Search

CN-121982776-A - Rope skipping wireless air-jump detection method based on vision

CN121982776ACN 121982776 ACN121982776 ACN 121982776ACN-121982776-A

Abstract

The invention discloses a rope skipping wireless air-jump detection method based on vision, which comprises the steps of acquiring rope skipping videos by arranging a camera above a rope skipping test area, and acquiring 19 key point coordinates of a tester based on a human key point recognition model trained by a YOLO frame; the method comprises the steps of calculating a wrist position score through shoulder-wrist vectors, judging whether the wrist position is normal, constructing a human body central line through shoulder-crotch connecting lines, calculating left and right wrist symmetry scores, constructing an arm gesture plane through shoulder-elbow-wrist three points, calculating an upper limb gesture score, calculating a skip score of each rope skipping action by integrating wrist position, symmetry, upper limb gesture and time sequence jitter characteristics, counting suspected skip times and occupation ratio, judging that wireless skip exists when a preset threshold value is exceeded, and outputting an alarm. The method can accurately identify the cordless empty jump behavior in the rope jump test, improves the fairness and accuracy of the sports test, and is suitable for scenes such as school sports examination and physical ability test.

Inventors

  • WU SHURUI
  • ZHANG FENGDONG
  • LIANG FAN

Assignees

  • 广东先知大数据股份有限公司

Dates

Publication Date
20260505
Application Date
20260129

Claims (10)

  1. 1. The rope skipping wireless air-skip detection method based on vision is characterized by comprising the following steps of: 1) A camera is arranged above the rope skipping test area, rope skipping videos of the tested person are collected, and images are extracted frame by frame; 2) Based on a human body key point recognition model, 19 key point coordinates of a tester in the jumping process are obtained, wherein the coordinates comprise left shoulder, right shoulder, left elbow, right hand, left wrist, right wrist, left palm, right palm, left crotch and right crotch, and the coordinates are obtained through training by a YOLO frame; 3) The method comprises the steps of starting to acquire coordinates of key points of a tester in each frame i of an image of a video, wherein the coordinates comprise a left shoulder coordinate (slx i ,sly i ), a right shoulder coordinate (srx i ,sry i ), a left elbow coordinate (elx i ,ely i ), a right elbow coordinate (erx i ,ery i ), a left wrist coordinate (hlx i ,hly i ), a right wrist coordinate (hrx i ,hry i ), a left palm coordinate (alx i ,aly i ), a right palm coordinate (arx i ,ary i ), a left crotch coordinate (ulx i ,uly i ) and a right crotch coordinate (urx i ,ury i ), acquiring the number n 1 of skipping ropes of the tester after the test is finished, acquiring a time frame t j ,j=1,...,n 1 when the tester finishes one skipping action as a sequence number corresponding to each time frame, and defining t 0 =0; 4) Calculating a left wrist position score g11 i and a right wrist position score g12 i based on the shoulder-wrist vector, wherein when g11 i and g12 i fall into a preset threshold value interval (ts 1 ,ts 2 ), the wrist position of the ith frame is normal; 5) Constructing a human body centerline l i by a shoulder and crotch connecting line, solving the symmetry point of the right wrist about l i , and judging by calculating a left wrist symmetry score g3 i and a right wrist symmetry score g31 i of each frame i to obtain a wrist symmetry deviation score, presetting a threshold ts 3 , and judging that the personnel are symmetrical with two hands when g31 i ＜ts 3 ; 6) Constructing an arm posture plane by three points of shoulder, elbow and wrist, respectively calculating a left arm posture score g41 i 、g42 i and a right arm posture score g4 i , wherein the multiplication of the two scores is the upper limb posture score g41 5748, and the arm posture meets the basic requirement of rope skipping only when the left arm posture score g41 i and the right arm posture score g42 i meet the basic requirement at the same time; 7) Acquiring the total number n 1 of rope skipping completed by a tested person and a start-stop frame sequence t j -1+1～t j ,j=1,...,n 1 corresponding to each rope skipping action; 8) And (3) for the j-th rope skipping action, integrating the results of the steps 4) to 6) and the time jitter characteristics, and calculating a null jump score g7 j : a) Calculating the average value of g1 i ·g3 i in the action period, presetting a threshold ts 5 , and judging the j action as normal jump if the average value is less than or equal to ts 5 ; b) Calculating the average value of g41 i in the action period, presetting a threshold ts 6 , and if the average value is less than or equal to ts 6 , judging the j-th upper limb gesture as normal jump, otherwise judging the j-th upper limb gesture as suspected empty jump; c) Respectively calculating frame-level variances of the left palm-left wrist relative distance and the right palm-right wrist relative distance in the action period, presetting a threshold ts 7 , if the variances are less than or equal to ts 7 , judging that the left wrist and the right wrist do not shake obviously and have the possibility of wireless skip, otherwise, judging that the left wrist and the right wrist are suspected skip; d) g7 j =g71 j ·g72 j ·g73 j ·g74 j , when any one of g71 j ,g72 j ,g73 j ,g74 j has an abnormality, judging that the action is a suspected cordless empty jump, otherwise, judging that the action is a normal jump; 9) Counting the suspected skip times g n and the duty ratio g p of all n 1 actions, presetting a threshold ts 8 ,ts 9 , if g n ＞ts 8 or g p ＞ts 9 , judging that the tested person has the wireless skip in the rope skipping test, and outputting an alarm.
  2. 2. The rope skipping wireless air-jump detection method based on vision is characterized in that training data of the 19-point model in the step 2) are real rope skipping videos collected under multi-person, multi-scene and multi-illumination conditions, left palm key points and right palm key points are marked manually after frame-by-frame interception, marking frames with confidence degrees of more than 0.95 are used as positive samples, and a YOLO frame is adopted for training until loss converges.
  3. 3. The method for detecting rope skipping wireless air-skip according to claim 1, wherein in the step 4), the left wrist position score g11 i and the right wrist position score g12 i are calculated, so that the wrist position score g1 i is obtained, and the judgment result is: ; The method comprises the steps of acquiring a plurality of images of a hand-held skipping rope, acquiring cosine values of included angles formed by the wrist, the shoulder and a normal line of a vertical ground by capturing skipping rope video of the hand-held skipping rope, and calculating according to a confidence coefficient 0.95; When g1 i =1, the wrist position of the ith frame is normal, and otherwise, the wrist position of the ith frame is abnormal.
  4. 4. The rope skipping wireless air-jump detection method based on vision according to claim 3, wherein in the step 5), the left and right wrist symmetry score g3 i is calculated by the following formula: ; wherein g3 i is a wrist symmetry deviation score, which indicates the deviation degree of the two wrists relative to the line symmetry of the human body, ; Wherein, (hrx i ' ,hry i ') is the point of symmetry of the right wrist coordinate (hrx i ,hry i ) with respect to the human midline l i ：A i x+B i y+C i =0, ; The ts 3 is a set third judgment threshold value, represents the maximum value of the wrist offset degree, and is obtained by collecting a large number of hand-held skipping rope videos, intercepting each frame, calculating the ratio of offset pixel distance to trunk pixel distance through the wrist and the central line, obtaining the wrist offset degree range according to the range of the ratio, and calculating the maximum value; When g i =1, the person is judged to be symmetrical with both hands, otherwise, the person is judged to be asymmetrical.
  5. 5. The rope skipping wireless air-jump detection method based on vision according to claim 4, wherein in the step 6), the upper limb posture score g4 i is calculated by the following formula: ; the ts 4 is a set fourth judgment threshold value, represents the maximum value of the arm posture offset degree, and is obtained by collecting handheld skipping rope video, intercepting each frame, calculating the maximum value of the ratio of the straight line distance from the wrist to the elbow to the shoulder to the length of the large arm; When g i =1, the arm gesture accords with the basic requirement of the rope skipping action, and otherwise, the arm gesture does not accord with the basic requirement of the rope skipping action.
  6. 6. The method for detecting rope skipping cordless air-skip according to claim 5, wherein in step 8, the air-skip score g7 j of the jth rope skipping action is calculated by the following formula: ; The method comprises the steps of setting ts 5 as a fifth judgment threshold, setting ts 6 as a sixth judgment threshold, setting ts 7 as a seventh judgment threshold, and calculating the minimum value of jitter degree during wrist jitter by collecting the minimum value of fluctuation variance of wrist pixel distance in the whole process of each rope skipping in a handheld rope skipping video, wherein m is the same as i in sequence number; g7 j denotes a wrist standard score in the j-th action, g j =1 denotes that there is a possibility of cordless skip of the wrist; g72 j denotes an upper limb posture standard score in the j-th action, and g j =1 denotes that there is a possibility of cordless skip in the arm posture; g73 j denotes a wrist shake score in the j-th action, g j =1 denotes that there is a possibility of cordless skip without significant shake of the left wrist; g74 j =1 indicates that there is a possibility of cordless skip without significant jitter on the right wrist.
  7. 7. The rope skipping wireless air-hop detection method based on vision as claimed in claim 6, wherein in the step 9), the number g n and the duty ratio g p of the suspected air-hops are respectively calculated by the following formulas: ; Wherein ts 8 is a set eighth judgment threshold, ts 9 is a set ninth judgment threshold, calculating the number of the null hops and the duty ratio by collecting videos with wireless null hops, and obtaining the minimum value of the number and the duty ratio range according to the distribution and the confidence coefficient of 0.95; When g n ＞ts 8 or g p ＞ts 9 , it is determined that a person has a cordless empty jump in the jump rope test.
  8. 8. The method for detecting rope skipping wireless air-skip according to any one of claims 1-7, wherein when the rope skipping wireless air-skip is judged to exist, the system pops up a red 'air-skip warning' word on a display screen of the body measurement all-in-one machine after the test is finished, and synchronously uploads violation marks, key frame pictures and g n 、g p data to a teacher terminal.
  9. 9. The rope skipping wireless air-jump detection method based on vision according to any one of claims 1-7, wherein the installation height of the camera is 2.2-2.5m, the overlooking angle is 30-45 degrees, the short side of the visual field covers the rope skipping area more than or equal to 1.5m multiplied by 1.5m, and a diffusion cover is additionally arranged in front of a lens to eliminate reflection.
  10. 10. The method for detecting rope skipping wireless air-hop based on vision as claimed in claim 7, wherein all thresholds ts 1 -ts 9 are supported to be configured in a hierarchical mode in an integrated machine management menu according to a learning segment, gender or test grade by an administrator, and the configuration is completed and then issued to an edge computing box through a local encryption file or cloud API to take effect in real time.

Description

Rope skipping wireless air-jump detection method based on vision Technical Field The invention relates to the technical field of physical ability testing, in particular to a rope skipping wireless air-jump detection method based on vision. Background With the continuous development of sports education and the wide application of intelligent technology, a jump rope is an important physical exercise item, and the intelligent detection and counting technology has become a research hotspot. The traditional rope skipping test mainly relies on manual counting and supervision, and has the problems of low efficiency, insufficient accuracy and the like in a collective teaching environment. In recent years, rope skipping detection technology based on computer vision has been rapidly developed. Chinese patent CN114100103B discloses a rope skipping counting detection system and method based on key point recognition, which collects video images of rope skipping motion through an image acquisition module, recognizes image frames by using an image recognition module and performs state judgment, thereby realizing automatic counting of rope skipping motion. Chinese patent CN116416551a proposes a tracking algorithm-based video image multi-person self-adaptive rope skipping intelligent counting method, which obtains identity by triggering face recognition by lifting manual, tracks position in real time in combination with the tracking algorithm, and performs waveform analysis based on skeletal key points to obtain rope skipping count. Chinese patent CN113440789B discloses an intelligent counting method and system for multi-person rope skipping test, which verifies identity through face recognition, obtains skeletal key point information based on test images, and stores the skipping track as track waveform for analysis and statistics. Chinese patent CN115346149A relates to a rope skipping counting method based on a space-time diagram convolutional network, which utilizes the space-time diagram convolutional neural network to analyze the human body posture estimation result, and obtains the rope skipping state of a rope skipping person to accurately count. Chinese patent CN116311523B discloses a fancy rope skipping recognition algorithm based on image recognition, which solves the problem that the counting accuracy is affected by the shielded key points and the too small human body occupation ratio by dynamically selecting the key points and scaling technique. However, the prior art mainly focuses on the identification and counting of rope skipping actions, and obvious defects still exist for the detection of cheating behaviors in rope skipping tests, particularly the identification and prevention of cordless air-skip behaviors. In practical physical education and test environments, students may adopt cheating means such as wireless air jumping to obtain false test results, and teachers have limited manpower, so that each student is difficult to continuously monitor the whole course, the cheating behaviors cannot be effectively prevented, and the authenticity and the effectiveness of teaching evaluation are difficult to guarantee. Meanwhile, the existing visual detection method mainly focuses on the identification of jumping actions, lacks comprehensive analysis of key features such as hand actions, upper limb gestures and the like, is difficult to accurately distinguish real rope jumping and wireless air jumping actions, and has the problem that detection accuracy and reliability in complex actual application scenes are to be improved. Disclosure of Invention In order to solve the technical problems that in the traditional rope skipping training, the manpower of a teacher is limited, and the continuous whole-course supervision is difficult for each student, and the cheating actions such as wireless air-skipping and the like cannot be effectively prevented from being adopted by the students, and the dynamic rope skipping is difficult to technically and is easily influenced by interference factors, the technical effects of accurately judging whether the students actually use the rope skipping, guaranteeing the teaching evaluation fairness and improving the teaching efficiency of the teacher are achieved, and the rope skipping wireless air-skipping detection method based on vision is provided. The invention aims at realizing the technical scheme that the rope skipping wireless air-jump detection method based on vision comprises the following steps of: 1) A camera is arranged above the rope skipping test area, rope skipping videos of the tested person are collected, and images are extracted frame by frame; 2) Based on a human body key point recognition model, 19 key point coordinates of a tester in the jumping process are obtained, wherein the coordinates comprise left shoulder, right shoulder, left elbow, right hand, left wrist, right wrist, left palm, right palm, left crotch and right crotch, and the coordinates are obtained thr