Search

CN-122024314-A - Marathon running gesture correction method based on MEDIAPIPE and random forest algorithm

CN122024314ACN 122024314 ACN122024314 ACN 122024314ACN-122024314-A

Abstract

The invention provides a Marathon running gesture correcting method based on MEDIAPIPE and a random forest algorithm, which relates to the technical field of computer vision and mode recognition, and comprises the steps of firstly detecting 33 key skeleton nodes of a human body, calculating knee bending angles, arm swing amplitudes, trunk forward tilting angles, emptying heights and moving speeds, optimizing the knee bending angles, the arm swing amplitudes, the trunk forward tilting angles, the emptying heights and the moving speeds through feature importance analysis, feature degradation and the like, training and evaluating a model by adopting the random forest algorithm, testing and packaging to form a UI interface, constructing a real-time feedback APP capable of being implanted with a sports watch, and realizing real-time feedback and correction of the Marathon running gesture.

Inventors

  • YAN QIUFENG
  • LIU ZHILING
  • JIANG MENGYAO

Assignees

  • 南通大学

Dates

Publication Date
20260512
Application Date
20260126

Claims (10)

  1. 1. A marathon running posture correcting method based on MEDIAPIPE and random forest algorithm is characterized by comprising the following steps: s1, collecting multi-angle human body posture images; s2, detecting key points of human body and human body posture; s3, geometric feature extraction and data preprocessing; S4, training the extracted features by using a random forest algorithm, obtaining a classification model, and analyzing the importance of the features, evaluating the model and optimizing the model; s5, identifying the running gesture problem of the marathon athletes on the runway through an identification model; and S6, feeding back the running gesture problem in real time.
  2. 2. The marathon running gesture correcting method based on MEDIAPIPE and random forest algorithms according to claim 1, wherein the collection process in step S1 mainly includes that the athlete performs marathon daily training on the runway, unmanned aerial vehicles are arranged around the athlete, and the running gesture image of the athlete performing training is collected.
  3. 3. The marathon running gesture correcting method based on MEDIAPIPE and random forest algorithms according to claim 2, wherein the step S2 specifically includes the following steps: S21, detecting a human face by utilizing BlazeFace models; S22, predicting the middle points of the two shoulder joints and the middle points of the hip joints of the human body by using BlazePose model, and connecting the two points for the alignment operation of the whole human body; s23, predicting 33 key points of the human body.
  4. 4. A marathon running gesture correction method based on MEDIAPIPE and random forest algorithms as claimed in claim 3, wherein the geometric features in step S3 mainly include knee bending angle, arm swing angle, trunk forward tilting angle, flight height, and moving speed.
  5. 5. The marathon running gesture correcting method based on MEDIAPIPE and random forest algorithms according to claim 4, wherein the data preprocessing in step S3 includes loading raw data, grouping by category, deleting repeated samples, processing missing values, outlier processing, and saving processing results.
  6. 6. The method for correcting the martensi karsch running gesture based on MEDIAPIPE and the random forest algorithm according to claim 5, wherein the training of the martensi karsch running gesture geometric feature data flow by the random forest algorithm in the step S4 comprises the steps of martensi karsch running gesture geometric feature data processing, data set division, model construction and training, model evaluation and parameter adjustment.
  7. 7. The marathon running gesture correcting method based on MEDIAPIPE and random forest algorithm as claimed in claim 6, wherein the step S5 specifically includes collecting running gesture images of athletes during training in real time, reasoning the extracted characteristic data through a trained random forest model, outputting running gesture abnormal types and abnormal confidence, combining multi-angle images collected by unmanned aerial vehicles in multiple directions, cross-verifying abnormal judgment results, and avoiding misjudgment caused by single-angle shielding.
  8. 8. The marathon running gesture correcting method based on MEDIAPIPE and random forest algorithms according to claim 7, wherein the step S6 specifically includes building a lightweight running gesture real-time correcting APP supporting the implantation of a sports watch or a mobile phone end, and feeding back running gesture problems in real time.
  9. 9. The marathon running gesture correcting method based on MEDIAPIPE and random forest algorithm according to claim 2, wherein in the step S1, an unmanned aerial vehicle is arranged in each of the 4 directions of the athlete, namely, front, back, left and right, and the training athlete is subjected to 4-angle running gesture image acquisition.
  10. 10. The marathon running gesture correcting method based on MEDIAPIPE and random forest algorithms according to claim 9, wherein the step S4 includes processing marathon running gesture geometric feature data, processing abnormal values and repeated values in a dataset, dividing the dataset into a training set by using professional athlete standard running gesture geometric feature data stored after preprocessing, and using collected common runners or running gesture geometric feature data with nonstandard postures as a testing set.

Description

Marathon running gesture correction method based on MEDIAPIPE and random forest algorithm Technical Field The invention relates to the technical field of computer vision and pattern recognition, in particular to a marathon running gesture correction method based on MEDIAPIPE and a random forest algorithm. Background Marathon is used as a sports event with wide participation and strong ornamental value, and attracts a plurality of professional athletes and common masses to participate. Proper running posture is a key factor in the improvement of marathon performance. Irregular running postures (such as incorrect foot placement, inner buckles of knees, excessive trunk forward inclination and the like) not only can influence the improvement of marathon achievements, but also can cause sports injuries such as tibial stress syndrome, iliotibial band syndrome, plantar fasciitis and the like, and seriously influence the sports experience and health and safety of runners. The traditional marathon running gesture correcting mode mainly comprises two modes of manual guidance and simple equipment assistance, wherein the manual guidance relies on site observation of a professional coach, the problems of strong subjectivity, high cost and difficulty in covering whole-course movement exist, the traditional equipment assistance mode (such as a sport bracelet and a pressure sensing insole) can only collect partial movement data (such as step frequency and plantar pressure), the gesture characteristics of a runner cannot be comprehensively captured, accurate recognition and real-time correction feedback capability for running gesture abnormality are lacking, and meanwhile, the wearing of sensor equipment can influence the exertion of athletes to further influence the running gesture. Along with the development of computer vision and artificial intelligence technology, MEDIAPIPE algorithm is used as an open source posture estimation framework deduced by google, has the advantages of strong real-time performance and accurate bone node identification, and can rapidly extract human motion posture data. Currently, MEDIAPIPE algorithms are temporarily unavailable for use in marathon running gesture correction applications, which is a major penalty for popularization of marathon sports, athlete performance improvement, and physical health problems. Disclosure of Invention Aiming at the problems, the invention provides a Marathon running gesture correcting method based on MEDIAPIPE and a random forest algorithm, which realizes real-time accurate identification, abnormal judgment and personalized correction of running gestures, realizes high-efficiency training of Marathon sports, improves the performance of athletes and avoids injuries, and firstly selects three-dimensional coordinates of 33 key skeleton nodes of the whole body of the Marathon athletes, calculates knee bending angles, arm swing amplitude, trunk forward tilting angles, vacation heights and moving speeds, and takes the three-dimensional coordinates as characteristic values. Through a feature importance evaluation function built in a random forest algorithm, a plurality of decision trees are constructed, the splitting contribution degree of each feature in the decision trees is counted, the importance score of each feature is obtained, five features with the largest score are deleted, the feature is subjected to dimension reduction processing, the stability and accuracy of marathon running gesture recognition are finally improved, and technical support is provided for daily training of professional athletes and marathon fans. The invention provides a marathon running gesture correcting method based on MEDIAPIPE and a random forest algorithm, which comprises the following steps of: Step 1, acquiring human body posture images; Step 2, detecting key points of human body and human body posture; step 3, extracting features; step4, training and evaluating a random forest model; Step 5, identifying a running gesture error problem; and 6, feeding back the running gesture problem in real time. Further, the step 1 comprises the steps of arranging unmanned aerial vehicles in the front, rear, left and right directions of a training athlete, collecting running gesture images of the training athlete at 4 angles, and setting the unmanned aerial vehicles to a following function to fly along the advancing direction of the athlete. Further, step2 utilizes BlazePose module MEDIAPIPE to process the input picture or video frame in real time, and based on the lightweight neural network architecture, three-dimensional coordinates of 33 key skeleton nodes of the human body are detected rapidly and accurately. Further, step 3 calculates knee bending angle, arm horizontal swing angle, arm vertical swing angle, trunk forward tilting angle, flight height and moving speed according to three-dimensional coordinates of 33 key skeleton nodes of the human body, and takes the knee bending angle, the a