Thermal infrared fixed focus camera calibration method and system based on deep neural network

Abstract

The invention discloses a thermal infrared fixed focus camera calibration method and system based on a deep neural network. The thermal infrared fixed-focus camera calibration method based on the deep neural network comprises the following steps: identifying pixel coordinates of angular points in a checkerboard picture by using a pre-trained target detection model based on the deep neural network; integrally correcting the pixel coordinates of all the angular points by using a homography matrix to obtain corrected pixel coordinates; and performing camera calibration by using a Zhang Zhengyou camera calibration method according to the corrected pixel coordinates of each angular point to obtain an internal reference matrix and a distortion coefficient of the camera. According to the method, the angular point position in the image can be extracted robustly by using the deep learning model subjected to extensive learning, the problem of imaging blurring caused by low resolution, low signal-to-noise ratio and fixed focal length of the infrared image is solved, and a stable and reliable calibration method is provided for a commonly-used fixed-focus infrared camera.

Description

Technical Field

[0001] The present invention relates to the technical field of camera calibration, and in particular to a thermal infrared fixed-focus camera calibration method and system based on a deep neural network. Background Art

[0002] Infrared thermal imaging is an advanced imaging method that captures the invisible infrared radiation emitted by an object and converts it into a visible image. The core of this technology lies in the direct correlation between the temperature of an object and the infrared radiation energy it emits. All objects radiate infrared energy according to their own temperature, and infrared thermal imaging cameras can sensitively capture these subtle radiations and convert them into electrical signals, ultimately generating an image showing the temperature distribution on the surface of the object. Although infrared imaging technology can provide rich temperature data, in order to accurately determine the specific temperature of each location, these data must be combined with the geometric information of the image. In order to obtain accurate geometric information, the camera's intrinsic parameter matrix and distortion coefficients must be determined through a geometric calibration process. This process ensures the accuracy and reliability of infrared images, allowing us to interpret and apply these images more accurately. However, infrared imaging usually has low resolution, contrast, and signal-to-noise ratio, which significantly affects the accurate detection of corners in infrared images, and thus affects the calibration accuracy. More importantly, most thermal infrared cameras are non-zoomable, and out-of-focus situations are inevitable during the calibration image acquisition process. Therefore, a robust, reliable and high-precision camera calibration method is urgently needed in the field of infrared camera calibration. At present, deep learning algorithms have been widely used in the field of image processing, and a large number of studies have shown that image processing methods based on deep learning have gradually become a stable and reliable technology. Camera calibration, as a method of mining information from a large number of calibration object images to estimate camera parameters, has inherent similarities with data-driven deep learning technology. However, how to implement thermal infrared fixed-focus camera calibration based on deep neural networks is still a key technical problem that needs to be solved urgently. Summary of the invention

[0003] Technical problem to be solved by the present invention: In view of the above-mentioned problems of the prior art, a thermal infrared fixed-focus camera calibration method and system based on deep neural network are provided. The present invention aims to utilize a deep learning model that has been extensively learned to robustly extract the corner positions in the image, thereby overcoming the problems of low resolution, low signal-to-noise ratio and imaging blur caused by fixed focal length of infrared images, and providing a stable and reliable calibration method for commonly used fixed-focus infrared cameras.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: A thermal infrared fixed-focus camera calibration method based on a deep neural network comprises the following steps: S1, using a pre-trained target detection model based on a deep neural network to identify the pixel coordinates of the corner points in a chessboard image; S2, using a homography matrix to perform overall correction on the pixel coordinates of all corner points to obtain corrected pixel coordinates; S3, using Zhang Zhengyou's camera calibration method to perform camera calibration according to the corrected pixel coordinates of each corner point to obtain the camera's intrinsic parameter matrix and distortion coefficient.

[0005] Optionally, the target detection model is an improved model of the YOLOv8 target detection model. The improved model of the YOLOv8 target detection model is based on the YOLOv8 target detection model, and a serial local window attention mechanism module LWA is added to the output end of the backbone network of the YOLOv8 target detection model to improve the network's ability to mine local information of the image.

[0006] Optionally, the local window attention mechanism module LWA processes the input features from the output end of the backbone network by first dividing the input feature map into non-overlapping windows, then splicing each window in sequence on the depth channel, performing attention mechanism calculation with the center of the window as the anchor point, and finally returning the calculation result to the corresponding position of the anchor point in the original feature map.

[0007] Optionally, before step S1, a step of training a target detection model using manually labeled checkerboard images is also included.

[0008] Optionally, in step S2, the coordinates of all corner points are corrected as a whole using the homography matrix to obtain the corrected corner point coordinates, including: S2.1, according to the pixel coordinates of n corner points in the chessboard image and its corresponding world coordinates Substituting the equations shown below, we get a total of 2n equations: , In the above formula, , , , , , , and is the homography matrix The elements in , and have: ; S2.2, solve the 2n equations by the least squares method to obtain the homography matrix Elements in S2.3, the homography matrix Substitute the elements in into the following formula to correct all corner point coordinates as a whole to obtain the corrected corner point coordinates: , In the above formula, The pixel coordinates of the corner points after correction.

[0009] Optionally, in step S3, using the Zhang Zhengyou camera calibration method to perform camera calibration according to the corrected pixel coordinates of each corner point to obtain the camera's intrinsic parameter matrix and distortion coefficient means solving the optimization problem shown in the following formula to finally obtain the camera's intrinsic parameter matrix and distortion coefficient: , In the above formula, is the number of chessboard images, is the number of corner points in each chessboard image, is the pixel coordinate of the corrected corner point of the jth corner point in the i-th chessboard image, is the reprojection point coordinate of the jth corner point in the i-th chessboard image, is the camera’s intrinsic parameter matrix, is the rotation matrix of the i-th chessboard image, is the translation vector of the i-th chessboard image, is the world coordinate of the jth corner point in the i-th chessboard image, is the distortion coefficient, including radial distortion coefficient and tangential distortion coefficient.

[0010] In addition, the present invention also provides a thermal infrared fixed-focus camera calibration system based on a deep neural network, comprising: A corner point recognition program unit is used to use a pre-trained deep neural network-based target detection model to identify the coordinates of corner points in the chessboard picture; A corner point coordinate correction program unit is used to use a homography matrix to perform overall correction on all corner point coordinates to obtain corrected corner point coordinates; The parameter calibration program unit is used to calibrate the camera using Zhang Zhengyou's camera calibration method according to the corrected corner point coordinates to obtain the camera's intrinsic parameter matrix and distortion coefficients.

[0011] In addition, the present invention also provides a thermal infrared fixed-focus camera calibration system based on a deep neural network, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the thermal infrared fixed-focus camera calibration method based on a deep neural network.

[0012] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program or instruction, and the computer program or instruction is programmed or configured to execute the thermal infrared fixed-focus camera calibration method based on a deep neural network through a processor.

[0013] In addition, the present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the thermal infrared fixed-focus camera calibration method based on a deep neural network through a processor.

[0014] Compared with the prior art, the present invention mainly has the following advantages: the present invention is based on Zhang Zhengyou's camera calibration method, uses deep learning target detection technology to detect corner points in the image, and designs a corner point coordinate correction method based on the homography matrix according to the mathematical principle of camera calibration. The deep learning model that has been widely learned can robustly extract the corner point positions in the image, overcoming the problems of imaging blur caused by low resolution, low signal-to-noise ratio and fixed focal length of infrared images. It has the advantages of being robust and accurate, and provides a stable and reliable calibration method for commonly used fixed-focus infrared cameras. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of the basic flow of the method of the embodiment of the present invention.

[0016] Figure 2 Schematic diagram of the network structure of the improved model of the YOLOv8 target detection model in an embodiment of the present invention.

[0017] Figure 3 Schematic diagram of the network structure of the local window attention mechanism module LWA in an embodiment of the present invention.

[0018] Figure 4 Schematic diagram of an ideal imaging model of a camera in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] like Figure 1As shown, the thermal infrared fixed-focus camera calibration method based on a deep neural network in this embodiment includes the following steps: S1, using a pre-trained target detection model based on a deep neural network to identify the pixel coordinates of the corner points in the chessboard image (such as Figure 1 As shown in the circle in the figure); S2, use the homography matrix to correct the pixel coordinates of all corner points as a whole to obtain the corrected pixel coordinates (as shown in the figure). Figure 1 S3, according to the pixel coordinates of each corner point after correction, use Zhang Zhengyou camera calibration method to calibrate the camera to obtain the camera's internal parameter matrix and distortion coefficient. Figure 1 As shown, before step S1, this embodiment also includes a step of training the target detection model using manually marked chessboard images.

[0021] As an optional implementation, Figure 2 As shown, the target detection model in this embodiment is an improved model of the YOLOv8 target detection model. The improved model of the YOLOv8 target detection model is based on the YOLOv8 target detection model, and a local window attention mechanism module LWA (Local Window Attention) is added to the output end of the backbone network of the YOLOv8 target detection model to improve the network's ability to mine local information of the image. The YOLOv8 target detection model is mainly composed of a backbone network, a neck network and a detection network. Figure 2 Except for the local window attention mechanism module LWA, the other modules are existing modules of the YOLOv8 target detection model, so they are not described in detail here.

[0022] like Figure 3 As shown, the processing of the input features from the output end of the backbone network by the local window attention mechanism module LWA in this embodiment includes: firstly dividing the input feature map into non-overlapping windows, then splicing each window in turn on the depth channel, performing attention mechanism calculation with the center of the window as the anchor point, and finally returning the calculation result to the corresponding position of the anchor point in the original feature map. This method combines the feature extraction capability of the attention mechanism and reduces the amount of model calculation by dividing the non-overlapping windows.

[0023] Ideal pixel coordinates, that is, pixel coordinates after corner correction It can be transformed from the world coordinates by a homography matrix H Obtain, that is: , In the above formula, is the pixel coordinate of the corner point after correction, , , , , , , and is the homography matrix The elements in the homography matrix The degree of freedom is 8, and the following two equations can be obtained from the above formula: , If you want to solve the 8 parameters of the homography matrix, you only need to establish 8 equations for the coordinates of 4 pairs of corner points. In practice, the number of corner points provided by the calibration plate is much greater than 4, that is, the number of equations is greater than the number of unknowns. Since the camera parameters are fixed, the homography matrix H obtained by solving the 8-grid equations constructed by any 4 corner points and their world coordinates should be the same. If different solutions appear, the reason is only due to the error in the corner point coordinates. Therefore, in order to correct the error in the corner point coordinates, the homography matrix is ​​used in step S2 of this embodiment to perform an overall correction on all corner point coordinates to obtain the corrected corner point coordinates, including: S2.1, according to the pixel coordinates of n corner points in the chessboard image and its corresponding world coordinates Substituting the equations shown below, we get a total of 2n equations: , In the above formula, , , , , , , and is the homography matrix The elements in , and have: ; S2.2, solve the 2n equations by the least squares method to obtain the homography matrix Elements in S2.3, the homography matrix Substitute the elements in into the following formula to correct all corner point coordinates as a whole to obtain the corrected corner point coordinates: , In the above formula, The pixel coordinates of the corner points after correction.

[0024] Figure 4 This is a schematic diagram of the ideal imaging model of the camera. The ideal imaging model of the camera shows the geometric relationship between the world coordinate system, the camera coordinate system, the image coordinate system, and the pixel coordinate system. The symbols are the coordinate axis and the point coordinates. The world coordinate system The real point below, is the corner point, is the camera coordinate system. Figure 1 In the ideal imaging model shown in the figure, the process of projecting a 3D world coordinate point in the world coordinate system to a 2D pixel coordinate point in the pixel coordinate system can be described as follows: , In the above formula, is the scale factor, is the coordinate of the 2D pixel coordinate point in the pixel coordinate system, and is the camera’s extrinsic matrix, is a 3×3 camera intrinsic parameter matrix, and has: , in, and is the focal length of the camera (in pixels), and is the position of the camera principal point. In addition, due to errors in the manufacturing process and other reasons, the image will be distorted. The relationship between the pixel positions before and after the distortion is as follows: , in, and are the pixel position coordinates before and after distortion, , , is the radial distortion coefficient, , is the tangential distortion coefficient. The essence of camera calibration is to use a large number of image corner points And its corresponding real point The mapping between them estimates the camera's intrinsic parameter matrix and distortion coefficient , , , , The most widely used method is the Zhang Zhengyou camera calibration method, which models the camera calibration as an optimization problem, and the optimization goal is to minimize the reprojection error. In step S3 of this embodiment, the camera is calibrated using the Zhang Zhengyou camera calibration method according to the pixel coordinates corrected by each corner point to obtain the camera's intrinsic parameter matrix and distortion coefficient, which means solving the optimization problem shown in the following formula to finally obtain the camera's intrinsic parameter matrix and distortion coefficient: , In the above formula, is the number of chessboard images, is the number of corner points in each chessboard image, is the pixel coordinate of the corrected corner point of the jth corner point in the i-th chessboard image, is the reprojection point coordinate of the jth corner point in the i-th chessboard image, is the camera’s intrinsic parameter matrix, is the rotation matrix of the i-th chessboard image, is the translation vector of the i-th chessboard image, is the world coordinate of the jth corner point in the i-th chessboard image, is the distortion coefficient, including radial distortion coefficient and tangential distortion coefficient. The camera intrinsic parameters and distortion coefficient are continuously optimized in the iterative process of minimizing the above expression until the termination condition is reached, that is, the solution is completed.

[0025] In order to verify the thermal infrared fixed-focus camera calibration method based on deep neural network in this embodiment, experimental verification was carried out in this embodiment, and the specific steps include: 1) Data collection. In order to obtain high-contrast infrared images, we use a portable heating platform to heat the calibration plate (with a checkerboard picture) with different emissivity materials. The size of the calibration plate is smaller than the heating area, so it is covered with insulation material around to prevent exposure in infrared imaging. The imaging device used is the thermal infrared camera on the gimbal of DJI Mavic 3T drone. 2) Data division. The collected data is divided into two parts: training set and test set, and the labels are manually made. The training set is used to train the deep learning target detection model, which can then be used in a wide range of checkerboard corner detection tasks. The test set is used to verify the performance of the target detection model and the subsequent camera calibration process. In this embodiment, 200 training pictures and 50 test pictures are collected, and the spatial positions of the samples have the same distribution range. 3) Model training. The training starting weights of the improved YOLOv8 target detection model use YOLOv8n.pt provided by the publisher Ultralytics. The number of training rounds is 200, the batch size is 16, and the SGD optimizer is used. The mosaic data enhancement is turned off in the last 10 rounds to make the network converge. In addition, the original YOLOv8 model is trained in the same way for experimental comparison to prove the effectiveness of the improvement. 4) Corner detection. The pre-trained deep network detection model is used to detect the corners of the test image, and the rectangular box of the area where the corners are located is obtained. The center of the rectangular box is used as the initial corner coordinates. 5) Corner coordinate correction. The initial corner coordinates are corrected using the proposed homography matrix correction method to obtain the corrected coordinates. 6) Camera calibration. The corrected corner coordinates are used for camera calibration to obtain the final camera intrinsic parameters, distortion coefficients, and reprojection errors.

[0026] In the corner detection results, the evaluation criteria we focus on are mainly the corner missed detection rate, the number of false positives, and the number of rejected images that cannot be used for calibration. The missed detection rate and the number of false positives reflect the robustness of the detection method. The number of images that cannot be used for calibration reflects the impact of the detection method on the camera calibration task. In the camera calibration results, we used the root mean square error (RMSE) and the mean reprojection error (MRE), which are calculated as follows: , , In the above formula, is the number of chessboard images, is the number of corner points in each chessboard image, is the pixel coordinate of the corrected corner point of the jth corner point in the i-th chessboard image, is the reprojection point coordinate of the jth corner point in the i-th chessboard image. In addition, the maximum reprojection error and the standard deviation of the reprojection error are also counted. These indicators reflect the accuracy of camera calibration more comprehensively. The corner point detection results of our method and the camera calibration tools in OpenCV and MATLAB for 50 test images are shown in Table 1.

[0027] Table 1 Corner detection results of different methods

[0028] In Table 1, the improved YOLOv8 is an improved model of the YOLOv8 target detection model in this embodiment. As shown in Table 1, in the corner detection results of the original YOLOv8 model, there are 7 false positive corners in one image, which makes it impossible to use for camera calibration. In contrast, the improved YOLOv8 successfully detects all corners without false positives. This shows that the local window attention (LWA) module can effectively improve the robustness of corner detection.

[0029] After the comparison of corner point detection effects is completed, the camera calibration is performed using the corner point coordinates obtained by different methods. In order to verify the effectiveness of the homography matrix correction method proposed by us, the camera calibration is performed using the initial corner point coordinates and the corrected corner point coordinates, and the corresponding method numbers are 1 and 2, respectively. In order to verify the effectiveness of the network model improvement, the same test is performed using the original YOLOv8 model trained under the same conditions. The final calibration error evaluation indicators of the six methods are shown in Table 2, and the camera parameters and distortion coefficients obtained by the six methods are shown in Table 3.

[0030] Table 2 Calibration errors of different methods

[0031] Table 3 Calibration results of different methods

[0032] See also Figure 2 and Figure 3 It can be seen that the method we proposed has obtained the best value in all four evaluation indicators. When other conditions remain unchanged, the method using the original YOLOv8 model is suboptimal in all four indicators, which shows that the local window attention (LWA) module can effectively improve the accuracy of corner detection. The parameters are closest to the center position of the image (320, 256) (the image resolution is 640×512). It can be seen that the thermal infrared fixed-focus camera calibration method based on the deep neural network in this embodiment can effectively improve the robustness of corner point detection, and the corner point coordinate correction method based on the homography matrix can effectively improve the corner point detection accuracy and reduce the reprojection error of camera calibration. The actual experimental results based on thermal infrared camera calibration show that the thermal infrared fixed-focus camera calibration method based on the deep neural network in this embodiment has been greatly improved in terms of camera calibration accuracy compared with the existing mainstream camera calibration tools (OpenCV and MATLAB), providing a practical, stable and reliable calibration method for commonly used fixed-focus thermal infrared cameras.

[0033] In addition, this embodiment also provides a thermal infrared fixed-focus camera calibration system based on a deep neural network, including: A corner point recognition program unit is used to use a pre-trained deep neural network-based target detection model to identify the coordinates of corner points in the chessboard picture; A corner point coordinate correction program unit is used to use a homography matrix to perform overall correction on all corner point coordinates to obtain corrected corner point coordinates; The parameter calibration program unit is used to calibrate the camera using Zhang Zhengyou's camera calibration method according to the corrected corner point coordinates to obtain the camera's intrinsic parameter matrix and distortion coefficients.

[0034] In addition, this embodiment also provides a thermal infrared fixed-focus camera calibration system based on a deep neural network, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the thermal infrared fixed-focus camera calibration method based on a deep neural network.

[0035] In addition, this embodiment also provides a computer-readable storage medium, which stores a computer program or instruction, and the computer program or instruction is programmed or configured to execute the thermal infrared fixed-focus camera calibration method based on a deep neural network through a processor.

[0036] In addition, this embodiment also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the thermal infrared fixed-focus camera calibration method based on a deep neural network through a processor.

[0037] Those skilled in the art should understand that the technical solutions provided by the embodiments of the present application may be in the form of methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0038] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A thermal infrared fixed-focus camera calibration method based on deep neural network, characterized in that: The method comprises the following steps: S1, using a pre-trained target detection model based on a deep neural network to identify the pixel coordinates of the corner points in the checkerboard image; S2, using a homography matrix to perform an overall correction on the pixel coordinates of all corner points to obtain the corrected pixel coordinates; S3, using the Zhang Zhengyou camera calibration method to perform camera calibration according to the corrected pixel coordinates of each corner point to obtain the camera's intrinsic parameter matrix and distortion coefficient.

2. The thermal infrared fixed-focus camera calibration method based on deep neural network according to claim 1, characterized in that: The target detection model is an improved model of the YOLOv8 target detection model. The improved model of the YOLOv8 target detection model is based on the YOLOv8 target detection model, and a serial local window attention mechanism module LWA is added to the output end of the backbone network of the YOLOv8 target detection model to improve the network's ability to mine local information of the image.

3. The thermal infrared fixed-focus camera calibration method based on deep neural network according to claim 2 is characterized in that: The local window attention mechanism module LWA processes the input features from the output end of the backbone network by first dividing the input feature map into non-overlapping windows, then splicing each window in sequence on the depth channel, performing attention mechanism calculation with the center of the window as the anchor point, and finally returning the calculation result to the corresponding position of the anchor point in the original feature map.

4. The thermal infrared fixed-focus camera calibration method based on deep neural network according to claim 3 is characterized in that: Before step S1, there is also a step of training the target detection model using the manually labeled checkerboard images.

5. The thermal infrared fixed-focus camera calibration method based on deep neural network according to claim 1, characterized in that: In step S2, the coordinates of all corner points are corrected as a whole using the homography matrix to obtain the corrected corner point coordinates including: S2.1, according to the pixel coordinates of n corner points in the chessboard image and its corresponding world coordinates Substituting the equations shown below, we get a total of 2n equations: ， In the above formula, , , , , , , and is the homography matrix The elements in , and have: ； S2.2, solve the 2n equations by the least squares method to obtain the homography matrix Elements in S2.3, the homography matrix Substitute the elements in into the following formula to correct all corner point coordinates as a whole to obtain the corrected corner point coordinates: ， In the above formula, The pixel coordinates of the corner points after correction.

6. The thermal infrared fixed-focus camera calibration method based on deep neural network according to claim 4, characterized in that: In step S3, the camera is calibrated using the Zhang Zhengyou camera calibration method according to the corrected pixel coordinates of each corner point to obtain the camera's intrinsic parameter matrix and distortion coefficient, which means solving the optimization problem shown in the following formula to finally obtain the camera's intrinsic parameter matrix and distortion coefficient: ， In the above formula, is the number of chessboard images, is the number of corner points of each chessboard image, is the pixel coordinate of the corrected corner point of the jth corner point in the i-th chessboard image, is the reprojection point coordinate of the jth corner point in the i-th chessboard image, is the camera’s intrinsic parameter matrix, is the rotation matrix of the i-th chessboard image, is the translation vector of the i-th chessboard image, is the world coordinate of the jth corner point in the i-th chessboard image, is the distortion coefficient, including radial distortion coefficient and tangential distortion coefficient.

7. A thermal infrared fixed-focus camera calibration system based on deep neural network, characterized in that: include: A corner point recognition program unit is used to use a pre-trained deep neural network-based target detection model to identify the coordinates of corner points in the chessboard picture; A corner point coordinate correction program unit is used to use a homography matrix to perform overall correction on all corner point coordinates to obtain corrected corner point coordinates; The parameter calibration program unit is used to calibrate the camera using Zhang Zhengyou's camera calibration method according to the corrected corner point coordinates to obtain the camera's intrinsic parameter matrix and distortion coefficients.

8. A thermal infrared fixed-focus camera calibration system based on a deep neural network, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the thermal infrared fixed-focus camera calibration method based on deep neural network as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program or instruction stored therein, characterized in that: The computer program or instruction is programmed or configured to execute the thermal infrared fixed-focus camera calibration method based on deep neural network as described in any one of claims 1 to 6 through a processor.

10. A computer program product comprising a computer program or instructions, characterized in that The computer program or instruction is programmed or configured to execute the thermal infrared fixed-focus camera calibration method based on deep neural network as described in any one of claims 1 to 6 through a processor.

Images