adjusting the different positions and postures of the target calibration plate, the image acquisition device can acquire images of the calibration target. Figure 2In the optical path system structure shown in the figure, the four image acquisition devices are respectively camera I to camera IV. Based on α target calibration plate images, the Zhang Zhengyou calibration method is used to calibrate the intrinsic and extrinsic parameters of the α image acquisition devices. As the position of the target calibration plate changes, the extrinsic parameters of the image acquisition device will change accordingly. Therefore, the extrinsic parameters of the image acquisition device need to be calibrated separately. However, the change of the extrinsic parameters does not affect the determination of the properties of the image acquisition device itself.

[0102] Taking the calibration of camera I as an example, the gradient method based on the change of image pixel grayscale value extracts any four corner points on the calibration chessboard in the camera shooting field of view, and sets the calibration image to be at Z in the world coordinate system. w = 0, the ideal imaging model of the camera can be written as follows:

[0103]

[0104] Among them, s is a non-zero scale transformation factor, γ mainly represents the inclination between the x and y axes of the image, K represents the intrinsic parameter of the camera, and H = [r1, r2, t] T Represents the external parameters of the camera, m = [u, v, 1] T Expressed as pixel coordinate value, M = [X w ,Y w ,Z w , 1] represents the value in the world coordinate system, let H = [h1, h2, h3] = qK[r1, r2, t] T ', then: sm = HM, where r1 T r2=0,||r1||=||r2||=1,then we can get the following internal parameter constraint relation:

[0105]

[0106] Two equations can be obtained for each corner point. Select more than four corner points from the calibration target image, and three of them satisfy the non-collinearity to obtain the matrix H. Therefore, an H matrix can be obtained for each calibration target image, that is:

[0107]

[0108] Can be rewritten as:

[0109]

[0110] Further we can get:

[0111]

[0112] Let h' = [h 11 ,h 12 ,h13 ,h 21 ,h 22 ,h 23 ,h 31 ,h 32 ,h 33 ] T , then we have:

[0113]

[0114] Use the orthogonal matrix constraints of the internal parameters to construct the matrix, that is, let B = C -T C -1 The internal and external parameters of camera I can be obtained. From the above, the internal parameter of camera I can be obtained as K1, and the external parameter is C1. Similarly, according to the above calibration steps, it can be known that the internal parameters of camera II, camera III, and camera IV are K2, K3, and K4, and the external parameters of camera II, camera III, and camera IV are C2, C3, and C4, that is, the internal parameters of the four image acquisition devices determined after calibration are respectively expressed as {K1, K2, K3, K4}, and the external parameters are respectively expressed as {C1, C2, C3, C4}.

[0115] In step S4, the image is preprocessed before extracting the image ROI to remove redundant information. Suppose the image ROI extracted from α target calibration plate images is represented as {I1, I2, ..., I α}, in the process of extracting the image region of interest from each target calibration plate image, for each target calibration plate image, when the gray value of a pixel is less than the gray mean square error of the target calibration plate image, the gray value of the pixel defaults to 0; otherwise, the gray value of the pixel remains unchanged, and the process satisfies the expression:

[0116]

[0117] Where f(x,y) represents the grayscale value of a pixel, E(∑(f(x,y)-I)) 2 Represents the grayscale mean square error of the target calibration plate image, Represents the average grayscale value of the target calibration plate image, u*v represents the pixel of the target calibration plate image; according to the above formula, the output region of interest of image I corresponding to camera I to camera IV is I1, the region of interest of image II is I2, the region of interest of image III is I3, and the region of interest of image IV is I4.

[0118] The process of outputting the feature point set of the image area of ​​interest is:

[0119] The SIFT method is used to find the extreme points in the region of interest of each target calibration plate image. Each pixel in the region of interest of a target calibration plate image is traversed, and the grayscale value of the pixel is compared with all its adjacent points. When its grayscale value is greater than or less than the grayscale values ​​of all adjacent points in the image domain and scale domain, it is regarded as a maximum point; and each pixel is traversed and searched in the scale space and two-dimensional image space to ensure that the extreme points are fully detected. Then the feature point sets of the region of interest of each target calibration plate image are {P1, P2, ..., P α In this embodiment, the target calibration plate images acquired by the four image acquisition devices are Image1, Image2, Image3, and Image4, and the output image feature point sets are P1, P2, P3, and P4, respectively.

[0120] Assume that the internal parameters {K1, K2, ..., K α}Pixel values ​​are represented as {(u1, v1), (u2, v2), ..., (u α , v α )}, the image size is represented as {(U1, V1), (U2, V2), ..., (U α , V α )},U1≠U2,...,≠U α , V1≠V2,...,≠V α ; That is, in this embodiment, Image1, Image2, Image3, and Image4 respectively represent the images of the calibration plate under the same field of view captured by camera I, camera II, camera III, and camera IV at the same time, wherein the pixels occupied by the Image1 calibration plate are (u1, v1), the pixels occupied by the Image2 calibration plate are (u2, v2), the pixels occupied by the Image3 calibration plate are (u3, v3), and the pixels occupied by the Image4 calibration plate are (u4, v4), wherein Image1, Image2, Image3 The image sizes of Image1, Image2, Image3 and Image4 are (U1, V1), (U2, V2), (U3, V3) and (U4, V4) respectively. Since the resolution of each camera is different, the image sizes taken by each camera are inconsistent, that is, U1≠U2≠U3≠U4, V1≠V2≠V3≠V4. Therefore, the image pixels occupied by the calibration plate are inconsistent. To ensure the accuracy of subsequent image registration, the four images should be uniformly resized at the same resolution. This patent uses a bilinear interpolation method to perform image scale transformation on heterogeneous images to ensure that the pixel accuracy of the calibration plate images is consistent at the same time. The target image is based on the lowest common multiple of the resolution of the four calibration plate images at the same time:

[0121] ku = [u1, u2, u3, u4], k v =〔v1,v2,v3,v4〕;

[0122] Among them, k u represents the least common multiple of the long sides of the four target calibration plate images, k v Represents the least common multiple of the wide edges in the four target calibration plate images;

[0123] Assume that the pixels occupied by the target calibration plate image after size transformation are represented as {(u1', v1'), (u2', v2'), ..., (u α ',v α ')}, the image size is expressed as {(U 1-1 , V 1-1 )、(U 2-1 , V 2-1 ),...,(U α-1 , V α-1 )}, the target calibration plate image pixels after scale transformation and the target calibration plate image pixels before scale transformation have the following relationship:

[0124] u1'=u2'=....=u α '=k u

[0125] v1'=v2'=...=v α '=k v

[0126] That is, the pixels of the calibration plate image after scale transformation and the pixels of the calibration plate image before scale transformation have the following relationship:

[0127] u1'=u2'=u3'=u4'=k u

[0128] v1′=v2′=v3′=v4′=k v

[0129] For any target calibration plate image λ in the α target calibration plate images, λ=1,2,...,α; the size of the target calibration plate image after the scale transformation and the size of the target calibration plate image before the scale transformation have the following relationship:

[0130] U λ-1 =U λ *(k u / u λ ), V λ-1 =V λ *(k v / v λ ).

[0131] That is, in this embodiment, the scale change is as follows according to the relationship between the sizes of the previous and next images:

[0132] U 1-1 =U1*(k u / u1), V 1-1 =V1*(k v / v1)

[0133] U 2-1 =U2*(k u / u2), V 2-1 =V2*(k v / v2)

[0134] U 3-1 =U3*(k u / u3), V 3-1 =V3*(k v / v3)

[0135] U 4-1 =U4*(k u / u4), V 4-1 =V4*(k v / v4)

[0136] The intrinsic parameter K1 of camera I is known, where the intrinsic pixel value of camera I is (u1, v1), that is, the pixel size of Image1 occupied by a grid in the calibration plate area on Image1 is (u1 / n, v1 / n), as follows Figure 2 As shown in , similarly, according to the above calibration results, the calibration results of camera II, camera III, and camera IV are K2, K3, and K4 respectively, which means that the pixel sizes occupied by one grid of the calibration plate on Image2, Image3, and Image4 are (u2 / n, v2 / n), (u3 / n, v3 / n), and (u4 / n, v4 / n) respectively. The pixels on Image1, Image2, Image3, and Image4 are expanded to the same size, that is, the number of pixels occupied by each grid of the calibration plate on Image1, Image2, Image3, and Image4 are the same. The output result is as follows Figure 3 This process is implemented by linear interpolation method, and bilinear interpolation method is used to scale the four target calibration plate images at the same time, so that the number of pixels occupied by each chessboard grid in the four target calibration plate images after size transformation is consistent. The number of pixels occupied by a grid on the target image is (k u / (2n+1),k v / (2n+1)), after the image scale transformation, the pixel change of the four target calibration plate images corresponding to a single grid is (k u / (2n+1),k v / (2n + 1)), for any one of the four target calibration plate images λ, where λ = 1, 2, 3, 4, the difference between the pixels corresponding to a single grid and the pixels occupied by a grid on the target image is |u λ - u λ '|, then |u λ - u λ '| times of linear interpolation must be performed on the target calibration plate image λ. Taking camera I as an example, the process of scale transformation between Image1 and the target image is as follows:

[0137] S51. Refer to Figure 4 and Figure 5 , define the upper left corner of the target calibration plate image λ and the target image as the coordinate origin ( Figure 5 the dot in), take the point coordinates Q 11 (x1, y1), Q 12 (x2, y2), Q 21 (x3, y3), Q 22 (x4, y4) of the target calibration plate image λ, and the gray value function f of Q 11 , Q 12 , Q 21 , Q 22 , R1, R2, Z ∈ P1, where P1 represents the set of feature points output from the region of interest of the target calibration plate image Image1;

[0138] S52. Perform linear interpolation in the x - direction, and the relationship is as follows:

[0139]

[0140]

[0141] Then perform interpolation in the y - direction, and the relationship is as follows:

[0142]

[0143] S53. For a point T(x, y) on the target calibration plate image Image1, where x1 < x < x2, y1 < y < y2, its gray value f(x, y) is obtained by the following formula:

[0144]

[0145] S54. Align the geometric centers between the interpolated target calibration plate image Image1 and the target image:

[0146] It is known that the side - length ratios between the target calibration plate image Image1 and the target image are respectively Assume that the pixel of a certain point on the target calibration plate image Image1 is (i, j), and the pixel point with the coordinates of the point (i', j') on the target image is mapped to the pixel point on the target image through the side length ratio, that is, the relationship between the pixel point on the image to be converted and the pixel point on the original image after the scale transformation is as follows:

[0147]

[0148]

[0149] At this point, the target calibration plate image Image1 and the target image have achieved the same size transformation. According to the above steps, the scale transformation between Image2, Image3, Image4 and the target image can be achieved. The bilinear interpolation method is used to transform the image scale of heterogeneous images to ensure the consistency of the pixel accuracy of the target calibration plate image at the same time.

[0150] Example 3

[0151] In step S6, when registering the scale-transformed α target calibration plate images, a registration model between the scale-transformed α target calibration plate images is established based on the characteristic base point identification and space transformation method, so as to transform them into the same coordinate system;

[0152] Assume that the image plane coordinate system of any image acquisition device λ among the α image acquisition devices is O λ -X λ Y λ , the image plane coordinate system of any image acquisition device η is O η -X η Y η , the coordinate transformation relationship between the image acquisition device λ and the image acquisition device η is given by (R λη 、T λη ) indicates that R λη represents the rotation matrix between the image plane coordinate system of the image acquisition device λ and the image plane coordinate system of the image acquisition device η, T λη represents the translation matrix between the image plane coordinate system of the image acquisition device λ and the image plane coordinate system of the image acquisition device η; T λ 、T η Respectively represent the translation matrices of the image acquisition device λ and the image acquisition device η, R λ , R η denote the rotation matrices of the image acquisition device λ and the image acquisition device η respectively;

[0153] In this embodiment, the coordinate system O1-X1Y1 is the imaging plane coordinate system of camera I; the coordinate system O2-X2Y2 is the imaging plane coordinate system of camera II, the coordinate system O3-X3Y3 is the imaging plane coordinate system of camera III; the coordinate system O4-X4Y4 is the imaging plane coordinate system of camera IV; the mapping points of the spatial point P to be measured on the imaging I, imaging II, imaging III, and imaging IV coordinate systems are P1, P2, P3, and P4, and A1B1C1D1, A1B1C1D2, A1B1C1D3, and A1B1C1D4 are the positions of the calibration plate in the imaging coordinate system respectively; wherein the coordinate transformation relationship between the imaging plane of camera I and the imaging plane of camera II is given by (R 12 、T 12 ) represent the rotation matrix and translation matrix between the image coordinate system O1-X1Y1 and the image coordinate system O2-X2Y2 respectively; similarly, (R 13 、T 13 ) respectively represent the conversion relationship between the image coordinate system O1-X1Y1 and the image coordinate system O3-X3Y3; (R 14 、T 14 ) represent the conversion relationship between the image coordinate system O1-X1Y1 and the image coordinate system O4-X4Y4 respectively; point P is a point on the surface of the actual object to be measured (the imaging plane is two-dimensional), and its world coordinate is P w .

[0154] Among them, the mapping relationship between the spatial point and the plane coordinates and the conversion relationship between different imaging coordinate systems satisfy: for the point Q on the surface of the actual object to be measured, its world coordinate is Q w , the coordinates of point Q in the image plane coordinate system of the image acquisition device λ and the image acquisition device η are expressed as:

[0155] Q λ =R λ Q w +T λ

[0156] Q η =R η Q w +T η

[0157] The image acquisition device λ is set as the target coordinate system, then:

[0158] Q λ =R λη Q η +T λη

[0159] in,

[0160] The feature point sets of the regions of interest of the target calibration plate image λ and the target calibration plate image η acquired by the image acquisition device λ and the image acquisition device η are P respectively. λ With P η , and its transformation relationship is expressed as:

[0161] P η (x,y)=kR λη Pη(x,y)+T λη .

[0162] The purpose of registration is to transform the target plate image to the same coordinate system. The matching points on the image coordinate system can be transformed to overcome the errors caused by the installation of the imaging device and ensure the imaging quality.

[0163] Then, for the known camera I and camera II, their translation matrices are T1 and T2, and their rotation matrices are R1 and R2. Then the coordinate systems of point P in the two coordinate systems can be expressed as

[0164]

[0165] By setting the left camera coordinate system to the target coordinate system, we can get:

[0166] P1=R 12 P2+T 12

[0167] in,

[0168]

[0169] From the above results, we can know the mapping relationship between spatial points and plane coordinates and the conversion relationship between different imaging coordinate systems. Therefore, the spatial point P completes the conversion between the imaging coordinate systems O1-X1Y1 and O2-X2Y2, that is, the matching points on the image coordinate system can be converted.

[0170] The above is the conversion relationship of camera II relative to camera I. Similarly, the conversion process of the spatial position relationship of camera III and camera IV relative to camera I is the same as above.

[0171] The registration of the scale-transformed α target calibration plate images also includes:

[0172] Establish chessboard coordinates, obtain the coordinates of the four outer vertices of the target calibration plate under the transformed posture, and after obtaining the coordinates of the four outer vertices, establish the local coordinate system of the outer vertices, so as to establish the correspondence between the outer vertices of the target calibration plate image and the target calibration plate encoding, and according to the correspondence between the world coordinate number of the inner corner point of the chessboard and the matrix coordinate system number of the inner corner point, achieve the matching of the target calibration plate image landmark points, and obtain the degree of registration between the α target calibration plate images after scale transformation, the corresponding discrete signal of the image to be tested is expanded, and the registration reference point is in the same position in the corresponding two-dimensional array of the expanded full-frame image. This process can improve the accuracy of feature point matching between images, and coordinates can be adaptively obtained for simple multi-deformation vertices.

[0173] Since there is a special conversion relationship among the feature points, the conversion relationship between the feature points of Image2-1 to Image4-1 and Image1-1 can be performed according to the following steps:

[0174] Since the spatial transformation of an image has the general characteristics of converting parallel lines into parallel lines and mapping finite points to finite points, translation, mirroring, scaling, etc. are special cases of two-dimensional spatial transformation. Assuming that a point (x, y) in space is the mirror point (x′, y′) about the line ax+by+c=0, based on the above transformation special cases, the coordinates of a point after translation, mirroring, and scaling can be expressed by spatial transformation as:

[0175]

[0176]

[0177] That is simplified to:

[0178] p′(x′,y′)=kRp(x,y)+t

[0179] Where p′(x′,y′) and p(x,y) are coordinate vectors, and k is the scaling factor.

[0180] That is, the spatial feature point set transformation of Image1-1 and Image2-1 is expressed as:

[0181] p2(x,y)=kR 12 p1(x,y)+T 12

[0182] To improve the accuracy of feature point matching between images, see Figure 6 ,For the acquisition method of simple multi-deformation vertices, the method of adaptively acquiring the coordinates of the four outer vertices of the chessboard with irregular posture changes is as follows:

[0183] (1) With the upper left point of the image as the origin o1, the horizontal direction as the u axis, and the vertical direction as the v axis, a coordinate system o1uv is established. With the upper left vertex of the chessboard as the origin o2, the line connecting the upper left vertex and the upper right vertex as the r axis, and the line connecting the upper left vertex and the lower left vertex as the s axis, a coordinate system o2rs is established.

[0184] (2) Assume that the checkerboard coordinate system and the image coordinate system completely coincide with each other, and their four vertices are A1(x1, y1), A2(x m ,y1)、A3(x m ,y n )、A4(x1,y n );A ij (x ij ,y ij ) is the i-th corner point in the r direction and the j-th corner point in the s direction in the chessboard coordinate system, and 1≤i≤m, 1≤j≤n, m,n,i,j∈Z;

[0185] (3) When the position of the chessboard changes irregularly in the field of view, assume that the translation matrix between the coordinate systems o1uv and o2rs is T = [a x ,a y ,0] T , rotation matrix R = [0,0,θ] T , where 0≤a x , 0≤a y , -45≤θ≤45 (the rest of the angles can be regarded as symmetrical, mirror images or integer periods separated by 360 degrees, so only this angle range is discussed); ε x , ε y represents the length of the chessboard in the x and y directions as a constant. When the distortion is not considered, ε x =ε y =ε, then A ij The positioning coordinates in the chessboard coordinate system o2rs are:

[0186]

[0187] (4) When the position of the chessboard changes in the field of view, according to the mapping relationship between the chessboard coordinates and the image coordinates, any point A on the chessboard ij (r ij ,s ij ) is the coordinate A in the image coordinate system ij ′(u ij ,v ij ) satisfies the following relationship:

[0188]

[0189] Right now:

[0190]

[0191] From this we can see that:

[0192]

[0193] Since -45≤θ≤45, cosθ+sinθ≥0, cosθ-sinθ≥0. In addition, when the chessboard position is determined, a x 、a y , θ, ε are all constants, that is:

[0194] ①When i=1、j=1,u ij +v ij Take the minimum value, which is the outer vertex A1 of the chessboard;

[0195] ①When i=m、j=n,u ij +v ij Take the minimum value, which is the outer vertex A3 of the chessboard.

[0196] ①When i=m、j=1,u ij -v ij Take the maximum value, which is the outer vertex A4 of the chessboard;

[0197] ①When i=1、j=n,u ij -v ij Take the minimum value, which is the outer vertex A2 of the chessboard;

[0198] From the above theoretical reasoning results, we can know that the calculation rules of the four outer vertices of the chessboard are as follows:

[0199]

[0200] After obtaining the image positioning coordinates of the outer vertices, it is necessary to establish a corresponding relationship between the outer vertices of the image and the chessboard calibration plate encoding. To this end, a local coordinate system of the outer vertices is established, and the positioning information of the outer four vertices is used to match the landmarks:

[0201] The world coordinates of the inner corner points of the chessboard are numbered from left to right and from top to bottom;

[0202] (2) Assume that the inner corner point is B p (u,v)(1≤p≤196, and p∈z), through the above-mentioned chessboard vertex recognition algorithm, the four outer vertex coordinates of the inner corner point are automatically obtained, B1′ (upper left), B2′ (upper right), B3′ (lower right), and B4′ (lower left), and the local chessboard coordinate system O3u3v3 is established with B1′ as the origin, B1′B4′ as the v-axis, and B1′B2′ as the u-axis.

[0203] (3) The inner corner point B in the image coordinate system p (u,v) is transformed into the local coordinate system O3u3v3 to obtain the new coordinates C of the inner corner point p (u,v), its changing relationship is as follows:

[0204] C p =R3T3B p

[0205] Where R3 is the rotation matrix between the outer vertex coordinate system and the inner corner coordinate system, and T3 is the translation matrix between the outer vertex coordinate system and the inner corner coordinate system. The expressions are as follows:

[0206]

[0207] in,

[0208] (4) Based on corner point C p The u and v values ​​in the (u, v) image positioning bid are sorted. The sorting rule is: the u values ​​with the same size are arranged in a row. On this basis, the v values ​​are sorted from small to large in a row from left to right to obtain the new corner point image positioning coordinates C. q (u,v)(1≤q≤196, and q∈z);

[0209] (5) Locate the coordinates C according to the inner corner image q (u,v) The rules are numbered from left to right and from top to bottom. The numbering result is as follows Figure 7 As shown;

[0210] (6) According to the correspondence between the world coordinate numbers of the inner corner points of the chessboard and the matrix coordinate system numbers of the inner corner points, the matching of the landmark points is achieved.

[0211] (7) According to the results of the above landmark matching, it is assumed that the feature point sets in the calibration plate or the feature point sets of the images between different images are P1, P2, P3, and P4, that is:

[0212]

[0213] E is an m-dimensional column vector whose elements are all 1.

[0214] Use the least squares method to find the minimum value of the standard deviation between Image1-2 and the point set to be registered

[0215]

[0216] Where i = 2, 3, 4. The conditions for minimizing the above equation are as follows:

[0217]

[0218] make

[0219]

[0220] (kR) * =Q′Q -1

[0221] Solve (kR) * You can know the degree of registration between Image1-2, Image2-2, Image3-2, and Image4-2.

[0222] Example 4

[0223] Suppose that the target calibration plate image λ acquired by the image acquisition device λ in the α image acquisition devices is image a after size transformation and registration, and the target calibration plate image η acquired by the image acquisition device η is image b after size transformation and registration, respectively. Figure 8 and Fig. 9 As shown, the discrete feature two-dimensional point sets of image a and image b are P1, P2, f A , f B are the grayscale values ​​of image a and image b respectively, then:

[0224]

[0225]

[0226] Where, i = 1, 2, 3, 4, 5, ..., n;

[0227] Since the image to be tested is expanded based on the image registration technology, the registration reference points are located at the same position in the two-dimensional array corresponding to the expanded full-frame image after the discrete signal is expanded, a simple and effective parallel fusion algorithm can be used for processing. Suppose the effective feature space of image a is F A , the effective feature space of image b is F B , the feature space of the target image is Ω, for any sample ξ∈Ω, and the two feature vectors corresponding to ξ are g∈F A ,β∈F B , the parallel fusion formula is defined as:

[0228] ξ=g+β

[0229] Based on the above formula, image a and image b are fused to obtain a two-dimensional array image corresponding to the full frame of the target image, such as Fig.10 As shown; the array calculation is as follows:

[0230]

[0231] In this way, the α sub-images at the same moment can be spliced, and then effective image information can be obtained, providing a basis for determining the welding direction trajectory at the next moment.

[0232] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for imaging with a high dynamic range, characterized in that It includes the following steps: S1. Build an optical path system for high-dynamic-range illumination light propagation. Along the propagation of light in the optical path system, arrange α image acquisition devices in parallel; S2. Place a target calibration board on the plane to be measured, change the pose of the target calibration board, and use high-dynamic-range illumination to irradiate the target calibration board on the plane to be measured. After the optical path system divides the calibration board under high-dynamic-range illumination into α imaging optical paths, it irradiates the α image acquisition devices; S3. Use the α image acquisition devices to separately acquire images of the target calibration board, obtain α target calibration board images, and calibrate the internal and external parameters of the α image acquisition devices based on the α target calibration board images to determine the internal and external parameters of the α image acquisition devices; S4. Perform feature extraction on the α target calibration board images respectively, extract the regions of interest in the images, and output the feature point sets of the regions of interest in the images; S5. Based on the internal and external parameters of the α image acquisition devices and the feature point sets of the regions of interest in the images, use the bilinear interpolation method to perform scale transformation on the α target calibration board images at the same moment; S6. Register the α target calibration board images after scale transformation; S7. Perform parallel fusion on the two-dimensional arrays corresponding to the α target calibration board images after registration to obtain the array corresponding to the fused full-frame image, and then locate the targets within the field of view.

2. The imaging method with high dynamic range according to claim 1, characterized in that: In step S1, the built optical path system includes a filter, an α prism set directly above the filter and with a bottom area matching the area of the filter, and a plane mirror. α is an integer not less than 3; there are α plane mirrors in total. One end of each plane mirror stands on the plane where the filter is located, and each plane mirror is placed obliquely from the end located on the plane where the filter is located to the other end. The α plane mirrors surround the α prism, and the α mirror surfaces face one prism surface of the α prism respectively. The high-dynamic-range illumination light is incident through the filter and then through the filter to the α prism. The α prism splits the incident light into α light paths on the same plane and enters the α plane mirrors respectively. The α plane mirrors reflect the α light paths respectively.

3. The imaging method with high dynamic range according to claim 2, characterized in that: In step S2, the target calibration board is a checkerboard calibration board with a size of (2n + 1)×(2n + 1), where n represents a variable. The pose of the target calibration board can be changed 5 to 6 times within six degrees of freedom of the field of view of the α image acquisition devices. Each time the pose of the target calibration board changes, the α image acquisition devices simultaneously acquire the target calibration board images 1 to 2 times.

4. The imaging method with high dynamic range according to claim 3, characterized in that: In step S3, based on α target calibration plate images, the Zhang Zhengyou calibration method is used to calibrate the intrinsic parameters and extrinsic parameters of α image acquisition devices. The intrinsic parameters of the α image acquisition devices determined after calibration are respectively expressed as {K1, K2, ..., K α }, the external parameters are represented as {C1, C2, ..., C α }.

5. The imaging method with high dynamic range according to claim 4, characterized in that: In step S4, the image region of interest extracted from the α target calibration plate images is represented as {I1, I2, ..., I α }, in the process of extracting the image region of interest from each target calibration plate image, for each target calibration plate image, when the gray value of a pixel is less than the gray mean square error of the target calibration plate image, the gray value of the pixel defaults to 0; otherwise, the gray value of the pixel remains unchanged, and the process satisfies the expression: Where f(x,y) represents the grayscale value of a pixel, E(∑(f(x,y)-I)) 2 Represents the grayscale mean square error of the target calibration plate image, represents the average gray value of the target calibration plate image, and u*v represents the pixel of the target calibration plate image; The process of outputting the feature point sets of the regions of interest in the images is as follows: The SIFT method is used to find the extreme points in the region of interest of each target calibration plate image. Each pixel in the region of interest of a target calibration plate image is traversed, and the grayscale value of the pixel is compared with all its adjacent points. When its grayscale value is greater than or less than the grayscale values ​​of all adjacent points in the image domain and scale domain, it is regarded as a maximum point; and each pixel is traversed and searched in the scale space and two-dimensional image space to ensure that the extreme points are fully detected. Then the feature point sets of the region of interest of each target calibration plate image are {P1, P2, ..., P α }.

6. The imaging method with high dynamic range according to claim 5, characterized in that: Assume that the internal parameters {K1, K2, ..., K α } is represented by {(u1, v1), (u2, v2), ..., (u α , v α )}, the image size is represented as {(U1, V1), (U2, V2), ..., (U α , V α )},U1≠U2,...,≠U α , V1≠V2,...,≠V α ; Let the target image be based on the least common multiple of the resolutions of the α target calibration board images at the same moment, that is: k u =〔u1,u2,...,u α 〕,k v =〔v1,v2,...,v α 〕; Among them, k u represents the least common multiple of the long sides of the α-frame target calibration plate images, k v Represents the least common multiple of the wide edges in the α-width target calibration plate image; Assume that the pixels occupied by the target calibration plate image after size transformation are represented as {(u1', v1'), (u2', v2'), ..., (u α ',v α ')}, the image size is expressed as {(U 1-1 , V 1-1 )、(U 2-1 , V 2-1 ),...,(U α-1 , V α-1 )}, the target calibration plate image pixels after scale transformation and the target calibration plate image pixels before scale transformation have the following relationship: u1'=u2'=....=u α '=k u v1'=v2'=...=v α '=k v For any one of the α target calibration board images λ in the α target calibration board images, λ = 1, 2,..., α; the relationship between the size of the target calibration board image after scale transformation and the size of the target calibration board image before scale transformation is as follows: U λ-1 =U λ *(k u / u λ ),V λ-1 =V λ *(k v / v λ )。 7. The imaging method with high dynamic range according to claim 6, characterized in that: The pixel size occupied by each checkerboard grid in the α-frame target calibration plate image is expressed as {(u1 / n,v1 / n), (u2 / n,v2 / n), ..., (u α / n,v α / n)}, and use bilinear interpolation to scale the α target calibration images at the same time, so that the number of pixels occupied by each chessboard grid in the α target calibration images after size transformation is consistent. The number of pixels occupied by a grid on the target image is (k u / (2n+1),k v / (2n+1)), after the image scale transformation, the pixel change of a single grid of the α-frame target calibration plate image is (k u / (2n+1),k v / (2n+1)), for any target calibration plate image λ in the α target calibration plate images, λ=1,2,...,α, the difference between the pixels of a single grid and the pixels occupied by a grid on the target image is |u λ -u λ '|, then the target calibration plate image λ must be |u λ -u λ '| Sublinear difference, the process of obtaining α images of the target calibration plate after scale transformation is: S51. Define the target calibration plate image λ and the upper left corner of the target image as the coordinate origin, and take the coordinate Q of the target calibration plate image λ 11 (x1,y1),Q 12 (x2,y2),Q 21 (x3,y3),Q 22 The gray value function f of (x4,y4), and Q 11 , Q 12 , Q 21 , Q 22 , R1, R2, Z∈P λ , P λ Represents the feature point set outputted from the region of interest of the target calibration plate image λ; S52. Perform linear interpolation in the x direction, and the relationship is as follows: Then perform interpolation in the y direction, and the relationship is as follows: For a point T(x, y) on the target calibration board image λ, where x1 < x < x2 and y1 < y < y2, its gray value f(x, y) is obtained by the following formula: S54. Align the geometric center between the interpolated target calibration plate image λ and the target image: It is known that the side length ratios between the target calibration plate image λ and the target image are Assume that the pixel of a certain point on the target calibration plate image λ is (i, j), and the pixel point with the coordinates of the point (i', j') on the target image is mapped to the pixel point on the target image through the side length ratio, that is, the relationship between the pixel point on the image to be converted and the pixel point on the original image after the scale transformation is as follows: At this point, the size transformation between the target calibration plate image λ and the target image is unified.

8. The imaging method with high dynamic range according to claim 7, characterized in that: In step S6, when registering the scale-transformed α target calibration plate images, a registration model between the scale-transformed α target calibration plate images is established based on the characteristic base point identification and space transformation method, so as to transform them into the same coordinate system; Assume that the image plane coordinate system of any image acquisition device λ among the α image acquisition devices is O λ -X λ Y λ , the image plane coordinate system of any image acquisition device η is O η -X η Y η , the coordinate transformation relationship between the image acquisition device λ and the image acquisition device η is given by (R λη 、T λη ) indicates that R λη represents the rotation matrix between the image plane coordinate system of the image acquisition device λ and the image plane coordinate system of the image acquisition device η, T λη represents the translation matrix between the image plane coordinate system of the image acquisition device λ and the image plane coordinate system of the image acquisition device η; T λ 、T η Respectively represent the translation matrices of the image acquisition device λ and the image acquisition device η, R λ , R η denote the rotation matrices of the image acquisition device λ and the image acquisition device η respectively; Among them, the mapping relationship between space points and plane coordinates and the transformation relationship between different imaging coordinate systems satisfy: For a point Q on the surface of the actual object to be measured, its world coordinate is Q w , the coordinates of point Q in the image plane coordinate system of the image acquisition device λ and the image acquisition device η are expressed as: Q λ =R λ Q w +T λ Q η =R η Q w +T η The image acquisition device λ is set as the target coordinate system, then: Q λ =R λη Q η +T λη in, The feature point sets of the regions of interest of the target calibration plate image λ and the target calibration plate image η acquired by the image acquisition device λ and the image acquisition device η are P respectively. λ With P η , and its transformation relationship is expressed as: P η (x,y)=kR λη Pη(x,y)+T λη 。 9. The imaging method with high dynamic range according to claim 8, characterized in that: The registration of the scale-transformed α target calibration plate images also includes: Establish chessboard coordinates, obtain the coordinates of the four outer vertex points of the chessboard of the target calibration plate in the transformed posture, and after obtaining the coordinates of the four outer vertices, establish the local coordinate system of the outer vertices, so as to establish the correspondence between the outer vertices of the target calibration plate image and the target calibration plate encoding, and according to the correspondence between the world coordinate number of the inner corner point of the chessboard and the matrix coordinate system number of the inner corner point, realize the matching of the target calibration plate image landmark points, and obtain the degree of registration between the α target calibration plate images after scale transformation, the corresponding discrete signal of the image to be tested is expanded, and the position of the registration reference point in the corresponding two-dimensional array of the expanded full-frame image is the same.

10. The imaging method with high dynamic range according to claim 9, characterized in that: Suppose that the target calibration plate image λ acquired by the image acquisition device λ in the α image acquisition devices is image a after size transformation and registration, and the target calibration plate image η acquired by the image acquisition device η is image b after size transformation and registration. The discrete feature two-dimensional point sets of image a and image b are P1, P2, f respectively. A , f B are the grayscale values ​​of image a and image b respectively, then: Where, i=1, 2, 3, 4, 5..., n, m are pixel coordinate values; Assume that the effective feature space of image a is F A , the effective feature space of image b is F B , the feature space of the target image is Ω, for any sample ξ∈Ω, and the two feature vectors corresponding to ξ are g∈F A ,β∈F B , the parallel fusion formula is defined as: ξ=g+β Based on the above formula, image a and image b are fused to obtain a two-dimensional array image corresponding to the full frame of the target image; the array calculation is as follows:

Images