Outdoor sports teaching method, device and equipment and storage medium

Abstract

The invention relates to the technical field of video processing, in particular to an outdoor sports teaching method, device and equipment and a storage medium, and the method comprises the steps: inputting initial multi-modal teaching data into a reinforcement learning task model for processing, and obtaining a target teaching video; comparing the target teaching video with the student action video to obtain a comparison result, and generating a teaching suggestion based on the comparison result; the training of the reinforcement learning task model comprises the following steps: training a basic learning cognitive model based on initial multi-modal teaching data to obtain a multi-modal closed-loop simulation generation model; processing the initial multi-modal teaching data based on a multi-modal closed-loop simulation generation model to obtain a to-be-compared teaching video; comparing the to-be-compared teaching video with the model teaching video to determine a target point location action video segment in the model teaching video; and training a multi-modal closed-loop simulation generation model based on the target point location action video segment and the initial multi-modal teaching data to obtain a reinforcement learning task model. And the accuracy of video output and teaching suggestions of the model can be improved.

Description

Technical Field

[0001] This application relates to the field of video processing technology, and in particular to an outdoor sports teaching method, apparatus, equipment and storage medium. Background Technology

[0002] With the continuous development of intelligent technology in outdoor sports, outdoor sports assistive robots have been deeply involved in outdoor sports teaching activities; for example, golf caddy robots can help students learn the correct hitting motion.

[0003] The outdoor sports assistive robot has a pre-set teaching video generation model. The teaching video generated by the model contains standard movements. This teaching video can not only be used by students to learn, but can also be compared with the students' actual practice videos to identify the movements that the students have not mastered and to provide teaching suggestions.

[0004] However, the aforementioned instructional video generation model is generally trained using only athlete motion data. Due to the difficulty in obtaining such data, the amount of data is low. This results in poor accuracy of the video output of the instructional video generation model, which may lead to substandard movements in the generated instructional videos. Consequently, the accuracy of subsequent recognition of students' substandard movements is low, and the accuracy of the instructional suggestions given is also poor. Summary of the Invention

[0005] To improve the accuracy of video output from the teaching video generation model and the accuracy of teaching suggestions provided by the outdoor sports assistive robot, this application provides an outdoor sports teaching method, device, equipment, and storage medium.

[0006] Firstly, this application provides an outdoor sports teaching method, including:

[0007] The initial multimodal teaching data is input into the reinforcement learning task model for processing to obtain the target teaching video;

[0008] The target teaching video is compared with the student's action video to obtain the comparison results, and teaching suggestions are generated based on the comparison results;

[0009] The training steps of the reinforcement learning task model include:

[0010] Based on the initial multimodal teaching data, a basic learning cognitive model is trained to obtain a multimodal closed-loop simulation generation model. The initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0011] The initial multimodal teaching data is processed based on the multimodal closed-loop simulation generation model to obtain the teaching videos to be compared.

[0012] By comparing the teaching video to be compared with the model teaching video, the target point action video segment in the model teaching video is determined;

[0013] Based on the target location action video segment and the initial multimodal teaching data, the multimodal closed-loop simulation generation model is trained to obtain the reinforcement learning task model.

[0014] Secondly, this application provides an outdoor sports teaching device, comprising:

[0015] The video generation module is used to input the initial multimodal teaching data into the reinforcement learning task model for processing, and obtain the target teaching video;

[0016] The module is recommended to compare the target teaching video with the student action video to obtain comparison results, and generate teaching suggestions based on the comparison results.

[0017] The training steps of the reinforcement learning task model include:

[0018] Based on the initial multimodal teaching data, a basic learning cognitive model is trained to obtain a multimodal closed-loop simulation generation model. The initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0019] The initial multimodal teaching data is processed based on the multimodal closed-loop simulation generation model to obtain the teaching videos to be compared.

[0020] By comparing the teaching video to be compared with the model teaching video, the target point action video segment in the model teaching video is determined;

[0021] Based on the target location action video segment and the initial multimodal teaching data, the multimodal closed-loop simulation generation model is trained to obtain the reinforcement learning task model.

[0022] Thirdly, this application provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the method described above.

[0023] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method.

[0024] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0025] The aforementioned outdoor sports teaching method, apparatus, equipment, and storage medium, through inputting initial multimodal teaching data into a reinforcement learning task model for processing, obtains a target teaching video; compares the target teaching video with student action videos to obtain comparison results, and generates teaching suggestions based on the comparison results; wherein, the training steps of the reinforcement learning task model include: training a basic learning cognitive model based on the initial multimodal teaching data to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images; processing the initial multimodal teaching data based on the multimodal closed-loop simulation generation model to obtain a teaching video to be compared; comparing the teaching video to be compared with a model teaching video to determine the target point action video segment in the model teaching video; and training the multimodal closed-loop simulation generation model based on the target point action video segment and the initial multimodal teaching data to obtain a reinforcement learning task model. Through the above implementation, compared with the existing technology that only uses the limited amount of real athlete motion data that is difficult to collect to train the teaching video generation model, firstly, this embodiment uses easily collected multimodal teaching data with a large amount of data to train the teaching video generation model, so as to initially achieve the goal of improving the accuracy of the model's video output; secondly, based on the first step, this embodiment further identifies the target point motion video segments corresponding to the video segments in the teaching videos to be compared where the movements are not in place from the model teaching videos, and then focuses on using the target point motion video segments to continue to train the teaching video output model a second time, thereby further improving the accuracy of the model's video output and the accuracy of the teaching suggestions given by the outdoor sports assistive robot.

[0026] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of an outdoor sports teaching method provided in the embodiments of this application;

[0029] Figure 2 This is a schematic diagram illustrating the secondary training process of a video generation model provided in this application embodiment;

[0030] Figure 3 This is a schematic diagram of the structure of an outdoor sports teaching device provided in the embodiments of this application;

[0031] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application;

[0032] Figure 5 This is an internal structural diagram of a computer-readable storage medium provided in an embodiment of this application. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this disclosure.

[0034] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings herein are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0035] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0036] Example 1

[0037] Figure 1 This is a flowchart of an outdoor sports teaching method provided in Embodiment 1 of this application, for reference. Figure 1 The method can be executed by a device that performs the method, which can be implemented in software and / or hardware, and the method includes:

[0038] S110. Input the initial multimodal teaching data into the reinforcement learning task model for processing to obtain the target teaching video.

[0039] It's worth noting that as people's pursuit of quality of life continues to improve, outdoor sports are becoming increasingly popular, such as golf, baseball, skiing, and ice skating. Beginners need to learn and master the basic movements and positions in these sports during the initial learning phase. Taking golf as an example, the basic movements include stance, body rotation, swing, follow-through, and putting. To help beginners quickly master these basic movements, outdoor sports often provide participants with corresponding outdoor sports assistance robots. These robots not only carry sports equipment but also provide instructional guidance. For instance, in golf, a caddie robot can carry equipment and provide instructional guidance.

[0040] To provide instruction to trainees, the outdoor sports assistive robot is equipped with a reinforcement learning task model. This model has been trained and has achieved good video output accuracy. The reinforcement learning task model is used to output target teaching videos, which contain one or more standard motion point teaching videos. These motion point teaching videos can be used as a reference for trainees to improve their accuracy in performing motion points.

[0041] It should be noted that, in order to improve the accuracy of the video output of the reinforcement learning task model, this embodiment pre-sets initial multimodal teaching data. The initial multimodal teaching data is used to input into the reinforcement learning task model for processing to generate the target teaching video. The initial multimodal teaching data consists of teaching data of various different data types, such as text teaching data, voice teaching data, video teaching data, and image teaching data.

[0042] Among them, since the initial multimodal teaching data is multimodal data, it has a rich variety of data types and a high amount of data, which makes it easier to effectively improve the accuracy of the target teaching video output. That is, the various action points in the target teaching video have a high degree of standard, such as the swing height can reach the standard height.

[0043] S120. Compare the target teaching video with the student action video to obtain the comparison result, and generate teaching suggestions based on the comparison result.

[0044] It should be noted that the outdoor sports assistive robot is equipped with a camera, which is used to capture video of the student learning the movement points, and the video is recorded as the student's movement video.

[0045] In this case, some movement points in the student's action video may not meet the standards, while the target teaching video output by the reinforcement learning task model has a high standard for each movement point. Therefore, by comparing the student's action video with the target teaching video, the analysis model in the outdoor sports assistive robot can further analyze the student's current non-compliant movement points and the reasons for non-compliance, and provide corresponding teaching suggestions. These teaching suggestions can be displayed to the student through the display screen and text on the outdoor sports assistive robot, or through the speaker on the outdoor sports assistive robot to display the student's voice.

[0046] Among them, see Figure 2 The training steps for the reinforcement learning task model include:

[0047] A110. Based on the initial multimodal teaching data, train the basic learning cognitive model to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0048] It should be noted that, in traditional solutions, video generation models are trained using only real-person athlete motion data (training data), which is difficult to collect. Due to the low amount of real-person athlete motion data, the accuracy of the video output from the trained video generation model is often poor, and many motion points in the output video do not meet the standards.

[0049] In this embodiment, the first measure taken to improve the accuracy of the video output of the video generation model is to significantly increase the amount of training data. In this embodiment, the untrained video generation model is referred to as the basic learning cognitive model. Regarding the amount of training data for the basic learning cognitive model, this embodiment uses easily collected multimodal data to train the basic learning cognitive model, and this multimodal data is referred to as the initial multimodal teaching data.

[0050] In this embodiment, the initial multimodal teaching data specifically includes: teaching text, teaching audio data, teaching video, and teaching images; wherein, the teaching text is text data, specifically text describing the standard requirements for each movement point; the teaching audio data is audio data, specifically audio data of open-source or authorized live instructors providing movement point guidance to live students; the teaching video is video data, specifically video data of open-source or authorized live instructors providing movement point guidance to live students; the teaching images are image data, specifically images of open-source or authorized live instructors providing key movement point guidance to live students; wherein, the teaching images can also be directly extracted from the keyframes of the teaching video, and there is no specific limitation; in other embodiments, the specific types of data contained in the initial multimodal teaching data are not specifically limited.

[0051] Specifically, the basic learning cognitive model can be iteratively trained using the initial multimodal teaching data mentioned above, thereby obtaining a preliminarily trained video generation model, denoted as the multimodal closed-loop simulation generation model.

[0052] It should be noted that, compared to the training data in traditional solutions, firstly, the initial multimodal teaching data significantly increases the variety of training data, which helps improve the accuracy of the video output of the video generation model; secondly, the modal data (teaching text, teaching audio data, teaching video, and teaching images) in the initial multimodal teaching data are all easily accessible and can be acquired in large quantities to significantly increase the amount of training data, which facilitates further improvement in the accuracy of the video output of the video generation model.

[0053] A120. Based on the multimodal closed-loop simulation generation model, process the initial multimodal teaching data to obtain the teaching videos to be compared.

[0054] It should be noted that the initial multimodal teaching data can be used not only as training data, but also as input data for the model after the model has been trained.

[0055] Specifically, the initial multimodal teaching data is input into the multimodal closed-loop simulation generation model for processing, which yields teaching videos of the action points, denoted as the teaching videos to be compared. It should be noted that since the multimodal closed-loop simulation generation model is a preliminarily trained model, although the accuracy of its video output has significantly improved compared to the untrained version, the current video output accuracy may not yet meet the standard for use. Therefore, further secondary training of the multimodal closed-loop simulation generation model is needed to further improve the model's video output accuracy.

[0056] A130. Compare the teaching video to be compared with the model teaching video to determine the target point action video segment in the model teaching video.

[0057] It should be noted that the current video output accuracy of the multimodal closed-loop simulation generation model may not yet meet the standard for use, meaning that some motion points in the teaching videos to be compared may not meet the standard. If videos corresponding to the non-compliant motion points can be obtained, and the multimodal closed-loop simulation generation model is further trained using these videos, then the corresponding motion points in the subsequent teaching videos output by the multimodal closed-loop simulation generation model can be made to meet the standard as much as possible.

[0058] To obtain videos corresponding to the substandard movement points, this embodiment pre-sets a model teaching video. This model teaching video is an instructional video taught by an authorized professional coach and / or top athlete to teach various movement points in a certain outdoor sport (such as golf). All movement points in the model teaching video meet the standards. If the teaching video to be compared, generated by the multimodal closed-loop simulation model, is compared with the model teaching video, the substandard movement points in the teaching video to be compared can be identified. Furthermore, the video segment corresponding to the substandard movement point in the model teaching video can be obtained, and this video segment is recorded as the target movement video segment. The movement points shown in the target movement video segment meet the standards.

[0059] A140. Based on the target point action video segment and the initial multimodal teaching data, train the multimodal closed-loop simulation generation model to obtain the reinforcement learning task model.

[0060] It should be noted that, since the action points corresponding to the target point action video segments in the teaching videos to be compared output by the multimodal closed-loop simulation generation model are not up to standard, while the action points shown in the target point action video segments are up to standard, if the target point action video segments are included in the training process of the multimodal closed-loop simulation generation model, the accuracy of the video output of the video generation model can be effectively improved.

[0061] Specifically, the target point action video segments and the initial multimodal teaching data are used together as training data to train the multimodal closed-loop simulation generation model, and the trained model is denoted as the reinforcement learning task model.

[0062] It should be noted that the initial multimodal teaching data is mainly used to ensure the richness and sufficient amount of training data, while the target point motion video segments are mainly focused on making the multimodal closed-loop simulation generation model output teaching videos that meet the standards for each motion point. These teaching videos include motion points that were originally not up to standard but are now up to standard. The new model obtained after the multimodal closed-loop simulation generation model has been trained is denoted as the reinforcement learning task model.

[0063] It should also be noted that this embodiment processes initial multimodal teaching data into a reinforcement learning task model to obtain a target teaching video; compares the target teaching video with student action videos to obtain comparison results, and generates teaching suggestions based on the comparison results; wherein, the training steps of the reinforcement learning task model include: training a basic learning cognitive model based on the initial multimodal teaching data to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images; processing the initial multimodal teaching data based on the multimodal closed-loop simulation generation model to obtain a teaching video to be compared; comparing the teaching video to be compared with a model teaching video to determine the target point action video segment in the model teaching video; and training the multimodal closed-loop simulation generation model based on the target point action video segment and the initial multimodal teaching data to obtain a reinforcement learning task model. Through the above implementation, compared with the existing technology that only uses the limited amount of real athlete motion data that is difficult to collect to train the teaching video generation model, firstly, this embodiment uses easily collected multimodal teaching data with a large amount of data to train the teaching video generation model, so as to initially achieve the goal of improving the accuracy of the model's video output; secondly, based on the first step, this embodiment further identifies the target point motion video segments corresponding to the video segments in the teaching videos to be compared where the movements are not in place from the model teaching videos, and then focuses on using the target point motion video segments to continue to train the teaching video output model a second time, thereby further improving the accuracy of the model's video output and the accuracy of the teaching suggestions given by the outdoor sports assistive robot.

[0064] Example 2

[0065] This application provides a second embodiment of an outdoor sports teaching method, which optimizes the step in the first embodiment of "comparing the teaching video to be compared with the model teaching video to determine the target point action video segment in the model teaching video". It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments. The method includes:

[0066] S210. Input the initial multimodal teaching data into the reinforcement learning task model for processing to obtain the target teaching video.

[0067] S220. Compare the target teaching video with the student action video to obtain the comparison result, and generate teaching suggestions based on the comparison result.

[0068] The training steps of the reinforcement learning task model include:

[0069] A210. Based on the initial multimodal teaching data, train the basic learning cognitive model to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0070] A220. Based on the multimodal closed-loop simulation generation model, process the initial multimodal teaching data to obtain the teaching videos to be compared.

[0071] A231. Extract the video segments of the points to be compared from the teaching videos to be compared, and extract the video segments of the model points from the model teaching videos, wherein the video segments of the points to be compared and the video segments of the model points correspond to the same action points.

[0072] It should be noted that an outdoor sport generally includes multiple action points. Taking golf as an example, the action points it includes at least are: stance, body rotation, swing, putting, and follow-through.

[0073] The instructional videos to be compared are outdoor sports instructional videos generated by a multimodal closed-loop simulation generation model. Each instructional video consists of video segments corresponding to different action points. Taking golf as an example, the instructional videos to be compared consist of standing posture video segments, body rotation video segments, swing video segments, putting video segments, and follow-through video segments. It should be noted that, because the video output accuracy of the multimodal closed-loop simulation generation model may not meet the requirements, the action points shown in some video segments corresponding to certain action points in the instructional videos to be compared may not meet the standards, such as the swing height not reaching the standard height.

[0074] To determine the video segments corresponding to the substandard movement points in the instructional videos to be compared, this embodiment also includes a pre-set model instructional video, which is an open-source and / or licensed video of professional athletes / coaches teaching outdoor sports (such as golf, skiing, etc.). By comparing the video segments corresponding to the same movement points in the instructional videos to be compared with those in the model instructional videos, the video segments corresponding to the substandard movement points in the instructional videos to be compared can be determined. To this end, video segments corresponding to the same movement points can be extracted from both the instructional videos to be compared and the model instructional videos. The video segments extracted from the instructional videos to be compared are recorded as the video segments of the movement points to be compared, and the video segments extracted from the model instructional videos are recorded as the video segments of the movement points to be compared.

[0075] Specifically, this embodiment pre-sets a deep learning model for extracting video segments. For example, the deep learning model is a trained action classification model, such as TSN, SlowFast, etc., without any specific limitation. The teaching video to be compared and the model teaching video are respectively input into the action classification model for processing, thereby obtaining the video segments of the points to be compared corresponding to each action point in the teaching video to be compared, and the video segments of the model points corresponding to each action point in the model teaching video.

[0076] A232. Compare the video segment of the point to be compared with the corresponding video segment of the model point to determine the target point action video segment in the model teaching video.

[0077] In this case, taking a set of video segments with the same action points to be compared and a video segment with the model action points as an example, the degree of difference between the action points in the two video segments is first compared. If the degree of difference is large, it means that the action points corresponding to the video segments to be compared do not meet the standard.

[0078] Specifically, taking a set of video segments with the same action points as an example, the video segments with the same action points to be compared and the video segments with the same action points are first processed by dynamic time adjustment or linear interpolation to ensure that they have the same number of video frames. Next, in both video segments, the coordinates of all key points are converted to relative coordinates with the center of the human hip as the origin, thereby eliminating the influence of the person's position in the frame. Furthermore, scaling is performed according to the length of the human torso or the standard deviation of all key points to eliminate the influence of height and shooting distance, thus obtaining the key point sequence corresponding to each video frame in the two video segments. Then, the pose distance between all video frames corresponding to the same time in the two video segments is calculated. This pose distance can be the average or sum of the Euclidean distances between all corresponding key points in the two video frames. Finally, the pose distances corresponding to each time in the video segment are summed to obtain the video segment distance.

[0079] If the distance of the video segment is greater than the preset distance threshold, it is determined that the difference between the video segment of the point to be compared and the video segment of the model point is large. In this case, the video segment of the model point is taken as the action video segment of the target point.

[0080] It should be noted that the action points corresponding to the target point action video segment itself meet the standards, while the action points corresponding to the comparison point video segment do not meet the standards. The comparison point video segment was generated by a multimodal closed-loop simulation generation model. If the target point action video segment corresponding to the comparison point video segment is subsequently added to the training process of the multimodal closed-loop simulation generation model, the accuracy of the multimodal closed-loop simulation generation model's video output can be effectively improved, making the action points corresponding to the video segments output by the multimodal closed-loop simulation generation model after training more standard.

[0081] A240. Based on the target point action video segment and the initial multimodal teaching data, train the multimodal closed-loop simulation generation model to obtain the reinforcement learning task model.

[0082] Example 3

[0083] This application provides a third embodiment of an outdoor sports teaching method, which optimizes the "comparing the video segment of the point to be compared with the corresponding video segment of the model point to determine the target point action video segment in the model teaching video" in the second embodiment. It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments. The method includes:

[0084] S310. Input the initial multimodal teaching data into the reinforcement learning task model for processing to obtain the target teaching video.

[0085] S320. Compare the target teaching video with the student action video to obtain the comparison result, and generate teaching suggestions based on the comparison result.

[0086] The training steps of the reinforcement learning task model include:

[0087] A310. Based on the initial multimodal teaching data, train the basic learning cognitive model to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0088] A320. Based on the multimodal closed-loop simulation generation model, process the initial multimodal teaching data to obtain the teaching videos to be compared.

[0089] A331. Extract the video segments of the points to be compared from the teaching videos to be compared, and extract the video segments of the model points from the model teaching videos, wherein the video segments of the points to be compared and the video segments of the model points correspond to the same action points.

[0090] A332A. Calculate the degree of video difference between the video segment of the point to be compared and the corresponding video segment of the model point.

[0091] In addition to the method shown in step A232 of Example 2, the degree of video difference can also be calculated using deep learning.

[0092] Specifically, the video segments of the points to be compared and the video segments of the model points corresponding to the same action are first input into a preset pose sequence extraction model for processing, to obtain the pose sequences corresponding to the two video segments, denoted as the first pose sequence and the second pose sequence; for example, the pose sequence extraction model is OpenPose, MediaPipe Pose, etc., and there is no specific limitation; then, through the processing in a preset feature vector calculation model, 512-dimensional feature vectors corresponding to the two pose sequences are obtained, denoted as the first feature vector and the second feature vector, where the feature vector is used to reflect the "fingerprint" or "embedding" of the action, for example, the feature vector calculation model is a pre-trained CRT-GCN model; finally, the cosine similarity or Euclidean distance between the first feature vector and the second feature vector is calculated as the degree of video difference.

[0093] In other embodiments, the specific method for calculating the degree of video difference is not limited.

[0094] A332B. In response to the video difference being greater than a preset difference threshold, the model point video segment corresponding to the video segment of the point to be compared is taken as the target point action video segment.

[0095] In this embodiment, a threshold for the degree of difference is preset to measure the magnitude of the video difference. If the degree of difference in the video is greater than the threshold, it is determined that the degree of difference in the video is too large. At this time, the model point video segment corresponding to the video segment of the point to be compared is taken as the target point action video segment.

[0096] It should be noted that the target point action video segment itself corresponds to the standard action point. If the target point action video segment is subsequently added to the training process of the multimodal closed-loop simulation generation model, the accuracy of the multimodal closed-loop simulation generation model's video output can be effectively improved, making the action points corresponding to the video segments output by the multimodal closed-loop simulation generation model after training more standard.

[0097] A340. Based on the target point action video segment and the initial multimodal teaching data, train the multimodal closed-loop simulation generation model to obtain the reinforcement learning task model.

[0098] Example 4

[0099] This application provides a fourth embodiment of an outdoor sports teaching method, which optimizes the "calculation of the video difference between the video segment of the point to be compared and the corresponding model point video segment" in the third embodiment. It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments. The method includes:

[0100] S410. Input the initial multimodal teaching data into the reinforcement learning task model for processing to obtain the target teaching video.

[0101] S420. Compare the target teaching video with the student action video to obtain the comparison result, and generate teaching suggestions based on the comparison result.

[0102] The training steps of the reinforcement learning task model include:

[0103] A410. Based on the initial multimodal teaching data, train the basic learning cognitive model to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0104] A420. Based on the multimodal closed-loop simulation generation model, process the initial multimodal teaching data to obtain the teaching videos to be compared.

[0105] A431. Extract the video segments of the points to be compared from the teaching videos to be compared, and extract the video segments of the model points from the model teaching videos, wherein the video segments of the points to be compared and the video segments of the model points correspond to the same action points.

[0106] A432A1. Perform feature extraction on the video segment of the point to be compared to obtain a first feature matrix; perform feature extraction on the video segment of the model point corresponding to the video segment of the point to be compared to obtain a second feature matrix.

[0107] In this embodiment, a feature extraction model is pre-defined. For example, the feature extraction model can be SlowFast, X3D model, etc., and the specific model is not limited. This model can effectively capture the dynamic information of the video in the spatiotemporal dimension, rather than just the static information of a single frame image. Through this feature extraction model, features can be extracted from the video segment of the point to be compared and the video segment of the model point, respectively, and a first feature matrix corresponding to the video segment of the point to be compared and a second feature matrix corresponding to the video segment of the model point can be obtained.

[0108] A432A2. Determine the eigenvector distribution corresponding to the first feature matrix to obtain the first eigenvector distribution; determine the eigenvector distribution corresponding to the second feature matrix to obtain the second eigenvector distribution.

[0109] It should be noted that the above feature matrices generally follow a Gaussian distribution, and the Gaussian distribution corresponding to the first feature matrix is ​​denoted as the first eigenvector distribution N(μ_g,Σ_g), and the Gaussian distribution corresponding to the second feature matrix is ​​denoted as the second eigenvector distribution N(μ_r,Σ_r).

[0110] Where μ_g is the mean vector (4096-dimensional) of the video segment to be compared, and μ_r is the mean vector (4096-dimensional) of the video segment of the model point; Σ_g is the covariance matrix (4096×4096-dimensional) of the video segment to be compared, and Σ_r is the covariance matrix (4096×4096-dimensional) of the video segment of the model point.

[0111] A432A3. Calculate the degree of video difference based on the first feature vector distribution and the second feature vector distribution.

[0112] It should be noted that the video difference degree D can be calculated using the first and second feature vector distributions described above. The formula for calculating the video difference degree D is as follows:

[0113] D = ||μ_r - μ_g||² + Tr(Σ_r + Σ_g - 2(Σ_r·Σ_g)^(1 / 2));

[0114] Where, ||μ_r - μ_g||² is the square of the Euclidean distance between the two mean vectors; Tr(Σ_r +Σ_g - 2(Σ_r·Σ_g)^(1 / 2)) is the trace of the matrix (the sum of the diagonal elements), used to measure the similarity in shape between the two distributions; (Σ_r·Σ_g)^(1 / 2) is the geometric mean of the two covariance matrices.

[0115] A432B, In response to the video difference being greater than a preset difference threshold, the model point video segment corresponding to the video segment of the point to be compared is taken as the target point action video segment.

[0116] A440. Based on the target point action video segment and the initial multimodal teaching data, train the multimodal closed-loop simulation generation model to obtain the reinforcement learning task model.

[0117] Example 5

[0118] This application provides a fifth embodiment of an outdoor sports teaching method, which optimizes the "calculation of the video difference between the video segment of the point to be compared and the corresponding model point video segment" in the third embodiment. It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments. The method includes:

[0119] S510. Input the initial multimodal teaching data into the reinforcement learning task model for processing to obtain the target teaching video.

[0120] S520. Compare the target teaching video with the student action video to obtain the comparison result, and generate teaching suggestions based on the comparison result.

[0121] The training steps of the reinforcement learning task model include:

[0122] A510. Based on the initial multimodal teaching data, train the basic learning cognitive model to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0123] A520. Based on the multimodal closed-loop simulation generation model, process the initial multimodal teaching data to obtain the teaching videos to be compared.

[0124] A531. Extract the video segments of the points to be compared from the teaching videos to be compared, and extract the video segments of the model points from the model teaching videos, wherein the video segments of the points to be compared and the video segments of the model points correspond to the same action points.

[0125] A532A1. Calculate the structural similarity index between the video segment of the point to be compared and the corresponding video segment of the model point.

[0126] It should be noted that Example 4 determines the degree of difference between two video segments based on their distribution. This example aims to measure the degree of difference between two video segments based on their differences in brightness, contrast, and structure.

[0127] Specifically, this embodiment includes a pre-set structural similarity index calculation algorithm. This algorithm is used to process the two video segments separately to obtain the structural similarity index between the two video segments. The structural similarity index is used to measure the degree of difference in brightness, contrast, and structure between the corresponding video frames in the two video segments.

[0128] A532A2. Based on the structural similarity index, the degree of video difference is obtained.

[0129] The degree of video difference is the sum of the structural similarity indices of the corresponding video frames in each group of the two video segments.

[0130] A532B, In response to the video difference being greater than a preset difference threshold, the model point video segment corresponding to the video segment of the point to be compared is taken as the target point action video segment.

[0131] A540. Based on the target point action video segment and the initial multimodal teaching data, train the multimodal closed-loop simulation generation model to obtain the reinforcement learning task model.

[0132] Example 6

[0133] This application provides a method for teaching outdoor sports, which optimizes the "comparing the target teaching video with the student's action video to obtain a comparison result, and generating teaching suggestions based on the comparison result" in embodiment one. It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments. The method includes:

[0134] S610. Input the initial multimodal teaching data into the reinforcement learning task model for processing to obtain the target teaching video.

[0135] S621. Extract motion points from the target teaching video to obtain a target motion point set; extract motion points from the student's motion video to obtain a student motion point set.

[0136] It should be noted that the target instructional video consists of multiple video frames, each of which can show the posture of the corresponding action point; when the outdoor sports assistive robot is practicing the action points, it can use its own camera to collect the student's action video. Similarly, the student's action video also consists of multiple video frames, each of which can show the posture of the student's action point.

[0137] The preset motion point extraction model can process each video frame in the target teaching video to obtain the motion points of the posture in each video frame, and the set of motion points corresponding to each video frame is recorded as the target motion point set; similarly, the motion point extraction model can also process each video frame in the student's action video to obtain the motion points of the posture in each video frame, and the set of motion points corresponding to each video frame is recorded as the student motion point set.

[0138] S622. Calculate the overlap of the target action point set and the student action point set in the preset spatial coordinate system as the comparison result.

[0139] It should be noted that the target instructional video is generated by the reinforcement learning task model that has completed the secondary training process. All action points in the target instructional video are standard action points, and the positions of the action points corresponding to the target instructional video are standard positions. If the positions of the action points in the student action point set corresponding to the student action video have a high degree of overlap with the positions of the corresponding action points in the target action point set, it indicates that the student's action point execution is relatively standard. Otherwise, it indicates that the action point execution is not up to standard.

[0140] It should also be noted that, since the positions of the characters in the target teaching video and the student action video may be different, in order to prevent the position from adversely affecting the calculation of overlap, after obtaining the target action point set and the student action point set, the target action point set and the student action point set are respectively transformed into a preset spatial coordinate system to obtain a first point set and a second point set; then the overlap of action points in the first point set and the second point set is calculated as the comparison result; wherein, the overlap is used to reflect the distance between any two corresponding action points in the first point set and the second point set.

[0141] S623. Analyze the comparison results and generate teaching suggestions.

[0142] The comparison results are used to reflect the standard of each movement point of the trainees, and can also be used to generate corresponding teaching suggestions.

[0143] Specifically, this embodiment has a preset language model. By inputting the above comparison results and preset prompts into the language model, the language model can generate corresponding teaching suggestions. For example, the prompt is "Please analyze the comparison results and give corresponding teaching suggestions". Taking golf as an example, the teaching suggestions output by the language model can be "Your swing height still needs to be improved. Keep up the good work!"

[0144] The training steps of the reinforcement learning task model include:

[0145] A610. Based on the initial multimodal teaching data, train the basic learning cognitive model to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0146] A620. Based on the multimodal closed-loop simulation generation model, process the initial multimodal teaching data to obtain the teaching videos to be compared.

[0147] A630. Compare the teaching video to be compared with the model teaching video to determine the target point action video segment in the model teaching video.

[0148] A640. Based on the target point action video segment and the initial multimodal teaching data, train the multimodal closed-loop simulation generation model to obtain the reinforcement learning task model.

[0149] Example 7

[0150] This application provides a method for teaching outdoor sports, which optimizes the "analysis of the comparison results and generation of teaching suggestions" in embodiment six. It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments. The method includes:

[0151] S710. Input the initial multimodal teaching data into the reinforcement learning task model for processing to obtain the target teaching video.

[0152] S721. Extract motion points from the target teaching video to obtain a target motion point set; extract motion points from the student motion video to obtain a student motion point set.

[0153] S722. Calculate the overlap between the target action point set and the student action point set in a preset spatial coordinate system, and use this as the comparison result. The comparison result includes sub-comparison results corresponding to different motion point positions.

[0154] It should be noted that outdoor sports generally consist of multiple action points, and each action point has a corresponding set of sub-points in both the target action point set and the trainee action point set. In this embodiment, when comparing the point sets, different action points correspond to two sub-point sets in both the target action point set and the trainee action point set, and are compared separately. That is, the overlap of action points in the two sub-point sets corresponding to different action points is calculated separately, which serves as the comparison sub-results for different action points.

[0155] S723A. Analyze the results of the different comparison sub-results to obtain the teaching sub-suggestions corresponding to each comparison sub-result.

[0156] For example, taking the comparison sub-result (action point overlap) corresponding to a certain action point in outdoor sports, by inputting the comparison sub-result and the preset prompt words into a preset language model for processing, teaching sub-suggestions corresponding to the comparison sub-result can be obtained; for example, the teaching sub-suggestion is "Your swing height still needs to be improved, keep up the good work!"

[0157] S723B. Based on the aforementioned teaching sub-suggestions, teaching suggestions are obtained.

[0158] The teaching suggestions consist of sub-suggestions for different action points.

[0159] The training steps of the reinforcement learning task model include:

[0160] A710. Based on the initial multimodal teaching data, train the basic learning cognitive model to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0161] A720. Based on the multimodal closed-loop simulation generation model, process the initial multimodal teaching data to obtain the teaching videos to be compared.

[0162] A730. Compare the teaching video to be compared with the model teaching video to determine the target point action video segment in the model teaching video.

[0163] A740. Based on the target point action video segment and the initial multimodal teaching data, train the multimodal closed-loop simulation generation model to obtain the reinforcement learning task model.

[0164] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0165] Example 8

[0166] Based on the same inventive concept, this embodiment also provides an outdoor sports teaching device for implementing the outdoor sports teaching method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the outdoor sports teaching device provided below can be found in the limitations of the outdoor sports teaching method described above, and will not be repeated here.

[0167] In this embodiment, as Figure 3 As shown, an outdoor sports teaching device is provided, comprising:

[0168] The video generation module is used to input the initial multimodal teaching data into the reinforcement learning task model for processing, and obtain the target teaching video;

[0169] The module is recommended to compare the target teaching video with the student action video to obtain comparison results, and generate teaching suggestions based on the comparison results.

[0170] The training steps of the reinforcement learning task model include:

[0171] Based on the initial multimodal teaching data, a basic learning cognitive model is trained to obtain a multimodal closed-loop simulation generation model. The initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images.

[0172] The initial multimodal teaching data is processed based on the multimodal closed-loop simulation generation model to obtain the teaching videos to be compared.

[0173] By comparing the teaching video to be compared with the model teaching video, the target point action video segment in the model teaching video is determined;

[0174] Based on the target location action video segment and the initial multimodal teaching data, the multimodal closed-loop simulation generation model is trained to obtain the reinforcement learning task model.

[0175] Each module in the aforementioned outdoor sports teaching device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0176] It should be noted that this embodiment processes initial multimodal teaching data into a reinforcement learning task model to obtain a target teaching video; compares the target teaching video with student action videos to obtain comparison results, and generates teaching suggestions based on the comparison results; wherein, the training steps of the reinforcement learning task model include: training a basic learning cognitive model based on the initial multimodal teaching data to obtain a multimodal closed-loop simulation generation model, wherein the initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images; processing the initial multimodal teaching data based on the multimodal closed-loop simulation generation model to obtain a teaching video to be compared; comparing the teaching video to be compared with a model teaching video to determine the target point action video segment in the model teaching video; and training the multimodal closed-loop simulation generation model based on the target point action video segment and the initial multimodal teaching data to obtain a reinforcement learning task model. Through the above implementation, compared with the existing technology that only uses the limited amount of real athlete motion data that is difficult to collect to train the teaching video generation model, firstly, this embodiment uses easily collected multimodal teaching data with a large amount of data to train the teaching video generation model, so as to initially achieve the goal of improving the accuracy of the model's video output; secondly, based on the first step, this embodiment further identifies the target point motion video segments corresponding to the video segments in the teaching videos to be compared where the movements are not in place from the model teaching videos, and then focuses on using the target point motion video segments to continue to train the teaching video output model a second time, thereby further improving the accuracy of the model's video output and the accuracy of the teaching suggestions given by the outdoor sports assistive robot.

[0177] In an optional embodiment, comparing the teaching video to be compared with the model teaching video to determine the target point action video segment in the model teaching video includes:

[0178] Extract the video segments of the points to be compared from the teaching videos to be compared, and extract the video segments of the points to be compared from the model teaching videos, wherein the video segments of the points to be compared and the video segments of the points to be compared correspond to the same action points;

[0179] By comparing the video segment of the point to be compared with the corresponding video segment of the model point, the target point action video segment in the model teaching video is determined.

[0180] In an optional embodiment, the step of comparing the video segment of the point to be compared with the corresponding video segment of the model point to determine the target point action video segment in the model teaching video includes:

[0181] Calculate the degree of video difference between the video segment of the point to be compared and the corresponding video segment of the model point;

[0182] In response to the video difference being greater than a preset difference threshold, the model point video segment corresponding to the video segment of the point to be compared is taken as the target point action video segment.

[0183] In an optional embodiment, calculating the degree of video difference between the video segment of the point to be compared and the corresponding video segment of the model point includes:

[0184] Feature extraction is performed on the video segment of the point to be compared to obtain a first feature matrix; feature extraction is performed on the video segment of the model point corresponding to the video segment of the point to be compared to obtain a second feature matrix;

[0185] Determine the eigenvector distribution corresponding to the first feature matrix to obtain the first eigenvector distribution; determine the eigenvector distribution corresponding to the second feature matrix to obtain the second eigenvector distribution;

[0186] The degree of video difference is calculated based on the first feature vector distribution and the second feature vector distribution.

[0187] In an optional embodiment, calculating the degree of video difference between the video segment of the point to be compared and the corresponding video segment of the model point includes:

[0188] Calculate the structural similarity index between the video segment of the point to be compared and the corresponding video segment of the model point;

[0189] Based on the structural similarity index, the degree of video difference is obtained.

[0190] In an optional embodiment, the comparison between the target instructional video and the student's action video yields a comparison result, and the generation of instructional suggestions based on the comparison result includes:

[0191] Motion points are extracted from the target teaching video to obtain a target motion point set; motion points are extracted from the student's motion video to obtain a student motion point set.

[0192] Calculate the overlap between the target action point set and the student action point set in a preset spatial coordinate system, and use this as the comparison result;

[0193] The comparison results are analyzed to generate teaching suggestions.

[0194] In an optional embodiment, the comparison results include comparison sub-results corresponding to different motion points;

[0195] Accordingly, the analysis of the comparison results to generate teaching suggestions includes:

[0196] Analyze the results of the different comparison sub-results to obtain the teaching sub-suggestions corresponding to each comparison sub-result;

[0197] Based on the aforementioned teaching sub-suggestions, teaching suggestions are derived.

[0198] Example 9

[0199] In this embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows. Figure 4 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements an outdoor sports teaching method.

[0200] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the computer device to which the present disclosure is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0201] Example 10

[0202] In this embodiment, a computer-readable storage medium is provided, such as... Figure 5 As shown, a computer program is stored thereon, and when the computer program is executed by the processor, it implements the steps in the above-described method embodiments.

[0203] Example 11

[0204] In this embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0205] It should be noted that the information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and it does not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0206] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this disclosure can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this disclosure may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this disclosure may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0207] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0208] The embodiments described above are merely illustrative of several implementations of this disclosure, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent disclosure. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this disclosure, and these all fall within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the appended claims.

Claims

1. An outdoor sports teaching method, characterized in that, include: The initial multimodal teaching data is input into the reinforcement learning task model for processing to obtain the target teaching video; The target teaching video is compared with the student's action video to obtain the comparison results, and teaching suggestions are generated based on the comparison results; The training steps of the reinforcement learning task model include: Based on the initial multimodal teaching data, a basic learning cognitive model is trained to obtain a multimodal closed-loop simulation generation model. The initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images. The initial multimodal teaching data is processed based on the multimodal closed-loop simulation generation model to obtain the teaching videos to be compared. By comparing the teaching video to be compared with the model teaching video, the target point action video segment in the model teaching video is determined; Based on the target location action video segment and the initial multimodal teaching data, the multimodal closed-loop simulation generation model is trained to obtain the reinforcement learning task model.

2. The method according to claim 1, characterized in that, The step of comparing the teaching video to be compared with the model teaching video to determine the target point action video segment in the model teaching video includes: Extract the video segments of the points to be compared from the teaching videos to be compared, and extract the video segments of the points to be compared from the model teaching videos, wherein the video segments of the points to be compared and the video segments of the points to be compared correspond to the same action points; By comparing the video segment of the point to be compared with the corresponding video segment of the model point, the target point action video segment in the model teaching video is determined.

3. The method according to claim 2, characterized in that, The step of comparing the video segment of the point to be compared with the corresponding video segment of the model point to determine the target point action video segment in the model teaching video includes: Calculate the degree of video difference between the video segment of the point to be compared and the corresponding video segment of the model point; In response to the video difference being greater than a preset difference threshold, the model point video segment corresponding to the video segment of the point to be compared is taken as the target point action video segment.

4. The method according to claim 3, characterized in that, The calculation of the video difference between the video segment of the point to be compared and the corresponding video segment of the model point includes: Feature extraction is performed on the video segment of the point to be compared to obtain a first feature matrix; feature extraction is performed on the video segment of the model point corresponding to the video segment of the point to be compared to obtain a second feature matrix; Determine the eigenvector distribution corresponding to the first feature matrix to obtain the first eigenvector distribution; determine the eigenvector distribution corresponding to the second feature matrix to obtain the second eigenvector distribution; The degree of video difference is calculated based on the first feature vector distribution and the second feature vector distribution.

5. The method according to claim 3, characterized in that, The calculation of the video difference between the video segment of the point to be compared and the corresponding video segment of the model point includes: Calculate the structural similarity index between the video segment of the point to be compared and the corresponding video segment of the model point; Based on the structural similarity index, the degree of video difference is obtained.

6. The method according to claim 1, characterized in that, The comparison between the target instructional video and the student's action video yields a comparison result, and instructional suggestions are generated based on the comparison result, including: Motion points are extracted from the target teaching video to obtain a target motion point set; motion points are extracted from the student's motion video to obtain a student motion point set. Calculate the overlap between the target action point set and the student action point set in a preset spatial coordinate system, and use this as the comparison result; The comparison results are analyzed to generate teaching suggestions.

7. The method according to claim 6, characterized in that, The comparison results include comparison sub-results corresponding to different motion points; Accordingly, the analysis of the comparison results to generate teaching suggestions includes: Analyze the results of the different comparison sub-results to obtain the teaching sub-suggestions corresponding to each comparison sub-result; Based on the aforementioned teaching sub-suggestions, teaching suggestions are derived.

8. An outdoor sports teaching device, characterized in that, The device includes: The video generation module is used to input the initial multimodal teaching data into the reinforcement learning task model for processing, and obtain the target teaching video; The module is recommended to compare the target teaching video with the student action video to obtain comparison results, and generate teaching suggestions based on the comparison results. The training steps of the reinforcement learning task model include: Based on the initial multimodal teaching data, a basic learning cognitive model is trained to obtain a multimodal closed-loop simulation generation model. The initial multimodal teaching data includes at least: teaching text, teaching audio data, teaching video, and teaching images. The initial multimodal teaching data is processed based on the multimodal closed-loop simulation generation model to obtain the teaching videos to be compared. By comparing the teaching video to be compared with the model teaching video, the target point action video segment in the model teaching video is determined; Based on the target location action video segment and the initial multimodal teaching data, the multimodal closed-loop simulation generation model is trained to obtain the reinforcement learning task model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images