A Study of University Sports Training Methods Assisted by Artificial Intelligence
Published Online: Mar 24, 2025
Received: Oct 21, 2024
Accepted: Feb 04, 2025
DOI: https://doi.org/10.2478/amns-2025-0788
Keywords
© 2025 Ye Han, published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Physical education training is not a static process, on the contrary, it is a dynamic process. Specifically, the physical education teacher according to the training program designed before the class and set the training objectives, organizing teaching activities to be carried out smoothly, through the guidance and supervision to guide the students in the process of completing the teaching objectives to improve the individual physical quality, strengthen the individual psychological quality, and ultimately to achieve a truly healthy and comprehensive development [1-4]. In general, physical training has physical training and sports skills training [5]. Physical education teachers usually design appropriate teaching plans and arrange reasonable teaching contents according to the syllabus and the actual situation of the students in the physical education classroom, and gradually teach the students various physical education theories and skills from easy to difficult according to the various links of the teaching plan [6-8]. Different from physical education, sports training in the process of its most significant feature is competitive, physical education teachers pay more attention to students in daily learning and competition results, in the content arrangement also pay more attention to the leakage of fill in the gaps, that is, for the student’s deficiencies focus on training, so as to ensure that the students can realize the overall development of the students can be a breakthrough in the process of actual training [9-11].
With the rapid development of science and technology, artificial intelligence has been effectively applied in different fields, in order to give full play to the role of sports training, we need to optimize and innovate the sports training methods and management, and form a professional systematic training, in order to promote the effective improvement of the level of sports training, in order to complete the training objectives [12-15]. At the same time, sports training also involves more training content, data analysis and processing demand is relatively large, through the analysis of data, sports training can provide targeted scientific training methods to avoid unnecessary injuries [16-18]. Therefore, artificial intelligence can be used in sports training to improve the effectiveness and scientific nature of training, and to achieve the corresponding sports training goals.
AI has a wide range of applications in sports science, including motion analysis, injury prevention, injury rehabilitation strategies, training plans, and event strategy adjustments [19], with the main purpose of improving the level of sports training. Specifically, with the support of artificial intelligence, Literature [20] established a motion database by capturing movements, constructed a three-dimensional model of the human body, and formed a motion simulation training system, realizing the visualization of sports training data. Similarly, the literature [21] uses the capture of athletes’ training movements to assist athletes in understanding their own movement defects and improving their training in a targeted manner. Similarly, a motion training management information system was established with the support of the action recognition model of convolutional neural network in Literature [22], which is superior to traditional methods in terms of response speed. Literature [23] developed an expert system for sports training in colleges and universities using human-computer interaction techniques with artificial intelligence to show the training content searched by students in less than 5 seconds to assist in training. Literature [24] designed a data mining system for physiological parameters of sports training under artificial intelligence and association rules in the perspective of health science training, which was shown to be 90% accurate under Markov logic training. Literature [25] used artificial intelligence technology to monitor athletes’ training in real time according to their status, characteristics, plans, physiological indicators, injury history, nutritional conditioning, etc., and analyzed the training rules of athletes under each parameter in order to guide training. Literature [26] established a three-dimensional athletic index data set and knowledge system for athletes, analyzed the data with the support of artificial intelligence, and combined with sports biomechanics, graphic imaging, human physiology, etc. to form a multi-targeted feedback training method, which helps to improve the level of athlete training. Literature [27] used artificial intelligence technology to construct a target tracking model for sport functional exercise training, which is a local image block for sparse coding method, optimizing the scientific and intelligent functional physical training.
In summary, the study, artificial intelligence through the capture of athletes’ training actions, visualization of training data, while real-time monitoring, assisting coaches and athletes to understand their own training status, and then combined with the individual physiological and biochemical indicators and knowledge of sports training theories, to achieve the intelligent management of sports training and information query, and improve the level of sports training. At the same time, artificial intelligence has assisted in the development of sports training programs through various types of data analysis. However, none of them have introduced specific training methods for AI to promote sports training, and the school’s consideration for the introduction of AI into the teaching guidance of sports training includes feasibility and practicability in addition to cost, therefore, a practical and concrete method is needed.
In this paper, the principle of TOF technology is applied to process the data related to students’ training and movements obtained from the Kinect sensor during university sports training. A filtering algorithm is used to process the skeletal data in the sports training action data to ensure smooth and accurate processing. The three-dimensional spatial angles of different sports training actions are first established, and the angles of each skeletal joint point are calculated according to the space vector method to extract the direct angle skeletal features of the joint points. The parameters in the skeletal features are recognized using the Hidden Markov Model, and the results of sports training action recognition are obtained after model training. Subsequently, the intelligent system for sports training is designed by combining the sensing device acquisition module, the skeletal point extraction module and the action recognition module to generate a scientific training program through action analysis. Finally, this study applies the intelligent system to the sports training of a university, and analyzes the changes in the physical quality and functional movement scores of college students under the assistance of the intelligent system and the traditional training mode, to explore the auxiliary effect of the intelligent system in university sports training.
In this paper, Kinect sensing device is utilized to acquire data during the training process of university sports, KinectV2 data flow is shown in Fig. 1.The data acquisition process of KinecV2 is to firstly acquire the data source from the Kinect sensor which is turned on, turn on the reader from the data source, and then acquire the frames from the reader, and finally acquire the data from the frames in the Kinect, the data sources include six kinds, which are infrared image, color image, depth image, human index, human skeleton and sound [28].
Kinectv2 data flow diagram
The data stream acquisition method of KinectV2 described above is different compared to the acquisition method of the first generation product.The Source of Kinect V2, all have a 1 data stream that acquires Color and Depth from Kinect.The function of Kinect V1 to take the data stream directly is the same as the function in V2.Kinect SDK 2.0 added the Reader can open the same Source many times, through this configuration, multi-threaded applications, do not need to copy the acquired data to other threads for processing. It is also possible for multiple applications to obtain data from the same sensor. This way the utilization of the device is increased.
KinectV2 uses the TOF light time-of-flight measurement technology, TOF is through the infrared transmitter projected by the modulated near infrared light, when irradiated to the environment of the object infrared light will be reflected, infrared camera to receive the reflected infrared light, the calculation of the time difference of the light or the phase difference, the depth of the object can be obtained (i.e., the object to the depth of the depth of the depth camera distance).
During university athletic training data collection, the Kinect emits a high-frequency modulated pulse of connected light through an infrared transmitter, and the light wave is reflected and accepted by a depth camera. Due to the near speed of light (3*108 m/s), it is difficult to accurately measure depth information by directly measuring small differences in time of flight. For example, in the range of Kinect (0.5m to 4.5m), if the time-of-flight difference is measured directly to calculate the depth information, the acquisition frequency of the camera needs to be up to 3*108 frames/second, which is almost impossible to achieve with the high hardware requirements. And through the depth of the camera in the TOF photoreceptor chip for each pixel emitted light waves to and from the camera and the object between the phase of the specific differences were recorded, and then according to the pattern of change of the brightness of each pixel through the data processing unit to extract the phase difference of Δ
KinectV2 using the principle of TOF technology [29] can for real-time fast calculation of the output depth information, to 30FPS 512 × 424 resolution data, structured light need to use complex correlation algorithms, processing speed is relatively slow, for the somatosensory interaction of the time-sensitive requirements of the application of the application of high significance of the scene. At the same time, the depth information calculation of TOF will not be affected by the object surface features and grayscale, but the use of binocular stereo vision technology requires the target to have a good feature change, otherwise it will not be able to calculate the depth information. Therefore, TOF technology can carry out very accurate three-dimensional detection, at the same time, the depth calculation accuracy of TOF is not affected by the change of distance, and has a certain degree of anti-interference ability for strong and weak light environments, and can be stabilized at the level of cm, and KinectV2 can be optimized to reach the level of
After filtering and processing the skeletal data from the university sports training data collected by KinectV2, the output of smooth skeletal data filtered out the peaks and noises, so that the skeletal data is more accurate in describing the real skeletal characteristics during the exercise process. The calculated relative angles of joints in three-dimensional space are also smoother, so this system selects the relative angles of multiple joints in three-dimensional space as the skeletal space features for the differences of different sports such as pull-ups, squats, etc. The method used for sports training data recognition is described in detail below.
The first method of calculating the angle in three-dimensional space is the traditional analytical geometry method. To analytical geometry methods necessarily need to consider the boundary conditions, such as parallel, overlap, perpendicular, intersection, etc., to these boundary conditions need to be targeted processing, the realization of a large amount of computational procedures are more complex. It will bring a lot of unexpected problems, however, the use of space vector method does not need to deal with the boundary conditions, so this paper selects the use of space vector method to calculate the angle of the joints.
The spatial vector method [30] requires the use of a conventional mathematical coordinate system, however, the KinectV2 adopts a different coordinate system from the general spatial coordinate system, its
Using the above method, the vectors between the joints of the human body can be formed, and then calculate the angle between the two vectors, which can be regarded as the angle between the joints.
At this point, you only need to calculate the angle
In the same way the angle of rotation of the shoulder
The normal vector of the plane XOY is:
This results in a shoulder rotation angle
From the above spatial vector method of extracting direct angular skeletal features at joint points, a series of this features can be extracted to reflect the postural skeletal features at different moments. These features are utilized as inputs to the motion recognition model for accurate and complete motion recognition.
The core problem in the implementation of training movement recognition in college sports is the need to find out the hidden states in the Hidden Markov Model [31]. The initial choice of a good or bad model will have a direct impact on the training results and thus on the motion recognition accuracy. To reduce the motion recognition process, all the action sequences are sliced equally into
By slicing and dicing the sequence of movements in university sports training, a sequence of state transitions for each movement is obtained, and using these state transition sequences the initialization parameters for each movement can be calculated. The number of times the
The above derived formula represents the probability that the first frame of such action is in state
From the analysis, the transfer probability of the initial state
In this paper, the Baum-welch algorithm [32] is used for parameter learning, after initializing the parameters in the previous section to obtain the initialization parameter
The technical architecture of the intelligent system for university sports training in this study is shown in Figure 2. The intelligent system collects the athlete’s bone data in real time through the KinectV2 camera, filters the bone data to obtain smooth bone data through coordinate mapping, and then uses the spatial vector method to calculate the angle characteristics between the points of the bone joints, collects the motion feature data for HMM model training, and realizes a set of highly reliable and low-latency sports training intelligent assistance system by selecting the maximum output probability. The data acquisition layer obtains the skeletal data stream of the sportsman using the Microsoft KinectV2 intelligent camera, and uses it to recognize skeletal motion. Skeletal algorithm processing layer is necessary to filter the data from the data acquisition layer in order to obtain smooth skeletal data, and utilize the spatial vector method to calculate the angle characteristics of these skeletal joints. After collecting a large number of predefined motion data in the skeletal algorithm processing layer, then calculate the initial parameters of the HMM, train the motion model, and obtain a lot of Hidden Markov Model output probabilities, and then compare and select the largest output probability as the result of motion recognition. In the application layer, the motion recognition layer recognizes the different motions of the sportsmen participating in sports training, and the data generated during the whole process of the motion, including the motion time, the number of motions, and the result settlement, are stored, and finally the data are uploaded to the server, scientifically analyzed and managed, and the university sports training program that meets each student’s own characteristics is generated.
Physical training system framework
This paper learned before the experiment that the first-year students’ physical education foundation is weak, which is not conducive to the operation of this experiment, so this thesis selects the experimental subjects from the second year of college A. At the beginning of the sophomore semester, the university conducted a physical fitness test for the sophomore students, and by looking at the results of the physical fitness test, it was learned that the physical fitness of the sophomore physical education majors (5) and (7) two classes of students are closer to each other, which is conducive to this experiment. Therefore, with the consent of the school, these two classes were selected for the experiment. In order to ensure the fairness of the experiment, before the experiment, we excluded 4 students who were temporarily unfit for sports due to physical diseases, and finally screened out 62 college students, with 31 students in both the experimental group (Group A) and the control group (Group B).
Experimental grouping Before conducting the experiment, 62 college students were tested for physical fitness and basic skills in sports, and according to the ranking order of the test results, a serpentine grouping method was used to divide them into experimental and control groups to ensure that the two groups of students had a uniform level before the experiment. Experimental time and place The location of the experiment was A gymnasium of N city college, and the time of the experiment was September 1, 2023-December 31, 2023, with a total of 24 hours in 17 weeks, the first and last hours were for testing, and the remaining 22 hours were for teaching. Control of physical education and sports training experiment In this experiment, in order to avoid the influence of the difference of teacher’s teaching on the effect of physical training, the students in group A and group B were taught by the same teacher, and the two classes, except for the different methods of physical education and sports training, were kept uniform in the content of physical education learning, learning time, learning progress, and evaluation standards. During the experimental period, the subjects were reminded not to participate in other sports activities as much as possible, except for the two weekly sessions of the experiment. If there are more than two (including two) absences due to injury or illness, etc., then the experimental subjects will not participate in the final test. In order to avoid experimental errors, the experiment was conducted using the single-blind method, i.e., the two groups of students were not aware of the purpose of this teaching experiment. And throughout the experiment, students’ psychology will be paid attention to in time and more verbal encouragement will be used to avoid students practicing negatively for some reasons. At the end of the experiment, three physical education teachers from college A in N city were asked to evaluate the teaching effectiveness of these 62 students. They did not know the experimental grouping beforehand to ensure the fairness and impartiality of the evaluation. The average score of the 3 teachers’ ratings was taken at the end, so as to reduce subjective errors in the ratings of the evaluating teachers. Experimental variables The independent variable is that Group A adopts the sports training intelligent system assisted training method, while the control group adopts the traditional sports training method. The dependent variables are the results of physical fitness tests and the results of sports functional movement skills tests.
Based on relevant research and seeking advice from sports-related expert teachers, this paper selects sports test items suitable for college students to perform according to the National Physical Fitness Standard for Students, including standing long jump, 50-meter run, 800/1000-meter run, sit-ups and seated forward bends.
Before the experiment, in order to exclude the influence of different physical quality factors on the experimental results between Group A and Group B, the college students were tested on five physical quality items, namely, standing long jump, 50-meter run, 800/1000-meter run, sit-up and sitting forward bend, and the results of the pre-test of physical quality are shown in Table 1. T-test P-value of sitting forward bends were 0.526>0.05, 0.845>0.05, 0.663>0.05, 0.263>0.05 and 0.241>0.05 respectively. The P-value of the five physical fitness test results of the two groups of students was greater than 0.05, so it means that there is no significant difference in the physical fitness of the students in the two groups, which is in line with the requirements of the current experiment, and it is possible to carry out the next step of the experimental study.
The physical quality analysis of the students before the experiment
| Test item | Group A | Group B | t | P |
|---|---|---|---|---|
| Fixed jump | 185.26±15.62cm | 188.49±16.54cm | -0.956 | 0.526 |
| 50m run | 7.42±0.56s | 7.39±0.68s | 0.362 | 0.845 |
| 800/1000m run | 3.56±0.26min | 3.57±0.21min | -0.163 | 0.663 |
| Sit-ups | 30.26±5.64 | 29.68±5.87 | 0.529 | 0.263 |
| Preflexion | 13.24±3.62cm | 14.25±4.52cm | -0.887 | 0.241 |
After 17 weeks and 24 teaching experimental courses, the difference test was conducted to compare and analyze whether there was a significant difference between the changes in physical fitness level of students in Group A and Group B before and after the experiment. The specific results of the comparative analysis between Group A and Group B in the post-experiment test are shown in Table 2, the results of the comparison between Group A’s pre- and post-tests are shown in Table 3, and the results of the comparison between Group B’s pre- and post-tests are shown in Table 4. From the results of the data analysis, it can be seen that after 17 weeks of physical training based on the sports training intelligent system, the physical quality test indexes of the students in group A have improved compared with those before the experiment. From the mean value of each physical quality test index, among them, standing long jump improved by 11cm, paired samples T-test result P=0.002>0.05, 50m run and 800/1000m run improved by 0.49s and 0.37min respectively, paired samples T-test result showed P=0.021, 0.042<0.05. Before and after the experiment, the students of group A’s physical quality T-test of all the indicators of the test P is less than 0.05. The above data analysis results show that after 17 weeks of intelligent system-assisted physical training, the average score of the physical fitness test of the students in group A increased significantly. In addition, there was also a significant difference between the students of group A and group B in terms of physical fitness in the post-test of the experiment. Therefore, it can be assumed that the intelligent system-assisted physical education and sports methods have a more significant effect on the physical fitness of university students.
After the physical quality experiment (A VS B)
| Test item | Group A | Group B | t | P |
|---|---|---|---|---|
| Fixed jump | 196.26±16.95cm | 189.63±17.52cm | 2.123 | 0.001 |
| 50m run | 6.93±0.42s | 7.21±0.35s | 2.922 | 0.022 |
| 800/1000m run | 3.19±0.35min | 3.51±0.56min | 2.507 | 0.034 |
| Sit-ups | 35.64±6.95 | 31.25±7.48 | 2.866 | 0.017 |
| Preflexion | 17.52±2.84cm | 15.26±5.29cm | 2.689 | 0.034 |
After the physical quality experiment (A VS A)
| Test item | Before | After | t | P |
|---|---|---|---|---|
| Fixed jump | 185.26±15.62cm | 196.26±16.95cm | 2.481 | 0.002 |
| 50m run | 7.42±0. 56s | 6.93±0.42s | 2.557 | 0.021 |
| 800/1000m run | 3.56±0.26min | 3.19±0.35min | 2.797 | 0.042 |
| Sit-ups | 30.26±5.64 | 35.64±6.95 | 2.779 | 0.007 |
| Preflexion | 13.24±3.62cm | 17.52±2.84cm | 2.939 | 0.017 |
After the physical quality experiment (B VS B)
| Test item | Before | After | t | P |
|---|---|---|---|---|
| Fixed jump | 188.49±16.54cm | 189.63±17.52cm | 1.417 | 0.137 |
| 50m run | 7.39±0.68s | 7.21±0.35s | 1.104 | 0.142 |
| 800/1000m run | 3.57±0.21min | 3.51±0.56min | 0.371 | 0.124 |
| Sit-ups | 29.68±5.87 | 31.25±7.48 | 1.405 | 0.144 |
| Preflexion | 14.25±4.52cm | 15.26±5.29cm | 1.054 | 0.098 |
Functional Movement Screening (FMS) is a method designed and proposed by American orthopedic training specialists and training experts to assess athletes’ movement patterns, which can easily identify individuals’ functional limitations and asymmetrical development. It consists of seven movements and is widely used to evaluate basic movement abilities of various people, including shoulder and hip flexibility and core muscle group stability evaluation. Functional Movement Screening is designed to provide a simpler and quicker understanding of a person’s movement quality, and is therefore a comprehensive assessment of the quality of physical movement. Along with the popularization of FMS, its assessment is widely used in rehabilitation training and physical training to measure the quality of movement and plays an important role in sports. Functional Movement Screening (FMS) is a test of seven basic movements, which is designed to screen out the weak links in body functions that may cause injury and to fully understand the flexibility and coordination of the body of the tested person. Therefore, it is suitable to use this study to determine the improvement effect of functional basic skills on college students’ sports training under the application of an intelligent assistive system for sports training.
Functional movement tests were performed on 62 students in Group A and Group B respectively before the experiment, and the results of the comparative analysis of the FMS tests of the two groups are shown in Fig. 3, where FMS1-FMS6 stand for deep squat, upward step over the bar rack, straight line lunge, shoulder mobility, supine leg raise, and push-ups, respectively. It was found that there was not much difference in the FMS functional movement assessment scores between the two groups of subjects, with the test scores for straight line lunge (2.12 and 2.13) and push-ups (1.94 and 1.95) being very close to each other. Utilizing the obtained data using independent samples t-test, it was found that the p-value of the mean score of the single assessment of functional movements and the total mean score of the students in groups A and B were greater than 0.05, indicating that the FMS assessment scores between the students in groups A and B before the experiment were non-significantly different. Functional training of students can to a certain extent reflect the level of students’ basic physical function, according to the results of data analysis, it can be seen that the students of group A and group B are at the same level of basic physical function, therefore, the selected experimental group and the control group of students meet the requirements of the experiment.
Analysis of FMS test for college students
In order to further explore the greatest advantages of the sports training intelligent system to assist university sports training, the stage assessment of movements was conducted for students in Group A and Group B in the middle and late stages of the experiment to analyze the changes produced by the experimental intervention. The results of analyzing the scores of the stage-specific movement assessment are shown in Figure 4, with (a)-(f) representing the results of the functional movement test scores in the first, fourth, seventh, tenth, thirteenth and seventeenth weeks of the experiment, respectively. Independent samples t-tests were conducted on the stage movement assessment data of Group A and Group B students respectively, and the results showed that the p-value in the data of the 7th week was 0.032, which was less than 0.05 with significant difference, indicating that the beginning stage of the formation of sports training movement skills of the students in Group A began to change. In the 10th week of movement assessment the mean and standard deviation of group A and group B were 3.40 and 2.87 points, and the mean value of group A was much higher than that of group B. The mean value of group A was 3.40 and 2.87 points, and the mean value of group B was much higher. The p-value of the two groups was 0.008 and p<0.01, which is a highly significant difference, indicating that the students in group A showed more significant changes in the middle stage of the experiment. In the 13th and 17th weeks of the functional movement assessment of physical education and sports, the average value of the movement assessment of students in group A reached 3.81 and 3.98 points, respectively, and group B was 3.25 and 3.31 points, respectively, with a P value of 0.001, 0.000<0.01, and a highly significant difference between the two groups, which indicates that the mastery of the movement skills of the students in the experimental group was superior to that of the students in the control group. By analyzing the assessment results of every 3 weeks, it can be seen that the changes in the assessment scores of functional movements in sports training between students of group A and group B are large, especially in the middle and late stages of the experiment, and the initial period does not have a significant difference. This indicates that the application of intelligent systems in university sports training can assist in calibrating functional movements and improving assessment scores of university students.
Analysis of stage action evaluation
This paper utilizes sensing technology and bone tracking technology, comprehensively considers and analyzes the actual needs of university sports training, and develops a set of university sports training intelligent auxiliary system based on action recognition technology. The effectiveness of the intelligent system has been explored, and the results show that:
After 17 weeks of sports training based on the sports training intelligent system, the physical fitness test indexes of the second-year students majoring in physical education in college A were improved compared with those before the experiment. 50-meter run and 800/1000-meter run were improved by 0.49s and 0.37min, respectively, and the results of the paired-sample t-test before and after the experiment showed that P=0.021, 0.042<0.05, which is a significant difference. Under the intelligent system-assisted training, the mean values of the students in the experimental class reached 3.81 and 3.98 points in the examination of functional movements of sports in the 13th and 17th weeks, which were much higher than those of the students in the control class under the traditional training program. It can be assumed that the intelligent system-assisted university sports methods, based on intelligent systems, have more significant effects on college students’ physical fitness and functional training movement enhancement.
“Artificial intelligence + sports training” is the direction of the development of university sports training, the future of sports training is inevitably inseparable from the assistance of intelligent means, this study can make the development of intelligent training to be realized at an early date. Future research can consider the establishment of a more complete model of human movement characteristics, so that it can be adjusted to most actual sports scenarios.