Standing long jump examination system based on AI vision
Through the AI vision-based standing long jump examination system, the limitations of image processing and human posture recognition in the existing technology are solved, and the accurate recognition and detailed analysis of opposite long jump actions are realized, detailed action reports are provided, and measurements are improved. Accuracy and user experience.
Patent Information
- Application Number
- CN202411130343.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing standing long jump test system has limitations in image processing, human posture recognition, motion analysis and data storage, resulting in inaccurate measurement results, single application scenarios, and inability to provide detailed motion analysis reports.
Adopt a standing long jump examination system based on AI vision, including image acquisition module, image processing module, AI analysis module, data storage module and result output module. Images are acquired through high-definition cameras, image distortion correction and horizontal perspective transformation are used to combine AI The human posture recognition and action recognition model calculates key parameters such as the start and landing of the candidate's movements, and generates a detailed action analysis report.
It improves the processing accuracy of image data and the accuracy of human posture recognition, realizes comprehensive evaluation and accurate measurement of opposite long jump movements, provides intuitive motion analysis reports and training guidance, and enhances the fairness and user experience of the system.
Smart Images

Figure CN119068413A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of AI vision algorithms, in particular to a standing long jump test system based on AI vision. Background Art
[0002] Standing long jump is one of the common items in physical examinations, which has high requirements on the physical fitness and coordination ability of candidates. However, the traditional standing long jump test mainly relies on manual measurement and scoring, which is not only inefficient but also easily affected by human factors, resulting in inaccurate measurement and scoring. Therefore, it is particularly important to develop a system that can automatically and accurately conduct standing long jump tests.
[0003] Most of the data in the existing standing long jump training and testing process is collected through manual measurement and infrared sensor equipment, and then manually entered into the information system. This process is inefficient and has problems such as manual measurement errors and cumbersome equipment deployment. In addition, data recording requires a lot of manual recording and subsequent verification. In addition, it can only realize standing long jump performance judgment and action violation judgment, and cannot provide standing long jump process action analysis, and the application scenario is single.
[0004] With the rapid development of artificial intelligence and computer vision technology, a standing long jump test system based on AI vision has become possible. This system can use high-definition cameras to obtain the action images and videos of candidates, and automatically identify and calculate the candidates' long jump scores through image processing and analysis technology. This method can not only improve the efficiency of the test, but also reduce the interference of human factors and improve the fairness and accuracy of the test.
[0005] The existing background technology still has the following shortcomings:
[0006] 1. There are still certain limitations in image processing. Affected by factors such as ambient light, camera angle and resolution, the quality of the collected images is poor. Subsequent image processing technology cannot completely eliminate problems such as image distortion and perspective transformation, which affects the accuracy of the measurement results to a certain extent.
[0007] 2. The existing standing long jump test system still needs to be improved in terms of human posture recognition and motion analysis. Since the standing long jump action is complex and changeable, involving multiple stages such as take-off, aerial posture and landing, more sophisticated and accurate human posture recognition technology is needed to capture and analyze these actions. The system also needs to have stronger adaptive capabilities for the individual differences and action characteristics of different candidates.
[0008] 3. The existing standing long jump test system also has certain deficiencies in data storage and user display. Most of the existing systems do not have storage functions and cannot track the growth and changes of candidates in real time. In terms of user display, the existing systems often only provide simple score displays and action violation prompts, and lack detailed and intuitive action analysis reports. Summary of the invention
[0009] In order to solve the above technical problems, the present invention provides a standing long jump test system based on AI vision to solve the problems in the prior art.
[0010] The technical solution is as follows:
[0011] A standing long jump test system based on AI vision, comprising:
[0012] Image acquisition module: The image acquisition module uses a high-definition camera to acquire images and videos of the standing long jump test area;
[0013] Image processing module: The image processing module is connected to the image acquisition module in communication, receives the image and video information collected by the image acquisition module, and uses opencv related functions to perform image distortion correction and horizontal perspective transformation operations;
[0014] AI analysis module: The AI analysis module is connected to the image processing module in communication, performs AI human posture recognition and AI object recognition according to the information transmitted by the image processing module, and performs related analysis and calculation;
[0015] Data storage module: The data storage module is communicatively connected to the AI analysis module, receives data information and report information transmitted by the AI analysis module for storage;
[0016] Result output module: The result output module is communicatively connected to the AI analysis module, receives information transmitted by the AI analysis module, and outputs test results, data analysis reports, and training instructions in real time.
[0017] Preferably, the image acquisition module uses a camera to obtain image information of the venue and the examinee during the examinee's standing long jump test.
[0018] Preferably, the image processing module includes an image distortion correction unit and a horizontal perspective transformation unit:
[0019] The image distortion correction unit is used to perform distortion correction on the collected image information;
[0020] The horizontal perspective transformation unit is used to perform horizontal perspective transformation on the image.
[0021] Preferably, the image distortion correction unit first identifies the logo of the standing jump field, 4 squares, according to the image recognition algorithm logic, and then corrects the image distortion size according to the camera's own parameters. The opencv built-in functions initUndistortRectifyMap and initUndistortRectifyMap perform image distortion correction;
[0022] The initUndistortRectifyMap function calculates the distortion-corrected position of each pixel in the image based on the camera's intrinsic parameter matrix and distortion coefficients:
[0023] First, the intrinsic parameter matrix of the camera is usually expressed as:
[0024]
[0025] Among them, x and y represent horizontal and vertical, f x and f y is the focal length in the horizontal and vertical directions, c x and c y is the principal point coordinate, which represents the coordinate of the intersection of the camera optical axis and the imaging plane in the pixel coordinate system; the distortion coefficient includes the radial distortion coefficient k 1 , k 2 and the tangential distortion coefficient p 1 、p 2 ;
[0026] The camera distortion coefficient, camera principal point coordinates and focal length can be obtained from the relevant camera manufacturer's documents;
[0027] For any pixel coordinate point (x, y) in the image, the coordinates of its original coordinates after distortion can be expressed as (x d ,y d ), obtained by the following calculation:
[0028] x d =x(1+k 1 r 2 +k 2 r 4 )+2p 1 (xc x )(yc y )+p 2 (r 2 +2(xc x ) 2 )
[0029] y d =y(1+k 1 r 2 +k 2 r 4 )+2p2 (xc x )(yc y )+p 1 (r 2 +2(yc y ) 2 )
[0030] in
[0031] The initUndistortRectifyMap function generates the horizontal mapping table mapx and the vertical mapping table mapy based on these calculations; for each pixel position (x, y), the value mapx(x, y) in the horizontal mapping table is the corrected coordinate x of the pixel in the horizontal direction. c , expressed as:
[0032]
[0033] Among them, z represents the distance from the camera to the photographed object, and the value mapy(x, y) in the vertical mapping table is the corrected coordinate y of the pixel in the vertical direction c , expressed as:
[0034]
[0035] When correcting a distorted image, by looking up these mapping tables, each pixel is moved to the corresponding correction position according to the value in the mapping table, thereby achieving distortion correction for the entire image and obtaining an undistorted image.
[0036] Preferably, the horizontal perspective transformation unit changes the presentation of the image in the horizontal direction through a specific operation according to the visual law of “near objects are larger and far objects are smaller”;
[0037] In OpenCV, it is called: "PerspectiveTransformation";
[0038] For the corrected data point (x d ,y d ), after horizontal perspective transformation, it becomes (x′, y′), which can be described by the following mathematical relationship:
[0039]
[0040] Among them, by adjusting the element a related to the horizontal direction 11 、a 12 and a 13 The value of can realize various deformations of image stretching, compression, tilt in the horizontal direction, simulating the visual difference when observing objects from different horizontal perspectives;
[0041] In order to achieve a specific horizontal perspective effect, based on the known coordinates of corresponding points on the original image and the target image, a set of equations is constructed to solve the elements of the transformation matrix; once this transformation matrix is determined, it can be applied to the entire original image to obtain a new image after the horizontal perspective transformation, so that it shows the expected perspective effect in the horizontal direction.
[0042] Preferably, the A1 analysis module includes an action recognition model unit and a landing point recognition unit;
[0043] The action recognition model unit is used to establish a recognition model, and judge the tester's jumping and landing actions based on the logic of human action skeleton recognition;
[0044] The landing point recognition unit recognizes the landing point of the shoe according to the recognition model, calculates the displacement of the examinee during the entire project test, and calculates the distance of the standing jump by reference to the coordinate system;
[0045] Preferably, the action recognition model unit uses a neural network to establish a model, and based on the logic of human action skeleton recognition, determines the tester's actions such as jumping and landing; constructs a neural network model:
[0046] Input layer: receives input image data and establishes a three-dimensional coordinate system based on the image information;
[0047] Feature extraction layer: used to automatically extract meaningful features from images, which can reflect the structure and posture of the human body;
[0048] Output layer: Output includes key point coordinates, take-off angle, take-off height and other information;
[0049] Calculation process: Match the input image with the pre-defined human body template, determine the key point position according to the matching degree, and the detected human body key point coordinate set is {p i (x i ,y i ,z i )}, where i represents different key points, each of which has its specific coordinates in three-dimensional space;
[0050] The judgment is made by analyzing the height change of the key point; for the key point P (x, y, z), the change rate of its z-axis coordinate in the time series is calculated When the rate of change changes rapidly from negative to positive and then from positive to negative within a period of time, the take-off and landing moments are accurately determined;
[0051] The take-off angle is calculated based on the shoulder and hip joint key points, and the vector between these two key points is measured. Then define the ground vector The cosine of the angle is obtained by calculating the vector dot product and the modulus length:
[0052]
[0053] Finally, the take-off angle is obtained through inverse trigonometric functions;
[0054] Calculation of take-off height: Compare the height of the ankle key point at the moment of take-off with the height when initially standing, and the difference is the take-off height;
[0055] Calculation of the jump range: determine the highest point of the body during the jump, and calculate the difference between its height and the take-off height as the jump range;
[0056] For the landing angle, it is calculated by the angle between the line connecting the ankle and knee key points and the ground;
[0057] Illegal action detection, for hand touching the ground violation, the z-axis coordinate of the hand key point P (x, y, z) is compared with the set ground, and if z < 0, it is judged as a hand touching the ground violation.
[0058] The violation of exceeding the take-off line is determined by comparing the horizontal position of the key point of the foot before and at the moment of take-off with the position of the take-off line. If it is greater than or equal to the take-off line, it is a violation of exceeding the take-off line.
[0059] Preferably, the landing point recognition unit calculates the displacement of the person during the entire project test process according to the determined shoe landing point position, and calculates the standing jump distance through the reference of the coordinate system;
[0060] The landing point recognition unit accurately obtains the key point of the heel in the three-dimensional coordinate system, does not recognize the style, color and size of the shoe, and calculates the distance by using the difference between the horizontal coordinate of the final landing position of the key point of the heel and the horizontal coordinate of the take-off line;
[0061]
[0062] That is, first calculate the distance component in the horizontal direction, and then take the square root to get the actual distance value.
[0063] Preferably, the result output module generates a user visualization interface, and the content of the user visualization interface includes the candidate's score evaluation and the candidate's action analysis report, and the candidate's action analysis report includes the take-off angle, take-off height, jump range, landing angle and illegal actions.
[0064] Compared with the prior art, the present invention has the following beneficial effects:
[0065] 1. The present invention corrects the distortion and transforms the image horizontally, so that the present invention can better process and analyze the image data in the standing long jump process, thereby realizing the accurate recognition and evaluation of the examinee's movements. By introducing the AI analysis module, the system can accurately identify the examinee's key movements such as take-off and landing, and establish a model through the human posture estimation algorithm to realize the recognition of human movement skeleton. This not only improves the accuracy of recognition, but also provides a reliable data basis for subsequent movement analysis and performance evaluation.
[0066] 2. The present invention uses a landing point recognition unit to accurately determine the landing point of the heel, and calculates the displacement of the examinee during the entire project test process. With the help of a three-dimensional coordinate system and key point detection technology, the system can ignore the style, color and size of the shoes, and only calculate the horizontal displacement distance by identifying the key points of the heel, thereby achieving accurate measurement of the standing long jump distance and providing strong support for evaluating the examinee's performance.
[0067] 3. The present invention generates a user visualization interface through the result output module, and presents the candidate's score evaluation, candidate's action analysis report and other contents to the user in an intuitive and easy-to-understand manner. The candidate's action analysis report not only points out the problems in the action, but also provides targeted training guidance and suggestions, which helps candidates to improve their actions and scores in a targeted manner. BRIEF DESCRIPTION OF THE DRAWINGS
[0068] Figure 1 It is a work flow chart of the present invention.
[0069] Figure 2 It is the analysis flow chart of the AI analysis module of the present invention.
[0070] Figure 3 It is an overall device diagram of the present invention.
[0071] Figure 4 It is a comparative illustration of the effects before and after distortion correction of the present invention.
[0072] Figure 5 It is a schematic diagram of the necessity of perspective transformation of the present invention. DETAILED DESCRIPTION
[0073] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.
[0074] like Figure 1-5 As shown, the present invention provides a standing long jump test system based on AI vision, comprising:
[0075] Image acquisition module: The image acquisition module uses a high-definition camera to acquire images and videos of the standing long jump test area;
[0076] Image processing module: The image processing module is connected to the image acquisition module in communication, receives the image and video information collected by the image acquisition module, and uses opencv related functions to perform image distortion correction and horizontal perspective transformation operations;
[0077] AI analysis module: The AI analysis module is connected to the image processing module in communication, performs AI human posture recognition and AI object recognition according to the information transmitted by the image processing module, and performs related analysis and calculation;
[0078] Data storage module: The data storage module is communicatively connected to the AI analysis module, receives data information and report information transmitted by the AI analysis module for storage;
[0079] Result output module: The result output module is communicatively connected to the AI analysis module, receives information transmitted by the AI analysis module, and outputs test results, data analysis reports, and training instructions in real time.
[0080] Embodiment 1: Figure 1-5 As shown, in this embodiment, a candidate is taking a standing long jump test;
[0081] The image acquisition module uses a camera to obtain image information of the venue and the examinee during the standing long jump test.
[0082] The image processing module includes an image distortion correction unit and a horizontal perspective transformation unit:
[0083] The image distortion correction unit is used to perform distortion correction on the collected image information;
[0084] The horizontal perspective transformation unit is used to perform horizontal perspective transformation on the image.
[0085] The image distortion correction unit first identifies the logo of the standing jump site, 4 squares, according to the logic of the image recognition algorithm, and then corrects the image distortion size according to the camera's own parameters. The opencv built-in functions initUndistortRectifyMap and initUndistortRectifyMap perform image distortion correction;
[0086] The initUndistortRectifyMap function calculates the distortion-corrected position of each pixel in the image based on the camera's intrinsic parameter matrix and distortion coefficients:
[0087] First, the intrinsic parameter matrix of the camera is expressed as:
[0088]
[0089] Among them, x and y represent horizontal and vertical, f x and fy is the focal length in the horizontal and vertical directions, c x and c y is the principal point coordinate, which represents the coordinate of the intersection of the camera optical axis and the imaging plane in the pixel coordinate system; the distortion coefficient includes the radial distortion coefficient k 1 , k 2 and the tangential distortion coefficient p 1 、p 2 ;
[0090] The camera distortion coefficient, camera principal point coordinates and focal length can be obtained from the relevant camera manufacturer's documents;
[0091] For any pixel coordinate point (x, y) in the image, the coordinates of its original coordinates after distortion can be expressed as (x d ,y d ), obtained by the following calculation:
[0092] x d =x(1+k 1 r 2 +k 2 r 4 )+2p 1 (xc x )(yc y )+p 2 (r 2 +2(xc x ) 2 )
[0093] y d =y(1+k 1 r 2 +k 2 r 4 )+2p 2 (xc x )(yc y )+p 1 (r 2 +2(yc y ) 2 )
[0094] in
[0095] The initUndistortRectifyMap function generates the horizontal mapping table mapx and the vertical mapping table mapy based on these calculations; for each pixel position (x, y), the value mapx(x, y) in the horizontal mapping table is the corrected coordinate x of the pixel in the horizontal direction. c , expressed as:
[0096]
[0097] Among them, z represents the distance from the camera to the photographed object, and the value mapy(x, y) in the vertical mapping table is the corrected coordinate y of the pixel in the vertical direction c , expressed as:
[0098]
[0099] When correcting a distorted image, by looking up these mapping tables, each pixel is moved to the corresponding correction position according to the value in the mapping table, thereby achieving distortion correction for the entire image and obtaining an undistorted image.
[0100] The horizontal perspective transformation unit changes the horizontal presentation of the image through specific operations based on the visual law of "near objects are larger and far objects are smaller";
[0101] In OpenCV, it is called: "PerspectiveTransformation";
[0102] For the corrected data point (x d ,y d ), after horizontal perspective transformation, it becomes (x′, y′), which can be described by the following mathematical relationship:
[0103]
[0104] Among them, by adjusting the element a related to the horizontal direction 11 、a 12 and a 13 The value of can realize various deformations of image stretching, compression, tilt in the horizontal direction, simulating the visual difference when observing objects from different horizontal perspectives;
[0105] In order to achieve a specific horizontal perspective effect, based on the known coordinates of corresponding points on the original image and the target image, a set of equations is constructed to solve the elements of the transformation matrix; once this transformation matrix is determined, it can be applied to the entire original image to obtain a new image after the horizontal perspective transformation, so that it shows the expected perspective effect in the horizontal direction.
[0106] The present invention performs distortion correction and horizontal transformation on the image, so that the present invention can better process and analyze image data in the standing long jump process, thereby realizing accurate recognition and evaluation of the examinee's movements, and enhancing the accuracy of the system compared with the prior art.
[0107] After the image information of the examinee's examination process is subjected to distortion correction and horizontal perspective transformation, the AI analysis module analyzes the image content, including the take-off angle, take-off height, jump range, landing angle, illegal actions and horizontal landing distance.
[0108] The AI analysis module includes an action recognition model unit and a landing point recognition unit;
[0109] The action recognition model unit is used to establish a recognition model and judge the tester's jumping and landing actions based on the human action skeleton recognition logic;
[0110] The landing point recognition unit identifies the shoe landing point based on the recognition model, calculates the displacement of the examinee during the entire project test, and calculates the standing jump distance through the coordinate system reference;
[0111] The action recognition model unit uses a neural network to build a model. Based on the logic of human action skeleton recognition, it determines the tester's jumping, landing and other actions. Construct a neural network model:
[0112] Input layer: receives input image data and establishes a three-dimensional coordinate system based on the image information;
[0113] Feature extraction layer: used to automatically extract meaningful features from images, which can reflect the structure and posture of the human body;
[0114] Output layer: The output includes various information such as key point coordinates, take-off angle, and take-off height.
[0115] Calculation process: Match the input image with the pre-defined human body template, determine the key point position according to the matching degree, and the detected human body key point coordinate set is {p i (x i ,y i ,z i )}, where i represents different key points, each of which has its specific coordinates in three-dimensional space;
[0116] The judgment is made by analyzing the height change of the key point; for the key point P (x, y, z), the change rate of its z-axis coordinate in the time series is calculated When the rate of change changes rapidly from negative to positive and then from positive to negative within a period of time, the take-off and landing moments are accurately determined;
[0117] The following is the detailed calculation:
[0118] Take-off angle calculation: Take the shoulder and hip joint key points as the calculation, and calculate the vector between these two key points Then define the ground vector The cosine of the angle is obtained by calculating the vector dot product and the modulus length:
[0119]
[0120] Finally, the take-off angle is obtained by the inverse trigonometric function arccosα;
[0121] Calculation of take-off height: Assume that the height of the ankle key point at the moment of take-off is u, and the height of the ankle key point at the initial standing is v, then the take-off height h = uv.
[0122] Calculation of the leap range: Assume that the height of the highest key point of the body during the leap is E, then the leap range is E.
[0123] Landing angle calculation: The coordinates of the ankle key point are P(x,y,z), the coordinates of the knee key point are P(x,y,z), and the vector between these two key points is calculated Define the ground vector The cosine of the angle is obtained by calculating the vector dot product and the modulus length:
[0124]
[0125] Finally, the take-off angle is obtained by the inverse trigonometric function arccosβ;
[0126] Illegal action detection, for hand touching the ground violation, the z-axis coordinate of the hand key point P (x, y, z) is compared with the set ground, and if z < 0, it is judged as a hand touching the ground violation.
[0127] The violation of exceeding the take-off line is determined by comparing the horizontal position of the key point of the foot before and at the moment of take-off with the position of the take-off line. If it is greater than or equal to the take-off line, it is a violation of exceeding the take-off line.
[0128] The landing point recognition unit calculates the displacement of the person during the entire project test according to the shoe landing point, and calculates the distance of the standing jump through the coordinate system reference;
[0129] The landing point recognition unit accurately obtains the key point of the heel in the three-dimensional coordinate system, does not recognize the style, color and size of the shoe, and calculates the distance by using the difference between the horizontal coordinate of the final landing position of the key point of the heel and the horizontal coordinate of the take-off line;
[0130]
[0131] That is, first calculate the distance component in the horizontal direction, and then take the square root to get the actual distance value.
[0132] The introduction of the AI analysis module has achieved a comprehensive evaluation and precise analysis of the standing long jump movements of candidates compared to existing technologies. With the help of the action recognition model unit, the system can accurately identify the key actions of candidates such as take-off and landing, and establish a model through the human posture estimation algorithm to realize the recognition of human action bones. Through the landing point recognition unit, the landing point of the shoe is accurately judged, and the displacement of the candidate during the entire project test is calculated accordingly. With the help of the three-dimensional coordinate system and key point detection technology, the system can ignore the style, color and size of the shoe, and only calculate the horizontal displacement distance by identifying the key points of the heel, thereby realizing the accurate measurement of the standing long jump distance. The accuracy of this measurement method is as high as within 1cm, which provides strong support for the evaluation of candidates' scores. This not only improves the accuracy of recognition, but also provides a reliable data basis for subsequent action analysis and performance evaluation.
[0133] Preferably, the result output module generates a user visualization interface, and the content of the user visualization interface includes the candidate's score evaluation and the candidate's action analysis report, and the candidate's action analysis report includes the take-off angle, take-off height, jump range, landing angle and illegal actions.
[0134] The embodiments of the present invention are provided for the purpose of illustration and description. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present invention. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present invention.
Claims
1. A standing long jump test system based on AI vision, characterized in that: include: Image acquisition module: The image acquisition module uses a high-definition camera to acquire images and videos of the standing long jump test area; Image processing module: The image processing module is connected to the image acquisition module in communication, receives the image and video information collected by the image acquisition module, and uses opencv related functions to perform image distortion correction and horizontal perspective transformation operations; AI analysis module: The AI analysis module is connected to the image processing module in communication, performs AI human posture recognition and AI object recognition according to the information transmitted by the image processing module, and performs related analysis and calculation; Data storage module: The data storage module is communicatively connected to the AI analysis module, receives data information and report information transmitted by the AI analysis module for storage; Result output module: The result output module is communicatively connected to the AI analysis module, receives information transmitted by the AI analysis module, and outputs test results, data analysis reports, and training instructions in real time.
2. The AI vision-based standing long jump test system as claimed in claim 1, characterized in that: The image acquisition module uses a camera to obtain image information of the venue and the examinee during the examinee's standing long jump test.
3. The AI vision-based standing long jump test system as claimed in claim 1, characterized in that: The image processing module includes an image distortion correction unit and a horizontal perspective transformation unit: The image distortion correction unit is used to perform distortion correction on the collected image information; The horizontal perspective transformation unit is used to perform horizontal perspective transformation on the image.
4. The AI vision-based standing long jump test system as claimed in claim 3, characterized in that: The image distortion correction unit first identifies the logo of the standing jump field, 4 squares, according to the image recognition algorithm logic, and then corrects the image distortion size according to the camera's own parameters. The opencv built-in functions initUndistortRectifyMap and initUndistortRectifyMap perform image distortion correction; The initUndistortRectifyMap function calculates the distortion-corrected position of each pixel in the image based on the camera's intrinsic parameter matrix and distortion coefficients: The intrinsic parameter matrix of the camera is expressed as: Among them, x and y represent horizontal and vertical, f x and f y is the focal length in the horizontal and vertical directions, c x and c y is the principal point coordinate, which represents the coordinate of the intersection of the camera optical axis and the imaging plane in the pixel coordinate system; the distortion coefficient includes radial distortion coefficients k1, k2 and tangential distortion coefficients p1, p2; The camera distortion coefficient, camera principal point coordinates and focal length can be obtained from the relevant camera manufacturer's documents; For any pixel coordinate point (x, y) in the image, the coordinates of its original coordinates after distortion can be expressed as (x d ,y d ), obtained by the following calculation: x d =x(1+k1r 2 +k2r 4 )+2p1(x-c x )(y-c y )+p2(r 2 +2(x-c x ) 2 ) y d =y(1+k1r 2 +k2r 4 )+2p2(x-c x )(y-c y )+p1(r 2 +2(y-c y ) 2 ) in The initUndistortRectifyMap function generates the horizontal mapping table mapx and the vertical mapping table mapy based on these calculations; for each pixel position (x, y), the value mapx(x, y) in the horizontal mapping table is the corrected coordinate x of the pixel in the horizontal direction c , expressed as: Among them, z represents the distance from the camera to the photographed object, and the value mapy(x,y) in the vertical mapping table is the corrected coordinate y of the pixel in the vertical direction c , expressed as: When correcting a distorted image, by looking up these mapping tables, each pixel is moved to the corresponding correction position according to the value in the mapping table, thereby achieving distortion correction for the entire image and obtaining an undistorted image.
5. The AI vision-based standing long jump test system as claimed in claim 3, characterized in that: The horizontal perspective transformation unit changes the image presentation in the horizontal direction through specific operations according to the visual law of "near is big and far is small"; In OpenCV, it is called: "PerspectiveTransformation"; For the corrected data point (x d ,y d ), after horizontal perspective transformation, becomes (x', y'), which can be described by the following mathematical relationship: Among them, by adjusting the element a related to the horizontal direction 11 、a 12 and a 1% The value of can realize various deformations of image in horizontal direction, such as stretching, compression and tilt, and simulate the visual difference when observing objects from different horizontal perspectives; Based on the known coordinates of corresponding points on the original image and the target image, the elements of the transformation matrix are solved; the transformation matrix is determined and applied to the entire original image to obtain a new image after horizontal perspective transformation so that it shows the expected perspective effect in the horizontal direction.
6. The AI vision-based standing long jump test system as claimed in claim 1, characterized in that: The AI analysis module includes an action recognition model unit and a landing point recognition unit; The action recognition model unit is used to establish a recognition model and judge the tester's jumping and landing actions based on the human action skeleton recognition logic; The landing point recognition unit recognizes the landing point of the shoe according to the recognition model, calculates the displacement of the examinee during the entire project test, and calculates the distance of the standing jump through the reference of the coordinate system.
7. The AI vision-based standing long jump test system as claimed in claim 6, characterized in that: The action recognition model unit uses a neural network to establish a model, based on the human action skeleton recognition logic, to judge the tester's take-off and landing actions, and to build a neural network model: Input layer: receives input image data and establishes a three-dimensional coordinate system based on the image information; Feature extraction layer: used to automatically extract meaningful features from images, which can reflect the structure and posture of the human body; Output layer: Output includes key point coordinates, take-off angle, and take-off height information; Calculation process: Match the input image with the pre-defined human body template, determine the key point position according to the matching degree, and the detected human body key point coordinate set is {p ( (x ( ,y ( ,z ( )}, where i represents different key points, each of which has its specific coordinates in three-dimensional space; The judgment is made by analyzing the height change of the key point; for the key point P (x, y, z), the change rate of its z-axis coordinate in the time series is calculated When the rate of change changes rapidly from negative to positive and then from positive to negative within a period of time, the take-off and landing moments are accurately determined; Angle calculation, for the take-off angle, the shoulder and hip joint key points are used as the calculation, and the vector between these two key points is calculated Then define the ground vector (0,0,1), the cosine of the angle is obtained by vector dot product and modulus calculation: Finally, the take-off angle is obtained through inverse trigonometric functions; Calculation of take-off height: Compare the height of the ankle key point at the moment of take-off with the height when initially standing, and the difference is the take-off height; Calculation of the jump range: determine the highest point of the body during the jump, and calculate the difference between its height and the take-off height as the jump range; For the landing angle, it is calculated by the angle between the line connecting the ankle and knee key points and the ground; Illegal action detection, for hand touching the ground violation, the z-axis coordinate of the hand key point P (x, y, z) is compared with the set ground, and if z < 0, it is judged as a hand touching the ground violation; The violation of exceeding the take-off line is determined by comparing the horizontal position of the key point of the foot before and at the moment of take-off with the position of the take-off line. If it is greater than or equal to the take-off line, it is a violation of exceeding the take-off line.
8. The AI vision-based standing long jump test system as claimed in claim 6, characterized in that: The landing point recognition unit calculates the displacement of the person during the entire project test process according to the determined shoe landing point position, and calculates the standing jump distance through the reference of the coordinate system; The landing point recognition unit accurately obtains the key point of the heel in the three-dimensional coordinate system, does not recognize the style, color and size of the shoe, and calculates the distance by using the difference between the horizontal coordinate of the final landing position of the key point of the heel and the horizontal coordinate of the take-off line; That is, first calculate the distance component in the horizontal direction, and then take the square root to get the actual distance value.
9. The AI vision-based standing long jump test system as claimed in claim 1, characterized in that: The result output module generates a user visualization interface, and the content of the user visualization interface includes the candidate's score evaluation and the candidate's action analysis report. The candidate's action analysis report includes the take-off angle, take-off height, take-off range, landing angle and illegal actions.
Citation Information
Patent Citations
Self-closed hole flapping-wing aircraft
CN102407940A
Intelligent distance measuring method and system for standing long jump based on computer vision
CN114712769A
Motion evaluation method and device, heterogeneous AI chip, equipment and storage medium
CN117065310A
Human body action standard judgment method and device based on real-time image processing
CN117115926A
Cited By
Melting welding training examination system based on AI multi-action recognition
CN119477628A
Melting welding training and assessment system based on AI multi-action recognition
CN119477628B
Physical exercise AI standing long jump evaluation data management system
CN120242438A
A sports exercise AI standing long jump evaluation data management system
CN120242438B
