Pull-up force borrowing upper bar detection method based on visual detection

Abstract

The application discloses a chin-up object borrowing detection method based on visual detection. The method continuously acquires human body key point coordinates including shoulder, elbow, hand, crotch, knee, foot and the like through a camera. The system firstly confirms that a person enters a preparation state through calculation of a hand-raising judgment score, then calculates an upper bar score based on the longitudinal distance between the hands and the horizontal bar line in the test stage, calculates a falling bar score based on the longitudinal distance between the feet and the ground straight line, and determines a detection frame interval. The method calculates a single-foot suspension continuous score in the interval, identifies a suspension interval, calculates a person rising score for each suspension interval, judges whether there is a rising behavior by analyzing the longitudinal coordinate change of the shoulder and the crotch, and determines that there is a rule violation behavior of stepping on an object to borrow the upper bar in the chin-up process when it is detected that there is a rising behavior in at least one suspension interval. The method can automatically and accurately detect the rule violation behavior in the chin-up test, and improves the fairness and accuracy of the test.

Description

Technical Field

[0001] This invention relates to the field of physical fitness testing technology, specifically to a visual detection-based method for detecting pull-up leverage. Background Technology

[0002] Pull-ups, as an important physical fitness test item, are widely used in sports examinations, military training, and fitness assessments. Traditional pull-up tests mainly rely on manual supervision and counting. However, manual supervision suffers from limitations in perspective, distraction, and subjective judgment, making it difficult to guarantee accuracy and consistency. Especially in large-scale tests, supervisors cannot simultaneously monitor the proper form of multiple test subjects, easily leading to missed or incorrect judgments.

[0003] With the development of computer vision and artificial intelligence technologies, intelligent pull-up supervision systems based on visual detection are gradually emerging. Among existing technologies, Chinese patent CN117853984A discloses an intelligent pull-up judging system and method, which uses human posture estimation technology to locate key points on the human body and judge the standardization and counting of pull-up movements. Chinese patent CN114998986A proposes an intelligent recognition method for pull-up movement standardization based on computer vision, which determines whether the test subject has completed a standard pull-up by calculating the motion trajectory of the coordinate values ​​of key points on the human body. Chinese patent CN118767417B relates to a pull-up detection method that determines parameters such as the score for lifting off the ground, the score for gripping the bar, and the score for straight arms based on the coordinates of the test subject's key points. Chinese patent CN116306766B discloses a smart pull-up assessment and training system based on skeletal recognition technology, which counts movements by recognizing frame images of the lowest and highest points. Chinese patent CN118609201A proposes a visual detection-based method for counting and evaluating the quality of pull-ups, using the BlazePose model to obtain key points of human posture for action classification and scoring.

[0004] However, existing visual detection-based pull-up supervision techniques mainly focus on motion counting and basic compliance judgment, and still have significant shortcomings in detecting the important violation of using leverage to get onto the bar. Current leverage detection schemes based on object recognition technology have the following drawbacks: First, there are many types and shapes of objects that can be used for leverage, including boxes, stools, bricks, etc., making it difficult to establish a complete target object database for accurate identification; second, there are usually legal facilities and items around the testing area, and object detection methods are prone to misclassifying these legal items as leverage tools, leading to a high false alarm rate; third, when the subject's body obscures the assisting object, the object detection algorithm may fail to accurately identify the object, and the subject may use small, easily concealed objects, further reducing the detection accuracy. Therefore, existing technologies lack a reliable and accurate solution that can determine whether there is a violation of using leverage to get onto the bar based on the characteristics of human motion itself. Summary of the Invention

[0005] To address the limitations of traditional pull-up assist supervision methods, such as limited field of view and distraction, and the shortcomings of existing target recognition-based solutions, such as the difficulty in establishing a complete target object database due to the large variety of objects, the high false alarm rate due to misjudgment, and the low detection accuracy when the subject's body obscures the assist object, this invention provides a visual detection-based pull-up assist detection method to achieve more reliable and accurate detection of cheating behavior in pull-ups and improve the accuracy and consistency of supervision.

[0006] The technical solution adopted by this invention to solve its technical problem is: to provide a visual detection-based method for detecting pull-up leverage, comprising the following steps: S1. Set up a camera directly in front of the horizontal bar, and set the horizontal bar line l1: y=y1 and the ground line l2: y=y2 in the camera image; S2. When the system detects a person entering the detection area, it issues a voice prompt, "Raise both hands and wait for system confirmation to start the test," and continuously acquires the coordinates of key points on the human body starting from frame i=0 when the person appears. These key points include at least the coordinates of the left shoulder (slx). i sly i ), right shoulder coordinates (srx) i ,sry i Left elbow coordinate (elx) i ,ely i Right elbow coordinates (erx) i ery i Left-handed coordinates (hlx) i ,hly i Right-handed coordinates (hrx) i hry i ); S3. Starting from the n1th frame, calculate the hand-raising judgment score g11 based on the key points. i And the average score g1 of the hand-raising judgment in n1 consecutive frames. i When the value exceeds the second threshold ts2, the personnel are determined to be in a ready state, and a voice prompt "Please prepare, 3, 2, 1, start" is issued. At the same time, the video frame count is reset to i=0. S4. After entering the testing phase, continuously acquire key points of the human body, including: left shoulder coordinates (slx). i sly i ), right shoulder coordinates (srx) i ,sry i Left elbow coordinate (elx) i ,ely i Right elbow coordinates (erx) i ery i Left-handed coordinates (hlx) i ,hly i Right-handed coordinates (hrx) i hry i ), left hip coordinates (ulx) i uly i ), right hip coordinates (urx) i ,ury i Left knee coordinates (klx) i ,kly i ), right knee coordinates (krx) i ,kry i ), left foot coordinates (flx) i fly i ), right foot coordinate (frx) i fry i ); S5. Starting from frame n2, calculate the score g2 on the bar based on the vertical distance between the hands and the horizontal bar line l1. i The score g3 for landing on the bar is calculated based on the longitudinal distance l2 between the feet and the ground. i When g2 i When =1, it is determined that the person has completed the bar placement, and the first bar placement frame i1 is recorded; when g3 i When i = 1, it is determined that the person has completed the landing on the bar. The first landing frame i2 is recorded. i3 = min(i1, i2) is taken as the maximum detection frame of this landing process. S6. Within frame interval 1 to i3, execute for each frame i; S61, Calculate the score for sustained single-leg suspension (g41) i,mIf the displacement of the left or right foot in the subsequent mi frames is less than the eighth threshold ts8 and the height of the foot off the ground is greater than the ninth threshold ts9, then the foot is determined to be in a suspended state, and the suspension end frame g4 is recorded. i ; S62. After sequentially searching, n4 suspended intervals are obtained, and their starting frame sequence is denoted as {ix1, ix2, ..., ix}. n4}; S7. For each suspended interval j∈[1, n4], calculate the personnel's ascending score g5. j If the ratio of the difference between the mean shoulder ordinate at the end of the interval and the mean shoulder ordinate at the beginning of the interval to the ratio of the difference between the mean hip and shoulder ordinates at the beginning of the interval is greater than the tenth threshold ts 10 If so, it is determined that there is an upward trend in that interval; S8. If there is at least one suspended interval, the ascending score g5 j If the score is 1, it indicates that the person violated the rules by stepping on an object to use as leverage to get onto the bar during the pull-up, and a violation notice and invalid score will be output after the test.

[0007] Preferably, in step S3, g1 i It is calculated using the following formula: ; Where m is the same as i, representing the number of video frames; n1 is the set time frame length. If the time frame length exceeds this, it can be determined that the person is raising their hand, and the corresponding time range is 2-3 seconds.

[0008] Preferably, in step S3, the hand-raising determination score is g11. i It is calculated using the following formula: ; Wherein, ts1 is the first judgment threshold, which is calculated by the maximum value of the ratio of the horizontal offset distance of the arm to the vertical distance of the arm in the collected images of people raising their hands; ts2 is the second judgment threshold, which is calculated by the minimum value of the ratio of the number of frames of raised hands images to the total number of frames in the collected videos of people raising their hands.

[0009] Preferably, in step S5, the score g2 is achieved by lifting the bar. i It is calculated using the following formula: ; Where n2 is the set time frame length, exceeding this time frame length can determine whether the person is on the bar or off the bar, corresponding to a time range of 1-2 seconds; ts3 is the set third judgment threshold, which is calculated by collecting the minimum value of the ratio of the number of frames in the on-bar state to the number of frames in the on-bar state within the time window.

[0010] Preferably, in step S5, the score g3 is achieved by lifting the bar. i It is calculated using the following formula: ; Wherein, n2 is the set time frame length. If the time frame exceeds this time frame length, it can be determined that the person is on the bar or off the bar. The corresponding time range is 1-2 seconds. ts4 is the set fourth judgment threshold, which is calculated by collecting the minimum value of the ratio of the number of frames of images in the off-bar state within the time window to the number of frames.

[0011] Preferably, in step S5, the score g2 for the upper bar is calculated based on the longitudinal distance between the hands and the horizontal bar line l1. i This includes calculating the single-frame upstroke criterion for each frame m, and letting g21 m To score for the position on the bar, g21 m It is calculated using the following formula: ; Where ts5 is the set fifth judgment threshold, which is calculated by collecting the maximum pixel distance between the hands and the horizontal bar in the image under the condition of the bar; when g21 m =1 indicates an up bar state.

[0012] Preferably, in step S5, the score g3 for the drop test is calculated based on the longitudinal distance between the feet and the straight line l2 on the ground. i Includes: For each frame m, calculate the single-frame drop criterion, let g31 m To score for the landing position, g31 m =1 indicates the bar is in the down state, g31 m It is calculated using the following formula: ; Where ts6 is the set sixth judgment threshold, which is calculated by collecting the maximum pixel distance between the feet and the ground in the image under the condition of falling on the bar; when g31 m =1 indicates the state of the bar being dropped.

[0013] Preferably, in step S61, the suspended end frame g4 i It is calculated using the following formula: g4 i =max(g41) i, i+1 g41 i, i+2 ,...,g41 i, i3 ); g41 i,m The score is given for the continuous suspension from frame i to frame m. ; g42i,p The score for the dangling criterion is calculated from frame i to frame p. ; Among them, ts7 is the seventh judgment threshold, which is calculated by collecting the minimum value of the ratio of the number of frames in the suspended state to the total number of frames in the video of a person stepping on an object and causing their feet to dangle; ts8 is the eighth judgment threshold, which is calculated by collecting the maximum value of the pixel distance between the suspended feet in the video of a person stepping on an object and causing their feet to dangle; ts9 is the ninth judgment threshold, which is calculated by collecting the minimum value of the pixel distance between the suspended feet and the ground in the image of a person stepping on an object and causing their feet to dangle.

[0014] Preferably, in step S7, the score g5 increases. j It is calculated using the following formula: ; Where j is the index of the suspended interval; ts 10 The tenth judgment threshold is calculated by collecting the minimum value of the ratio of the upward displacement of a person to the distance from their torso during the process of rising by stepping on an object.

[0015] Preferably, in step S8, when If a participant is found to have used an object to gain leverage while getting onto the bar, the system will alert the participant on the screen after they get on or off the bar, and the result will be invalid.

[0016] The beneficial effects of this invention are as follows: By using a vision-based detection method, the standardization of actions is judged based on the characteristics of human body movements themselves, avoiding the problems of traditional target recognition technologies such as difficulty in establishing a complete object database, easy misjudgment, and low detection accuracy. This allows for more reliable and accurate detection of cheating behavior during pull-ups using leverage, improving the accuracy and consistency of supervision and solving the problems of limited perspective and distracted attention in traditional manual supervision methods. Compared with existing technologies, this invention does not require the establishment of a complete target object database, avoiding the identification difficulties caused by the wide variety and shapes of objects that can be used for leverage; by analyzing the changes in the coordinates of key human body points, the false alarm rate caused by legal facilities and objects around the testing site is effectively reduced; even when the subject's body obscures the auxiliary object or uses a highly concealed object, the violation can still be accurately identified through human body movement characteristics, significantly improving the reliability and accuracy of detection. Detailed Implementation

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments. The content mentioned in the embodiments is not intended to limit the present invention.

[0018] Example 1 This invention provides a visual detection-based method for detecting improper leverage during pull-ups. This method uses computer vision technology to automatically detect improper leverage during pull-ups, ensuring the accuracy and fairness of the test results.

[0019] S1. Set up a camera directly in front of the horizontal bar, and define the horizontal bar line l1: y=y1 and the ground line l2: y=y2 in the camera's view. The distance between the camera and the horizontal bar should be 2-3 meters to ensure that the test subject's full-body movements can be captured completely. Using image calibration technology, accurately mark the position of the horizontal bar in the camera view, setting it as the horizontal line l1 with the vertical coordinate y1; simultaneously mark the position on the ground, setting it as the horizontal line l2 with the vertical coordinate y2. These two baselines provide spatial reference for subsequent movement determination.

[0020] S2: When the system detects a person entering the detection area, it issues a voice prompt, "Raise both hands and wait for system confirmation to start the test," and continuously acquires the coordinates of key human points starting from frame i=0 when the person appears. The system uses a deep learning-based human pose estimation algorithm to detect and extract the coordinates of key human points in real time, including: left shoulder coordinates (slx). i sly i ), right shoulder coordinates (srx) i ,sry i Left elbow coordinate (elx) i ,ely i Right elbow coordinates (erx) i ery i Left-handed coordinates (hlx) i ,hly i Right-handed coordinates (hrx) i hry i The accuracy of key point detection reaches the pixel level, ensuring the accuracy of subsequent calculations.

[0021] S3: Starting from frame n1, calculate the hand-raising judgment score g11 based on key points. i And the average score g1 of the hand-raising judgment in n1 consecutive frames. i When the value exceeds the second threshold ts2, the judge enters the preparation state and issues a voice prompt, "Please prepare, 3, 2, 1, begin," while simultaneously resetting the video frame count to i=0. The hand-raising judgment score is g11. i Calculated using the following formula: ; Wherein, ts1 is the set first judgment threshold, which is obtained by calculating the maximum value of the ratio of the horizontal offset distance of the arm to the vertical distance of the arm in the collected images of people raising their hands, with a typical value of 0.3; ts2 is the set second judgment threshold, which is obtained by calculating the minimum value of the ratio of the number of raised hand image frames to the total number of frames in the collected video of people raising their hands, with a typical value of 0.8. g1 i Calculated using the formula: ; Where m, like i, is the number of video frames; n1 is the set time frame length, corresponding to a time range of 2-3 seconds, which is approximately 60-90 frames at 30fps.

[0022] S4: After entering the testing phase, continuously acquire human body key points. Further expansion of key points includes: left shoulder coordinates (slx). i sly i ), right shoulder coordinates (srx) i ,sry i Left elbow coordinate (elx) i ,ely i Right elbow coordinates (erx) i ery i Left-handed coordinates (hlx) i ,hly i Right-handed coordinates (hrx) i hry i ), left hip coordinates (ulx) i uly i ), right hip coordinates (urx) i ,ury i Left knee coordinates (klx) i ,kly i ), right knee coordinates (krx) i ,kry i ), left foot coordinates (flx) i fly i ), right foot coordinate (frx) i fry i Adding detection of key points in the lower body is to accurately determine the condition of a person's feet and changes in body posture. This data provides crucial information for subsequent leverage detection.

[0023] S5: Starting from frame n2, calculate the score g2 on the bar based on the vertical distance between the hands and the horizontal bar line l1. i The score g3 for landing on the bar is calculated based on the longitudinal distance l2 between the feet and the ground. i When g2 i When =1, it is determined that the person has completed the bar placement, and the first bar placement frame i1 is recorded; when g3 iWhen i = 1, the person is deemed to have completed the landing on the bar, and the first landing frame i2 is recorded. i3 = min(i1, i2) is taken as the maximum detection frame for this landing process. For each frame m, the single-frame landing criterion g21 is calculated. m , , ts5 is the fifth threshold value, calculated by collecting the maximum pixel distance between the hands and the horizontal bar in the image under the "up bar" state; a typical value is 15 pixels. When g21... m =1 indicates an up bar state. Up bar score: g2 i Calculated using the formula: .

[0024] For each frame m, calculate the single-frame drop criterion g31. m , , ts6 is the set sixth judgment threshold, which is calculated by collecting the maximum pixel distance between the feet and the ground in the image under the condition of falling on the bar, with a typical value of 20 pixels; when g31 m =1 indicates a "dropped bar" state. The score for a dropped bar is g3. i Calculated using the formula: .

[0025] Wherein, n2 is the set time frame length, corresponding to a time range of 1-2 seconds, which is approximately 30-60 frames at 30fps; ts3 is the set third judgment threshold, which is calculated by collecting the minimum value of the frame number to frame number ratio of images in the up-bar state within the time window, with a typical value of 0.7; ts4 is the set fourth judgment threshold, which is calculated by collecting the minimum value of the frame number to frame number ratio of images in the down-bar state within the time window, with a typical value of 0.6.

[0026] S6: Within the frame interval 1 to i3, perform dangling detection and interval search for each frame i.

[0027] S61: Calculate the score for sustained single-leg suspension (g41) i,m If the displacement of the left or right foot in the subsequent mi frames is less than the eighth threshold ts8 and the height of the foot off the ground is greater than the ninth threshold ts9, then the foot is determined to be in a suspended state, and the suspension end frame g4 is recorded. i End frame g4 (suspended) i Calculated using the formula: g4 i =max(g41) i, i+1 g41 i, i+2 ,...,g41 i, i3 ).

[0028] g41 i,m The score is calculated for the continuous suspension from frame i to frame m. .

[0029] g42 i,p The score for the dangling criterion is calculated from frame i to frame p. .

[0030] Among them, ts7 is the seventh judgment threshold, which is calculated by collecting the minimum value of the ratio of the number of frames in the suspended state to the total number of frames in the video of a person stepping on an object and causing their feet to dangle, with a typical value of 0.6; ts8 is the eighth judgment threshold, which is calculated by collecting the maximum value of the pixel distance between the suspended feet in the video of a person stepping on an object and causing their feet to dangle, with a typical value of 10 pixels; ts9 is the ninth judgment threshold, which is calculated by collecting the minimum value of the pixel distance between the suspended feet and the ground in the image of a person stepping on an object and causing their feet to dangle, with a typical value of 30 pixels.

[0031] S62: n4 suspended intervals are obtained by sequential search, and their starting frame sequence is denoted as {ix1, ix2, ..., ix}. n4 The system iterates through all detected suspended end frames, groups consecutive suspended states into a single suspended interval, and records the start and end frame positions for each interval.

[0032] S7: For each suspended interval j∈[1, n4], calculate the personnel's ascent score g5. j If the ratio of the difference between the mean shoulder ordinate at the end of the interval and the mean shoulder ordinate at the beginning of the interval to the ratio of the difference between the mean hip and shoulder ordinates at the beginning of the interval is greater than the tenth threshold ts 10 If so, then it is determined that there is an upward trend in that interval.

[0033] Rising score g5 j It is calculated using the following formula: , Where j is the index of the suspended interval; ts 10 The tenth judgment threshold is calculated by collecting the minimum value of the ratio of the upward displacement of a person to the distance from their torso during the process of rising by stepping on an object; a typical value is 0.4. This calculation method can effectively identify whether a person obtains upward assistance by stepping on an object.

[0034] S8: If there is at least one suspended interval, the ascending score g5 j =1 indicates that the person violated the rules by stepping on an object to use as leverage to get onto the bar during the pull-up, and a violation notice and invalid score will be displayed after the test. If a participant is found to have used an object to gain leverage while getting onto the bar, the system will display a warning on the screen after they get on or off the bar, and their score will be invalidated.

[0035] This detection method employs a multi-layered judgment mechanism, forming a complete automated detection process from preparation state recognition to violation detection. The system can accurately identify various improper leverage behaviors during pull-ups, especially instances where individuals use objects for additional support, ensuring the objectivity and impartiality of the test results. By setting multiple adjustable threshold parameters, the system can adapt to different testing environments and personnel characteristics, demonstrating good practicality and scalability.

[0036] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A visual detection-based method for detecting pull-up leverage, characterized in that: Includes the following steps: S1. Set up a camera directly in front of the horizontal bar, and set the horizontal bar line l1: y=y1 and the ground line l2: y=y2 in the camera image; S2. When the system detects a person entering the detection area, it issues a voice prompt, "Raise both hands and wait for system confirmation to start the test," and continuously acquires the coordinates of key human body points starting from frame i=0 when the person appears. These key points include at least the coordinates of the left shoulder (slx). i sly i ), right shoulder coordinates (srx) i ,sry i Left elbow coordinate (elx) i ,ely i Right elbow coordinates (erx) i ery i Left-handed coordinates (hlx) i ,hly i Right-handed coordinates (hrx) i hry i ); S3. Starting from the n1th frame, calculate the hand-raising judgment score g11 based on the key points. i And the average score g1 of the hand-raising judgment in n1 consecutive frames. i When the value exceeds the second threshold ts2, the personnel are determined to be in a ready state, and a voice prompt "Please prepare, 3, 2, 1, start" is issued. At the same time, the video frame count is reset to i=0. S4. After entering the testing phase, continuously acquire key points of the human body, including: left shoulder coordinates (slx). i sly i ), right shoulder coordinates (srx) i ,sry i Left elbow coordinate (elx) i ,ely i Right elbow coordinates (erx) i ery i Left-handed coordinates (hlx) i ,hly i Right-handed coordinates (hrx) i hry i ), left hip coordinates (ulx) i uly i ), right hip coordinates (urx) i ,ury i Left knee coordinates (klx) i ,kly i ), right knee coordinates (krx) i ,kry i ), left foot coordinates (flx) i fly i ), right foot coordinate (frx) i fry i ); S5. Starting from frame n2, calculate the score g2 on the bar based on the vertical distance between the hands and the horizontal bar line l1. i The score g3 for landing on the bar is calculated based on the longitudinal distance l2 between the feet and the ground. i When g2 i When =1, it is determined that the person has completed the bar placement, and the first bar placement frame i1 is recorded; when g3 i When i = 1, it is determined that the person has completed the landing on the bar. The first landing frame i2 is recorded. i3 = min(i1, i2) is taken as the maximum detection frame of this landing process. S6. Within frame interval 1 to i3, execute for each frame i; S61, Calculate the score for sustained single-leg suspension (g41) i,m If the displacement of the left or right foot in the subsequent mi frames is less than the eighth threshold ts8 and the height of the foot off the ground is greater than the ninth threshold ts9, then the foot is determined to be in a suspended state, and the suspension end frame g4 is recorded. i ; S62. After sequentially searching, n4 suspended intervals are obtained, and their starting frame sequence is denoted as {ix1, ix2, ..., ix}. n4 }; S7. For each suspended interval j∈[1, n4], calculate the personnel's ascending score g5. j If the ratio of the difference between the mean shoulder ordinate at the end of the interval and the mean shoulder ordinate at the beginning of the interval to the ratio of the difference between the mean hip and shoulder ordinates at the beginning of the interval is greater than the tenth threshold ts 10 If so, it is determined that there is an upward trend in that interval; S8. If there is at least one suspended interval, the ascending score g5 j If the score is 1, it indicates that the person violated the rules by stepping on an object to use as leverage to get onto the bar during the pull-up, and a violation notice and invalid score will be output after the test.

2. The method for detecting pull-up leverage based on visual detection according to claim 1, characterized in that: In step S3, g1 i It is calculated using the following formula: ； Where m is the same as i, representing the number of video frames; n1 is the set time frame length. If the time frame length exceeds this, it can be determined that the person is raising their hand, and the corresponding time range is 2-3 seconds.

3. The method for detecting pull-up leverage based on visual detection according to claim 1, characterized in that: In step S3, the score g11 is determined by raising hands. i It is calculated using the following formula: ， Wherein, ts1 is the first judgment threshold, which is calculated by the maximum value of the ratio of the horizontal offset distance of the arm to the vertical distance of the arm in the collected images of people raising their hands; ts2 is the second judgment threshold, which is calculated by the minimum value of the ratio of the number of frames of raised hands images to the total number of frames in the collected videos of people raising their hands.

4. The method for detecting pull-up leverage based on visual detection according to claim 1, characterized in that: In step S5, the score for the upper bar is g2. i It is calculated using the following formula: ； Where n2 is the set time frame length, exceeding this time frame length can determine whether the person is on the bar or off the bar, corresponding to a time range of 1-2 seconds; ts3 is the set third judgment threshold, which is calculated by collecting the minimum value of the ratio of the number of frames in the on-bar state to the number of frames in the on-bar state within the time window.

5. The method for detecting pull-up leverage based on visual detection according to claim 1, characterized in that: In step S5, the score for the upper bar is g3. i It is calculated using the following formula: ； Wherein, n2 is the set time frame length. If the time frame exceeds this time frame length, it can be determined that the person is on the bar or off the bar. The corresponding time range is 1-2 seconds. ts4 is the set fourth judgment threshold, which is calculated by collecting the minimum value of the ratio of the number of frames of images in the off-bar state within the time window to the number of frames.

6. The method for detecting pull-up leverage based on visual detection according to claim 1, characterized in that: In step S5, the score g2 for the upper bar is calculated based on the longitudinal distance between the hands and the horizontal bar line l1. i This includes calculating the single-frame upstroke criterion for each frame m, and letting g21 m To score for the position on the bar, g21 m It is calculated using the following formula: ； Where ts5 is the set fifth judgment threshold, which is calculated by collecting the maximum pixel distance between the hands and the horizontal bar in the image under the condition of the bar; when g21 m =1 indicates an up bar state.

7. The method for detecting pull-up leverage based on visual detection according to claim 1, characterized in that: In step S5, the score g3 for the drop test is calculated based on the longitudinal distance between the feet and the straight line l2 on the ground. i Includes: For each frame m, calculate the single-frame drop criterion, let g31 m To score for the landing position, g31 m =1 indicates the bar is in the down state, g31 m It is calculated using the following formula: ； Where ts6 is the set sixth judgment threshold, which is calculated by collecting the maximum pixel distance between the feet and the ground in the image under the condition of falling on the bar; when g31 m =1 indicates a bar-like shape.

8. The method for detecting pull-up leverage based on visual detection according to claim 1, characterized in that: In step S61, the suspended end frame g4 i It is calculated using the following formula: g4 i =max(g41 i, i+1 ,g41 i, i+2 ,...,g41 i, i3 ); g41 i,m The score is given for the continuous suspension from frame i to frame m. ； g42 i,p The score for the dangling criterion is calculated from frame i to frame p. ； Among them, ts7 is the seventh judgment threshold, which is calculated by collecting the minimum value of the ratio of the number of frames in the suspended state to the total number of frames in the video of a person stepping on an object and causing their feet to dangle; ts8 is the eighth judgment threshold, which is calculated by collecting the maximum value of the pixel distance between the suspended feet in the video of a person stepping on an object and causing their feet to dangle; ts9 is the ninth judgment threshold, which is calculated by collecting the minimum value of the pixel distance between the suspended feet and the ground in the image of a person stepping on an object and causing their feet to dangle.

9. The method for detecting pull-up leverage based on visual detection according to claim 1, characterized in that: In step S7, the score increases by g5. j It is calculated using the following formula: ； Where j is the index of the suspended interval; ts 10 The tenth judgment threshold is calculated by collecting the minimum value of the ratio of the upward displacement of a person to the distance from their torso during the process of rising by stepping on an object.

10. The method for detecting pull-up leverage based on visual detection according to claim 1, characterized in that: In step S8, when If a participant is found to have used an object to gain leverage while getting onto the bar, the system will alert the participant on the screen after they get on or off the bar, and the result will be invalid.

Images