Role of artificial intelligence algorithm for taekwondo teaching effect evaluation model

Abstract

The problems and disadvantages of the traditional teaching mode of Taekwondo in colleges and universities are obvious, which is not conducive to cultivating the interest of contemporary college students in learning Taekwondo. In order to improve the teaching effect of Taekwondo, based on the intelligent algorithm of human body feature recognition, this study uses support vector machine to construct a Taekwondo teaching effect evaluation model based on artificial intelligence algorithm. The model corrects the movement of the students by recognizing the movement characteristics of the students’ Taekwondo and can conduct the movement guidance and exercises through the simulation method. In order to verify the performance of the model in this study, this study set up control experiments and mathematical statistical methods to verify the performance of the model. The research results show that the model proposed in this paper has a certain effect and can be applied to teaching practice

Keywords

Artificial intelligence improved algorithm Taekwondo teaching effect evaluation and analysis

1 Introduction

The 21st century is a society in which talents compete. With the acceleration of the modernization process and the improvement of the economic level, health has become an important topic of human concern. Sports, as one of the important factors that affect the health level, has become or will become an indispensable social activity and lifestyle of human beings. The awareness of full participation in sports and lifelong sports has been deeply rooted in the hearts of the people, and the demand for sports-applied talents in social sports will continue to increase. As a cradle of talent training, how to train talents in line with the development of the market and how to improve the social competitiveness of talents have become the common topics of higher education innovation and reform today. The main purpose of higher education is to train high-skilled and high-tech talents, and application talents who can directly serve the operation, management and development of enterprises [1]. As an important part of higher vocational education, high-efficiency sports education shoulders important responsibilities in the field of social sports and leisure sports, such as national fitness programs, sunshine sports projects, and community sports services. How to carry out talent training program reform, curriculum system innovation, teaching quality evaluation, etc. around social and market needs has become an important subject that needs to be studied urgently. The traditional Taekwondo teaching mode only emphasizes the students’ mastery of basic skills such as leg technique, posture, and stunts. It ignores the guidance of students’ internal needs and life, survival, and concepts of life, and does not conform to the unity of nature and humanity in the spiritual connotation of Taekwondo. Moreover, it does not focus on the main purpose of self-cultivation and perception and the “Sansheng education” concept of Taekwondo teaching. Whether it is school education or club-type extracurricular education, there is a lack of attention to the internal needs of students, which leads to students’ weak sense of life, lack of survival skills, and low ability to live. Moreover, most Taekwondo gyms still remain at the stage of imparting Taekwondo skills, and have not established and guided the students’ life, survival, and concept of life, and have ignored the students’ increasing mental health needs and social adaptation needs [2].

The development of Taekwondo projects is inseparable from the innovation of the teaching model. The traditional Taekwondo teaching model that emphasizes competition and ignores the humanities greatly affects students’ enthusiasm for learning and hinders the future development of the Taekwondo industry. This research is the theoretical innovation and enrichment of the research on Taekwondo teaching mode and provides a reference for the follow-up research on the innovation of Taekwondo teaching mode. In addition, through the teaching reform, the coach truly regards students as the main body of learning in the teaching process, attaches importance to the students’ subjective feelings, and enriches the students’ classroom experience, which enables students to understand life more correctly, view setbacks, and promote healthy physical and mental growth and good social adaptation of students [3].

Therefore, it is necessary to pay attention to the application of artificial intelligence algorithms in Taekwondo teaching, effectively correct the problems that may exist in the teaching, effectively improve the teaching efficiency, and help students continue to improve their regularity and reaction speed in Taekwondo.

2 Related works

The first peak of artificial intelligence: After this meeting in 1956, artificial intelligence ushered in its first Happy Time [4].In the more than ten years since then, computers have been used to solve computational, geometric, and linguistic problems. Because the computing power of computers far exceeds that of humans, this has led many experts to learn with confidence to study how to make machines develop artificially. Some scholars even believe that: “In twenty years, machines will be able to accomplish everything humans can do.” The first trough of artificial intelligence [5]: Due to the lack of scientific research personnel’s estimates of artificial intelligence research and development, many cooperative plans failed, which confused everyone about the future of artificial intelligence and caused many scientific research projects to be interrupted. At that time, the technical bottlenecks faced by artificial intelligence were mainly three aspects. First, because of the lack of computer performance at that time, many artificial intelligence applications could not be used. Second, the actual situation is more complicated than expected. At that time, the application of artificial intelligence was only to solve the specified problem, and only the specified problem was considered during the design. When the actual problem was encountered, the difficulty of the problem increased, which caused the program to crash immediately. Third, artificial intelligence requires a lot of data. At that time, the amount of data was small, which caused the machine to be unable to learn and evolve through a large amount of data [6].

The literature [7] designed a set of “expert system” called XCON for digital equipment companies. This is an intelligent computer program system, which contains expert-level knowledge and experience, and the system performs reasoning and judgment based on these knowledge and experience. This system can save the company more than 4,000 dollars per year, and the birth of the expert system also ignited the flame of artificial intelligence development. After that, Apple and IMB produced desktop computers with superior performance, which far exceeded the computers equipped with expert systems. Since then, the expert system has ended [8]. Artificial intelligence rises again: Since the mid-1990 s, with the advancement of neural network technology, people have an objective understanding of artificial intelligence [9], and the development of artificial intelligence technology has entered a stable period [10]. The computer system “Dark Blue” made by IBM defeated the world chess champion Kasparov and once again brought artificial intelligence to the public’s vision. This is an important milestone in the development of artificial intelligence [11]. The literature [12] made a breakthrough in the field of deep learning of neural networks, which made humans see the possibility of machines achieving intelligence and surpassing humans once again. This is a landmark event in the advancement of artificial intelligence technology. Google’s AlphaGo in the field where humans believe that machines cannot beat humans, defeated South Korean chess player Li Shishi, which once again triggered a wave of artificial intelligence. As Internet giants such as Google, Baidu, Ali, and Tencent entered the battlefield of artificial intelligence research and development, another round of artificial intelligence frenzy was launched. At present, artificial intelligence technology is increasingly mature and applied to people’s daily life, and people are happy and need them to integrate into life. This wave of artificial intelligence may realize the intelligence of human beings and make science fiction movies into reality [13].

In the field of artificial intelligence and language applications, American scientists and scholars have made great contributions and guided the expansion of artificial intelligence technology research into the field of language teaching [14]. The literature [15 –18] demonstrated a special kind of “simultaneous interpreter”. The most interesting thing is that after the presenter spoke English to the system for about an hour, the system can not only translate the English into Mandarin immediately, but also simulate the speaker’s voice and intonation. During the one hour he speaks English, the system uses machine learning to identify the nuances in everyone’s voice, and builds the corresponding model, and finally presents a true simultaneous translation with the same phonetic tone as the expressor. The literature [16] combined various knowledge accumulated in English teaching and computer programming to discuss the application of artificial intelligence in second language teaching. Moreover, it particularly emphasized that in second language teaching, attention should be paid to the analysis of language learners’ errors to help learners correct their errors. The literature [17] introduced the application and implementation plan of “academic English intelligent teaching system”, and the content of the introduction is both practical and innovative. Moreover, the program not only provides certain technical ideas for researchers engaged in the field of artificial intelligence applications, but also provides practical experience for frontline workers in the second foreign language teaching.

3 Preprocessing of 3D motion capture data

In this paper, after studying the movement of the human body and considering various posture changes in human movements, the main human posture characteristic parameters selected in space are speed, displacement, acceleration and angle changes of key nodes of the human body, as shown in Table 1. The speed is a vector, which not only simply indicates the speed but also the direction, and the acceleration not only indicates the acceleration in the horizontal direction, but also indicates the acceleration in the vertical direction. The angle characteristics of the human body during exercise are very important. For example, when a person performs a running motion and a jumping motion, the angle of the legs changes differently, and the angle of the hands is also different. Therefore, the angular velocity of the knee joint (the angle between the thigh and the lower leg) and the angular velocity of the elbow joint (the angle between the elbow and the arm) are both selected as the characteristic parameters. In the motion capture raw data, there are the frames and the time interval between frames. The parameters in the time dimension can be obtained by the above two variables [19 –23].

Table 1
Characteristic parameters of human posture

Name Introduction

Speed Direction and speed of movement

Displacement Vector difference between action cutoff and start

Acceleration Horizontal and vertical acceleration

Angular velocity Joint angle change

Name	Introduction
Speed	Direction and speed of movement
Displacement	Vector difference between action cutoff and start
Acceleration	Horizontal and vertical acceleration
Angular velocity	Joint angle change

As shown above, suitable motion features are selected, but they cannot be directly calculated from the original BVH data, and feature space conversion is required, that is, conversion of BVH format files.It can be seen that the data stored in the BVH file is the rotation data of the Euler angle. Because it is necessary to analyze the relative positional relationship of the coordinate system of each joint of the human body in space, it is necessary to explain the commonly used coordinate system in 3D graphics before introducing how to obtain the desired features. This will be used later in this article when converting Euler angle data.

The data of the Euler angle data collected in the experiment is 66 columns per frame, corresponding to 21 nodes. The root node has six channels, so it corresponds to the first six columns, and the subsequent nodes correspond to three columns of data in turn. Because the 21 nodes between the human body are established through mathematical modeling, the Euler angle is used to store the offset data for this node. Euler angle is one of the commonly used methods to express rotation. It is proposed by the mathematician Euler. For a reference system in three-dimensional space, the orientation of any coordinate system can be expressed by three Euler angles. However, the Euler angle does not indicate the position of the human body in space, nor can it intuitively describe the human body’s movement state. Moreover, the effect of directly using the Euler angle data as the input feature to the platform is not ideal, so it is particularly important to convert the Euler angle data to data in the world coordinate system.

To transform the object coordinate system into the world coordinate system, the inertial coordinate system needs to be used as the intermediate transition coordinate system.:The specific idea is to assume a point (x₀, y₀, z₀), first transform the object coordinate system to inertial coordinate system (x₁, y₁, z₁)[24]: $[\begin{matrix} x_{1} \\ y_{1} \\ z_{1} \end{matrix}] = R [\begin{matrix} x_{0} \\ y_{0} \\ z_{0} \end{matrix}]$ (1)

Among them, R is the rotation matrix that needs to be constructed between the two points, and the data in the experiment are sequentially rotated around the ZXY axis. Then, we can get the matrix obtained by R along the XYZ axis, as shown in formula (2): $R = R_{z} R_{x} R_{y}$ (2)

Among them, R_z represents the rotation matrix obtained along the Z axis, as shown in formula (3). Similarly, the R_x and R_y rotation matrix can be obtained, as shown in formula (4) and formula (5): $R_{z} (θ) = [\begin{matrix} cos (θ) & - sin (θ) & 0 \\ sin (θ) & cos (θ) & 0 \\ 0 & 0 & 0 \end{matrix}]$ (3) $R_{x} (θ) = [\begin{matrix} 1 & 0 & 0 \\ 0 & cos (θ) & - sin (θ) \\ 0 & sin (θ) & cos (θ) \end{matrix}]$ (4) $R_{y} (θ) = [\begin{matrix} cos (θ) & 0 & sin (θ) \\ 0 & 1 & 0 \\ - sin (θ) & 0 & cos (θ) \end{matrix}]$ (5)

According to the above formula, the conversion from the object coordinate system (x₀, y₀, z₀) to the inertial coordinate system (x₁, y₁, z₁) is completed. Next, the world coordinate system (x₂, y₂, z₂) is obtained by translation [25]. Because the adjacent nodes of the BVH file are connected, the position of the child node can be obtained from the position of the parent node. If it is assumed that the inertial rotation matrix of all previous nodes of node J is R₁, R₂, ⋯ , R_r, then the offset of this node relative to its precursor node in the inertial coordinate system is shown in formula (6): $[\begin{matrix} x_{1} \\ y_{1} \\ z_{1} \end{matrix}] = R_{r} (R_{r - 1} (\dots (R_{2} ([\begin{matrix} x_{0} \\ y_{0} \\ z_{0} \end{matrix}]))))$ (6)

According to this formula, the position offset of any node relative to the root node can be calculated, which solves the rotation problem, and the coordinates in the world coordinate system can be obtained later by removing the translation.

A frame in a Taekwondo action video is just an image. However, an image cannot predict its next movement, that is, it cannot directly reflect the state of the movement. Therefore, this experiment uses 20 frames (One frame = 0.00666 seconds) as a unit to calculate the displacement, average speed, acceleration, and the angle between the knee joint and the elbow joint as input features. After obtaining the Euler angle and world coordinate system of the motion capture data, the corresponding displacement characteristics can be obtained accordingly. If it is assumed that the nodal world coordinate system acting at the previous moment is (x₀, y₀, z₀), and the nodal world coordinate system is (x₁, y₁, z₁) after ten frames, the displacement, velocity and acceleration are as shown in formula (7), formula (8), formula (9) [21]:

$s = \sqrt{{(x_{1} - x_{0})}^{2} + {(y_{1} - y_{0})}^{2} + {(z_{1} - z_{0})}^{2}}$ (7) $v = s / (frames \times frameTime)$ (8) $a = (v_{2} - v_{1}) / (frames \times frameTime)$ (9)

Among them, frame and frameTime exist in the data block part of the BVH file, which can be directly read and used.

The side length of the triangle formed by the three nodes of the human leg and the angle of the knee joint can be obtained as shown in formula (10), formula (11), formula (12), and formula (13): $a = \sqrt{{(x_{0} - x_{1})}^{2} + {(y_{0} - y_{1})}^{2} + {(z_{0} - z_{1})}^{2}}$ (10) $b = \sqrt{{(x_{1} - x_{2})}^{2} + {(y_{1} - y_{2})}^{2} + {(z_{1} - z_{2})}^{2}}$ (11) $c = \sqrt{{(x_{0} - x_{2})}^{2} + {(y_{0} - y_{2})}^{2} + {(z_{0} - z_{2})}^{2}}$ (12) $θ = arccos ((a^{2} + b^{2} + c^{2}) / 2 ab)$ (13)

4 Automatic segmentation algorithm

In the Taekwondo feature automatic generation platform based on direct segmentation, motion segmentation is a prerequisite for identifying Taekwondo movements. It is very important to find the start frame and end frame of the action in the motion sequence to separate the action in time to classify the action. The threshold-based segmentation algorithm is a direct segmentation algorithm, which separates segmentation and recognition, and performs segmentation first and then motion recognition. When the movement of the arm waving upward ends, the linear velocity becomes zero. In the same way, the linear velocity is the same result when waving the arm downward. Therefore, by setting the threshold value of the linear velocity to zero, the motion can be divided. If the world coordinate system of the action node at the beginning is (x₀, y₀, z₀), and the world coordinate system of the node at the end is (x₁, y₁, z₁), then the speed is as shown in formula (14): $v = \frac{\sqrt{{(x_{1} - x_{0})}^{2} + {(y_{1} - y_{0})}^{2} + {(z_{1} - z_{0})}^{2}}}{frames \times frameTime}$ (14)

Among them, frames represents the number of time frames that the action lasts, and frameTime represents the interval between each frame.

The advantage of segmentation algorithm based on threshold is that it is easy to understand and simple to calculate. As shown in Fig. 1, it is a frame diagram of motion speed after filtering and denoising. After setting the threshold, in chronological order, when it changes from below the threshold to above, it is the beginning of a movement, and when it changes from above the threshold to below the threshold, it is the end of a movement. Therefore, the segmentation algorithm based on threshold can realize motion segmentation.

Fig. 1

Motion speed frame.

5 Research on automatic generation algorithm of Taekwondo features based on extreme learning

In the Taekwondo feature automatic generation platform, selecting the appropriate motion recognition algorithm is a key step for automatically mapping human actions to feature extraction, and is also a core part of the Taekwondo feature automatic generation platform. This section mainly discusses the feasibility of the dynamic programming method, support vector machine, and extreme learning as the platform recognition algorithm for automatic Taekwondo feature generation.

In action recognition, time series is the representation form of data, which compares the similarity of two actions, that is, to compare the similarity of two sequences. In the time series, the lengths of two time series that need to be compared may not be equal, which indicates that in the field of motion recognition, different people behave at different speeds. Because the action signal is quite random, even if the same person does actions at different times, it is impossible to have a full length of time. In this case, the traditional Euclidean distance cannot effectively find the distance or similarity between two time series. That is to say, in most cases, the two sequences have very similar shapes as a whole, but these shapes are not aligned in the time series. In this case, the simplest alignment method to compare the similarity of two time series is linear scaling. That is, the short sequence is linearly enlarged to the same length as the long sequence and then the similar comparison is performed. However, this calculation does not take into account that the duration of each segment in the action segment will have a longer or shorter change in different situations, so the recognition effect may not be optimal. Therefore, more methods currently used are Dynamic Time Warping methods.

The action of the segmented action segment is single and relatively simple. Moreover, the distance between the two actions performs similarity matching according to the form of adaptively extracting key frames. Each video sequence is composed of many frames, so we first need to define the distance between two different frames, as shown in formula (15): $d (M, N) = \sqrt{\sum_{i = 1}^{21} {(m_{i} - n_{i})}^{2}}$ (15)

M and N represent two different frames of data, and i represents the i-th node. Since the three-dimensional model of the human body is 21 nodes, the maximum value of i is 21.The distance between frames is chosen to be the absolute distance of the current node, so that the situation that positive and negative values cancel each other will not occur. Similarly, the distance of each frame between two action sequences can be obtained, that is, the frame distance matrix of every two frames between two actions can be obtained, as shown in formula (16): $D_{f} = [\begin{matrix} d_{11} & \dots & d_{1 n} \\ ⋮ & ⋱ & ⋮ \\ d_{m 1} & \dots & d_{mn} \end{matrix}]$ (16)

After the construction of the inter-frame distance matrix is completed, the DTW method is used to find the optimal distance match. Because the time dimension in the action sequence is unidirectional, when the path is searched, there are only three possibilities for the next point, as shown in Fig. 2:

Fig. 2

Optimization path.

The inter-frame distance between two actions is shown in formula (17):

$D_{ij} = d_{ij} + min (D_{i - 1, j}, D_{i, j - 1}, D_{i - 1, j - 1})$ (17)

Among them, 1 < i < m + 1, 1 < j < n + 1, and m and n represent the number of rows and columns of the frame matrix of the template and the sequence to be recognized, that is, their frame lengths. As shown in Fig. 3 is a particularly similar DTW action matching example:

The Inter-frame distance is normalized to the range of 0–255, so that the grayscale image can be obtained, and the dynamically planned path can be intuitively seen.

Fig. 3

DTW action matching example.

Support vector machine (SVM) is the classification algorithm first proposed by Comma Cortes and Vapnik in 1995.The simplest version of SVM is a binary classification algorithm, which supports both linear and nonlinear classification. After evolution, multivariate classification can now be supported, and SVM is considered to be one of the best classification algorithms at present.

For linear separability, SVM attempts to find a straight line so that it can isolate binary data. In three-dimensional space or higher-dimensional space, it is trying to find a hyperplane to separate all binary classes. Among them, the case of two types of linear separability is relatively simple, so it is used as a representative, as shown in Fig. 5. sIn the figure, the two types of samples are represented by hollow points and solid points, respectively. In this case, we can find countless classification planes that correctly classify the samples, such as H₁H₂ and H₃ in the figure. However, it can be seen that the superiority of each classification plane is different. Some are only able to separate the two types of samples. If new samples appear in these two types of samples, they may not be accurately classified, that is, the generalization performance is different.

Fig. 4

Multiple hyperplane with accurate classification.

Fig. 5

Optimal classification plane.

The basic model of SVM is a linear classifier that seeks to separate hyperplanes with maximum spacing in the feature space. From the training data, SVM is divided into the following situations:

(1)When the training data is linearly separable, it is necessary to maximize the hard interval at this time. The hard interval is relative to the training data set or the feature space, which means that no classification errors are allowed. The model learned by the hard interval can be called a hard interval support vector machine.

(2)When the training data is approximately linearly separable, it is necessary to introduce slack variables. At this time, the soft interval should be maximized. The soft interval is relative to the hard interval, which allows a certain error to exist. For error measurement, a parameter needs to be defined. This parameter measures how big the data is and not several errors. The model learned by soft interval can be called soft interval support vector machine.

(3)When the training data is linear and inseparable, it is necessary to introduce a kernel function and maximize the soft interval. The trained model is a nonlinear support vector machine.

When the training data is linearly separable, there are countless separation hyperplanes in the space at this time to correctly separate the sample data. The perceptron uses the misclassification minimum strategy to find the separation hyperplane, but there are infinitely many solutions at this time. The linear separable support vector machine uses the maximum interval to find the optimal separation hyperplane. At this time, this hyperplane exists only. On the other hand, the classification results produced by separating hyperplanes at this time are the most robust.

Kernel functions are sometimes introduced in SVM. The reason is that when the original space is linearly inseparable, the sample can be mapped from the original space to a higher-dimensional feature space, so that the sample is linearly separable in this feature space.

In the above, this paper introduced in detail the content of SVM adopting the maximum interval, and the content that original problem of solving SVM is converted into the content of its dual problem and the content of the kernel function introduced by SVM. Next, this paper performs the derivation of linear SVM.

We assume that the data set has N samples x₁, x₂, ⋯ , x_N, which correspond to N labels y₁, y₂, ⋯ , y_N. Among them, y_i ∈ { - 1, 1 } , i = 1, 2, ⋯ , N. The optimal classification plane is shown in Fig. 5:

We assume that the hyperplane H with w · x + b = 0 can separate all the data, that is, the label is +1 when w · x + b > 0, and the label is -1 when w · x + b < 0. At the same time, there are two hyperplanes H₁ and H₂ parallel to H so that the positive and negative samples closest to H fall on H₁ and H₂ respectively. Then, the calculation formula of the physical distance from a sample to the classification interface is shown in formula (18): $\hat{y} = \frac{y_{n} (w^{T} x_{n} + b)}{∥ w ∥}$ (18)

What the SVM needs to do is to find the classification plane that maximizes the minimum physical interval, and the above problem can be replaced with a quadratic programming problem, as shown in formula (19): ${\begin{matrix} arg min \frac{1}{2} w^{2} \\ y_{n} (w^{T} x_{n} + b) \geq 1 \end{matrix}$ (19)

The constraints and optimization problems are obtained by combining Lagrange factors, as shown in the following formula (20): $L (w, b, a) = \frac{1}{2} w^{2} - \sum_{n = 1}^{N} a_{n} {y_{n} (w^{T} x_{n} + b) - 1}$ (20)

The solution here is to find the minimum physical interval first, and then maximize it, so that the partial derivative of L (w, b, a) to w, b is 0.The following two equations are obtained, as shown in formula (21) and formula (22): $\frac{\partial L (w, b, a)}{\partial w} = w - \sum_{n = 1}^{N} a_{n} y_{n} x_{n} = 0$ (21)

$\frac{\partial L (w, b, a)}{\partial b} = \sum_{n = 1}^{N} a_{n} y_{n} = 0$ (22)

By substituting (21) and (22) into (20), the following formula is obtained: ${\begin{matrix} max_{a} (\sum_{n = 1}^{N} a_{n} - \sum_{n = 1}^{N} \sum_{m = 1}^{N} a_{n} a_{m} y_{n} y_{m} x_{n}^{T} x_{n}) \\ s . t . \sum_{n = 1}^{N} a_{n} y_{n} = 0 \\ a_{n} \geq 0 \end{matrix}$ (23)

Finally, the linearly separable optimal plane parameters are obtained as follows, where x is the support vector: ${\begin{matrix} w = \sum_{n = 1}^{N} a_{n} y_{n} x_{n} \\ b = y_{n} - \sum_{n = 1}^{N} a_{n} y_{n} (x_{n} x_{m}) \end{matrix}$ (24)

The actions contained in the action database used in this article are divided into 16 types of leg movements, 20 types of hand movements, a total of 36 types of data, so it cannot be solved with the original SVM. The solution is to select the libsvm library in Matlab to classify Taekwondo movements. The multi-classification method of lib svm is realized by one-to-one method. Its specific idea is to design an SVM classifier between every two samples, so the action category in this experiment is 36 classes, which requires 630 classifiers.

The radial basis kernel function is used to classify and recognize the actions collected in the experiment, and the motion samples are cross-trained during training. In this experiment, 10-fold cross-validation was adopted, that is, the entire sample was divided into 10 parts, and 1 part was selected as the test sample in turn, and the remaining 9 parts were used as the training samples for training. Through cross-validation, the most efficient penalty parameter C = 5000 and the parameter g = 0.02127 in the radial basis kernel function can be obtained. Among them, what needs to be explained is that all the classifiers are tested separately during the identification, and then the results are obtained by voting. The process of using SVM to classify human actions in this experiment is shown in Fig. 6.

Fig. 6

Flowchart of generating Taekwondo features based on SVM.

6 Performance verification of Taekwondo teaching effect evaluation model based on artificial intelligence

The requirements of traditional colleges and universities for taekwondo courses tend to only improve students’ physical fitness and taekwondo skills. In the assessment of taekwondo courses, the school syllabus is also based on taekwondo techniques, and ignores the basic knowledge of Taekwondo theory. The model constructed in this study was applied to Taekwondo teaching. The performance analysis of the teaching model of this study was carried out through a control experiment. The number of students in the experimental group and the control group is 50, and the students’ scores on Taekwondo are scored before the experiment. The results are shown in Table 2 and Fig. 7.

Table 2
Statistical table of the scores of the students in the test group and the control group before the experiment

Serial number Control group Test group Serial number Control group Test group

1 79 78 26 54 58

2 78 78 27 53 56

3 78 78 28 53 56

4 77 77 29 52 56

5 76 76 30 51 56

6 73 76 31 50 53

7 72 75 32 50 53

8 70 73 33 50 53

9 68 73 34 47 52

10 67 72 35 46 50

11 62 70 36 45 46

12 60 68 37 45 44

13 60 67 38 44 42

14 60 66 39 43 42

15 59 66 40 40 41

16 59 65 41 39 41

17 59 65 42 37 39

18 58 65 43 37 37

19 57 64 44 36 36

20 57 62 45 36 36

21 57 62 46 34 34

22 57 62 47 34 34

23 56 59 48 34 33

24 56 59 49 33 31

25 56 58 50 31 31

Serial number	Control group	Test group	Serial number	Control group	Test group
1	79	78	26	54	58
2	78	78	27	53	56
3	78	78	28	53	56
4	77	77	29	52	56
5	76	76	30	51	56
6	73	76	31	50	53
7	72	75	32	50	53
8	70	73	33	50	53
9	68	73	34	47	52
10	67	72	35	46	50
11	62	70	36	45	46
12	60	68	37	45	44
13	60	67	38	44	42
14	60	66	39	43	42
15	59	66	40	40	41
16	59	65	41	39	41
17	59	65	42	37	39
18	58	65	43	37	37
19	57	64	44	36	36
20	57	62	45	36	36
21	57	62	46	34	34
22	57	62	47	34	34
23	56	59	48	34	33
24	56	59	49	33	31
25	56	58	50	31	31

Fig. 7

Statistical table of the scores of the students in the test group and the control group before the experiment.

From the results shown in Table 1 and Fig. 7, we can see that the test group and the control group have basically the same taekwondo performance before the test, the difference is not obvious, and there was no statistical significance. Therefore, it can be considered that the test group and the control group have the same starting point for Taekwondo performance before the test, and a controlled test can be carried out. After sixteen weeks of learning, the average value of each dimension of the students’ interest in the test and control groups has increased to varying degrees. The teaching methods and teaching time are relatively consistent. The difference is only that the text group adopts the artificial intelligence teaching model constructed in this paper for Taekwondo assisted teaching. Statistical methods were used to record and analyze various data during the experiment.

Figures 8 and 9 are the comparison of the scores of the control group and the test group before and after the experiment.

Fig. 8

Comparison diagram of the scores of the control group before and after the experiment.

Fig. 9

Comparison diagram mf the results of the text group before and after the experiment.

As shown in Figs. 8 and 9, from the figure we can see that the test group and the control group have a certain degree of improvement before and after the experiment. The difference in the degree of improvement between the test group and the control group cannot be clearly seen from Figs. 8 and 9.Therefore, the results of the test group and the control group before and after the experiment are compared together. The results of the test group and the control group are arranged from high to low. The results are shown in Table 3 and Fig. 10.

Table 3

Statistical table of scores of the test group and control group students after the experiment

Serial number	Control group	Test group	Serial number	Control group	Test group
1	86	94	26	56	64
2	84	93	27	56	63
3	84	91	28	54	63
4	81	90	29	54	63
5	80	87	30	54	62
6	78	86	31	54	61
7	74	84	32	51	61
8	73	83	33	51	59
9	73	82	34	51	56
10	69	81	35	51	54
11	67	80	36	47	53
12	65	78	37	46	49
13	64	78	38	45	48
14	63	78	39	44	48
15	62	77	40	44	46
16	61	74	41	43	44
17	61	70	42	39	44
18	61	70	43	39	42
19	61	69	44	38	42
20	61	69	45	37	42
21	60	69	46	37	40
22	60	67	47	36	38
23	58	66	48	36	36
24	57	66	49	35	33
25	57	65	50	32	33

Fig. 10

Statistical diagram of scores of the test group and control group students after the experiment.

As shown in Fig. 10, the two curves are the performance statistics of the text group and the control group after the experiment. In order to further describe the difference between the scores of the text group and the control group of students after the experiment, this study performed a difference treatment on the score difference between the two groups of students. The scores of the corresponding serial number students in the test group minus the scores of the corresponding students in the control group are obtained. The results are shown in Table 4 and Fig. 11.

Table 4

Statistical table of the difference between the scores of the students in the experimental group and the control group

Serial number	Difference	Serial number	Difference	Serial number	Difference
1	7	18	9	35	3
2	8	19	8	36	6
3	6	20	8	37	4
4	9	21	9	38	3
5	7	22	7	39	4
6	8	23	8	40	3
7	9	24	9	41	1
8	10	25	8	42	4
9	9	26	7	43	3
10	12	27	8	44	4
11	13	28	9	45	4
12	13	29	8	46	3
13	14	30	8	47	2
14	15	31	7	48	1
15	15	32	10	49	–1
16	12	33	8	50	1
17	9	34	6

Fig. 11

Statistical diagram of the difference between the scores of the students in the experimental group and the control group.

As shown in Table 4 and Fig. 11, in the statistical difference between the test group and the control group, except for the No. 49 student, the scores of the other students are higher in the test group than the control group. An analysis of the scores of student No. 49 found that the students with lower grades have attitude problems in their daily study, and he grades have not progressed from beginning to end, so this student do not affect the results of this experimental study. It can be seen that the evaluation model of Taekwondo teaching effect proposed in this paper has a certain effect in Taekwondo teaching.

7 Conclusion

The Taekwondo teaching effect evaluation model based on artificial intelligence algorithm can improve the disadvantages of the traditional teaching mode, reduce the student’s learning burden, optimize the classroom structure and enhance the teaching effect, strengthen the cultivation of students’ non-intelligent factors, maximize the benefit of the classroom, encourage students to cooperate in discussion and independent thinking, and promote the development of quality education. When the Taekwondo teaching effect evaluation model based on artificial intelligence algorithm is used in Taekwondo teaching, teachers conduct targeted teaching to understand the difficulties of student learning in teaching, mobilize the subjective initiative of students, enhance students’ interest in learning Taekwondo and learning effect, and improve students’ positive psychological quality. The evaluation of Taekwondo teaching effectiveness based on artificial intelligence algorithms should pay attention to embodying the core principles of student-centered teacher guidance, and teachers should encourage and guide students to explore independently, ask questions and solve problems, and break the thinking in the traditional teaching model. Moreover, teachers need to combine a variety of teaching methods and advanced teaching methods to establish a new classroom framework to correctly treat and guide students and teach students according to their aptitudes to all students.

References

Appelbaum

L.G.

and Erickson

, Sports vision training: A review of the state-of-the-art in digital training techniques[J], International Review of Sport and Exercise Psychology 11(1) (2018), 160–189.

Muthu

B.A.

, Sivaparthipan

C.B.

, Manogaran

, Sundarasekar

, Kadry

, Shanthini

and Dasel

, IOT based wearable sensor for diseases prediction and symptom analysis in healthcare sector [J], Peer-to-Peer Networking and Applications, 1–12.

Bennour

, Teaching Practices and Student Action in Physical Education Classes: Perspectives for Teacher Education[J], Creative Education 6(10) (2015), 934–944.

Bulat

, Can

N.K.

, Arslan

Y.Z.

, et al., Musculoskeletal Simulation Tools for Understanding Mechanisms of Lower-Limb Sports Injuries[J], Current Sports Medicine Reports 18(6) (2019), 210–216.

Cutforth

and Belansky

E.S.

, A Community-Engaged Approach to Translating Research into Practice: A Physical Education Story[J], Progress in Community Health Partnerships Research Education & Action 9(4) (2015), 571–582.

Niaz

, Khalid

, Ullah

, Aslam

, Raza

and Priyan

M.K.

, A bonded channel in cognitive wireless body area network based on IEEE 802.15.6 and internet of things [J], Computer Communications 150 (2020), 131–143.

Glazier

P.S.

and Mehdizadeh

, Challenging conventional paradigms in applied sports biomechanics research[J], Sports Medicine 49(2) (2019), 171–176.

Hwang

E.R.

and Kim

T.Y.

, Intensification of the education of public health, hygiene, and martial arts during the Japanese colonial period (1937–1945)[J], Journal of Exercise Rehabilitation 14(2) (2018), 160–167.

Ivashchenko

O.V.

and Kapkan

O.O.

, Simulation of process of 14–15 years old girls’ training of light athletic and gymnastic exercises[J], Pedagogics, psychology, medicalbiological problems of physical training and sports (8) (2015), 32–40.

10.

Khudolii

O.M.

, Ivashchenko

O.V.

, Iermakov

S.S.

, et al., Computer simulation of junior gymnasts’ training process[J], Science of Gymnastics Journal 8(3) (2016), 215–228.

11.

Lopatiev

, Ivashchenko

, Khudolii

, et al., Systemic approach and mathematical modeling in physical education and sports[J], Journal of Physical Education and Sport (JPES) 17(1) (2017), 146–155.

12.

Elhoseny

, Shehab

and Yuan

, Optimizing Robot Path in Dynamic Environments Using Genetic Algorithm and Bezier Curve [J], Journal of Intelligent & Fuzzy Systems 33(4) (2017), 2305–2316.

13.

Mullen

, Highton

and Twist

, The internal and external responses to a forward-specific rugby league simulation protocol performed with and without physical contact[J], International Journal of Sports Physiology and Performance 10(6) (2015), 746–753.

14.

Mulligan

, Lohse

K.R.

and Hodges

N.J.

, An action-incongruent secondary task modulates prediction accuracy in experienced performers: evidence for motor simulation[J], Psychological Research 80(4) (2016), 496–509.

15.

Murase

, Yokoyama

, Fujii

, et al., Baseball Catching Patterns Differ According to Task Constraints[J], Advances in Physical Education 06(3) (2016), 151–157.

16.

Owen

P.D.

and King

, Competitive balance measures in sports leagues: The effects of variation in season length[J], Economic Inquiry 53(1) (2015), 731–744.

17.

Payne

, Mitchell

, Bibb

, et al., The evaluation of new multi-material human soft tissue simulants for sports impact surrogates[J], Journal of the Mechanical Behavior of Biomedical Materials 41 (2015), 336–356.

18.

Prathik

, Anuradha

and Uma

, Prediction of Suitability of Soil for Different Crops using Spatial Data Mining [J], International Journal of Engineering and Advanced Technology 9(1) (2019), 2330–2337.

19.

Rein

and Memmert

, Big data and tactical analysis in elite soccer: future challenges and opportunities for sports science[J], SpringerPlus 5(1) (2016), 1–13.

20.

Ridderinkhof

K.R.

and Brass

, How kinesthetic motor imagery works: a predictive-processing theory of visualization in sports and motor expertise[J], Journal of Physiology-Paris 109(1-3) (2015), 53–63.

21.

Seo

S.W.

, Yun

, Kim

M.G.

, et al., Screen-Based Sports Simulation Using Acoustic Source Localization[J], Applied Sciences 9(15) (2019), 2970.

22.

Elandoulsi

SghaierDS.

, Mami

, et al., Physical Education Teacher’s Training in Swimming under the Joint Didactic Action[J], Creative Education 6(22) (2015), 2433–2437.

23.

de Souza

Í.M.M.

and Silva

C.L.D.

, Body Practices and Brazilian Culture: Pedagogical Contributions to Physical Education Professionals[J], Procedia Social & Behavioral Sciences 174 (2015), 434–440.

24.

Tivener

K.A.

and Gloe

D.S.

, The effect of high-fidelity cardiopulmonary resuscitation (CPR) simulation on athletic training student knowledge, confidence, emotions, and experiences[J], Athletic Training Education Journal 10(2) (2015), 103–112.

25.

Yang

, The Simulation of Table Tennis during the Course of Sports[J], Caribbean Journal of Science 52(4) (2019), 1561–1564.