A marker-based camera array calibration method, image stitching method, system, terminal, and readable storage medium

Abstract

The present invention discloses a calibration method for a camera array based on markers, an image stitching method, a system, a terminal, and a readable storage medium. The method includes: S1: arranging a camera array above a target area, where the shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras; S2: placing a medium with markers on the target area, and using the camera array to capture the medium to obtain sample images; S3: training an intelligent recognition model using the sample images captured by each camera in the camera array. Among them, the trained intelligent recognition model is used to calibrate the camera array, so that the linear distortion rate of the stitched image captured by the camera array is low, and the imaging speed meets the requirements of financial devices.

Description

Technical Field

[0001] The present invention belongs to the technical field of image shooting and processing, and particularly relates to a calibration method for a camera array based on markers, an image stitching method, a system, a terminal and a readable storage medium. Background Art

[0002] In the field of financial technology, with the rapid development of device technology and economy, in order to save labor costs, more and more intelligent devices have emerged, and a scanner is one of them. A scanner collects data through a device, uploads it to a PC, and then is processed manually. In the application of bank financial institutions, a scanner is an indispensable device and can be used for electronic archiving of paper documents such as customer signature materials, application forms, certificates, and mortgage documents. It has the advantage of centralized authorization at bank counters and can avoid transaction risks at the same time. There are two main categories of scanners on the market. One is a non-contact scanner, such as a high-speed camera. A high-speed camera uses a camera to take pictures of an object at a certain distance. In theory, as long as it is not higher than the triangular area between the camera and the ground plane, the high-speed camera can scan. The picture data obtained by scanning with the high-speed camera is exported through software and saved to the PC side. However, it occupies a large space, has requirements for the focal length, is not suitable for scenarios with space requirements, and the original image captured has a large distortion rate and requires corresponding algorithms to solve the distortion rate problem. The objects captured cannot be used as evidence, and at the same time, the price is relatively expensive. The other is a contact scanner, such as a CIS scanner, which uses a contact photosensitive element (photosensitive sensor) for photosensing. At a position 1 mm to 2 mm below the scanning device, a certain number of red, green, and blue three-color LED (light-emitting diode) sensors are tightly arranged together, and the required light source is generated by driving, and the image can be uploaded in its original proportion. However, since it must be close to the object itself during application, its usage scenarios are limited, and its scanning and transmission speeds are slow, and it is not suitable for scenarios with speed requirements.

[0003] Therefore, in the financial field, there are high requirements for the scanning speed and the image distortion rate, and the existing scanning means temporarily cannot meet these requirements. Therefore, it is extremely necessary to study a new scanning and shooting technical means. Using a camera for shooting has become an optional item, and then how to overcome the problem of linear distortion when using a camera for shooting needs further study. Summary of the Invention

[0004] The object of the present invention is to provide a calibration method for a camera array based on markers, an image stitching method, a system, a terminal and a readable storage medium for the problems existing in the prior art. The method of the present invention uses the cameras in the camera array to take pictures and stitch them. In order to overcome the problem of linear distortion of the stitched images, a medium provided with markers is used to correct the camera array, so that the linear distortion rate of the stitched images obtained by using the corrected camera array is low and the imaging speed is fast.

[0005] On the one hand, the present invention provides a calibration method for a camera array based on markers, comprising the following steps:

[0006] S1: A camera array is arranged above the target area, and the shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras;

[0007] S2: A medium provided with markers is placed on the target area, and the medium is photographed by the camera array to obtain a sample image;

[0008] S3: An intelligent recognition model is trained using the sample images captured by each camera in the camera array;

[0009] Wherein, the intelligent recognition model is used to correct the camera array.

[0010] Optionally, the method further comprises:

[0011] S4: The intelligent recognition model and the camera array are used for shooting to obtain a corrected image;

[0012] S5: The linear distortion rate of the stitched image is obtained, and it is determined whether the linear distortion rate is less than a preset threshold, or whether the linear distortion rate is less than or equal to the preset threshold; if not satisfied, the intelligent recognition model is updated until the linear distortion rate of the stitched image obtained by using the current intelligent recognition model is less than or equal to the preset threshold.

[0013] Optionally, the intelligent recognition model includes correction parameters for each camera, and the correction parameters include: A, k1, k2, R i , t i , X j , where A is the internal parameter coefficient of the camera,, X j is the optical center coordinate, k1, k2 are distortion coefficients, R i , t i are all external parameters.

[0014] Optionally, the process of obtaining the correction parameters in the intelligent recognition model is as follows:

[0015] Step 3.1: Obtain the sample images captured by each camera in the camera array;

[0016] Step 3.2: Solve the internal parameter coefficient A and the external parameters [R i t i of the camera without distortion; wherein, the internal parameter coefficient A is: (u0, v0) is the principal point coordinates, α and β are the u-axis and v-axis of the horizontal image, and γ is the skew parameter of the u-axis and v-axis;

[0017] Step 3.3: Use the overlapping area in the sample images taken by adjacent cameras and adopt the least squares method to estimate the distortion coefficients k1 and k2, and then recalculate the internal camera parameter A and the external parameters [R i t i , and perform cyclic updates to obtain several groups of parameters;

[0018] Step 3.4: Determine the values of all correction parameters with the following objective function, and the objective function is as follows:

[0019]

[0020] where n is the number of images, m is the number of pixel points on the image, x ij is the pixel point j on the i-th image, and x′(A, k1, k2, R i , t i , X j ) is the projection of the i-th image under the condition of following the distortion.

[0021] Optionally, the medium provided with the marker is a black and white grid paper.

[0022] In a second aspect, the present invention provides an image stitching method based on the above method, including the following steps:

[0023] S1-1: Arrange a camera array above the target area, the shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras;

[0024] S2-1: Obtain the intelligent recognition model of the same camera array, the same camera array corresponds to a unified intelligent recognition model, and the same camera array is characterized by the same configuration and the same installation height of the cameras;

[0025] S2-2: Use the intelligent recognition model and the camera array to take pictures to obtain corrected images;

[0026] S2-3: Then stitch the images taken by each camera to obtain a stitched image.

[0027] Optionally, the system further includes:

[0028] A shooting module, configured to place a medium provided with a marker on the target area and use the camera array to shoot the medium to obtain sample images;

[0029] Among them, a camera array is arranged above the target area, and the shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras;

[0030] An intelligent recognition model construction module, configured to train an intelligent recognition model by using the sample images captured by each camera in the camera array, wherein the intelligent recognition model is used to correct the camera array.

[0031] Optionally, the system further includes:

[0032] A correction module, which uses the intelligent recognition model and the camera array to capture corrected images;

[0033] A splicing module, which splices the real-time images of each camera based on image fusion technology to obtain a spliced image;

[0034] A judgment module, configured to obtain the linear distortion rate of the spliced image, and judge whether the linear distortion rate is less than a preset threshold, or whether the linear distortion rate is less than or equal to a preset threshold; if not satisfied, the intelligent recognition model construction module updates the intelligent recognition model until the linear distortion rate of the spliced image obtained by using the current intelligent recognition model is less than or equal to the preset threshold.

[0035] In a third aspect, the present invention provides a terminal, which includes a processor and a memory, the memory stores a computer program, and the processor calls the computer program to execute:

[0036] The steps of the above-mentioned image splicing method of the above-mentioned camera array calibration method based on markers.

[0037] In a fourth aspect, the present invention provides a readable storage medium, which stores a computer program, and the computer program is called by a processor to execute:

[0038] The steps of the above-mentioned image splicing method of the above-mentioned camera array calibration method based on markers.

[0039] Beneficial effects

[0040] Due to the fast scanning speed of the prior art, the image distortion rate is large. For example, for a high-speed document scanner, the scanning speed is fast, the image distortion rate is large, and the objects captured cannot be used as evidence, but only as archival materials and auxiliary materials; although the patch-type scanner has the advantage of low distortion rate, its scanning speed is slow; and for patch scanning of the scanned object, it is required that the physical surface must be flat; between the advantages and disadvantages of these two, the camera array calibration method and image splicing method proposed by the present invention can achieve fast imaging and low linear distortion rate of the imaged image.

[0041] Among them, the camera scanning and photographing is inherently faster than the patch scanning method. Only one complete picture needs to be taken. First, place the A4 paper with a special medium correctly for photographing to obtain a complete picture. The special medium on this A4 paper can be arranged regularly. That is, first use the special medium to calibrate the camera array to obtain a calibrated intelligent calibration model, and then match the actually photographed picture with the intelligent calibration model. In this way, not only the scanning speed is fast, but the calibration speed is also very fast. Secondly, the linear distortion rate of the stitched image obtained by using the calibrated camera array is lower than that of the traditional high-speed document scanner. Especially in the further preferred solution of the present invention, a threshold value is set for the image linear distortion rate, and the intelligent calibration model is continuously updated based on this threshold value, further ensuring that the linear distortion rate of the stitched image obtained by using the calibrated camera array is low. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is the main flow block diagram of the image stitching method with low linear distortion rate of the 3×4 camera array according to Embodiment 3 of the present invention;

[0043] Figure 2 is the main flow block diagram of the intelligent calibration and calibration model provided by the embodiment of the present invention;

[0044] Figure 3 is the main flow block diagram of the system optimization parameters of the image stitching method with low linear distortion rate of the 3×4 camera array provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] The present invention will be further described below in conjunction with embodiments.

[0046] Embodiment 1:

[0047] A marker-based camera array calibration method is provided in this embodiment, including the following steps:

[0048] S1: Arrange a camera array above the target area. The shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras.

[0049] In this embodiment, a 3×4 camera array is taken as an example. In other feasible embodiments, the number of cameras and the arrangement of the array can be adjusted. Regardless of the size of the camera array, its shooting range needs to meet: covering the target area; and there is an overlapping area in the shooting areas of adjacent cameras. Among them, the target area is the target shooting area.

[0050] S2: Place a medium with markers on the target area, and use the camera array to photograph the medium to obtain a sample image.

[0051] In this embodiment, the A4 paper printed with black and white grids is used as the medium, and the black and white grids are used as markers. In other feasible embodiments, the paper printed with special patterns can be used as the medium, and the special patterns are used as markers.

[0052] S3: Train an intelligent recognition model using the sample images captured by each camera in the camera array; wherein, the intelligent recognition model is used to correct the camera array. Wherein, the intelligent recognition model contains correction parameters, and the correction parameters include: A, k1, k2, R i , t i , X j , where A is the internal parameter coefficient of the camera,, X j is the center point coordinate of the image, k1 and k2 are distortion coefficients, and R i , t i are all external parameters.

[0053] Among them, the process of training the intelligent recognition model is actually an intersection correction process, which is used to obtain the internal parameters, external parameters, distortion coefficients, etc. of the camera. Among them, the imaging process of the camera is essentially a coordinate system conversion. First, the points in space are converted from the "world coordinate system" to the "camera coordinate system", then projected onto the imaging plane (image physical coordinate system), and finally the data on the imaging plane is converted to the image pixel coordinate system. However, due to the deviation of the lens manufacturing accuracy and the assembly process, distortion will be introduced, resulting in the distortion of the original image. The present invention only considers linear distortion. In the present invention, the internal parameters can be obtained from the set of data of the focal length and resolution of the camera, the external parameters are the parameters obtained by the conversion of the adjacent overlapping images and the coordinate system, and the distortion coefficients k1 and k2 are also calculated using the characteristics of the overlapping area. Since the above correction processes can all be realized by the prior art, no detailed description is given to them.

[0054] In this embodiment, the acquisition process of the correction parameters M, k1, k2, R i , t i , X j in the intelligent recognition model is as follows:

[0055] Step 3.1: Obtain the sample images captured by each camera in the camera array;

[0056] Step 3.2: Solve the internal parameter coefficient A and external parameters [R i t i of the camera without distortion;

[0057] Step 3.3: Use the overlapping area in the sample images captured by adjacent cameras and adopt the least square method to estimate the distortion coefficients k1 and k2, and then recalculate the internal parameter matrix A and the external parameters [R i t i , and perform cyclic update to obtain several groups of parameters;

[0058] Among them, different degrees of linear distortion, also known as radial distortion, will occur in the camera. If the ideal center coordinates without linear distortion are expressed as (u, v), then represents the coordinates corresponding to the actual linear distortion. For any point (x, y) and are the ideal undistorted and actual distorted normalized image coordinates respectively. According to the projection characteristics of the pinhole model, the following relational expressions can be obtained:

[0059]

[0060]

[0061] where (k1, k2) are the distortion coefficients; since the image distortion center point and the principal point axis are the same, has a certain linear relationship with (u0, v0), and the corresponding relational expression is:

[0062]

[0063]

[0064] The above formula is optimized as:

[0065]

[0066] It can be abbreviated as Dk = d;

[0067] Then, the linear least square method is used to obtain k = (D T D) -1 d; where k = [k1, k2] T is the external parameter.

[0068] Based on the above principle, the present invention first solves the internal parameter matrix A and the external parameters [R i t i of the camera without distortion; then uses the overlapping area in the sample images captured by adjacent cameras and adopts the least square method to estimate the distortion coefficients k1 and k2, and then recalculate the internal parameter matrix A and the external parameters [R i t i , and perform cyclic update to obtain several groups of parameters, and determine a set of optimal parameters based on the objective function in step S3.5.

[0069] Step 3.4: Determine the values of all calibration parameters using the following objective function, where the objective function is as follows:

[0070]

[0071] where n is the number of images, m is the number of pixel points on the image, and x ij is the pixel point j on the i-th image, and x′(A, k1, k2, R i , t i , X j ) is the projection of the i-th image under the condition of following distortion. That is, the above objective function can be understood as in camera calibration, the objective function uses minimizing the reprojection error, that is, projecting the spatial coordinates onto the image according to the estimated projection method to obtain the pixel estimated value, and the error between this estimated value and the actual observed value (the coordinates of the checkerboard corner points) is minimized.

[0072] It can be seen from this embodiment that after obtaining the calibration parameters A, k1, k2, R i , t i , X j in the intelligent recognition model, use them to calibrate the pictures taken by the camera array.

[0073] Embodiment 2:

[0074] Based on Embodiment 1, this embodiment further considers reducing the linear distortion rate. Therefore, a marker-based camera array calibration method provided in this embodiment includes the following steps:

[0075] S11: Deploy a camera array above the target area, where the shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras.

[0076] S21: Place a medium with markers on the target area, and use the camera array to capture the medium to obtain sample images.

[0077] S31: Train an intelligent recognition model using the sample images captured by each camera in the camera array.

[0078] S41: Use the intelligent recognition model and the camera array to capture corrected images;

[0079] S51: Based on image fusion technology, splice the real-time images of each camera to obtain a spliced image;

[0080] S61: Obtain the linear distortion rate of the stitched image, and determine whether the linear distortion rate is less than a preset threshold, or whether the linear distortion rate is less than or equal to the preset threshold; if not satisfied, update the intelligent recognition model until the linear distortion rate of the stitched image obtained by using the current intelligent recognition model is less than or equal to the preset threshold.

[0081] Among them, in this embodiment, the preset threshold is set to 1%. The process of updating the intelligent recognition model is as follows: Based on the currently determined distortion coefficient, update the internal parameters and external parameters. If the linear distortion rate requirement is still not satisfied, continue to update the distortion coefficient, internal parameters, and external parameters until the linear distortion rate requirement is satisfied. Then, use the final model for calibration shooting. This model can be applied to working conditions with the same model and the same installation environment.

[0082] Embodiment 3:

[0083] The purpose of this embodiment is that for a camera matrix of the same type, the trained intelligent recognition model is universal. Therefore, after the intelligent recognition model is trained, this model can be used to process the image stitching of a camera array of the same type. Therefore, an image stitching method based on a camera array calibration method provided in this embodiment includes the following steps:

[0084] S1-1: Deploy a camera array above the target area. The shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras;

[0085] S2-1: Obtain the intelligent recognition model of the same camera array. The same camera array corresponds to a unified intelligent recognition model. The same camera array is characterized by the same configuration and the same installation height of the cameras;

[0086] S2-2: Use the intelligent recognition model and the camera array to capture a corrected image;

[0087] S2-3: Then stitch the images captured by each camera to obtain a stitched image.

[0088] Embodiment 4:

[0089] This embodiment provides a system based on a camera array calibration method, which includes: a camera, a shooting module, and an intelligent recognition model construction module.

[0090] Among them, a camera array is deployed above the target area. The shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras;

[0091] The shooting module is used to place a medium with a marker on the target area and use the camera array to shoot the medium to obtain a sample image;

[0092] The intelligent recognition model construction module trains an intelligent recognition model using the sample images captured by each camera in the camera array. Among them, the intelligent recognition model is used to correct the camera array.

[0093] In some implementation manners, it further includes a splicing module, a correction module, and a judgment module.

[0094] Among them, the correction module uses the intelligent recognition model and the camera array to perform shooting to obtain a corrected image; the splicing module stitches the real-time images of each camera based on the image fusion technology to obtain a stitched image; the judgment module is used to obtain the linear distortion rate of the stitched image and judge whether the linear distortion rate is less than a preset threshold, or whether the linear distortion rate is less than or equal to a preset threshold; if not satisfied, the intelligent recognition model construction module updates the intelligent recognition model until the linear distortion rate of the stitched image obtained using the current intelligent recognition model is less than or equal to the preset threshold.

[0095] Among them, for the specific implementation process of each unit module, please refer to the corresponding process of the foregoing method. It should be understood that for the specific implementation process of the above unit module, refer to the method content, and the present invention will not elaborate here specifically. And the above division of the functional module units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. At the same time, the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0096] Embodiment 5:

[0097] This embodiment provides a terminal, which includes a processor and a memory. The memory stores a computer program, and the processor calls the computer program to execute: the steps of the foregoing method for calibrating a camera array based on a marker, specifically:

[0098] Use the camera array to shoot the medium to obtain a sample image;

[0099] Use the sample images captured by each camera in the camera array to train an intelligent recognition model.

[0100] Among them, a camera array is arranged above the target area and a medium with a marker is placed on the target area. Among them, the shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras.

[0101] In some implementations, the processor invokes a computer program to execute the steps of the marker-based camera array calibration method described above, specifically:

[0102] Obtain a sample image obtained by the camera array photographing the medium.

[0103] Train an intelligent recognition model using the sample images captured by each camera in the camera array.

[0104] Use the intelligent recognition model and the camera array to capture a corrected image. Based on image fusion technology, stitch the real-time images of each camera to obtain a stitched image.

[0105] Obtain the linear distortion rate of the stitched image, and determine whether the linear distortion rate is less than a preset threshold, or whether the linear distortion rate is less than or equal to a preset threshold; if not satisfied, update the intelligent recognition model until the linear distortion rate of the stitched image obtained using the current intelligent recognition model is less than or equal to the preset threshold.

[0106] In some implementations, the processor invokes a computer program to execute the steps of the image stitching method of the marker-based camera array calibration method, specifically:

[0107] Obtain the intelligent recognition model of the same camera array. Among them, a camera array is arranged above the target area, the shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras; the same camera array corresponds to a unified intelligent recognition model, and the same camera array is characterized by the same configuration and the same installation height of the cameras.

[0108] Use the intelligent recognition model and the camera array to capture a corrected image.

[0109] Based on image fusion technology, stitch the real-time images of each camera to obtain a stitched image.

[0110] Among them, for the specific implementation process of each step, please refer to the description of the foregoing method.

[0111] It should be understood that in the embodiments of the present invention, the so-called processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0112] Embodiment 6

[0113] This embodiment provides a readable storage medium that stores a computer program, and the computer program is called by a processor to execute: the steps of the method for calibrating a camera array based on markers, specifically:

[0114] Use the camera array to capture a sample image of the medium.

[0115] Use the sample images captured by each camera in the camera array to train an intelligent recognition model.

[0116] Among them, a camera array is arranged above the target area, and a medium provided with markers is placed on the target area. Among them, the shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras.

[0117] In some implementation manners, the computer program is called by a processor to execute: the steps of the method for calibrating a camera array based on markers, specifically:

[0118] Obtain the sample images captured by the camera array of the medium.

[0119] Use the sample images captured by each camera in the camera array to train an intelligent recognition model.

[0120] Use the intelligent recognition model and the camera array to capture a corrected image; based on image fusion technology, splice the real-time images of each camera to obtain a spliced image.

[0121] It also performs: obtaining the linear distortion rate of the stitched image, and determining whether the linear distortion rate is less than a preset threshold, or whether the linear distortion rate is less than or equal to the preset threshold; if not satisfied, updating the intelligent recognition model until the linear distortion rate of the stitched image obtained by using the current intelligent recognition model is less than or equal to the preset threshold.

[0122] In some implementation manners, the computer program is called by a processor to perform: the steps of an image stitching method for a camera array calibration method based on markers, specifically:

[0123] Obtain an intelligent recognition model of the same camera array. Among them, a camera array is arranged above the target area, the shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras; the same camera array corresponds to a unified intelligent recognition model, and the same camera array is characterized by the same configuration and the same installation height of the cameras.

[0124] Use the intelligent recognition model and the camera array to take pictures to obtain a corrected image.

[0125] Based on the image fusion technology, stitch the real-time images of each camera to obtain a stitched image.

[0126] Among them, the specific implementation process of each step can refer to the description of the foregoing method.

[0127] The readable storage medium is a computer-readable storage medium, which may be an internal storage unit of the controller described in any of the foregoing embodiments, such as the hard disk or memory of the controller. The readable storage medium may also be an external storage device of the controller, such as a plug-in hard disk equipped on the controller, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the readable storage medium may also include both the internal storage unit of the controller and the external storage device. The readable storage medium is used to store the computer program and other programs and data required by the controller. The readable storage medium may also be used to temporarily store data that has been output or will be output.

[0128] It should be emphasized that the examples described in the present invention are illustrative rather than restrictive. Therefore, the present invention is not limited to the examples described in the specific implementation manners. Any other implementation manners obtained by those skilled in the art according to the technical solutions of the present invention, without departing from the spirit and scope of the present invention, whether modified or replaced, also belong to the protection scope of the present invention.

Claims

1. A marker-based camera array calibration method, characterized in that: It includes the following steps: S1: Arrange a camera array above the target area. The shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras; S2: Place a medium with markers on the target area, and use the camera array to capture the medium to obtain a sample image; S3: Use the sample images captured by each camera in the camera array to train an intelligent recognition model; Wherein, the intelligent recognition model is used to correct the camera array; S4: Use the intelligent recognition model and the camera array to capture corrected images, and stitch the real-time images of each camera based on image fusion technology to obtain a stitched image; S5: Obtain the linear distortion rate of the stitched image, and determine whether the linear distortion rate is less than a preset threshold, or whether the linear distortion rate is less than or equal to the preset threshold; if not, update the intelligent recognition model until the linear distortion rate of the stitched image obtained by using the current intelligent recognition model is less than or equal to the preset threshold; The process of updating the intelligent recognition model is as follows: Based on the currently determined distortion coefficient, update the internal parameters and external parameters. If the linear distortion rate requirement is still not met, continue to update the distortion coefficient, internal parameters, and external parameters until the linear distortion rate requirement is met.

2. The method according to claim 1, wherein: The intelligent recognition model contains calibration parameters, and the calibration parameters include: A, k1, k2, R i , t i , X j , where A is the internal parameter coefficient of the camera, X j is the optical center coordinate, k1 and k2 are distortion coefficients, and R i , t i are both external parameters.

3. The method according to claim 2, wherein: The process of obtaining the correction parameters in the intelligent recognition model is as follows: Step 3.1: Obtain the sample images captured by each camera in the camera array; Step 3.2: Solve for the internal parameter matrix A and the external parameters [R i t i ; Among them, the internal reference coefficient A is as follows: (u0, v0) are the coordinates of the principal point, α and β are the u-axis and v-axis of the horizontal image, and γ is the skew parameter of the u-axis and v-axis; Step 3.3: Use the overlapping area in the sample images captured by adjacent cameras and adopt the least squares method to estimate the distortion coefficients k1 and k2, and then recalculate the internal camera parameter A and the external parameters [R i t i , and perform cyclic update to obtain several groups of correction parameters; Step 3.4: Determine the values of all correction parameters with the following objective function, and the objective function is as follows: where n is the number of images, m is the number of pixel points on the image, and x ij is the pixel point j on the i-th image, and x′(A, k1, k2, R i , t i , X j ) is the projection of the i-th image under the condition of following distortion.

4. The method according to claim 1, characterized in that: The medium with markers is a black-and-white grid paper.

5. An image stitching method based on the method described in claim 1, characterized in that: It includes the following steps: S2-1: Obtain the intelligent recognition model of the same camera array. The same camera array corresponds to a unified intelligent recognition model. The same camera array is characterized by the same camera configuration and the same installation height; S2-2: Use the intelligent recognition model and the camera array to capture corrected images; S2-3: Then stitch the images captured by each camera to obtain a stitched image.

6. A system based on the method according to any one of claims 1-4, characterized in that: It includes: A shooting module, which is used to place a medium with markers on the target area and use the camera array to capture the medium to obtain a sample image; Wherein, a camera array is arranged above the target area. The shooting range of the camera array covers the target area, and there is an overlapping area in the captured images of adjacent cameras; An intelligent recognition model construction module, which is used to train an intelligent recognition model with the sample images captured by each camera in the camera array. Among them, the intelligent recognition model is used to correct the camera array; A correction module, which uses the intelligent recognition model and the camera array to capture corrected images; A stitching module, which stitches the real-time images of each camera based on image fusion technology to obtain a stitched image; A judgment module, configured to obtain the linear distortion rate of the spliced image, and judge whether the linear distortion rate is less than a preset threshold, or whether the linear distortion rate is less than or equal to the preset threshold; if not satisfied, the intelligent recognition model construction module updates the intelligent recognition model until the linear distortion rate of the spliced image obtained by using the current intelligent recognition model is less than or equal to the preset threshold; The process of updating the intelligent recognition model is as follows: based on the currently determined distortion coefficient, update the internal parameters and external parameters. If the linear distortion rate requirement is still not satisfied, continue to update the distortion coefficient, internal parameters, and external parameters until the linear distortion rate requirement is satisfied.

7. A terminal, characterized in that: It includes a processor and a memory. The memory stores a computer program, and the processor calls the computer program to execute: The steps of the method for calibrating a camera array based on markers according to claim 1 or the steps of the image splicing method of the method for calibrating a camera array based on markers according to claim 5.

8. A readable storage medium, characterized in that: Stores a computer program, and the computer program is called by a processor to execute: The steps of the method for calibrating a camera array based on markers according to claim 1 or the steps of the image splicing method of the method for calibrating a camera array based on markers according to claim 5.

Images