An artificial intelligence-based sports teaching evaluation system

Abstract

The application discloses a sports teaching evaluation system based on artificial intelligence, and relates to the technical fields of sports teaching and artificial intelligence.The system comprises a data acquisition module, a posture estimation module, a motion template library, a template configuration interface, a motion matching and evaluation module, an error positioning and feedback generation module and an output module.The data acquisition module can acquire video streams in real time.The posture estimation module can extract a skeletal key point sequence.The motion template library can store four parts of structured data.The template configuration interface provides a graphical configuration interface for teaching personnel to add or modify motion standard templates.The motion matching and evaluation module can align the skeletal key point sequence with a template in a time axis and calculate a deviation value.The error positioning and feedback generation module can identify an error motion stage and a joint point, generate visual feedback containing a skeletal annotation image and a text correction prompt, and output feedback information.The application can extend multiple sports items without retraining a model, and the feedback is interpretable and convenient for teaching adjustment.

Description

Technical Field

[0001] This invention relates to the fields of physical education and artificial intelligence technology, and in particular to a physical education assessment system based on artificial intelligence. Background Technology

[0002] Movement assessment in physical education is crucial for evaluating students' skill levels and guiding instruction. Traditional manual assessment suffers from issues such as strong subjectivity and delayed feedback, leading to the widespread application of AI-based physical education assessment systems.

[0003] Existing technologies have disclosed several related solutions. Authorized patent CN120339004B uses YOLOv8 and OpenPose combined with an LSTM network to achieve action evaluation and real-time scoring; CN121214554A involves analysis and feedback for multi-object motion scenes; CN119559541A uses inter-frame difference and temporal convolutional networks for action recognition in the medicine ball project. In academic research, while deep learning-based human pose estimation and action quality assessment are hot topics, most rely on private datasets and have limited generalization capabilities.

[0004] However, the existing technologies mentioned above still have the following shortcomings: First, the coverage of assessment items is limited, and multiple models need to be trained and deployed separately for different sports, resulting in high expansion costs and difficulty in meeting the teaching needs of multiple sports; Second, the movement standards are fixed in the model weights, which cannot be easily adjusted by non-technical personnel, and the lack of configurable interfaces limits the flexibility of teaching; Third, the feedback information only outputs a comprehensive score or simple error prompts, which cannot accurately locate the body part, movement stage, and quantitative deviation of the error, thus reducing the value of teaching assistance.

[0005] Therefore, there is a need for an AI-based physical education teaching and assessment system that supports rapid expansion across multiple projects, has configurable action standards, and provides interpretable feedback. Summary of the Invention

[0006] The present invention aims to at least partially solve one of the technical problems in the above-mentioned technologies.

[0007] To achieve the above objectives, the first aspect of this invention proposes an artificial intelligence-based physical education teaching assessment system, comprising:

[0008] A data acquisition module is used to acquire video streams of the subject in real time; a posture estimation module, connected to the data acquisition module, is used to extract the skeletal key point sequence of the subject from the video stream; a motion template library is used to store motion standard templates for at least one sport, each motion standard template consisting of the following four parts of structured data: motion stage division: the complete motion is divided into at least two stages in chronological order; keyframe index: at least one keyframe moment is assigned to each stage, and each keyframe corresponds to a set of standard joint angle vectors; standard joint angle vectors and allowable deviation ranges: for each keyframe, for a preset set of body joint points, the standard joint angle value and the positive and negative allowable deviation values ​​of that angle are stored; temporal constraints: including at least one of the sequence constraints between stages and the time length range of each stage; a template configuration interface, providing a graphical configuration interface for receiving... The teaching staff can add or modify the action standard template. The operations include: marking keyframes from video samples, inputting or adjusting standard joint angle values ​​and their allowable deviation ranges; an action matching and evaluation module, connected to the posture estimation module and the action template library, used to align the extracted skeletal keypoint sequence with the selected action standard template along the time axis, and calculate the deviation between the actual joint angle and the standard joint angle at each keyframe; an error localization and feedback generation module, connected to the action matching and evaluation module, used to identify the erroneous action stage and specific body joints based on the deviation value exceeding the allowable deviation range, and generate visual feedback information including skeletal annotation images and text correction prompts, wherein the skeletal annotation images distinguishably display joints with excessive deviations; and an output module for outputting the visual feedback information.

[0009] In addition, the artificial intelligence-based physical education teaching assessment system proposed above according to the present invention may also have the following additional technical features:

[0010] Furthermore, the motion matching and evaluation module uses a dynamic time warping algorithm to non-linearly align the skeletal keypoint sequence with the keyframe sequence in the motion standard template on the time axis.

[0011] Furthermore, the preset set of body joint points includes at least six of the following joint points: neck, shoulder, elbow, wrist, hip, knee, and ankle, and the standard joint angle vector includes elbow flexion-extension angle, shoulder abduction angle, knee flexion-extension angle, hip flexion-extension angle, and trunk forward tilt angle.

[0012] Furthermore, the error localization and feedback generation module also includes a semantic mapping unit, which maps the out-of-standard deviation value to a predefined semantic label. The semantic label includes: joint name, error type identifier and corresponding correction suggestion text, and displays the correction suggestion text overlaid in the vicinity of the corresponding joint in the skeletal annotation image.

[0013] Furthermore, it also includes a scoring calculation module, which is connected to the motion matching and evaluation module, for calculating a comprehensive score based on the weighted sum of the joint angle deviation values ​​under each keyframe and the satisfaction of the temporal constraints. The output module also outputs the comprehensive score.

[0014] Furthermore, the graphical configuration interface of the template configuration interface supports: importing demonstration videos containing standard movements, receiving user operations to select keyframes on the timeline of the demonstration video, automatically extracting the skeletal keypoint sequence corresponding to the selected keyframe, using the joint angle values ​​in the sequence as the initial values ​​of the standard joint angle vector, and providing a slider control for users to adjust the allowable deviation range.

[0015] Compared with the prior art, the beneficial effects of the present invention are:

[0016] 1. Supports rapid expansion of multiple projects: With a configurable motion template library, new sports projects can be added without retraining the deep learning model. Only the stages, keyframes and angle standards need to be defined, which effectively reduces the system's adaptation cost to new projects.

[0017] 2. Provides interpretable feedback: It can pinpoint specific movement stages, body parts, and quantify deviations, generating visual skeletal diagrams and textual correction suggestions to help students understand the reasons for their errors;

[0018] 3. Facilitates teaching adjustments: Physical education teachers can independently adjust movement standards through a graphical configuration interface (such as reducing the allowable deviation range of a certain joint angle) to meet the teaching needs of students of different ages and levels;

[0019] 4. Lightweight system architecture: Compared with end-to-end deep neural network solutions, this solution uses template matching, which significantly reduces computational overhead and is conducive to deployment on edge devices in teaching settings. Under typical configurations, it can achieve real-time feedback with low latency. Attached Figure Description

[0020] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0021] Figure 1 This is a schematic diagram of the module structure of a sports teaching and assessment system based on artificial intelligence according to an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the data structure of a motion template library for an artificial intelligence-based physical education teaching and assessment system according to an embodiment of the present invention;

[0023] Figure 3 This is an operation flowchart of a graphical configuration interface for a sports teaching and assessment system based on artificial intelligence according to an embodiment of the present invention;

[0024] Figure 4 This is a schematic diagram of a skeletal annotation image generated by an artificial intelligence-based physical education teaching and assessment system according to an embodiment of the present invention;

[0025] Figure 5 This is a schematic diagram of the stage division and keyframes of a standing long jump action in an artificial intelligence-based physical education teaching and assessment system according to an embodiment of the present invention.

[0026] As shown in the figure:

[0027] 1. Data acquisition module; 2. Pose estimation module; 3. Action template library; 4. Template configuration interface; 5. Action matching and evaluation module; 6. Error localization and feedback generation module; 7. Semantic mapping unit; 8. Output module; 9. Scoring calculation module. Detailed Implementation

[0028] Embodiments of the present invention are described in detail below, with examples of the embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0029] Example 1

[0030] like Figure 1 As shown, this embodiment provides a sports teaching and evaluation system based on artificial intelligence, including: a data acquisition module 1, a posture estimation module 2, a motion template library 3, a template configuration interface 4, a motion matching and evaluation module 5, an error location and feedback generation module 6, and an output module 7.

[0031] Data acquisition module 1: Uses at least one high-definition camera to collect the motion video stream of the subject in real time.

[0032] It should be noted that the camera can be placed at multiple angles such as the front and side of the stadium, with a sampling frame rate of no less than 30fps, and the video stream is transmitted to the processing unit via USB or Ethernet.

[0033] Specifically, the aforementioned cameras are, for example, USB cameras or webcams.

[0034] Pose estimation module 2: Connected to data acquisition module 1, it uses a pre-trained human pose estimation network to extract the skeletal key point sequence of the subject from video frames.

[0035] Specifically, the pre-trained human pose estimation networks mentioned above are, for example, OpenPose, MediaPipe, or HRNet.

[0036] Taking OpenPose as an example, the algorithm can output the two-dimensional coordinates (x, y) and confidence scores of 18 or 25 key points of the human body. The key points extracted in this embodiment include: neck, left and right shoulders, left and right elbows, left and right wrists, left and right hips, left and right knees, left and right ankles, etc. For continuous video frames, the pose estimation module 2 outputs a time series T={P1, P2, ..., Pn}, where Pi is the set of coordinates of all key points in the i-th frame.

[0037] Action Template Library 3: Stores standard action templates for at least one sport, such as Figure 2 As shown, each action standard template is stored in a structured data format (such as JSON or XML), specifically including:

[0038] Action phase division: For example, the standing long jump is divided into four phases: "preparation - take-off - flight - landing".

[0039] Keyframe Index: Each phase specifies one or more keyframe moments. For example, the keyframe for the "take-off" phase is the instant of take-off, and the keyframe for the "airborne" phase is the moment when the body is fully extended. Keyframes are represented by the time offset (milliseconds) relative to the start of the action or the frame number.

[0040] Standard joint angle vector and allowable deviation range: For each keyframe, for a preset set of body joint points (such as left and right knee joints, left and right hip joints, torso, etc.), store standard joint angle values ​​(such as knee flexion and extension angle 140°) and their positive and negative allowable deviations (such as +10° / -5°).

[0041] Timing constraints include the sequence constraints between stages (e.g., the order of "preparation" → "jump" → "takeoff" → "landing" cannot be reversed) and the time range of each stage (e.g., the "takeoff" stage lasts 200ms to 400ms).

[0042] Template configuration interface 4: Provides a graphical configuration interface, allowing instructors to add or modify action standard templates. Specific operation procedures are as follows: Figure 3 As shown:

[0043] Instructors click the "Create Template" button and select the name of the sports event (such as "Standing Long Jump").

[0044] The system prompts you to import a demonstration video (a standard movement video recorded by a professional athlete or teacher).

[0045] Users play the demonstration video and click on the keyframes of each stage in sequence on the timeline (for example, pause at the moment of takeoff and click "Mark as takeoff keyframe").

[0046] The system automatically calls the pose estimation module 2 to extract the skeletal keypoint sequence of the current keyframe, calculates the angle values ​​of each joint, and automatically fills them with the initial values ​​of the standard joint angle vector.

[0047] Users can fine-tune the standard value and allowable deviation range of each joint angle using the slider control (e.g., the standard value of the knee flexion-extension angle is 140°, and the allowable deviation is upper limit +10° and lower limit -5°).

[0048] Users set the sequence constraints and time range between stages and save the template.

[0049] Specifically, the aforementioned teaching staff refers to physical education teachers who do not possess professional knowledge in training deep learning models.

[0050] Action matching and evaluation module 5: Connects the posture estimation module 2 and the action template library 3. First, the skeletal key point sequence of the person being evaluated is aligned with the selected action standard template on the time axis. Since the movement speed of different individuals is different, this embodiment uses the dynamic time warping (DTW) algorithm to achieve non-linear alignment, so that the two sequences are best matched on the time axis. After alignment, at each key frame, the deviation value between the actual joint angle and the standard joint angle is calculated. For example, in the "take-off" key frame, the actual knee flexion and extension angle is 125°, and the standard value is 140°, so the deviation is -15°.

[0051] Error localization and feedback generation module 6: Connects to action matching and evaluation module 5. Based on the deviation value exceeding the allowable deviation range, it identifies the erroneous action stage and specific body joint. This module includes a semantic mapping unit 61, which maps the excessive deviation value to predefined semantic labels. The following shows an example of the mapping between deviation range and semantic labels:

[0052] Key points Deviation range Error type identifier Correction suggestion text knee joint Deviation > +10° KNEE001 If the knees bend excessively during takeoff, the squatting depth should be reduced. knee joint Bias <-5° KNEE002 If the knees don't bend enough at takeoff, the squat depth should be increased. Hip joint Deviation > +8° HIP001 When in the air, avoid excessive hip extension; be sure to engage your core. trunk Lean angle >15° TRUNK001 If your upper body leans too far forward, keep your torso upright.

[0053] Error localization and feedback generation module 6 generates visual feedback information including skeletal annotated images and text correction prompts, such as... Figure 4 As shown, in the skeletal annotation image, joints with excessive deviations are displayed with distinctive markers (such as red highlights) (e.g., the knee joint is marked with a red dot), and correction suggestion text (such as "insufficient knee flexion") is overlaid on the adjacent position of the joint.

[0054] Output module 7: Outputs visual feedback information to a display screen (such as a tablet, mobile phone, or classroom screen) for students and teachers to view in real time.

[0055] Scoring Calculation Module 8: Connected to Action Matching and Evaluation Module 5, it calculates a comprehensive score based on the weighted sum of the joint angle deviations in each keyframe and the satisfaction of temporal constraints. For example, comprehensive score = 100 - Σ(wi × deviation rate i) - temporal deduction. The weights wi can be assigned according to the importance of the joints (e.g., knee joint weight 0.3, hip joint weight 0.2, etc.). Output Module 7 simultaneously outputs the comprehensive score.

[0056] Example 2: Specific application of the standing long jump event:

[0057] Taking the standing long jump as an example, such as Figure 5 The following diagram illustrates the operation process of this system:

[0058] The physical education teacher creates a new "Standing Long Jump" template through template configuration interface 4, imports a standard standing long jump demonstration video, and marks four keyframes on the video timeline:

[0059] Keyframe 1 (Preparation Phase): Stand with arms swinging back and knees slightly bent.

[0060] Keyframe 2 (takeoff phase): The moment the feet push off the ground, the body extends.

[0061] Keyframe 3 (Airborne Phase): The body is fully extended and the arms are swinging forward.

[0062] Keyframe 4 (landing phase): Both feet touch the ground, knees bend to cushion the impact.

[0063] The system automatically extracts joint angles from each keyframe: for example, in keyframe 2, the standard value for the knee flexion-extension angle is 140°, the standard value for the hip flexion-extension angle is 160°, and the standard value for the trunk forward tilt angle is 10°. The teacher sets the allowable deviations using the slider: knee ±8°, hip ±10°, trunk forward tilt angle ±5°, and sets the timing constraints: the order of preparation → take-off → flight → landing cannot be reversed, and the flight phase time is 200ms~400ms. The template is then saved.

[0064] A student performs a standing long jump, and data acquisition module 1 records the video of their movement. Posture estimation module 2 extracts the skeletal keypoint sequence. Action matching and evaluation module 5 aligns this sequence with a "standing long jump" template using DTW (Dynamic Time-Warping). In keyframe 2 (takeoff phase), the actual knee flexion-extension angle is measured to be 125°, with a deviation of -15°. This deviation exceeds the allowable lower limit (-8°), and error localization and feedback generation module 6 maps it to a semantic label: error type identifier "KNEE002", with a correction suggestion of "insufficient knee flexion at takeoff, squat depth should be increased". Simultaneously, the knee joint is marked in red in the skeletal annotation image, and the correction text is displayed. Output module 7 displays the image and a score of 85 on a tablet computer.

[0065] If teachers believe that the tolerance for insufficient knee flexion should be higher for lower grade students, they can adjust the lower limit of the allowable deviation of the knee joint from -8° to -12° through template configuration interface 4. After saving, the system will evaluate subsequent students according to the new standard.

[0066] Example 3: Template expansion for sprint events

[0067] If a school needs to add sprint events, teachers do not need to program or retrain the model; they can simply create a new template through the template configuration interface (4).

[0068] Movement Phase Division: The sprint movement is broken down into the following phases:

[0069] The starting phase: from hearing the signal to the moment the starting blocks leave the track.

[0070] Acceleration phase: from the start to reaching maximum speed.

[0071] During the middle stage: Maintain a steady pace at maximum speed.

[0072] The final sprint: the last few steps before the finish line.

[0073] Keyframe Index:

[0074] Keyframes during the starting phase: the moment the starting blocks leave the track (maximum forward lean angle of the torso, full extension of the push-off leg).

[0075] Keyframes for the acceleration phase: the moment of first step landing (arm swing amplitude, knee height).

[0076] Keyframe during the mid-race phase: the moment of maximum extension of the supporting leg (full extension of the hip, knee, and ankle).

[0077] Keyframe of the sprint phase: The moment the torso presses forward before the finish line.

[0078] Standard joint angle vector and permissible deviation range:

[0079] At the start: the standard value for the forward tilt angle of the trunk is 45°, with an allowable deviation of ±5°; the standard value for the knee flexion and extension angle of the push-off leg is 150°, with an allowable deviation of ±8°.

[0080] During the acceleration phase: the standard value for the elbow flexion and extension angle of the arm swing is 90°, with an allowable deviation of ±10°; the standard value for the knee height angle of the supporting leg is 90°, with an allowable deviation of ±10°.

[0081] During the running phase: the standard value of the hip extension angle of the supporting leg is 180° (fully extended), with an allowable deviation of ±5°; the standard value of the plantar flexion angle of the ankle joint is 140°, with an allowable deviation of ±10°.

[0082] Sprint phase: Standard value for forward trunk pressure angle is 30°, with an allowable deviation of ±5°.

[0083] Timing constraints:

[0084] Phase sequence constraint: The order of start → acceleration → mid-race → sprint cannot be reversed.

[0085] Time range: The acceleration phase is usually 2 to 4 seconds, while the running phase is set according to the distance.

[0086] Cycle constraint: During the middle running phase, the step frequency should be maintained between 3.5 and 4.5 steps per second.

[0087] Template saving and application:

[0088] After teachers complete the above parameter settings through the graphical configuration interface and save it as a "sprint" template, the system can then conduct real-time evaluation of students' sprint movements: capture video streams, extract skeletal key point sequences, align them with the sprint template using the DTW time axis, calculate the deviation between the actual joint angle and the standard angle at each keyframe, identify erroneous movements (such as insufficient forward lean angle at the start, insufficient hip extension during the run, etc.), generate skeletal annotation images and correction suggestions (such as "insufficient forward lean at the start, the forward lean angle should be increased"), and output a comprehensive score.

[0089] The entire process requires no coding, and ordinary physical education teachers can independently complete the configuration and evaluation of movement standards for sprint events.

[0090] Example 4: Edge Computing Deployment

[0091] To meet the real-time requirements of physical education teaching, this system can be deployed on edge computing devices (such as NVIDIA Jetson Nano, Raspberry Pi 4B, or industrial-grade edge gateways). Data acquisition module 1 connects to a USB camera, pose estimation module 2 uses a lightweight MediaPipe model (which can run in real time on ARM devices), and motion matching and evaluation module 5 and error localization and feedback generation module 6 both run on edge devices.

[0092] Under typical configurations (e.g., using NVIDIA Jetson Nano, input resolution 640×480, pose estimation using MediaPipe, template library containing no more than 10 keyframes), the overall processing latency for a single motion evaluation is typically between 50 and 150 milliseconds, which meets the typical real-time feedback requirements in sports teaching scenarios. Actual latency depends on hardware performance, pose estimation algorithm complexity, template data volume, and video acquisition and transmission. For projects with high real-time requirements (such as sprint start reaction detection), latency can be further reduced by optimizing the algorithm and using higher-performance edge devices.

[0093] The above latency range is derived from reasonable expectations based on technical parameters known in the art. Specifically, lightweight pose estimation models such as MediaPipe typically have a single-frame inference time of 15-25 milliseconds on ARM architecture processors, which can be reduced to 10-15 milliseconds on edge devices with GPU acceleration (such as NVIDIA Jetson series). The time complexity of the Dynamic Time Warping (DTW) algorithm for aligning two sequences of length N and M is O(N×M). For typical sports actions (video length 2-5 seconds, sampling rate 30fps, template keyframes 10-20), the matching calculation time is typically in the range of 20-50 milliseconds. The above data are within the range that can be reasonably obtained by those skilled in the art based on publicly available technical documents or conventional performance tests, without relying on specific experimental data. Therefore, an overall latency of 50-150 milliseconds is a reasonable expectation of those skilled in the art.

[0094] Example 5: Construction of Semantic Mapping Table

[0095] At the core of the semantic mapping unit 61 is a predefined set of mapping rules used to convert out-of-standard deviation values ​​into readable semantic labels in order to generate targeted corrective suggestions.

[0096] Each mapping rule must contain at least the following data fields:

[0097] Joint Identifier: Specifies the body joints (such as knee, hip, elbow, trunk, etc.) to which this rule applies.

[0098] Action Phase: Specifies the action phase to which this rule applies (such as the take-off phase, the airborne phase, the release phase, etc.).

[0099] Deviation condition: Specifies the triggering condition for the deviation value, including the comparison type (greater than, less than, or within a range) and the threshold.

[0100] Error type identifier: Used to uniquely identify the error type, facilitating internal system processing or statistics.

[0101] Correction suggestion text: Guidance text displayed to users to explain the cause of the error and how to improve it.

[0102] The following table shows a set of exemplary mapping rules:

[0103] Key points Action phase Deviation conditions Error type identifier Correction suggestion text knee joint take-off Deviation > +10° KNEE001 If the knees bend excessively during takeoff, the squatting depth should be reduced. knee joint take-off Deviation < -5° KNEE002 If the knees don't bend enough at takeoff, the squat depth should be increased. Hip joint Take off Deviation > +8° HIP001 When in the air, avoid excessive hip extension; be sure to engage your core. trunk take-off Lean angle >15° TRUNK001 If your upper body leans too far forward, keep your torso upright. elbow joint Take action Flexion-extension angle <150° ELBOW001 If the arm is not fully extended when releasing the ball, the pushing motion should be increased.

[0104] Mapping rule sets can be stored in the system in various data structures, such as relational database tables for easy querying and management, JSON configuration files for easy editing and expansion, and XML configuration files for easy data exchange with other systems.

[0105] Regardless of the specific storage format used, the mapping rules include the above data fields and support dynamic loading at runtime.

[0106] During actual operation, the error location and feedback generation module 6 performs the following steps:

[0107] Obtain the deviation values ​​of each joint in the current keyframe from the motion matching and evaluation module 5.

[0108] For each joint whose deviation value exceeds the allowable deviation range, obtain the name of the joint and its current motion stage.

[0109] Search the mapping rule set for rules that match "key point" and "action phase" and whose deviation conditions are met.

[0110] Extract the error type identifier and correction suggestion text from the matching rules.

[0111] The correction suggestion text is passed to output module 7 and overlaid on the corresponding joint in the skeletal annotation image.

[0112] The mapping rule set supports dynamic expansion without modifying the underlying system code. Instructors can use the expansion functionality of interface 4 in the template configuration to perform the following operations on the mapping rules:

[0113] New rules: Define new deviation conditions and correction suggestions for specific joint points and movement phases.

[0114] Modify rules: Adjust the deviation threshold or correction suggestion text of existing rules.

[0115] Delete rule: Remove rules that are no longer applicable.

[0116] Import / Export rule sets: Migrate rule configurations between different teaching scenarios.

[0117] For example, for lower-grade students, teachers can adjust the lower limit of the knee joint deviation condition from -5° to -10° to lower the evaluation standard; for higher-grade students, it can be restored to -5° to raise the requirements. All adjustments are made through a graphical interface without writing any code.

[0118] Example 6: Multiple Implementations of Distinctive Tags

[0119] Distinguishing marks are not limited to color highlighting; they can also be used in the following ways:

[0120] Draw circular, square, or arrow-shaped graphic markers around the joints that exceed the deviation limit, change the line style of the joint connection (such as dashed or thickened lines), display a flashing animation at the joints, and overlay the deviation value (such as "-15°").

[0121] In this embodiment, red dots are used to mark joints that exceed the standard, and green dots are used to mark joints that meet the standard. The deviation value is displayed in small print next to the joint.

[0122] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0123] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0124] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0125] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A physical education teaching and assessment system based on artificial intelligence, characterized in that, include: The data acquisition module (1) is used to acquire the video stream of the subject in real time; The pose estimation module (2) is connected to the data acquisition module (1) and is used to extract the skeletal key point sequence of the subject from the video stream. Action template library (3) is used to store action standard templates for at least one sport, each of which consists of the following four parts of structured data: Action phase division: Divide the complete action into at least two phases in chronological order; Keyframe Index: Specify at least one keyframe moment for each stage, and each keyframe corresponds to a set of standard joint angle vectors; Standard joint angle vector and allowable deviation range: For each keyframe, for a preset set of body joint points, store the standard joint angle value and the positive and negative allowable deviation values ​​of that angle; Temporal constraints include at least one of the following: sequential constraints between stages and time ranges for each stage. Template configuration interface (4) provides a graphical configuration interface for receiving the addition or modification operations of the teaching staff on the action standard template. The operations include: marking key frames from video samples, and inputting or adjusting standard joint angle values ​​and their allowable deviation ranges. The motion matching and evaluation module (5) connects the pose estimation module (2) and the motion template library (3) to align the extracted skeletal key point sequence with the selected motion standard template on the time axis, and calculate the deviation between the actual joint angle and the standard joint angle in each key frame. Error localization and feedback generation module (6) is connected to the action matching and evaluation module (5) and is used to identify the action stage and specific body joints where the error occurred based on the deviation value exceeding the allowable deviation range, and generate visual feedback information including skeletal annotation images and text correction prompts, wherein the skeletal annotation images display the joints with excessive deviations with distinctive markings. The output module (7) is used to output the visual feedback information.

2. The physical education teaching assessment system based on artificial intelligence according to claim 1, characterized in that, The motion matching and evaluation module (5) uses a dynamic time warping algorithm to non-linearly align the skeletal keypoint sequence with the keyframe sequence in the motion standard template on the time axis.

3. The physical education teaching assessment system based on artificial intelligence according to claim 1, characterized in that, The preset set of body joints includes at least six of the following joints: neck, shoulder, elbow, wrist, hip, knee, and ankle. The standard joint angle vector includes elbow flexion-extension angle, shoulder abduction angle, knee flexion-extension angle, hip flexion-extension angle, and trunk forward tilt angle.

4. The physical education teaching assessment system based on artificial intelligence according to claim 1, characterized in that, The error localization and feedback generation module (6) further includes a semantic mapping unit (61), which maps the deviation value that exceeds the standard to a predefined semantic label. The semantic label includes: joint name, error type identifier and corresponding correction suggestion text, and displays the correction suggestion text superimposed on the adjacent position of the corresponding joint in the skeletal annotation image.

5. The artificial intelligence-based physical education teaching assessment system according to claim 1, characterized in that, It also includes a scoring calculation module (8) and is connected to the motion matching and evaluation module (5) to calculate a comprehensive score based on the weighted sum of the joint angle deviation values ​​under each keyframe and the satisfaction of the temporal constraints. The output module (7) also outputs the comprehensive score.

6. The physical education teaching assessment system based on artificial intelligence according to claim 1, characterized in that, The graphical configuration interface of the template configuration interface (4) supports: importing demonstration videos containing standard actions, receiving user operations to select key frames on the timeline of the demonstration video, automatically extracting the skeletal key point sequence corresponding to the selected key frame, and using the joint angle values ​​in the sequence as the initial values ​​of the standard joint angle vector, while providing a slider control for the user to adjust the allowable deviation range.

Images