Camera self-calibration method and system in short-track speed skating scene

Abstract

The invention discloses a camera self-calibration method and system in a short-track speed skating scene, and belongs to the technical field of computer vision in a short-track speed skating motion scene. The method is used for solving the problem that a calibration object or other constraint conditions need to be prepared in advance in the existing camera calibration technology. The method comprises the following steps: acquiring two-dimensional feature points of an image; obtaining an initial value M0 of a projection matrix by using a corresponding relation between the two-dimensional feature points of the image and the known three-dimensional points; calculating a re-projection error between a predicted coordinate and an actual image coordinate according to M0, and minimizing the re-projection error to find an optimal solution M of a projection matrix, namely optimal solutions K, R and T of internal and external parameters of the self-calibration camera; according to the camera self-calibration method and system in the short-track speed skating scene, a more convenient and accurate calibration mode can be provided.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology in a short track speed skating sport scene, and more specifically to a camera self-calibration method and system in a short track speed skating scene. Background Art

[0002] With the vigorous development of short track speed skating in my country and the increasing demand for sports science analysis, computer vision calibration methods that are free from the constraints of geometric information of sports scenes and dependence on calibration objects have become the focus of technical research.

[0003] Chinese invention patent specification CN109448025A discloses a video-based method for automatically tracking and modeling the trajectory of short-track speed skaters. The method obtains the corresponding homography matrix based on the correspondence between the pixel plane and four non-collinear points on the ice surface, and can obtain the athlete's motion data without relying on wearable devices or sensors. However, the process requires the preparation of calibration objects in advance and requires the camera viewing angle to be fixed. Chinese invention patent specification CN105046691A discloses a camera self-calibration method based on orthogonal vanishing points, which can estimate camera parameters without calibration objects. However, three groups of mutually orthogonal parallel lines must exist in the scene. Such strict geometric constraints are difficult to meet in actual scenes, reducing the practicality of the method. Chinese invention patent specification CN111724481A discloses a camera calibration method based on OpenCV, which can help accurately reconstruct two-dimensional images for three-dimensional simulation. However, it is necessary to traverse each corner point of multiple groups of images to calculate the internal and external parameters of the camera, and there is a problem of heavy computing tasks caused by high data density.

[0004] Camera calibration in short track speed skating scenarios is a complex computer vision technology, which is complicated by camera lens distortion and parameter estimation. However, current calibration methods still face some challenges: object-based calibration methods require the preparation of calibration objects in advance, which has great restrictions on user scenarios and is not easy to apply to frequently changing camera intrinsic parameters; camera self-calibration methods based on scene geometry rely on image sequences and are easily affected by noise and sequence length; methods based on active vision require precise control platforms and are not suitable for scenarios with unfixed positions, and their algorithms have problems such as low space and time efficiency.

[0005] How to get rid of the geometric information constraints of motion scenes and the dependence on calibration objects has always been a bottleneck problem in the field of computer vision calibration. Therefore, it is of great significance to study a camera calibration method that does not require special calibration objects or complicated operation procedures for different short track speed skating scenes and camera configurations. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a camera self-calibration method and system in a short track speed skating scenario, so as to provide a more convenient and accurate camera calibration method.

[0007] A first aspect of the present invention provides a camera self-calibration method in a short track speed skating scenario, comprising the following steps:

[0008] S1. Establish the correlation between the image coordinate system and the world coordinate system;

[0009] S2, using a grandstand system layout scheme to collect and extract the coordinates of two-dimensional feature points of the image of the ice rink plane;

[0010] S3, obtain the initial value M0 of the projection matrix;

[0011] S4, minimize the reprojection error to find the optimal solution M of the projection matrix, that is, the optimal solution K, R, T of the internal and external parameters of the self-calibration camera;

[0012] Furthermore, the correlation between the image coordinate system and the world coordinate system is:

[0013]

[0014] Where λ is the unknown scale factor, (u,v) is the two-dimensional feature point P of the image I The coordinates of u 、f v is the focal length of the pixel coordinate system in the u-axis and v-axis directions, (u0, v0) is the principal point coordinate, R is the rotation matrix of the camera, T is the translation vector of the camera, K is the intrinsic parameter matrix of the camera (including the camera focal length and principal point coordinates), (X W ,Y W ,Z W )for The corresponding three-dimensional point P W The coordinates of .

[0015] Furthermore, the sample data: according to the venue design regulations for short track speed skating events, the three-dimensional point coordinates in the world coordinate system should strictly comply with the venue layout and facility requirements, cover the entire measurement ice surface, and use a single high-definition camera deployed on the horse track above the ice rink to cover the entire ice rink from a side-down angle, obtain multiple video data sets covering the entire field of athletes in short track speed skating competitions, and then randomly sample the data sets, pre-train to generate training data sets and test sets, and extract two-dimensional feature points in each frame of the image.

[0016] Furthermore, due to the above nonlinear camera imaging model, it is necessary to solve a projection matrix M containing an intrinsic parameter matrix K (camera focal length, distortion coefficient) and an extrinsic parameter matrix [RT] (camera rotation and translation), as follows:

[0017]

[0018] Let m = [m 11 ,m 12 ,…m 34 ] T , where m 11 ,m 12 ,…m 34 is the unknown quantity in the projection matrix M, and the image two-dimensional feature point P is used I With the known 3D point P W The corresponding relationship obtains the initial value of the projection matrix M0:

[0019] M0=K0[R0 T0]

[0020] Furthermore, the reprojection error between the predicted coordinates and the actual image coordinates is calculated based on M0, and the reprojection error is minimized to find the optimal solution M for the projection matrix. The specific formula for calculating the total loss function during the reprojection error training process is as follows:

[0021] L = ln(L 2D )+ln(L ΔP )+L PP

[0022] Among them, L is the total loss function in the training process of calculating the reprojection error, L 2D is the pixel distance between the projection point of the 3D point and the 2D observation point, L ΔP is the consistency error, L PP is the average Euclidean distance between the estimated principal point and the image center.

[0023] Furthermore, the projection matrix M is decomposed by QR to obtain the internal and external parameter estimation matrices K, R, T:

[0024] M=KR[I|-T]=[KR|-KRT]=[H,h]

[0025] Where [H,h] are known quantities, so we have:

[0026] Furthermore, for H -1 Perform QR decomposition to get R -1 and K -1 If the internal parameter element corresponding to the decomposed K matrix is ​​a negative number, a rotation matrix of 180° around the z-axis needs to be applied to the result.

[0027]

[0028] A second aspect of the present invention provides a camera self-calibration system for a short track speed skating scenario, comprising:

[0029] An extraction module, used to obtain two-dimensional feature points of the image of the ice rink plane in the short track speed skating scene;

[0030] The estimation module solves the initial value M0 of the projection matrix containing the estimated values ​​of the camera's internal and external parameters K0, R0, and T0;

[0031] The optimization module calculates the reprojection error between the predicted coordinates and the actual image coordinates based on the initial value M0 of the projection matrix, and minimizes the reprojection error to obtain the optimal solution M of the projection matrix, that is, the optimal solution K, R, T of the internal and external parameters of the self-calibrated camera.

[0032] A third aspect of the present invention provides a computer device, comprising a memory and a processor:

[0033] The memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the camera self-calibration method in the short track speed skating scenario as described in the first aspect of the present invention are implemented.

[0034] Compared with the prior art, the present invention has the following effects:

[0035] 1. This application collects two-dimensional feature points of the image in the short track speed skating scene, and uses the correspondence between the two-dimensional feature points of the image and the known three-dimensional points to obtain the initial value M0 of the projection matrix; based on M0, the reprojection error between the predicted coordinates and the actual image coordinates is calculated, and the reprojection error is minimized to find the optimal solution M of the projection matrix, that is, the optimal solution K, R, T of the internal and external parameters of the self-calibration camera; the camera self-calibration method and system in the short track speed skating scene proposed by the present invention can provide a more convenient and accurate calibration method;

[0036] 2. This application solves the defect that calibration objects or other constraints need to be prepared in advance in the existing short track speed skating scene, thereby improving the convenience of camera calibration; This application breaks through the limitation of the traditional calibration method that calibration objects or other constraints need to be prepared in advance, and realizes the application of camera self-calibration capability in the short track speed skating scene;

[0037] 3. This application solves the problem of heavy computing tasks caused by high data density in complex scenes of short track speed skating by solving a group of linear equations to calculate the initial value of the projection matrix M0. The optimal solution M of the projection matrix is ​​found by minimizing the reprojection error. After QR decomposition of the optimal solution M of the projection matrix, the optimal solution K, R, T of the internal and external parameters of the camera is obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a coordinate conversion diagram from the world coordinate system to the pixel coordinate system in a specific embodiment of the present invention

[0039] Figure 2 is a schematic top view of an international standard short track speed skating venue in a specific embodiment of the present invention;

[0040] Figure 3 It is a layout scheme design diagram of a short track speed skating rink plane calibration system according to a specific embodiment of the present invention;

[0041] Figure 4 It is a relationship diagram between various modules in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0043] It should be noted that the illustrations provided in the following embodiments are only used to illustrate the basic concept of the present invention in a schematic manner, and thus the illustrations only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0044] A camera calibration method in a short track speed skating scenario of this embodiment includes the following steps:

[0045] S1. Establish the correlation between the image coordinate system and the world coordinate system;

[0046] For any three-dimensional point P in space, its world coordinates are (X W ,Y W ,Z W ), the coordinates in the image coordinate system are (u, v), and the transformation relationship of point P from the world coordinate system to the pixel coordinate system is as follows Figure 1 As shown, the mapping relationship of point P from the world coordinate system to the image coordinate system is:

[0047]

[0048] Where λ is the unknown scale factor, (u,v) is the two-dimensional feature point P of the image I The coordinates of u 、f vis the focal length of the pixel coordinate system in the u-axis and v-axis directions, (u0, v0) is the principal point coordinate, R is the rotation matrix of the camera, T is the translation vector of the camera, K is the intrinsic parameter matrix of the camera (including the camera focal length and principal point coordinates), (X W ,Y W ,Z W ) is P I The corresponding three-dimensional point P W The coordinates of .

[0049] S2, using a grandstand system layout scheme to collect and extract the coordinates of two-dimensional feature points of the image of the ice rink plane;

[0050] According to the venue design regulations for short track speed skating events, the three-dimensional point coordinates in the world coordinate system should strictly comply with the venue layout and facility requirements, such as Figure 2 The figure below is a top view of the international standard short track speed skating venue. In order to cover the entire ice surface, a single high-definition camera is deployed on the bridle path above the ice rink. Figure 3 As shown, the entire ice rink is covered from a side-down perspective, and multiple video datasets covering the entire field of athletes in short track speed skating competitions are obtained. The datasets are then randomly sampled, pre-trained to generate training datasets and test sets, and the two-dimensional feature points in each frame are extracted.

[0051] The method for extracting two-dimensional feature points from an image includes: using an image pyramid to construct a multi-layer scale space, detecting brightness differences around pixels to quickly obtain feature points at different scales, and sorting and removing adjacent duplicate feature points according to response values ​​of the feature points to ensure that the feature points are evenly distributed, which is convenient for subsequent description and matching.

[0052] S3, obtain the initial value M0 of the projection matrix;

[0053] The training sample data is historical sample data in a short track speed skating competition scene or sample data extracted from a test video. This embodiment uses sample data extracted from a test video. In this embodiment, the sample data used is two-dimensional image feature points extracted from the test video, and the two-dimensional image feature points are used as training samples.

[0054] From the above nonlinear camera imaging model, we know that we need to solve an internal parameter matrix K that contains the focal length and distortion coefficient of the camera, which is only related to the internal structural characteristics of the camera and has nothing to do with environmental factors. There is also an external parameter matrix [RT] that contains the rotation and translation of the camera, which has nothing to do with the internal structure of the camera. The projection matrix M containing the internal and external parameters of the camera is established. The specific formula is as follows:

[0055]

[0056] Let m = [m 11 ,m12 ,…m 34 ] T , where m 11 ,m 12 ,…m 34 is the unknown quantity in the projection matrix M;

[0057] Substituting the projection matrix M into equation (1), let The specific formula is as follows:

[0058]

[0059] get:

[0060]

[0061] Among them, a pair of image two-dimensional feature points P I and its corresponding 3D point P W We can list two linear equations and construct a linear equation system to solve the unknown quantity m in the projection matrix M. The specific formula is as follows:

[0062]

[0063] in, n represents the number of matching point pairs (n=6);

[0064] The equations listed above are used to solve the projection matrix initial value M0 containing the estimated values ​​of the camera's internal and external parameters K0, R0, and T0. The specific formula is as follows:

[0065] M0=K0[R0 T0] (6)

[0066] S4, minimize the reprojection error to find the optimal solution M of the projection matrix, that is, the optimal solution K, R, T of the internal and external parameters of the self-calibration camera;

[0067] S41, minimizing the reprojection error to find the optimal solution M of the projection matrix;

[0068] According to the initial value of the projection matrix M0, the three-dimensional point P W The coordinates are projected onto the image plane to obtain the predicted position of the two-dimensional image point; the predicted position is compared with the actually observed image point, and the pixel distance between them, i.e., the reprojection error, is calculated.

[0069] Furthermore, the pixel distance between the projection point of the three-dimensional point and the two-dimensional observation point is calculated. The specific formula is as follows:

[0070]

[0071] Among them, L 2Dis the pixel distance between the projection point of the 3D point and the 2D observation point, F is the total number of frames of the collected video, j is the jth frame in the collected video, λ is the similarity transformation scale, are the two-dimensional coordinates of all feature points in the jth frame, are the 3D coordinates of all feature points in the jth frame, M j is the projection matrix of the jth frame.

[0072] Furthermore, the three-dimensional point P is estimated according to the camera internal and external parameter matrix W Project to pixel coordinates and calculate the pixel distance between the two-dimensional observation point. The error is used to measure the accuracy of the intrinsic and extrinsic parameters.

[0073] The average Euclidean distance of points in adjacent frames in the camera coordinate system is used as the consistency loss. This loss ensures that the motion of adjacent frames is smooth. The specific formula of the consistency loss is as follows:

[0074]

[0075] Among them, L ΔP is the consistency loss, F is the total number of frames of the captured video, j is the jth frame in the captured video, The feature points of the jth frame are transformed from world coordinates to camera coordinates through the rotation and translation matrix of the camera jth frame, and the Euclidean distance is calculated with the feature points in the camera coordinate system of the previous frame. λ is the similarity transformation scale. are the three-dimensional point coordinates corresponding to all feature points in the jth frame, M j is the projection matrix of the jth frame.

[0076] Furthermore, the principal point is usually near the center of the image. Therefore, in order to improve the accuracy of the principal point coordinates, the principal point error is introduced. The specific formula of the principal point error is as follows:

[0077]

[0078] Among them, L PP is to estimate the average Euclidean distance between the principal point and the image center, F is the total number of frames of the captured video, j is the jth frame in the captured video, and P X is the error of the image center point X, P Y is the error of the image center point Y, To estimate the principal point X j The average Euclidean distance of the pixel coordinates, To estimate the principal point Y j The average Euclidean distance of pixel coordinates;

[0079] Specifically, the specific formula for calculating the total loss function during the reprojection error training process is as follows:

[0080] L = ln(L 2D )+ln(L ΔP )+L PP (11)

[0081] Among them, L is the total loss function in the training process of calculating the reprojection error, L 2D is the pixel distance between the projection point of the 3D point and the 2D observation point, L ΔP is the consistency error, L PP is the average Euclidean distance between the estimated principal point and the image center.

[0082] Specifically, when the reprojection error is less than the threshold, the overall evaluation of the model is excellent, and the robustness is further verified and the model is applied to actual scenarios to ensure the stability of the model's performance in different scenarios and data sets, and to check whether the model responds to and processes real-time data in a timely manner;

[0083] Furthermore, when the reprojection error is equal to the threshold, the reprojection error is within the acceptable range, the model is comprehensively evaluated as good, and the generalization ability is verified; the generalization ability of the model in different data sets and environments is verified. If the reprojection error is less than the threshold after the data set is reintroduced for testing, the generalization ability of the model in different data sets and environments is good; if the reprojection error is greater than the threshold after the data set is reintroduced for testing, the introduced data set is marked and no longer used;

[0084] Furthermore, when the reprojection error is greater than the threshold, the reprojection error is high and the comprehensive evaluation of the model is poor. The reason for the high reprojection error will be analyzed for adjustment. If the reason for the high reprojection error is the model structure, it will be determined whether there is a situation where the two-dimensional feature points of the image are difficult to capture. If there is a situation where the two-dimensional feature points of the image are difficult to capture, a variety of training data will be introduced. If the reason for the high reprojection error is the training data, it will be analyzed whether the training data is insufficient. If the training data is insufficient, the complexity of the network will be increased, and the size and number of convolution kernels will be adjusted.

[0085] Specifically, the specific selection of the threshold usually depends on the requirements of the task and the expected level of the application scenario, and needs to be determined in combination with the actual situation.

[0086] Furthermore, the initial value M0 of the projection matrix is ​​substituted into the loss function, and back propagation is performed to obtain the optimal solution M of the projection matrix.

[0087] S42, performing QR decomposition on the optimal solution M of the projection matrix to obtain the optimal solutions K, R, T of the internal and external parameters;

[0088] The projection matrix M can be written as:

[0089] M=KR[I|-T]=[KR|-KRT]=[H,h] (12)

[0090] Where [H,h] are known quantities, so we have:

[0091] Furthermore, for H -1 Perform QR decomposition to get R -1 and K -1 Since K and R are homogeneous, the last element is changed to 1. In addition, if the internal parameter element corresponding to the decomposed K matrix is ​​a negative number, a rotation matrix of 180° around the z axis needs to be applied to the result.

[0092]

[0093] like Figure 4 As shown, a camera self-calibration system for a short track speed skating scenario in this embodiment includes:

[0094] An extraction module, used to obtain two-dimensional feature points of the image of the ice rink plane in the short track speed skating scene;

[0095] The estimation module solves the initial value M0 of the projection matrix containing the camera's internal and external parameters K0, R0, and T0;

[0096] The optimization module calculates the reprojection error between the predicted coordinates and the actual image coordinates based on the initial value M0 of the projection matrix, and minimizes the reprojection error to obtain the optimal solution M of the projection matrix, that is, the optimal solution K, R, T of the internal and external parameters of the self-calibrated camera.

[0097] The specific working process of the extraction module, estimation module and optimization module described in this embodiment refers to the step flow described in steps S1-4. The above modules can be different software programs to realize the corresponding functions of each module.

[0098] This embodiment also provides a computer device, which is applicable to the camera self-calibration method in a short track speed skating scenario, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to implement the camera self-calibration method based on the short track speed skating scenario in the above embodiment.

[0099] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0100] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the following steps are implemented: acquiring an image of a short track speed skating scene, collecting and extracting the coordinates of two-dimensional feature points of the image on the ice rink plane; using the correlation between the image coordinate system and the world coordinate system, calculating the reprojection error between the predicted coordinates and the actual image coordinates according to M0, minimizing the reprojection error to find the optimal solution M of the projection matrix, that is, the optimal solution K, R, T of the internal and external parameters of the self-calibration camera

[0101] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A camera self-calibration method for short track speed skating, characterized in that Specifically include the following steps: S1. Establish the correlation between the image coordinate system and the world coordinate system; S2, using a grandstand system layout scheme to collect and extract the coordinates of two-dimensional feature points of the image of the ice rink plane; S3, obtain the initial value M0 of the projection matrix; S4. Minimize the reprojection error to find the optimal solution M for the projection matrix, that is, the optimal solution K, R, T for the internal and external parameters of the self-calibrated camera.

2. The camera self-calibration method for short track speed skating according to claim 1, characterized in that: The correlation is: Where λ is the unknown scale factor, (u,v) is the two-dimensional feature point P of the image I The coordinates of u 、f v is the focal length of the pixel coordinate system in the u-axis and v-axis directions, (u0, v0) is the principal point coordinate, R is the rotation matrix of the camera, T is the translation vector of the camera, K is the intrinsic parameter matrix of the camera (including the camera focal length and principal point coordinates), (X W ,Y W ,Z W )for The corresponding three-dimensional point P W The coordinates of .

3. The camera self-calibration method for short track speed skating according to claim 1, characterized in that: The sample data described: According to the venue design regulations for short track speed skating events, the three-dimensional point coordinates in the world coordinate system should strictly comply with the venue layout and facility requirements, cover the entire measurement ice surface, and use a single high-definition camera deployed on the horse track above the ice rink to cover the entire ice rink from a side-down angle, obtain multiple video datasets covering the entire venue of athletes in short track speed skating competitions, and then randomly sample the datasets, pre-train to generate training datasets and test sets, and extract two-dimensional feature points in each frame of the image.

4. The camera self-calibration method for short track speed skating according to claim 1, characterized in that: The projection matrix M is: As can be seen from the above nonlinear camera imaging model, it is necessary to solve a projection matrix M containing an intrinsic parameter matrix K (camera focal length, distortion coefficient) and an extrinsic parameter matrix [RT] (camera rotation and translation), as follows: Let m = [m 11 ,m 12 ,…m 34 ] T , where m 11 ,m 12 ,…m 34 is the unknown quantity in the projection matrix M, and the two-dimensional feature points P of the image are used I With the known 3D point P W The corresponding relationship obtains the initial value of the projection matrix M0: M0=K0[R0 T0].

5. The camera self-calibration method for short track speed skating according to claim 1, characterized in that: The reprojection error between the predicted coordinates and the actual image coordinates is calculated based on M0, and the reprojection error is minimized to find the optimal solution M for the projection matrix. The specific formula for calculating the total loss function during the reprojection error training process is as follows: L=ln(L 2D )+ln(L ΔP )+L PP Among them, L is the total loss function in the training process of calculating the reprojection error, L 2D is the pixel distance between the projection point of the 3D point and the 2D observation point, L ΔP is the consistency error, L PP is the average Euclidean distance between the estimated principal point and the image center.

6. The camera self-calibration method for short track speed skating according to claim 1, characterized in that: Perform QR decomposition on the projection matrix M to obtain the internal and external parameter estimation matrices K, R, T: M=KR[I|-T]=[KR|-KRT]=[H,h] Where [H,h] are known quantities, so we have: Furthermore, for H -1 Perform QR decomposition to get R -1 and K -1 If the internal parameter element corresponding to the decomposed K matrix is ​​a negative number, a rotation matrix of 180° around the z-axis needs to be applied to the result.

7. A camera self-calibration system for short track speed skating, characterized in that: include, An extraction module, used to obtain two-dimensional feature points of the image of the ice rink plane in the short track speed skating scene; The estimation module solves the initial value M0 of the projection matrix containing the estimated values ​​of the camera's internal and external parameters K0, R0, and T0; The optimization module calculates the reprojection error between the predicted coordinates and the actual image coordinates based on the initial value M0 of the projection matrix, and minimizes the reprojection error to obtain the optimal solution M of the projection matrix, that is, the optimal solution K, R, T of the internal and external parameters of the self-calibrated camera.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a camera self-calibration method for a short track speed skating scenario according to any one of claims 1 to 6 are implemented.

Images