Transformation type athlete speed measurement method and system for local stadium

Abstract

The invention provides an athlete speed measurement method and system for a transformation type local stadium, and relates to the technical field of motion detection, and the method comprises the steps: obtaining the size of a current stadium, and obtaining the global flag position coordinates of a plurality of standard flag positions of the current stadium according to the size and a standard stadium lineation rule; carrying out zone bit identification and athlete coordinate identification on the real-time video stream shot in the current stadium at the same time to obtain local zone bit coordinates and local athlete coordinates; processing according to the local flag bit coordinates and the global flag bit coordinates to obtain a homography matrix, and processing according to the homography matrix and the local athlete coordinates to obtain global athlete coordinates of the athletes; and for each athlete, calculating to obtain a corresponding real-time speed through global athlete coordinates of the athlete among different frames in the real-time video stream. The method has the advantages that the problem of high subjectivity of data acquisition is solved; the defect of local picture positioning missing is made up; and insufficient limitation of dynamic data processing is broken through.

Description

Technical Field

[0001] The present invention relates to the technical field of motion detection, and in particular to a method and system for measuring the speed of athletes in a transformable local sports venue. Background Art

[0002] In live television broadcasts, multiple cameras are typically used simultaneously to capture the action from different angles to provide viewers with the best possible perspective. Each camera captures the scene the camera operator deems most important, and the director then selects the optimal footage for broadcast. In live sports broadcasts, commentators often quantify on-field phenomena, such as players' real-time speed and defenders' precise positioning. However, traditional technologies suffer from the following significant drawbacks:

[0003] Data collection is highly subjective: commentators rely primarily on visual observation and empirical inference to make data judgments, lacking objective quantitative evidence. This makes it difficult to ensure the accuracy and real-time nature of core data such as speed and position.

[0004] Loss of local image positioning: When the camera zooms in to capture a portion of the field, the correspondence between the image information and the field's global coordinates is lost, making it impossible to accurately determine the absolute position of the players on the field from a single frame.

[0005] Insufficient dynamic data processing: Traditional video analysis technology has difficulty in achieving spatiotemporal correlation of multiple frames and cannot calculate athlete displacement through the time difference of consecutive video streams;

[0006] Data visualization lags: The collected raw motion data lacks real-time fusion technology with video images, making it difficult to generate visualization effects such as velocity vectors and position coordinates in a timely manner, limiting the improvement of the viewing experience.

[0007] These deficiencies are particularly pronounced during football matches: With a pitch measuring 7,140 square meters (international standards), traditional technology cannot infer global coordinates from a camera zooming in to capture a partial image. Furthermore, the lack of real-time, accurate speed measurement for athletes at high speeds (instantaneous speeds of up to 30 km / h) leads to significant discrepancies between commentary data and the actual game. These issues severely hinder the accuracy of data-driven commentary and the implementation of augmented reality effects during live broadcasts. Summary of the Invention

[0008] In view of the problems existing in the prior art, the present invention provides a method for measuring athlete speed in a transformable local sports venue, comprising:

[0009] Step S1, obtaining the size of the current stadium, and then obtaining the global marker coordinates of multiple standard markers of the current stadium based on the size and standard stadium marking rules;

[0010] Step S2, obtaining a real-time video stream obtained by shooting the current stadium, and simultaneously performing marker recognition and player coordinate recognition on the real-time video stream to obtain local marker coordinates and local player coordinates;

[0011] Step S3, obtaining a homography matrix based on the plurality of local marker coordinates and the global marker coordinates, and obtaining the global athlete coordinates of each athlete in the current stadium by combining the homography matrix and the local athlete coordinates;

[0012] Step S4: For each of the athletes, the corresponding real-time speed is calculated using the global athlete coordinates of the athlete between different frames in the real-time video stream.

[0013] Preferably, the step S2 includes:

[0014] Step S21, pre-configuring a first image recognition process and a second image recognition process, and simultaneously assigning a real-time video stream to the first image recognition process and the second image recognition process to perform flag position recognition and athlete coordinate recognition, respectively;

[0015] Step S22: In the first image recognition process, for each frame of image, identify multiple markers contained in the image, and obtain the local marker coordinates of each marker in the image;

[0016] Step S23: In the second image recognition process, for each frame of image, multiple athletes included in the image are identified, the local athlete coordinates of each athlete in the image are obtained, and an ID is assigned.

[0017] Preferably, each of the athlete coordinates is also associated with an ID, and step S3 includes:

[0018] Step S31: Configure a third image recognition process to obtain the local marker coordinates, local player coordinates, and IDs obtained from the same frame of image recognition, and determine whether the number of marker coordinates is greater than a preset threshold.

[0019] If yes, go to step S32;

[0020] If not, return to step S31 to obtain the coordinates of each local marker and the coordinates and ID of each local athlete corresponding to the next frame of image;

[0021] Step S32, calculating the distances between all the local marker coordinates, then selecting the global marker coordinates of the markers corresponding to the multiple groups of local marker coordinates with the farthest distances and processing them to obtain the homography matrix;

[0022] Step S33: multiplying each local athlete coordinate by the homography matrix to obtain the corresponding global athlete coordinate of each athlete in the current stadium, and associating the global athlete coordinate with the corresponding ID.

[0023] Preferably, each of the athlete coordinates is also associated with an ID, and step S4 includes:

[0024] Step S41, for a player with the same ID, respectively obtaining the global player coordinates in different frame images;

[0025] Step S42 , calculating the corresponding real-time speed according to the time difference between different frame images and the distance between the corresponding global player coordinates.

[0026] The present invention also provides a system for measuring athlete speed in a transformable local sports venue, which uses the above-mentioned method for measuring athlete speed, and comprises:

[0027] A marker initialization module is used to obtain the size of the current stadium, and then obtain the global marker coordinates of multiple standard markers of the current stadium based on the size and standard stadium marking rules;

[0028] an image recognition module, connected to the landmark initialization module, for acquiring a real-time video stream shot at the current stadium, and simultaneously performing landmark recognition and athlete coordinate recognition on the real-time video stream to obtain local landmark coordinates and local athlete coordinates;

[0029] a coordinate calculation module, connected to the image recognition module, configured to obtain a homography matrix based on the coordinates of the plurality of local markers, and to obtain a global coordinate of each athlete in the current stadium by combining the homography matrix and the local athlete coordinates;

[0030] A speed calculation module is connected to the coordinate calculation module and is used to calculate the corresponding real-time speed of each athlete through the global athlete coordinates of the athlete between different frames in the real-time video stream.

[0031] Preferably, the image recognition module includes:

[0032] A process allocation unit is used to pre-configure a first image recognition process and a second image recognition process, and simultaneously allocate the real-time video stream to the first image recognition process and the second image recognition process for flag position recognition and athlete coordinate recognition, respectively;

[0033] a marker corresponding unit, configured to, in the first image recognition process, identify, for each frame of image, a plurality of markers contained in the image, and respectively correspond each of the markers to the standard marker of the current stadium to obtain a plurality of local marker coordinates;

[0034] The person recognition unit is configured to, in the first image recognition process, identify multiple athletes contained in each frame of image, obtain the local athlete coordinates of each athlete in the image, and assign an ID.

[0035] Preferably, each of the athlete coordinates is also associated with an ID, and the coordinate calculation module includes:

[0036] a valid frame screening unit, configured to configure a third image recognition process to obtain the local marker coordinates, the local player coordinates, and the IDs obtained from recognition of the same frame of image, generate a matrix calculation signal when it is determined that the number of the marker coordinates is greater than a preset threshold, and obtain the local marker coordinates, the local player coordinates, and the IDs corresponding to a next frame of image when it is determined that the number of the local marker coordinates is not greater than the preset threshold;

[0037] a matrix calculation unit, connected to the valid frame screening unit, for calculating the distances between all the local marker coordinates when receiving the matrix calculation signal, and then selecting the global marker coordinates of the markers corresponding to the multiple groups of local marker coordinates with the farthest distances and processing them to obtain the homography matrix;

[0038] A coordinate conversion unit is connected to the matrix calculation unit and is used to multiply each local athlete coordinate by the homography matrix to obtain the corresponding global athlete coordinate of each athlete in the current stadium and associate it with the corresponding ID.

[0039] Preferably, each of the athlete coordinates is also associated with an ID, and the speed calculation module includes:

[0040] A coordinate acquisition unit, configured to acquire, for a player with the same ID, the global player coordinates in different frame images;

[0041] The speed calculation unit is connected to the coordinate acquisition unit and is used to calculate the corresponding real-time speed according to the time difference between different frame images and the distance between the corresponding global player coordinates.

[0042] The above technical solution has the following advantages or beneficial effects:

[0043] 1. Addressing the subjective nature of data collection: Automatically extracting landmarks (such as corner flags and penalty area lines) and athlete outlines from video streams, eliminating human error and establishing an objective data collection framework. Compared to traditional methods that rely on visual observation and empirical inference, this invention provides a quantitative basis, significantly improving the accuracy and real-time performance of core data such as speed and position.

[0044] 2. Compensating for missing localized image positioning: A fixed pitch reference coordinate system based on FIFA standards is constructed to obtain global marker coordinates. The homography matrix is then calculated using an inverse perspective projection operation to accurately map local image pixels to the global pitch coordinates. Even when the camera zooms in on a specific area of the field, the player's absolute position on the pitch can be accurately determined from a single frame, resolving the positioning challenges associated with traditional technologies due to information loss.

[0045] 3. Overcoming the limitations of dynamic data processing: Using inter-frame motion vector modeling, the system differentially calculates the actual coordinate sequence of the same athlete in continuous video streams, achieving multi-scale data processing with dual spatial and temporal coupling. The system calculates the athlete's displacement by using the time difference between continuous video streams, addressing the inability of traditional technologies to achieve spatiotemporal correlation across multiple frames.

[0046] 4. Since the position of the athlete in the video can be identified, the calculated real-time speed of the athlete is marked at the corresponding position of the athlete in the video screen, and visual effects such as speed vector and position coordinates are generated in time, which improves the audience's viewing experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic diagram of the signage of a standard stadium in a preferred embodiment of the present invention;

[0048] Figure 2 1 is a flow chart of a method for measuring athlete speed in a transformable local sports venue in a preferred embodiment of the present invention;

[0049] Figure 3 Schematic diagram of a sub-flow chart of step S2 in a preferred embodiment of the present invention;

[0050] Figure 4 Schematic diagram of a sub-flow chart of step S3 in a preferred embodiment of the present invention;

[0051] Figure 5 Schematic diagram of a sub-flow chart of step S4 in a preferred embodiment of the present invention;

[0052] Figure 6 The figure is a schematic structural diagram of a system for measuring athlete speed in a transformable local sports venue in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0053] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. The present invention is not limited to this embodiment, and other embodiments may also fall within the scope of the present invention as long as they conform to the gist of the present invention.

[0054] In a preferred embodiment of the present invention, based on the above problems existing in the prior art, a method for measuring the speed of athletes in a transformable local sports field is provided. Figure 2 Shown include:

[0055] Step S1, obtaining the size of the current stadium, and then obtaining the global marker coordinates of multiple standard markers of the current stadium based on the size and standard stadium marking rules;

[0056] Step S2, obtaining a real-time video stream shot at the current stadium, and simultaneously performing marker recognition and player coordinate recognition on the real-time video stream to obtain local marker coordinates and local player coordinates;

[0057] Step S3, obtaining a homography matrix based on the multiple local marker coordinates and the global marker coordinates, and combining the homography matrix and the local athlete coordinates to obtain the global athlete coordinates of each athlete in the current stadium;

[0058] Step S4: For each athlete, the corresponding real-time speed is calculated by using the global athlete coordinates of the athlete between different frames in the real-time video stream.

[0059] Specifically, this embodiment accurately solves the four major technical defects in the background technology through the systematic design of three core steps, and achieves a technological breakthrough from local image to global data analysis.

[0060] (1) Solve the defect of "high subjectivity in data collection"

[0061] In the landmark recognition and athlete coordinate recognition of step S2, AI visual positioning is used to replace manual judgment. The landmarks (such as corner flags, penalty area lines, etc.) and athlete outlines in the video stream are automatically extracted through computer vision algorithms to eliminate manual experience errors.

[0062] Dual coordinate synchronous calibration: Parallel identification of local marker coordinates and local athlete coordinates (dynamic targets) to establish an objective data acquisition framework.

[0063] Technical effect: Improves the error source of speed / position data from "human eye observation ±1.5m level" to "pixel-level coordinate resolution".

[0064] (2) Solve the defect of "missing local image positioning"

[0065] In step S1 , a pitch fixed reference coordinate system (eg, 105 m long×68 m wide) based on FIFA standards is pre-constructed to establish a global pitch fixed reference coordinate system.

[0066] Inverse operation of perspective projection: Step S3 calculates the homography matrix H∈R by the correspondence between the local coordinates of the image and the global coordinates of at least 4 sets of landmark points 3 ×3, map any local screen pixel point (x, y) to the fixed reference coordinate system of the court, (X, Y) = H -1 (x,y).

[0067] (3) Solve the defect of "insufficient dynamic data processing"

[0068] The multi-frame spatiotemporal correlation analysis in step S4 uses inter-frame motion vector modeling. By performing differential calculation on the actual coordinate sequence {(X1, Y1, t1), (X2, Y2, t2) ...} of the same athlete (identified by ID) in a continuous video stream (e.g., 25 frames per second), the instantaneous velocity v = Δd / Δt, where Δd = √[(X2-X1)] 2 +(Y2-Y1) 2 ].

[0069] Even if the camera is zoomed in to a 10m×7m partial image (occupying only 1.5% of the field), the absolute positions of the players on the field can still be inferred through homography transformation.

[0070] By jointly calculating the homography matrix (spatial dimension) and the inter-frame difference (temporal dimension), we can break through the technical limitations of traditional video analysis that only focuses on space or a single time slice, and achieve dual space-time coupling.

[0071] It has multi-scale data processing capabilities. For different shooting scales such as panoramic / medium shot / close-up, it can recognize the local football field through corresponding recognition of landmarks. When the camera zooms in to see the details of the athletes, it can also quickly calculate the real-time position of the athletes and then calculate the corresponding speed.

[0072] Furthermore, since the position of the athlete in the video screen can be identified, the calculated real-time speed of the athlete is marked at the corresponding position of the athlete in the video screen, and visual effects such as speed vector and position coordinates are generated in time, which improves the audience's viewing experience.

[0073] Specifically, before the entire speed measurement process, the program needs to be initialized, mainly by inputting the length and width of the football field, accurate to the centimeter. With these two parameters, the system can combine the provisions of the standard field dividers to give the coordinates of all key points on the field and establish a global fixed reference coordinate system for the field.

[0074] You need to create a cross-section of a football field. Figure 1 As shown. Combining Table 1 of the "Standard Dimensions Table" for standard courts, assuming the coordinates (0, 0) are at the top left corner of the court (marked "1" on the diagram), and assuming 1 meter is 1 unit, it's easy to calculate the coordinates of all points on the diagram. For example, the serve point in the "center circle" (marked "20" on the diagram) has a global marker coordinate of (52.5, 34) in the court's fixed reference coordinate system. There are 39 such markers in total.

[0075] Table 1 Standard size table

[0076]

[0077] It should be noted that the parameters in the table above are for a standard soccer field. If the soccer field you want to calculate is non-standard, you only need to use the measured values when calculating these 39 points.

[0078] In a preferred embodiment of the present invention, step S2 is as follows: Figure 3 Shown include:

[0079] Step S21: pre-configure a first image recognition process and a second image recognition process, and simultaneously assign the real-time video stream to the first image recognition process and the second image recognition process to perform marker recognition and athlete coordinate recognition, respectively;

[0080] Step S22, in the first image recognition process, for each frame of image, multiple markers contained in the image are identified, local marker coordinates of each marker in the image are obtained, and each marker is respectively associated with the standard marker of the current stadium;

[0081] Step S23 , in the second image recognition process, for each frame of image, multiple athletes included in the image are identified, local athlete coordinates of each athlete in the image are obtained, and an ID is assigned.

[0082] Specifically, in this embodiment, the program starts receiving the same set of video streams (or opening video files) in three processes (or threads);

[0083] One process is responsible for identifying the landmarks of the stadium (either the entire stadium or a portion of it) in all images. Taking a specific frame of video as an example, its output is the "frame at a certain time, minute, and second" (or the total frame at a certain agreed-upon starting point) corresponding to that video image. This frame identifies n landmarks, measured in pixels, and specifies which of the 39 landmarks on a standard football field (with a fixed reference frame) each of these n landmarks corresponds to. Currently, there are numerous dedicated open-source research algorithms, such as the EVS Camera Calibration Challenge project, which provides several similar algorithms. The algorithm's output extracts landmarks, such as intersections and corners, from the current image, and determines the pixel coordinates of each landmark within the image.

[0084] Another process is responsible for identifying the athletes in all frames. Taking a specific frame of video image as an example, its output is the "frame number at a certain time, minute, and second" (or the total frame number at a certain agreed starting point) corresponding to the video image. m athletes are identified in the frame image, as well as the athlete's ID value. If it is possible to identify the exact person's name, it is better. If not, it is necessary to identify that a series of people in multiple frames of the same continuous video stream are the same person. The same athlete must use the same ID value in adjacent frames.

[0085] The identified athlete must be identified in the image using a specific identifier, which can be a point or a rectangular box. If the former is used, the point must represent the center of the athlete's feet as the coordinates of the athlete in the image. If the rectangular box is used, the midpoint of the bottom line of the rectangular box is used as the coordinates of the athlete in the image.

[0086] Currently, there are many open-source algorithms for identifying people in images, such as the YOLO series. After calling the algorithm, a rectangular box is generated, indicating that the identified person is within the box. To calibrate the person's position, this box needs to be converted to a point. Obviously, to represent the position of a player on a football field, the most appropriate point is the position of the player's feet, which is the center of the bottom edge of the rectangular box (not the exact center of the entire box). Furthermore, in real life, especially before a shot, players are often running at high speeds. At this time, the aspect ratio of the rectangular box is higher than when standing. In this case, the aspect ratio can be optimized to a certain ratio below the bottom center. To indicate that the player is in mid-air, if the person is standing at this time, the position of their feet would be somewhere below the current position of their feet in mid-air. This position represents the pixel coordinates of the player in the image.

[0087] In a preferred embodiment of the present invention, each athlete's coordinates are also associated with an ID, and step S3 is as follows: Figure 4 Shown include:

[0088] Step S31: Configure the third image recognition process to obtain the coordinates of each local marker, the coordinates of each local athlete, and the ID obtained from the recognition of the same frame image, and determine whether the number of the marker coordinates is greater than a preset threshold:

[0089] If yes, go to step S32;

[0090] If not, return to step S31 to obtain the coordinates of each local marker and the coordinates and ID of each local athlete corresponding to the next frame of image;

[0091] Step S32, calculating the distances between all pairs of local marker coordinates, then selecting the global marker coordinates of the standard markers corresponding to the multiple groups of local marker coordinates with the farthest distances and processing them to obtain a homography matrix;

[0092] Step S33: multiply each local athlete coordinate by the homography matrix to obtain the corresponding global athlete coordinate of each athlete in the current stadium, and associate it with the corresponding ID.

[0093] Specifically, in this embodiment, the main thread receives the real-time values (local flag coordinates and local athlete coordinates and ID) sent by the first two processes (the image recognition process corresponding to the flag identification and the image recognition process corresponding to the athlete coordinate identification), and aligns them according to "a certain time, a certain minute, and a certain second, a certain frame" or "a certain frame in total at a certain agreed starting point".

[0094] First, process the data from the first process and read the n local markers obtained from the frame. If n < 4, the judgment of this frame fails and the algorithm exits and proceeds to the next frame judgment. The algorithm must return at least four markers from the local markers identified in the image, and these four markers must not be on the same straight line.

[0095] If the number of markers returned by the algorithm is greater than 4, the preferred principle is that the horizontal and vertical distances of the closed area formed by these 4 markers are as large as possible, which is determined by the distance between the markers in the image. Select the 4 local markers with the farthest distance (calculate the distance between each two, for example, the pixel coordinates of the two markers a and b in the image are a(xa, ya) and b(xb, yb), and the distance between them is Calculate the distances between all markers (arrange them from largest to smallest) and obtain the coordinate values of the four local markers.

[0096] Combined with the algorithm in this step, the four local markers in the image are identified. No matter which four local markers they are, four sets of "one-to-one correspondence" image local coordinates-field global coordinate relationship tables can be established according to their same positions from the 39 global markers of a standard stadium. According to the marker sequence number corresponding to the selected marker, the corresponding four sets of global marker coordinate values are taken out from the 39 global markers of the fixed reference coordinate system of the stadium, and the getPerspectiveTransform function of OpenCV is called to obtain the "homography matrix H".

[0097] Through this matrix, we can find the coordinates (x0, y0) of any point on the source image (i.e. the live image captured by the camera) and the corresponding coordinates (X1, Y1) of the target image. The target image is as shown above. Figure 1 The cross-section of a football field shown in the image is a cross-section of a football field used in real-life scenarios. The coordinates (x, y, in pixels) of each point on the image are the actual positions on the field (x, y in meters).

[0098] Multiply the local coordinates of the player identified in the "Image Recognition Process for Player Coordinate Identification" by the "homography matrix" generated in the "Image Recognition Process for Flag Identification" to obtain the player's actual position coordinates (X, Y) on the football field (all in meters).

[0099] The same method can also be used to determine the global coordinates (X', Y') of the other athletes in the image on the actual football field. Of course, if there are multiple such images, the same processing can obviously be performed on each frame, thereby obtaining the actual position coordinates of each athlete in each image on the football field. Combined with the fact that the men's 100m world record is 9.58 seconds, set by Jamaican sprinter Usain Bolt at the Berlin World Athletics Championships on August 16, 2009, it is possible to determine the corresponding relationship between each individual in two temporally adjacent frames within the same set of images.

[0100] In a preferred embodiment of the present invention, each of the athlete coordinates is also associated with an ID, and step S4 is as follows: Figure 5 Shown include:

[0101] Step S41, for a player with the same ID, obtaining the global coordinates of each player in different frame images;

[0102] Step S42 , calculating the corresponding real-time speed based on the time difference between different frame images and the distance between the corresponding global player coordinates.

[0103] Specifically, in this embodiment, assume there are "frame n" and "frame m." After the calculations in the preceding embodiment, the homography matrix of "frame n" is An, and the homography matrix of "frame m" is Am. Clearly, the time interval S between the two frames is (mn) * 0.04 seconds. While the image rate captured by each camera varies, the time interval between two adjacent frames is fixed. For example, most common cameras capture 25 frames per second, so the time interval between any two adjacent frames is 40 milliseconds (1 * 1000 / 25).

[0104] There is an athlete with ID 5 in both "frame n" and "frame m", and the coordinates in the two frames are (xm, ym) and (xn, yn) respectively. The distance D between the two points is Divide the distance D by the time interval S to calculate the speed D / S, which is expressed in meters per second. To convert to another speed, follow the formula above.

[0105] For example, if a frame is the mth frame at a certain time, minute, and second in the entire video, and the coordinates of a player in that frame are (x0, y0), then through the "homography matrix" transformation, the player's position on the actual field is obtained as (X1, Y1). The same method can be used to calculate the player's position on the actual field at the nth frame at a certain time, minute, and second (X2, Y2). Then, his speed at that time is:

[0106]

[0107] speed2=speed1 / 3.6 (unit: km / h)

[0108] speed3 = speed1 * 2.24 (unit: miles per hour)

[0109] speed4=100 / speed1 (unit: seconds per hundred meters)

[0110] The numerator in the formula is the straight-line distance between the two points (in meters), and the denominator indicates that the interval between each two frames is 0.04 seconds (ie 40 milliseconds), and there are a total of (nm) frame intervals. Therefore, the speed unit calculated by the above formula is "meters / second". If you want to convert it to "kilometers / hour", just divide the value by 3.6 (or multiply by 0.277778). Some parts of Europe use "miles / hour", which can be multiplied by 2.24. However, in actual application, the above units are commonly used units in life. When using them to express speed, most people cannot intuitively feel the speed. It is recommended to use "hundred meters in seconds", that is, divide 100 by "meters / second". Anyone who has run 100 meters can quickly feel the difference between that person's speed and their own speed, and thus feel how fast or slow they are.

[0111] The purpose of this invention is to scientifically and intelligently calculate the athlete's characteristic values, such as speed and top speed, by identifying all of the above steps. These objective values not only help spectators truly experience the visual impact of the sport but also assist coaches and athletes in objectively analyzing the objective factors that determine success or failure.

[0112] The present invention also provides a system for measuring athlete speed in a transformable local sports venue, which uses the above-mentioned method for measuring athlete speed. Figure 6 Shown, including:

[0113] A marker initialization module 1 is used to obtain the size of the current stadium, and then obtain the global marker coordinates of multiple markers of the current stadium based on the size and standard stadium marking rules;

[0114] The image recognition module 2 is connected to the marker initialization module 1 and is used to obtain a real-time video stream obtained by shooting the current stadium, and simultaneously perform marker recognition and athlete coordinate recognition on the real-time video stream to obtain local marker coordinates and local athlete coordinates;

[0115] A coordinate calculation module 3, connected to the image recognition module 2, is configured to obtain a homography matrix based on the coordinates of the plurality of local markers, and to obtain a global coordinate of each athlete in the current stadium by combining the homography matrix and the local athlete coordinates;

[0116] The speed calculation module 4 is connected to the coordinate calculation module 3 and is used to calculate the corresponding real-time speed of each athlete through the global athlete coordinates of the athlete between different frames in the real-time video stream.

[0117] Preferably, the image recognition module 2 includes:

[0118] A process allocation unit 21 is used to pre-configure a first image recognition process and a second image recognition process, and simultaneously allocate the real-time video stream to the first image recognition process and the second image recognition process for flag position recognition and athlete coordinate recognition, respectively;

[0119] The marker corresponding unit 22 is used to identify multiple markers contained in each frame of the image in the first image recognition process, obtain multiple local marker coordinates, and respectively correspond each marker to a standard marker of the current stadium;

[0120] The person recognition unit 23 is configured to, in the first image recognition process, identify multiple athletes contained in each frame of image, obtain local athlete coordinates of each athlete in the image, and assign an ID.

[0121] Preferably, the coordinate calculation module 3 includes:

[0122] a valid frame screening unit 31 configured to configure a third image recognition process to obtain the local marker coordinates, local player coordinates, and IDs obtained from the same frame of image recognition, generate a matrix calculation signal when determining that the number of the marker coordinates is greater than a preset threshold, and obtain the local marker coordinates, local player coordinates, and IDs corresponding to the next frame of image when determining that the number of the local marker coordinates is not greater than the preset threshold;

[0123] a matrix calculation unit 32 connected to the valid frame screening unit 31, configured to calculate the distances between all the local marker coordinates when receiving the matrix calculation signal, and then select the global marker coordinates of the standard markers corresponding to the multiple groups of local marker coordinates with the farthest distances and process them to obtain the homography matrix;

[0124] The coordinate conversion unit 33 is connected to the matrix calculation unit 32 and is used to multiply each local athlete coordinate by the homography matrix to obtain the corresponding global athlete coordinate of each athlete in the current stadium, and associate it with the corresponding ID.

[0125] Preferably, the speed calculation module 4 includes:

[0126] A coordinate acquisition unit 41 is configured to acquire, for a player with the same ID, the global player coordinates in different frame images;

[0127] The speed calculation unit 42 is connected to the coordinate acquisition unit 41 and is used to calculate the corresponding real-time speed according to the time difference between different frame images and the distance between the corresponding global player coordinates.

[0128] The above are only preferred embodiments of the present invention and do not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of this specification and illustrations should be included in the protection scope of the present invention.

Claims

1. A method for measuring athlete speed in a transformable local sports venue, characterized in that: include: Step S1, obtaining the size of the current stadium, and then obtaining the global marker coordinates of multiple standard markers of the current stadium based on the size and standard stadium marking rules; Step S2, obtaining a real-time video stream obtained by shooting the current stadium, and simultaneously performing marker recognition and player coordinate recognition on the real-time video stream to obtain local marker coordinates and local player coordinates; Step S3, obtaining a homography matrix based on the plurality of local marker coordinates and the global marker coordinates, and obtaining the global athlete coordinates of each athlete in the current stadium by combining the homography matrix and the local athlete coordinates; Step S4: For each of the athletes, the corresponding real-time speed is calculated using the global athlete coordinates of the athlete between different frames in the real-time video stream.

2. The method for measuring athlete speed according to claim 1, wherein: The step S2 comprises: Step S21, pre-configuring a first image recognition process and a second image recognition process, and simultaneously assigning a real-time video stream to the first image recognition process and the second image recognition process to perform flag position recognition and athlete coordinate recognition, respectively; Step S22, in the first image recognition process, for each frame of image, multiple markers contained in the image are identified, the local marker coordinates of each marker in the image are obtained, and each marker is respectively associated with the standard marker of the current stadium; Step S23: In the second image recognition process, for each frame of image, multiple athletes included in the image are identified, the local athlete coordinates of each athlete in the image are obtained, and an ID is assigned.

3. The method for measuring athlete speed according to claim 1, wherein: Each of the athlete coordinates is also associated with an ID, and step S3 includes: Step S31: Configure a third image recognition process to obtain the local marker coordinates, local player coordinates, and IDs obtained from the same frame of image recognition, and determine whether the number of marker coordinates is greater than a preset threshold. If yes, go to step S32; If not, return to step S31 to obtain the coordinates of each local marker and the coordinates and ID of each local athlete corresponding to the next frame of image; Step S32, calculating the distances between all the local marker coordinates, then selecting the global marker coordinates of the standard markers corresponding to the multiple groups of local marker coordinates with the farthest distances and processing them to obtain the homography matrix; Step S33: multiplying each local athlete coordinate by the homography matrix to obtain the corresponding global athlete coordinate of each athlete in the current stadium, and associating the global athlete coordinate with the corresponding ID.

4. The method for measuring athlete speed according to claim 1, wherein: Each of the athlete coordinates is also associated with an ID, and step S4 includes: Step S41, for a player with the same ID, respectively obtaining the global player coordinates in different frame images; Step S42 , calculating the corresponding real-time speed according to the time difference between different frame images and the distance between the corresponding global player coordinates.

5. A system for measuring athlete speed in a transformable local sports venue, characterized in that: The method for measuring athlete speed according to any one of claims 1 to 4 is applied, comprising: A marker initialization module is used to obtain the size of the current stadium, and then obtain the global marker coordinates of multiple markers of the current stadium based on the size and standard stadium marking rules; an image recognition module, connected to the landmark initialization module, for acquiring a real-time video stream shot at the current stadium, and simultaneously performing landmark recognition and athlete coordinate recognition on the real-time video stream to obtain local landmark coordinates and local athlete coordinates; a coordinate calculation module, connected to the image recognition module, configured to obtain a homography matrix based on the coordinates of the plurality of local markers, and to obtain a global coordinate of each athlete in the current stadium by combining the homography matrix and the local athlete coordinates; A speed calculation module is connected to the coordinate calculation module and is used to calculate the corresponding real-time speed of each athlete through the global athlete coordinates of the athlete between different frames in the real-time video stream.

6. The athlete speed measurement system according to claim 5, characterized in that: The image recognition module includes: A process allocation unit is used to pre-configure a first image recognition process and a second image recognition process, and simultaneously allocate the real-time video stream to the first image recognition process and the second image recognition process for flag position recognition and athlete coordinate recognition, respectively; a marker corresponding unit, configured to, in the first image recognition process, for each frame of image, identify a plurality of markers contained in the image, obtain a plurality of local marker coordinates, and respectively correspond each of the markers to the standard marker of the current stadium; The person recognition unit is configured to, in the first image recognition process, identify multiple athletes contained in each frame of image, obtain the local athlete coordinates of each athlete in the image, and assign an ID.

7. The athlete speed measurement system according to claim 5, characterized in that: Each of the athlete coordinates is also associated with an ID, and the coordinate calculation module includes: a valid frame screening unit, configured to configure a third image recognition process to obtain the local marker coordinates, the local player coordinates, and the IDs obtained from recognition of the same frame of image, generate a matrix calculation signal when it is determined that the number of the marker coordinates is greater than a preset threshold, and obtain the local marker coordinates, the local player coordinates, and the IDs corresponding to a next frame of image when it is determined that the number of the local marker coordinates is not greater than the preset threshold; a matrix calculation unit, connected to the valid frame screening unit, for calculating the distances between all the local marker coordinates when receiving the matrix calculation signal, and then selecting the global marker coordinates of the standard markers corresponding to the multiple groups of local marker coordinates with the farthest distances and processing them to obtain the homography matrix; A coordinate conversion unit is connected to the matrix calculation unit and is used to multiply each local athlete coordinate by the homography matrix to obtain the corresponding global athlete coordinate of each athlete in the current stadium and associate it with the corresponding ID.

8. The athlete speed measurement system according to claim 5, characterized in that: Each of the athlete coordinates is also associated with an ID, and the speed calculation module includes: A coordinate acquisition unit, configured to acquire, for a player with the same ID, the global player coordinates in different frame images; The speed calculation unit is connected to the coordinate acquisition unit and is used to calculate the corresponding real-time speed according to the time difference between different frame images and the distance between the corresponding global player coordinates.

Images