AI standing long jump evaluation system supporting automatic calibration
By arranging Arucomarkers and high-definition cameras in the standing long jump area, combined with specific algorithms, automatic calibration and long jump action detection are achieved, solving the problems of bulky infrared devices and complex deployment of AI systems, and improving the efficiency of tests and the accuracy and fairness of the results. .
Patent Information
- Application Number
- CN202411590084.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-08
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing infrared long jump equipment is bulky and inadequate, unable to provide video playback, difficult to detect violations, affecting the fairness and accuracy of the test; the deployment of the AI stand-alone long jump evaluation system is complex, relies on professionals and environmental factors affect the identification accuracy, resulting in Test reliability and stability are reduced.
Four Arucomarkers proofreading square pads arranged along four corners of the vertical long jump area are used to combine high-definition cameras and specific algorithms to establish an automatic calibration site model, and use the site model to build modules, personnel determination modules and evaluation modules to detect long jump actions in real time and calculate results. , and store and organize results through cloud databases.
It realizes fast and accurate site calibration, improves testing efficiency, reduces equipment preparation time, enhances the accuracy and impartiality of performance assessment, and is suitable for various venues, including irregularly shaped venues, meeting professional sports events and schools. Accuracy requirements for physical education teaching.
Smart Images

Figure CN119075288A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of sports facilities, and in particular to an AI standing long jump evaluation system supporting automatic calibration. Background Art
[0002] In the field of standing long jump evaluation, there are currently two main types of equipment: infrared and AI. Infrared long jump equipment is more common in the market. It detects the long jump results of testers by emitting and receiving infrared rays. However, this type of equipment has many defects. On the one hand, its structure is complex, including multiple infrared transmitting and receiving devices, as well as corresponding control and processing units, which makes the equipment bulky as a whole. This makes it extremely inconvenient to carry and deploy in actual use, and requires a lot of manpower and time. For example, in large-scale sports tests organized by schools, it is necessary to frequently transfer equipment between different test sites. If the equipment is bulky, it will seriously affect the test efficiency. On the other hand, infrared long jump equipment has obvious functional deficiencies. During the test, it cannot provide a video playback function, which makes it very difficult to review and analyze the test results. Once the tester has doubts about the results, it is difficult to verify accurately due to the lack of video records as a basis. In addition, it is difficult for infrared equipment to effectively judge the violations of personnel during the process, such as being unable to accurately detect whether the tester has violated the rules, such as crossing the line, stepping and jumping, etc., so that cheating cannot be effectively avoided, affecting the fairness and accuracy of the test.
[0003] The application of AI technology in standing long jump evaluation also faces some problems. Among them, the most prominent is the problem of complex deployment. AI technology relies on cameras for image acquisition and analysis, and has high requirements for camera correction, resolution and placement. In the actual deployment process, professionals are required to accurately set camera parameters and adjust the position to ensure that the long jump action information of the tester can be accurately obtained. However, in many actual application scenarios, such as schools, small sports clubs, etc., there is often a lack of professional technicians, which makes it difficult to deploy equipment and achieve ideal test results. Moreover, even after professional debugging, the quality of the image collected by the camera may be disturbed due to environmental factors (such as light changes, complex venue background, etc.), thereby affecting the accuracy of AI algorithm recognition and analysis of long jump actions, further reducing the reliability and stability of the test. Therefore, an AI standing long jump evaluation system that supports automatic calibration is proposed to address the above problems. Summary of the invention
[0004] The purpose of the present invention is to provide an AI standing long jump evaluation system supporting automatic calibration to solve the problems raised in the above-mentioned background technology.
[0005] To achieve the above object, the present invention provides the following technical solutions:
[0006] An AI standing long jump evaluation system supporting automatic calibration, comprising:
[0007] Four square mats of Arucomarkers calibration codes arranged in a rectangular shape along the four corners of the standing long jump area;
[0008] Data collection end: The data collection end is a high-definition camera, which collects video images of the standing long jump area;
[0009] Cloud database; the cloud database includes a tester data storage unit and a score storage unit; the tester data storage unit stores the basic information and face information of the tester;
[0010] Evaluation end: The evaluation end includes a site model building module, a personnel determination module, an evaluation module, and a data interaction module;
[0011] The site model construction module extracts a video stream containing four Arucomarkers proofreading code square pads from the video screen, extracts several frames of images containing four Arucomarkers proofreading code square pads from the video stream, preprocesses the extracted images, uses the Arucomarkers detection algorithm to identify the four Arucomarkers proofreading code square pads in the image, and extracts the coordinates of the four corner points of the four Arucomarkers proofreading code square pads; uses the Euclidean distance formula to calculate the distance between adjacent Arucomarkers proofreading code square pads to obtain the size of the standing long jump area rectangle; draws a virtual take-off point at one end of the standing long jump area rectangle; the site model construction module establishes a virtual coordinate system extending from the virtual take-off point to the landing point on the video screen of the standing long jump area based on the coordinates of the four Arucomarkers proofreading code square pads, calculates the long jump scale line coordinates in the virtual coordinate system according to the known standing long jump area rectangle size, and draws virtual scale lines on the video screen according to the calculated scale line coordinates to obtain a site model;
[0012] The personnel identification module identifies the face information of the person who enters the virtual starting point in the video screen, and exchanges data with the test personnel data storage unit based on the face information to determine the identity of the test personnel;
[0013] The evaluation module receives all the motion images of the tester during the standing long jump, tracks the tester, and detects violations based on the target tracking images; after the tester starts to jump, the key points of the heel are detected in real time, each frame of data is recorded, the take-off frame and landing frame of the tester are determined, the shortest landing point is located, the coordinates of the shortest landing point are determined, the coordinates of the landing point are brought into the site model, and the relative distance between the take-off point and the landing point is calculated as the tester's standing long jump score;
[0014] The data interaction module receives the identity of the tester and the standing long jump result of the tester, sorts them out, and transmits them to the result storage unit.
[0015] As a preferred solution, it also includes a user end, through which the tester logs into the tester data storage unit to build an account and register as a tester, and the registration content includes basic information and facial information; after the score storage unit receives the tester's identity and the tester's standing long jump score, it organizes the data to obtain the tester's score sheet, and sends the score sheet to the user end.
[0016] As a preferred solution, assuming that the four Arucomarkers proofreading code square pads distributed at the four corners of the standing long jump area rectangle are A, B, C, and D, and the length and width of the standing long jump area rectangle are L and W, respectively, determining the coordinates of the four corner points of the four Arucomarkers proofreading code square pads includes the following steps:
[0017] The coordinates of the four corners of the standing long jump area rectangle are defined by the following formulas: A=(0,0), B=(L,0), C=(L,W), D=(0,W);
[0018] Step S1, extracting a video stream containing four Arucomarkers proofreading code square pads from a video screen, and extracting several frames of images containing four Arucomarkers proofreading code square pads from the video stream;
[0019] Step S2: preprocess the extracted image:
[0020] Step S2-1, convert each frame of image into a grayscale image using the following formula: GrayFrame = ConvertToGray (Frame), where GrayFrame is the converted grayscale image data, Frame is the current frame image data, and ConvertToGray is a function for converting a color image to a grayscale image;
[0021] Step S2-2, use Gaussian blur to reduce image noise, and put the converted grayscale image data GrayFrame into the following formula: BlurredFrame=GaussianBlur(GrayFrame,(5,5));
[0022] Among them, BlurredFrame is the image data after blur processing, (5,5) is the convolution kernel size of Gaussian blur, and GaussianBlur is the function of image filtering operation;
[0023] Step S3: Use the detection algorithm in the Aruco library to identify the Arucomarkers proofreading code square pad in the image. The detection algorithm formula in the Aruco library is as follows: (MarkerCorners,MarkerIDs)=DetectArucoMarkers(BlurredFrame);
[0024] Among them, MarkerCorners is the corner coordinates of the detected Arucomarkers proofreading code square pad, MarkerIDs is the ID of the detected Arucomarkers proofreading code square pad, BlurredFrame is the input blurred image data, DetectArucoMarkers is the detection algorithm function in the Aruco library, which is used to detect the Arucomarkers proofreading code square pad in the input blurred image data BlurredFrame;
[0025] Step S4, extract the coordinates of the four corner points of each Arucomarker proofreading code square pad, expressed by the following formula: , , , ;
[0026] Among them, the coordinates ( , )、( , )、( , )、( , ) are the coordinates of the corner points of the rectangle that fits the standing long jump area, and the coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad A, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad B, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad C, which rotate clockwise from the corner points of the rectangular area that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad D, which rotate clockwise with the corner points of the rectangle that fits the standing long jump area as the starting point.
[0027] As a preferred solution, the calculation of the rectangular size of the standing long jump area includes the following steps:
[0028] Step S5, calculating the distance between adjacent Arucomarkers calibration code square pads;
[0029] Step S5-1: Calculate the pixel length of the standing long jump area rectangle using the Euclidean distance formula and pixel width , the formula is as follows: , ;
[0030] Get the pixel length of the standing long jump area rectangle and pixel width ;
[0031] Step S5-2: Detect the center position C of the Arucomarkers proofreading code square pad and the four corner points of the Arucomarkers proofreading code square pad , , , ;
[0032] Step S5-3, calculate the actual horizontal width and pixel ratio ScaleW and the actual height and pixel ratio ScaleH of each Arucomarkers proofreading code square pad: ScaleW=width×2 / (dist( , )+dist( , )); ScaleH=height×2 / (dist( , )+dist( , ));
[0033] Among them, dist represents the pixel Euclidean distance between two points, width and height are the actual width and height set when customizing the Arucomarkers calibration code square pad. When customizing the code plate, ensure that width=height and the length is fixed;
[0034] Step S5-4, calculate the length L and W of the standing long jump area rectangle by the following formula: ;in, ; ;in, ;
[0035] in, and Respectively represent the actual height and pixel ratio of Arucomarkers calibration code square pads A and B, and Respectively represent the actual horizontal width and pixel ratio of the Arucomarkers calibration code square pads A and B.
[0036] As a preferred solution, the site model construction includes the following steps:
[0037] Step S6, determining the position of the virtual take-off point at one end of the standing long jump area rectangle, and setting the coordinates of the virtual take-off point as the origin, that is, JumpPoint=(0,0);
[0038] Step S7, taking the virtual take-off point as the origin, establishing a virtual coordinate system extending from the virtual take-off point to the landing point, and calculating the coordinates of the scale lines in the virtual coordinate system according to the length and width of the known standing long jump area rectangle, that is, ScaleLinei=JumpPoint+i×ScaleDistance, i=1,2,…,n; wherein ScaleDistance is the distance between each scale line, n is the number of scale lines, and i is an index variable indicating the serial number of the scale line;
[0039] Step S8: Draw virtual scale lines on the video screen according to the calculated scale line coordinates, combine the virtual starting point, Arucomarkers coordinates and the drawn scale lines to obtain a complete site model, and map the site model to the video screen of the high-definition camera.
[0040] As a preferred solution, the shortest landing point positioning includes the following steps:
[0041] Step S9, taking the image frame of the landing moment and the image frame of the area before the take-off, and performing frame difference calculation;
[0042] Step S9-1, set Indicates the pixel value of the image frame at the moment of landing. The frame difference image representing the pixel value of the jump frame is calculated by the following formula: ;
[0043] in, Represents a closed operation, and obtains the frame difference map ;
[0044] Step S10, positioning range interception; using the heel key point coordinates in the virtual coordinate system And foot length Get the location range ;in, , , to capture the area map ;
[0045] Step S11: Extraction The middle edge and gradient direction are calculated, and the corresponding distance of the edge point coordinates is calculated. Through the analysis of the edge points, the coordinates corresponding to the minimum distance are selected as the landing point;
[0046] Step S11-1: For edge points , calculate its distance to the preset reference point ,in, is the coordinate of the reference point, the reference point is the starting point ;turn up The minimum value in This is the coordinate of the landing point that is initially determined;
[0047] Step S12: Combine the scale information in the virtual coordinate system to calculate between which two scales the landing point coordinates are located. Assuming that the two adjacent scales α and β, the distances from the landing point to the corresponding straight lines of the left and right scales α and β are respectively , , then the final landing point distance Calculated by the following formula: , get the shortest landing point position.
[0048] It can be seen from the technical solution provided by the present invention that the AI standing long jump evaluation system supporting automatic calibration provided by the present invention has the following beneficial effects:
[0049] 1. By using four Arucomarkers calibration code square mats arranged in a rectangular shape along the four corners of the standing long jump area, combined with high-definition cameras and specific algorithms, the automatic calibration of the venue is realized; simply by placing the calibration code square mats and turning on the system, an accurate venue model can be quickly established, which greatly shortens the venue preparation time and improves the test efficiency; whether it is in a school physical education class, a sports event site or other places where standing long jump tests are required, the venue layout work can be completed quickly, reducing the delays in the test process caused by untimely venue preparation. Whether it is an indoor gymnasium, classroom, or an outdoor playground, open space, or even some irregularly shaped venues, the system can be used to implement standing long jump tests, broadening the application scenarios of standing long jump tests and improving the versatility of the system;
[0050] 2. The high-definition camera can capture video images of the standing long jump area at a high frame rate, providing a rich data basis for subsequent detailed analysis; the evaluation module uses advanced image processing and analysis technology to conduct all-round monitoring of the long jump action of the tester; from the moment of take-off to the landing process, by real-time detection of the heel key points, analysis of the motion trajectory of the lower limb key points, and accurate determination of the take-off frame and the landing frame, combined with the virtual coordinate system and scale lines, the relative distance between the take-off point and the landing point can be accurately calculated, thereby obtaining accurate standing long jump results; compared with traditional manual measurement or simple sensor measurement methods, the present invention effectively reduces measurement errors, improves the accuracy and fairness of performance evaluation, and can meet the strict requirements of professional sports events, athlete training, and school physical education for performance accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 The present invention is a schematic diagram of the overall structure of an AI standing long jump evaluation system that supports automatic calibration. DETAILED DESCRIPTION
[0052] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0053] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0054] like Figure 1As shown, an embodiment of the present invention provides an AI standing long jump evaluation system supporting automatic calibration, comprising four Arucomarkers calibration code square pads arranged in a rectangular shape along the four corners of a standing long jump area, a data collection terminal, a cloud database, and an evaluation terminal;
[0055] The data collection end is a high-definition camera, which collects video images of the standing long jump area;
[0056] The cloud database includes a tester data storage unit and a score storage unit; the tester data storage unit stores the basic information and face information of the tester;
[0057] In this embodiment, the evaluation end includes a site model construction module, a personnel determination module, an evaluation module, and a data interaction module;
[0058] The site model construction module extracts a video stream containing four Arucomarkers proofreading code square pads from the video screen, extracts several frames of images containing four Arucomarkers proofreading code square pads from the video stream, preprocesses the extracted images, uses the Arucomarkers detection algorithm to identify the four Arucomarkers proofreading code square pads in the image, and extracts the coordinates of the four corner points of the four Arucomarkers proofreading code square pads; uses the Euclidean distance formula to calculate the distance between adjacent Arucomarkers proofreading code square pads to obtain the size of the standing long jump area rectangle; draws a virtual take-off point at one end of the standing long jump area rectangle; the site model construction module establishes a virtual coordinate system extending from the virtual take-off point to the landing point on the video screen of the standing long jump area based on the coordinates of the four Arucomarkers proofreading code square pads, calculates the long jump scale line coordinates in the virtual coordinate system according to the known standing long jump area rectangle size, and draws virtual scale lines on the video screen according to the calculated scale line coordinates to obtain a site model;
[0059] Further, assuming that the four Arucomarkers proofreading code square pads distributed at the four corners of the standing long jump area rectangle are A, B, C, and D, and the length and width of the standing long jump area rectangle are L and W, respectively, determining the coordinates of the four corner points of the four Arucomarkers proofreading code square pads includes the following steps:
[0060] The coordinates of the four corners of the standing long jump area rectangle are defined by the following formulas: A=(0,0), B=(L,0), C=(L,W), D=(0,W);
[0061] Step S1, extracting a video stream containing four Arucomarkers proofreading code square pads from a video screen, and extracting several frames of images containing four Arucomarkers proofreading code square pads from the video stream;
[0062] Step S2: preprocess the extracted image:
[0063] Step S2-1, convert each frame of image into a grayscale image using the following formula: GrayFrame = ConvertToGray (Frame), where GrayFrame is the converted grayscale image data, Frame is the current frame image data, ConvertToGray is a function that converts a color image to a grayscale image. ConvertToGray calculates the grayscale value by taking a weighted average of the three color channels of the color image: red (R), green (G), and blue (B). The calculation formula is: ;
[0064] Where Gray represents the gray value, R, G, and B represent the red, green, and blue channel values of the corresponding pixel in the color image respectively;
[0065] Step S2-2, use Gaussian blur to reduce image noise, and put the converted grayscale image data GrayFrame into the following formula: BlurredFrame=GaussianBlur(GrayFrame,(5,5));
[0066] Among them, BlurredFrame is the image data after blur processing, (5,5) is the convolution kernel size of Gaussian blur, and GaussianBlur is the function of image filtering operation. The specific calculation steps are as follows (the coordinates in the calculated image are For example, the blurred value of a pixel:
[0067] 1. Determine the convolution kernel: Calculate each position in the convolution kernel based on the given convolution kernel size (such as (5,5)) and Gaussian standard deviation for The weight corresponding to the convolution kernel , the weight calculation is based on the above Gaussian function , and usually all weights are normalized so that their sum is 1;
[0068] 2. Weighted summation: For coordinates The pixel value of its surrounding neighborhood (determined by the convolution kernel size) And the corresponding convolution kernel weight Multiply and add all the results to get the blurred value of the pixel , the calculation formula is as follows:
[0069] in, The coordinates in the original image are The pixel value of
[0070] Step S3: Use the detection algorithm in the Aruco library to identify the Arucomarkers proofreading code square pad in the image. The detection algorithm formula in the Aruco library is as follows: (MarkerCorners,MarkerIDs)=DetectArucoMarkers(BlurredFrame);
[0071] Among them, MarkerCorners is the corner coordinates of the detected Arucomarkers proofreading code square pad, MarkerIDs is the ID of the detected Arucomarkers proofreading code square pad, BlurredFrame is the input blurred image data, DetectArucoMarkers is the detection algorithm function in the Aruco library, which is used to detect the Arucomarkers proofreading code square pad in the input blurred image data BlurredFrame;
[0072] Step S4, extract the coordinates of the four corner points of each Arucomarker proofreading code square pad, expressed by the following formula: , , , ;
[0073] Among them, the coordinates ( , )、( , )、( , )、( , ) are the coordinates of the corner points of the rectangle that fits the standing long jump area, and the coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad A, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad B, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad C, which rotate clockwise from the corner points of the rectangular area that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad D, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area;
[0074] The calculation of the standing long jump area rectangle dimensions includes the following steps:
[0075] Step S5, calculating the distance between adjacent Arucomarkers calibration code square pads;
[0076] Step S5-1: Calculate the pixel length of the standing long jump area rectangle using the Euclidean distance formula and pixel width , the formula is as follows: , ;
[0077] Get the pixel length of the standing long jump area rectangle and pixel width ;
[0078] Step S5-2: Detect the center position C of the Arucomarkers proofreading code square pad and the four corner points of the Arucomarkers proofreading code square pad , , , ;
[0079] Step S5-3, calculate the actual horizontal width and pixel ratio ScaleW and the actual height and pixel ratio ScaleH of each Arucomarkers proofreading code square pad: ScaleW=width×2 / (dist( , )+dist( , )); ScaleH=height×2 / (dist( , )+dist( , ));
[0080] Among them, dist represents the pixel Euclidean distance between two points, width and height are the actual width and height set when customizing the Arucomarkers calibration code square pad. When customizing the code plate, ensure that width=height and the length is fixed;
[0081] Step S5-3, calculate the length L and W of the standing long jump area rectangle by the following formula: ;in, ; ;in, ;
[0082] in, and Respectively represent the actual height and pixel ratio of Arucomarkers calibration code square pad A and B, and Respectively represent the actual horizontal width and pixel ratio of the Arucomarkers calibration code square pads A and B.
[0083] As a preferred solution, the site model construction includes the following steps:
[0084] Step S6, determining the position of the virtual take-off point at one end of the standing long jump area rectangle, and setting the coordinates of the virtual take-off point as the origin, that is, JumpPoint=(0,0);
[0085] Step S7, taking the virtual take-off point as the origin, establishing a virtual coordinate system extending from the virtual take-off point to the landing point, and calculating the coordinates of the scale lines in the virtual coordinate system according to the length and width of the known standing long jump area rectangle, that is, ScaleLinei=JumpPoint+i×ScaleDistance, i=1,2,…,n; wherein ScaleDistance is the distance between each scale line, n is the number of scale lines, and i is an index variable indicating the serial number of the scale line;
[0086] Step S8, draw virtual scale lines on the video screen according to the calculated scale line coordinates, combine the virtual starting point, Arucomarkers coordinates and the drawn scale lines to obtain a complete site model, and map the site model to the video screen of the high-definition camera;
[0087] The personnel identification module identifies the face information of the person who enters the virtual starting point in the video screen, and exchanges data with the test personnel data storage unit based on the face information to determine the identity of the test personnel;
[0088] The evaluation module receives all the motion images of the tester during the standing long jump, tracks the tester, and detects violations based on the target tracking images; after the tester starts to jump, the key points of the heel are detected in real time, each frame of data is recorded, the take-off frame and landing frame of the tester are determined, the shortest landing point is located, the coordinates of the shortest landing point are determined, the coordinates of the landing point are brought into the site model, and the relative distance between the take-off point and the landing point is calculated as the tester's standing long jump score;
[0089] Further, the shortest landing point positioning includes the following steps:
[0090] Step S9, taking the image frame of the landing moment and the image frame of the area before the take-off, and performing frame difference calculation;
[0091] Step S9-1: Set Indicates the pixel value of the image frame at the moment of landing. The frame difference image representing the pixel value of the jump frame is calculated by the following formula: ;
[0092] in, Represents a closed operation, and obtains the frame difference map ;
[0093] Step S10, positioning range interception; using the heel key point coordinates in the virtual coordinate system And foot length Get the location range ;in, , , to capture the area map ;
[0094] Step S11: Extraction The middle edge and gradient direction are calculated, and the corresponding distance of the edge point coordinates is calculated. Through the analysis of the edge points, the coordinates corresponding to the minimum distance are selected as the landing point;
[0095] Step S11-1: For edge points , calculate its distance to the preset reference point ,in, is the coordinate of the reference point, the reference point is the starting point ;turn up The minimum value in This is the coordinate of the landing point that is initially determined;
[0096] Step S12: Combine the scale information in the virtual coordinate system to calculate between which two scales the landing point coordinates are located. Assuming that the two adjacent scales α and β, the distances from the landing point to the corresponding straight lines of the left and right scales α and β are respectively , , then the final landing point distance Calculated by the following formula: , get the shortest landing point position;
[0097] The data interaction module receives the identity of the tester and the standing long jump result of the tester, sorts them out, and transmits them to the result storage unit.
[0098] In this embodiment, it also includes a user terminal. The tester logs in to the tester data storage unit through the user terminal to build an account to register as a tester. The registration content includes basic information and facial information. After the score storage unit receives the tester's identity and the tester's standing long jump score, it organizes the data to obtain the tester's score sheet and sends the score sheet to the user terminal.
[0099] The present invention will be further described in detail below based on the above system according to the test personnel's standing long jump test steps:
[0100] 1. Site Preparation
[0101] Accurately place four Arucomarkers calibration code square mats at the four corners of the selected standing long jump test site, ensuring that they form a standard rectangular layout; the positions and angles of these four mats need to be precisely adjusted to provide an accurate benchmark for the subsequent site model construction;
[0102] 2. System startup and initialization
[0103] Turn on the HD camera to ensure that it works properly and can clearly capture the entire standing long jump area, including the four Arucomarkers calibration code square mats and the tester's activities in the area; the camera establishes a stable connection with the evaluation end to transmit video data in real time;
[0104] Start the evaluation end, and the site model building module starts to obtain video images from the camera;
[0105] 3. Site model construction
[0106] Video streaming and image extraction:
[0107] The site model building module extracts continuous video data containing four Arucomarkers calibration code square mats from the video stream collected by the high-definition camera;
[0108] Further filter and extract several frames of representative images from the video stream, which will be used for subsequent precise processing;
[0109] Image preprocessing:
[0110] Grayscale conversion: For each frame of the extracted image, use the formula GrayFrame=ConvertToGray(Frame) to convert it into a grayscale image; this step simplifies the image data by removing the color information in the image, highlights the contrast between the Arucomarkers calibration code square pad and the background, and facilitates subsequent feature extraction and analysis;
[0111] Gaussian blur: The converted grayscale image data GrayFrame is put into the formula BlurredFrame=GaussianBlur(GrayFrame,(5,5)) for Gaussian blur processing; by setting the convolution kernel size to (5,5), the image is smoothed to effectively reduce the noise interference in the image, making the edge of the Arucomarkers calibration code square pad clearer and easier to detect accurately;
[0112] Arucomarkers detection and corner point coordinate extraction:
[0113] Using the detection algorithm in the Aruco library (MarkerCorners, MarkerIDs) = DetectArucoMarkers (BlurredFrame) analyzes the preprocessed image; the algorithm can accurately identify the position and characteristics of the Arucomarkers calibration code square pad in the blurred image BlurredFrame;
[0114] The detection algorithm returns the corner point coordinates MarkerCorners and unique ID MarkerIDs of each Arucomarker proofreading square pad; for the four Arucomarkers proofreading square pads, the coordinates of their four corner points are extracted respectively. Assuming that the four Arucomarkers proofreading square pads distributed at the four corners of the standing long jump area are A, B, C, and D respectively, their corner point coordinates are expressed as , , , , where the coordinates ( , )、( , )、( , )、( , ) are the coordinates of the corner points of the rectangle that fits the standing long jump area, and the coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad A, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad B, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad C, which rotate clockwise from the corner points of the rectangular area that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad D, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area;
[0115] Calculation of rectangular dimensions for standing long jump area:
[0116] Calculate the distance between adjacent Arucomarkers calibration code square pads;
[0117] Calculate the pixel length of the standing long jump area rectangle using the Euclidean distance formula and pixel width , the formula is as follows: , ;
[0118] Get the pixel length of the standing long jump area rectangle and pixel width ;
[0119] Detect the center position C of the Arucomarkers proofreading code square pad and the four corner points of the Arucomarkers proofreading code square pad , , , ;
[0120] Calculate the actual horizontal width and pixel ratio ScaleW and the actual height and pixel ratio ScaleH of each Arucomarkers calibration code square pad: ScaleW=width×2 / (dist( , )+dist( , )); ScaleH=height×2 / (dist( , )+dist( , ));
[0121] Among them, dist represents the pixel Euclidean distance between two points, width and height are the actual width and height set when customizing the Arucomarkers calibration code square pad. When customizing the code plate, ensure that width=height and the length is fixed;
[0122] The lengths L and W of the standing long jump area rectangle are calculated using the following formulas: ;in, ; ;in, ;
[0123] in, and Respectively represent the actual height and pixel ratio of Arucomarkers calibration code square pad A and B, and Respectively represent the actual horizontal width and pixel ratio of the Arucomarkers calibration code square pads A and B;
[0124] Virtual starting point and virtual coordinate system establishment and scale line drawing:
[0125] Determine the position of the virtual take-off point at one end of the standing long jump area rectangle close to the take-off direction, and set its coordinates as the origin, that is, JumpPoint=(0,0);
[0126] With the virtual take-off point as the origin, a two-dimensional virtual coordinate system is established, which extends from the virtual take-off point to the long jump landing point. According to the known length L and width W of the standing long jump area rectangle, and the preset distance ScaleDistance between each scale line, the coordinates of each scale line in the virtual coordinate system are calculated by the formula ScaleLinei=JumpPoint+i×ScaleDistance, i=1,2,…,n (i=1,2,…,n; where n is the number of scale lines), i is an index variable, indicating the serial number of the scale line.
[0127] According to the calculated scale line coordinates, virtual scale lines are drawn on the video screen. These scale lines will serve as an accurate reference for the subsequent calculation of long jump results. Finally, the virtual take-off point, Arucomarkers coordinates and drawn scale lines are integrated together to form a complete site model, and the site model is mapped to the video screen captured in real time by the high-definition camera, so that the virtual site model perfectly matches the actual scene, laying the foundation for accurate long jump evaluation.
[0128] 4. Personnel identification
[0129] When the tester enters the standing long jump area and approaches the virtual take-off point, the high-definition camera captures the video image containing the tester's face and transmits it to the person determination module;
[0130] The personnel identification module uses advanced face recognition algorithms to extract features from faces in the video and obtain facial feature information of the test personnel;
[0131] Compare and match the extracted facial feature information with the facial information stored in the tester data storage unit in the cloud database; search for the most similar facial record by traversing the data in the database;
[0132] Once the matching facial information is found, the identity of the tester is determined based on the associated records, and the identity information is associated with the subsequent evaluation process to ensure that the results are accurately recorded under the name of the corresponding tester;
[0133] 5. Long jump action assessment
[0134] Target tracking and violation detection (before takeoff):
[0135] The evaluation module receives the motion images of the tester in the standing long jump area transmitted by the high-definition camera, and uses the target tracking algorithm to track the tester in real time, locking his position and motion trajectory in the image;
[0136] When the tester is preparing to jump, the system detects 26 key points of his body posture, analyzes the positional relationship and movement trend of these key points, and determines whether there are any illegal actions such as crossing the line, single-leg hopping, step-step hopping, front-back hopping, and run-up hopping. At the same time, the image segmentation algorithm is used to conduct a detailed analysis of the 100×100 area around the key points of the toes of the posture to accurately detect whether there is any line stepping at the moment of take-off. If any illegal behavior is detected, the system immediately records the type and time of the violation, and can prompt the tester through sound or image.
[0137] Start frame determination:
[0138] After the tester starts jumping, the evaluation module detects the position change of the key point of the heel in real time and records the coordinate data of the key point of the heel in each frame of the image;
[0139] By analyzing the changes in the vertical position of the key points of the ankle, knee and hip of the lower limbs over time, the kinematic formula is used to calculate the vertical velocity (in is the vertical speed, is the change in the vertical position of the lower limb key points between adjacent frames, is the time interval between adjacent frames);
[0140] Continuously monitor the change of vertical speed. When the vertical speed gradually changes from a positive value to close to zero or reverses, determine the corresponding video frame as the take-off frame, and record the number of the take-off frame and the coordinates of the key point of the heel at the moment of take-off.
[0141] Landing frame determination:
[0142] During the long jump, the evaluation module continues to track the tester's movement status, focusing on the movement trajectory of the key point of the heel. During the long jump, the movement trajectory of the key point of the heel presents a parabola-like wave shape.
[0143] By analyzing the hip motion curve, the stable trough in the curve is found; the specific method is to perform curve fitting on the change of the coordinates of the key points of the hip over time, determine the extreme points of the curve, and judge the appearance of the stable trough in combination with the kinematic law; when it is determined that the hip motion curve has a stable trough, it is determined that the test person has landed, and the corresponding video frame at this time is the landing frame, and the number of the landing frame and the approximate position of the key point of the heel at the moment of landing are recorded;
[0144] Shortest landing point positioning:
[0145] Frame difference map calculation: After determining the take-off frame and landing frame, the evaluation module takes the image frame of the landing moment (i.e., the image frame of the long jump area with the shortest contact point) and the image frame of the area before the take-off to calculate the frame difference; Indicates the pixel value of the image frame at the moment of landing. Indicates the pixel value of the jump frame and the frame difference image By formula Calculate, where Represents a closed operation, which is used to reduce noise interference in the image and obtain a clear frame difference image ,The frame difference image can highlight the difference area between the landing moment and the picture before take-off, which helps to locate the landing point;
[0146] Positioning range interception: The coordinates of the key heel points at the moment of landing are determined in the virtual coordinate system and the known length of the sole Based on this, get the positioning range ;in , , according to this positioning range in the frame difference map Intercepted area map , the area map is considered to contain the most likely effective information of the landing point, which narrows the search range of subsequent calculations and improves the calculation efficiency and accuracy;
[0147] Landing point determination: intercepted area map Perform edge detection to extract edge and gradient direction information; for each edge point in the region map , calculate its distance to the preset reference point (starting point ) (in is the reference point coordinate); by comparing the distances from all edge points to the reference point, select the edge point coordinate corresponding to the minimum distance As the coordinates of the landing point determined initially, because in a normal long jump action, the landing point is usually near a position that has a certain relationship with the take-off point, based on this principle, the landing point position that best conforms to the long jump movement law is found through distance calculation;
[0148] Distance conversion: Combine the scale information in the virtual coordinate system to determine which two adjacent scales α and β the landing point coordinates are between; calculate the distances from the landing point to the corresponding straight lines of the left and right scales α and β respectively. , , and then by the formula Calculate the final landing point distance , get the exact coordinates of the shortest landing point, which will be used for subsequent score calculations;
[0149] Calculation of results:
[0150] The evaluation module obtains the coordinates of the take-off point (virtual take-off point JumpPoint=(0,0)) and the shortest landing point determined by the above steps;
[0151] Using the distance formula (in is the starting point coordinate, The relative distance between the take-off point and the landing point is calculated (the coordinates of the landing point). This distance is the standing long jump score of the tester, accurate to centimeters or smaller units to meet the requirements for score accuracy in different scenarios;
[0152] 6. Data Storage and Transmission
[0153] The data interaction module receives the tester's identity information and the calculated standing long jump score from the evaluation module;
[0154] The data is collated and formatted to ensure the integrity and accuracy of the data; the data is then transferred to the score storage unit of the cloud database for storage; the score storage unit associates the scores with the corresponding tester identities, and classifies and archives them according to certain rules to facilitate subsequent query and statistical analysis;
[0155] If the system is equipped with a user terminal, the score storage unit will further process the tester's score table after completing the data storage, such as generating a visual score report, statistical charts, etc.; then send this information to the user terminal, and the tester can view his or her own scores, historical test records, score trend analysis and other detailed information by logging into the user terminal, thereby realizing effective interaction with the evaluation system.
[0156] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. An AI standing long jump evaluation system supporting automatic calibration, characterized in that: include: Four square mats of Arucomarkers calibration codes arranged in a rectangular shape along the four corners of the standing long jump area; Data collection terminal; The data acquisition end is a high-definition camera, which collects video images of the standing long jump area; Cloud database; the cloud database includes a tester data storage unit and a score storage unit; The tester data storage unit stores the basic information and facial information of the tester; Evaluation end; the evaluation end includes a site model building module, a personnel determination module, an evaluation module and a data interaction module; The site model construction module extracts a video stream containing four Arucomarkers proofreading code square pads from the video screen, extracts several frames of images containing four Arucomarkers proofreading code square pads from the video stream, pre-processes the extracted images, uses the Arucomarkers detection algorithm to identify the four Arucomarkers proofreading code square pads in the image, and extracts the coordinates of the four corner points of the four Arucomarkers proofreading code square pads; uses the Euclidean distance formula to calculate the distance between adjacent Arucomarkers proofreading code square pads to obtain the size of the standing long jump area rectangle; draws a virtual take-off point at one end of the standing long jump area rectangle; the site model construction module establishes a virtual coordinate system extending from the virtual take-off point to the landing point on the video screen of the standing long jump area based on the coordinates of the four Arucomarkers proofreading code square pads, calculates the long jump scale line coordinates in the virtual coordinate system according to the known standing long jump area rectangle size, and draws virtual scale lines on the video screen according to the calculated scale line coordinates to obtain a site model; The personnel determination module identifies the face information of the person who enters the virtual starting point in the video screen, and communicates with the test personnel data storage unit according to the face information to determine the identity of the test personnel; The evaluation module receives all the motion pictures of the tester during the standing long jump, tracks the tester, and detects violations based on the target tracking pictures; after the tester starts to jump, the heel key points are detected in real time, each frame of data is recorded, the take-off frame and the landing frame of the tester are determined, the shortest landing point is located, the coordinates of the shortest landing point are determined, the coordinates of the landing point are brought into the site model, and the relative distance between the take-off point and the landing point is calculated as the standing long jump score of the tester; The data interaction module receives the identity of the tester and the standing long jump score of the tester, sorts them out, and transmits them to the score storage unit.
2. The AI standing long jump evaluation system supporting automatic calibration according to claim 1, characterized in that: It also includes a user end, through which the tester logs into the tester data storage unit to build an account to register the tester, and the registration content includes basic information and facial information; after the score storage unit receives the tester's identity and the tester's standing long jump score, it organizes the data to obtain the tester's score sheet, and sends the score sheet to the user end.
3. The AI standing long jump evaluation system supporting automatic calibration according to claim 1, characterized in that: Assuming that the four Arucomarkers calibration code square pads distributed at the four corners of the standing long jump area rectangle are A, B, C, and D, and the length and width of the standing long jump area rectangle are L and W, respectively, determining the coordinates of the four corner points of the four Arucomarkers calibration code square pads includes the following steps: The coordinates of the four corners of the standing long jump area rectangle are defined by the following formulas: A=(0,0), B=(L,0), C=(L,W), D=(0,W); Step S1, extracting a video stream containing four Arucomarkers proofreading code square pads from a video screen, and extracting several frames of images containing four Arucomarkers proofreading code square pads from the video stream; Step S2: preprocess the extracted image: Step S2-1, convert each frame of image into a grayscale image using the following formula: GrayFrame = ConvertToGray (Frame), where GrayFrame is the converted grayscale image data, Frame is the current frame image data, and ConvertToGray is a function for converting a color image to a grayscale image; Step S2-2, use Gaussian blur to reduce image noise, and put the converted grayscale image data GrayFrame into the following formula: BlurredFrame=GaussianBlur(GrayFrame,(5,5)); Among them, BlurredFrame is the image data after blur processing, (5,5) is the convolution kernel size of Gaussian blur, and GaussianBlur is the function of image filtering operation; Step S3: Use the detection algorithm in the Aruco library to identify the Arucomarkers proofreading code square pad in the image. The detection algorithm formula in the Aruco library is as follows: (MarkerCorners,MarkerIDs)=DetectArucoMarkers(BlurredFrame); Among them, MarkerCorners is the corner coordinates of the detected Arucomarkers proofreading code square pad, MarkerIDs is the ID of the detected Arucomarkers proofreading code square pad, BlurredFrame is the input blurred image data, DetectArucoMarkers is the detection algorithm function in the Aruco library, which is used to detect the Arucomarkers proofreading code square pad in the input blurred image data BlurredFrame; Step S4, extract the coordinates of the four corner points of each Arucomarker proofreading code square pad, expressed by the following formula: , , , ; Among them, the coordinates ( , )、( , )、( , )、( , ) are the coordinates of the corner points of the rectangle that fits the standing long jump area, and the coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad A, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad B, which rotate clockwise from the corner points of the rectangle that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad C, which rotate clockwise from the corner points of the rectangular area that fits the standing long jump area. The coordinates ( , )、( , )、( , ) are the coordinates of the other three corner points of the Arucomarker calibration code square pad D, which rotate clockwise with the corner points of the rectangle that fits the standing long jump area as the starting point.
4. The AI standing long jump evaluation system supporting automatic calibration according to claim 3, characterized in that: The calculation of the rectangular size of the standing long jump area comprises the following steps: Step S5, calculating the distance between adjacent Arucomarkers calibration code square pads; Step S5-1: Calculate the pixel length of the standing long jump area rectangle using the Euclidean distance formula and pixel width , the formula is as follows: , ; Get the pixel length of the standing long jump area rectangle and pixel width ; Step S5-2: Detect the center position C of the Arucomarkers proofreading code square pad and the four corner points of the Arucomarkers proofreading code square pad , , , ; Step S5-3, calculate the actual horizontal width and pixel ratio ScaleW and the actual height and pixel ratio ScaleH of each Arucomarkers proofreading code square pad: ScaleW=width×2 / (dist( , )+dist( , )); ScaleH=height×2 / (dist( , )+dist( , )); Among them, dist represents the pixel Euclidean distance between two points, width and height are the actual width and height set when customizing the Arucomarkers calibration code square pad. When customizing the code plate, ensure that width=height and the length is fixed; Step S5-4, calculate the length L and W of the standing long jump area rectangle by the following formula: ;in, ; ;in, ; in, and Respectively represent the actual height and pixel ratio of Arucomarkers calibration code square pad A and B, and Respectively represent the actual horizontal width and pixel ratio of the Arucomarkers calibration code square pads A and B.
5. The AI standing long jump evaluation system supporting automatic calibration according to claim 4, characterized in that: The site model construction includes the following steps: Step S6, determining the position of the virtual take-off point at one end of the standing long jump area rectangle, and setting the coordinates of the virtual take-off point as the origin, that is, JumpPoint=(0,0); Step S7, taking the virtual take-off point as the origin, establishing a virtual coordinate system extending from the virtual take-off point to the landing point, and calculating the coordinates of the scale lines in the virtual coordinate system according to the length and width of the known standing long jump area rectangle, that is, ScaleLinei=JumpPoint+i×ScaleDistance, i=1,2,…,n; wherein ScaleDistance is the distance between each scale line, n is the number of scale lines, and i is an index variable indicating the serial number of the scale line; Step S8: Draw virtual scale lines on the video screen according to the calculated scale line coordinates, combine the virtual starting point, Arucomarkers coordinates and the drawn scale lines to obtain a complete site model, and map the site model to the video screen of the high-definition camera.
6. The AI standing long jump evaluation system supporting automatic calibration according to claim 5, characterized in that: The shortest landing point positioning comprises the following steps: Step S9, taking the image frame of the landing moment and the image frame of the area before the take-off, and performing frame difference calculation; Step S9-1: Set Indicates the pixel value of the image frame at the moment of landing. The frame difference image representing the pixel value of the jump frame is calculated by the following formula: ; in, Represents a closed operation, and obtains the frame difference map ; Step S10, positioning range interception; using the heel key point coordinates in the virtual coordinate system And foot length Get the location range ;in, , , to capture the area map ; Step S11: Extraction The middle edge and gradient direction are calculated, and the corresponding distance of the edge point coordinates is calculated. Through the analysis of the edge points, the coordinates corresponding to the minimum distance are selected as the landing point; Step S11-1: For edge points , calculate its distance to the preset reference point ,in, is the coordinate of the reference point, the reference point is the starting point ;turn up The minimum value in This is the coordinate of the landing point that is initially determined; Step S12: Combine the scale information in the virtual coordinate system to calculate between which two scales the landing point coordinates are located. Assuming that the two adjacent scales α and β, the distances from the landing point to the corresponding straight lines of the left and right scales α and β are respectively , , then the final landing point distance Calculated by the following formula: , get the shortest landing point position.
Citation Information
Patent Citations
Long jump and triple jump foul detector system
CA2056898A1
Free standing long jump testing method and device
CN110624203A
Standing long jump distance measuring method and system based on target detection and image processing
CN116392800A
Binocular camera-based standing long jump distance measurement method
CN117663996A
Indoor parking lot digital twin system based on two-dimensional code and monitoring method thereof
CN118645003A
Cited By
Physical exercise AI standing long jump evaluation data management system
CN120242438A
