Action recognition system and method for aerobics auxiliary learning

Abstract

The invention relates to the technical field of action recognition, in particular to an action recognition system and method for aerobics auxiliary learning. Comprising an aerobics image collection unit, an image adjustment unit, an action simulation unit, a body node simulation unit and an action recognition and analysis unit, the aerobics image collection unit is used for acquiring standard body data of the trainee and performing human body posture adjustment on a standard image of aerobics according to the standard body data; the image adjusting unit is used for collecting a real-time learning image of the student and performing deviation angle analysis on the real-time learning image in combination with the standard body data; a complete process of blind area comparison-logic completion-accurate judgment is constructed through a body node simulation unit and an action recognition analysis unit, and firstly, a blind area body node comparison module quickly eliminates non-standard actions caused by abnormal shielding positions by comparing types and relative positions of real-time blind area nodes and standard blind area nodes.

Description

Technical Field

[0001] This invention relates to the field of motion recognition technology, and more specifically, to a motion recognition system and method for assisting learning aerobics. Background Technology

[0002] In the field of aerobics teaching and learning, the standardization of movements is the core element for improving learning effectiveness. Existing aerobics learning aids mostly revolve around movement reference and comparison, aiming to help students correct movement deviations independently by providing standard movement templates, thereby reducing their dependence on professional coaches.

[0003] Currently, in the teaching process, both the figures in the teaching videos and the standard action templates used for comparison adopt uniform body proportions and presentation angles, without considering the individual differences in the students' body shapes. Moreover, there is no dynamic correction for environmental factors such as shooting angle and distance. When the student's facial orientation or shooting position changes, it is easy to cause misjudgment of the action comparison due to perspective deviation. For the blind spots of body nodes formed by limb occlusion, existing technologies mostly use the method of ignoring or simply removing blind spot nodes, which cannot fully analyze the student's action state and further reduces the accuracy of action recognition. Therefore, an action recognition system and method for aerobics-assisted learning are proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a motion recognition system and method for aerobics-assisted learning, in order to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, one of the objectives of this invention is to provide a motion recognition system for aerobics-assisted learning, including an aerobics image collection unit, an image adjustment unit, a motion simulation unit, a body node simulation unit, and a motion recognition and analysis unit. The aerobics image collection unit is used to acquire the trainees' standard body data and adjust the human posture of the aerobics standard images based on the standard body data. The image adjustment unit is used to acquire real-time learning images of trainees, combine the real-time learning images with standard body data for offset angle analysis, and simultaneously use a skeleton tracking algorithm to identify body nodes in the real-time learning images and scale the teaching figures in the standard images according to the identified body nodes. The motion simulation unit is used to perform aerobics stage analysis based on real-time learning images, then combine the aerobics stage with the offset angle to perform motion simulation, and identify blind zone body nodes in the standard image in the motion simulation results. The body node simulation unit is used to compare blind zone body nodes by combining real-time learning images with standard images. After the comparison is consistent, the relevant blind zone body nodes of the aerobics stage are extracted, and the relevant blind zone body nodes are simulated and analyzed based on the body nodes identified by the real-time learning images. The action recognition and analysis unit is used to input simulated relevant blind zone body nodes into the real-time learning image, and compare the body nodes by combining the input real-time learning image with the standard image, and perform action standard recognition based on the comparison results.

[0006] As a further improvement to this technical solution, the aerobics image collection unit acquires standard body data before the aerobics training begins. The trainee stands parallel to the camera, and the camera collects images of the trainee's limbs to determine the trainee's body proportions. Then, the limb images are compiled as standard body data.

[0007] As a further improvement to this technical solution, the aerobics image collection unit extracts standard images from the aerobics teaching video, making the teaching video consist of multiple standard images. It then simulates and extracts the teaching figures from the standard images and adjusts the body proportions of the teaching figures in the standard images according to standard body data, so that the body proportions of the teaching figures in the aerobics teaching video are consistent with the body proportions of the students.

[0008] As a further improvement to this technical solution, the image adjustment unit includes a real-time acquisition module, an offset analysis module, and an image scaling module; The real-time acquisition module is used to capture the process of students learning aerobics in real time through a camera, and obtain real-time learning images of the students; The offset analysis module is used to acquire facial data of real-time learning images, and then perform difference angle analysis by combining the facial orientation of real-time learning images with the facial orientation of standard body data, and use the difference angle obtained from the analysis as the deviation angle between the student and the standard body data. The image scaling module is used to identify body nodes in real-time learning images using a skeleton tracking algorithm, obtain the body nodes of students in real-time learning images, and then scale the teaching figures in the standard image proportionally based on the obtained body nodes, so that the teaching figures and students in the real-time learning images can fit together perfectly.

[0009] As a further improvement to this technical solution, the image scaling module acquires the body nodes of the learner in the real-time learning image, including blind zone body nodes and non-blind zone body nodes. First, all body nodes are determined through standard body data. Then, in the real-time learning image, the determined body nodes are combined to perform an analysis of undisplayed data. Body nodes not displayed in the real-time learning image are taken as blind zone body nodes, while body nodes displayed in the real-time learning image are taken as non-blind zone body nodes.

[0010] As a further improvement to this technical solution, the motion simulation unit includes a stage analysis module and a blind spot body node extraction module; The stage analysis module is used to divide the aerobics instructional video into stages based on the continuity of the movements, so that the instructional video is divided into multiple stages, and then the real-time learning images are combined with the instructional video to determine the aerobics stage. The blind spot body node extraction module is used to extract the aerobics stage determined in the stage analysis stage, and then combine the teaching movements of the aerobics stage with the offset angle to simulate the movements, so that the teaching figure simulates the completion of the teaching movements after the offset angle, and then the blind spot body nodes are identified in the standard image in the movement simulation results.

[0011] As a further improvement to this technical solution, the body node simulation unit includes a blind zone body node comparison module and a node simulation module; The blind spot body node comparison module is used to compare the blind spot body nodes of the real-time learning image with the blind spot body nodes of the standard image. When the comparison result shows that the blind spot body nodes are consistent, the node simulation module is entered. Otherwise, when the comparison result shows that the blind spot body nodes are inconsistent, the student's action is judged to be non-standard and the student is reminded. The node simulation module is used to extract blind spot body nodes related to the movements in the current aerobics stage, and then combine the non-blind spot body nodes identified by the real-time learning image with the relevant blind spot body nodes for simulation analysis. Based on the non-blind spot body nodes, the module performs movement logic analysis on the relevant blind spot body nodes, thereby simulating the relevant blind spot body nodes.

[0012] As a further improvement to this technical solution, the action recognition and analysis unit includes a simulated input module and a final determination module; The simulation input module is used to input the relevant blind zone body nodes simulated by the node simulation module into the real-time learning image, and to compare the body nodes of the real-time learning image with the standard image after input, so as to obtain the body node difference data between the real-time learning image and the standard image. The final judgment module is used to perform posture deviation analysis based on body node difference data, and then provide auxiliary reminders to trainees through posture deviation analysis; If the body node difference data shows no difference, the student's movement is judged to be standard, and no reminder is given.

[0013] The second objective of this invention is to provide a motion recognition method for aerobics-assisted learning, comprising the motion recognition system for aerobics-assisted learning described in any one of the above-mentioned methods, including the following steps: S1. Obtain the trainees' standard body data, adjust the human posture of the standard aerobics image based on the standard body data, then collect the trainees' real-time learning images, analyze the offset angle by combining the real-time learning images with the standard body data, and simultaneously identify body nodes in the real-time learning images using a skeleton tracking algorithm, and scale the teaching figure in the standard image based on the identified body nodes. S2. Analyze the aerobics stage based on the real-time learning images, then simulate the movements by combining the aerobics stage with the offset angle. In the simulation results, identify blind-zone body nodes in the standard image. Compare the blind-zone body nodes between the real-time learning images and the standard image. If a match is found, extract the relevant blind-zone body nodes for the current aerobics stage, and perform simulation analysis on the relevant blind-zone body nodes based on the body nodes identified in the real-time learning images. S3. Input the simulated relevant blind zone body nodes into the real-time learning image, and compare the body nodes by combining the real-time learning image with the standard image. Based on the comparison results, perform action standard recognition.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. A motion recognition system and method for aerobics-assisted learning, which constructs a complete process of blind spot comparison, logical completion, and accurate judgment through a body node simulation unit and a motion recognition analysis unit. First, the blind spot body node comparison module quickly eliminates non-standard movements caused by abnormal occlusion positions by comparing the type and relative position of real-time blind spot nodes with standard blind spot nodes. Second, the node simulation module completes the blind spot node information based on the laws of human kinematics. The motion recognition analysis unit combines the complete node information to perform a full-body node comparison and clearly judges the standard of the movement based on the difference data. This not only avoids the recognition omissions caused by blind spot nodes, but also provides targeted auxiliary reminders, realizing an upgrade from partial recognition to comprehensive analysis, making motion assistance more accurate and instructive, and significantly improving the learning efficiency of learners.

[0015] 2. A motion recognition system and method for aerobics-assisted learning: This system uses an aerobics image collection unit to capture the body proportions of learners before learning and adjusts the body proportions of figures in the teaching images accordingly. This ensures that the standard images perfectly match the learners' own characteristics. This design fundamentally solves the problem of motion reference deviation caused by differences in the size of the teaching figures and the learners in traditional teaching. It allows each learner to have their own personalized standard image library. Simultaneously, the system breaks down the teaching video into independent standard images at fixed intervals or key nodes, providing a unified and stable benchmark for subsequent real-time comparison. This avoids the problem of blurred motion references in dynamic video footage, ensuring high accuracy in motion comparison from the benchmark level and laying a reliable foundation for subsequent recognition and analysis.

[0016] 3. A motion recognition system and method for aerobics-assisted learning: The offset analysis module calculates the deviation angle between the student's real-time facial orientation and the standard orientation, dynamically corrects the angle of the teaching image, and ensures that the student and the standard image are compared from the same perspective, avoiding misjudgment of motion due to shooting angle deviation. The image scaling module distinguishes between blind and non-blind area body nodes through a skeleton tracking algorithm, and proportionally scales the standard image based on the size of the non-blind area nodes, so that the teaching figure and the student's body shape are completely matched. At the same time, the motion simulation unit adjusts the orientation of the teaching figure based on the deviation angle, making the standard motion more realistic from the student's perspective, further ensuring the logical rationality of real-time comparison and reducing recognition interruption or misjudgment caused by environmental factors. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the motion recognition method for aerobics-assisted learning according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, one of the objectives of this invention is to provide a motion recognition system for aerobics-assisted learning, including an aerobics image collection unit, an image adjustment unit, a motion simulation unit, a body node simulation unit, and a motion recognition and analysis unit. The aerobics image collection unit is used to acquire the trainees' standard body data and adjust the human posture of the standard aerobics images based on the standard body data; By collecting standard body data (body proportions) of trainees and adjusting the body proportions of figures in teaching images accordingly, the standard images are matched with the trainees' own characteristics. This avoids motion reference deviations caused by differences in body shape between the teaching figures and trainees, improving the accuracy of motion comparison. At the same time, the teaching videos are broken down into standardized images to ensure that each motion posture has a clear reference template, providing a unified and stable benchmark for subsequent real-time comparison.

[0020] The aerobics image collection unit acquires standard body data. Before the aerobics training begins, the trainee stands parallel to the camera, and the camera collects images of the trainee's limbs to determine the trainee's body proportions (the ratio of upper limb length to lower limb length, the length ratio of different parts of the arm, the length ratio of different parts of the leg, etc.). The determined body proportions and the collected limb images and other relevant information are summarized to form the trainee's standard body data, which serves as the basis for subsequent aerobics learning movement recognition and adjustment.

[0021] The aerobics image collection unit extracts standard images from aerobics instructional videos. Multiple frames are extracted from the videos at fixed time intervals (or key movement nodes), breaking down the continuous instructional video into a series of independent standard images. Each frame corresponds to a standard movement posture, resulting in an instructional video composed of multiple standard images. The instructors in these standard images are then simulated and extracted. Based on standard body data, the body proportions of the instructors in the standard images are adjusted to match the body proportions of the trainees. After adjustment, the body proportions of the instructors are completely consistent with the trainees' standard body data, forming a standard image library adapted to the trainees' body types.

[0022] The image adjustment unit is used to acquire real-time learning images of trainees, combine the real-time learning images with standard body data for offset angle analysis, and simultaneously use a skeleton tracking algorithm to identify body nodes in the real-time learning images and scale the teaching figures in the standard images based on the identified body nodes. By acquiring student images in real time and analyzing the offset angle (facial orientation deviation), the angle and scale of the teaching images are dynamically corrected to ensure that the student's posture is compared with the standard image from the same viewpoint and scale, reducing errors caused by shooting angle and distance. At the same time, the skeleton tracking algorithm distinguishes between blind and non-blind zone nodes, clearly identifying visible and invisible body parts, providing a basis for subsequent blind zone processing, and avoiding recognition interruptions caused by limb occlusion.

[0023] The image adjustment unit includes a real-time acquisition module, an offset analysis module, and an image scaling module; The real-time acquisition module is used to capture the trainees' aerobics learning process in real time through a camera, serving as real-time learning images of the trainees; The offset analysis module is used to acquire facial data from real-time learning images. Then, it combines the facial orientation of the real-time learning images with the facial orientation of standard body data to perform a difference angle analysis. The resulting difference angle is used as the deviation angle between the learner and the standard body data. The steps are as follows: The system identifies and extracts the trainee's facial feature data (such as facial contours and the position of facial features), determines the specific position and orientation reference of the face in the image (such as the direction of the nose tip and the direction of the line connecting the eyebrows), and extracts the facial orientation information recorded during the standard body data acquisition phase (i.e., the reference facial orientation when the trainee stands parallel to the camera) from the pre-stored standard body data of the trainee. It then compares the facial orientation extracted from the real-time learning image with the standard facial orientation, and calculates the difference in deflection angle between the two using image angle analysis technology. This difference angle is the deviation angle between the trainee's current body data and the standard body data, as shown in the following formula: ; Where θ is the deviation angle between the trainee and the standard body data, θ1 is the trainee's facial orientation angle extracted from the real-time learning image, and θ0 is the facial reference orientation angle recorded in the standard body data.

[0024] The image scaling module is used to identify body nodes in real-time learning images using a skeleton tracking algorithm, and to obtain the body nodes of the learner in the real-time learning images (such as key nodes such as head, shoulder, elbow, wrist, hip, knee, and ankle). The image scaling module acquires the body nodes of the learner in the real-time learning image, including blind zone body nodes and non-blind zone body nodes. First, all body nodes are determined through standard body data. Then, the real-time learning image is combined with all the determined body nodes to perform undisplayed analysis. Body nodes not displayed in the real-time learning image are regarded as blind zone body nodes, and conversely, body nodes displayed in the real-time learning image are regarded as non-blind zone body nodes.

[0025] Then, based on the acquired body nodes, the teaching figure in the standard image is scaled proportionally so that the teaching figure and the students in the real-time learning image can fit together perfectly. Based on the identified non-blind zone body nodes (and combined with logical inference to blind zone nodes), key dimensions such as the body contour of the learner in the real-time learning image are determined. The teaching figure in the standard image is scaled proportionally so that the teaching figure and the learner in the real-time learning image are perfectly matched in body shape.

[0026] The motion simulation unit is used to perform aerobics stage analysis based on real-time learning images, then combine the aerobics stage with the offset angle to perform motion simulation, and identify blind zone body nodes in the standard image in the motion simulation results. Dividing instructional videos into coherent stages breaks down complex aerobics movements into units that can be compared step by step, reducing the complexity of real-time comparison and improving analysis efficiency. At the same time, by combining offset angles to simulate the standard movements of the instructors, the posture of the standard images is consistent with the current angle of the trainees, ensuring the authenticity of the standard movements from the trainees' perspective and enhancing the rationality of the comparison.

[0027] The motion simulation unit includes a phase analysis module and a blind spot body node extraction module; The stage analysis module is used to divide aerobics instructional videos into stages based on the continuity of movements. This results in multiple stages, each corresponding to a coherent sequence of movements with a clear objective. The module extracts the current posture features of the learner from real-time learning images and compares them with the standard movement features of each stage in the instructional video to determine the specific aerobics stage the learner is currently in. The blind spot body node extraction module is used to extract the aerobics stage determined in the stage analysis stage, and then combine the teaching movements of the aerobics stage with the offset angle to simulate the movement, so that the teaching figure simulates the completion of the teaching movement after the offset angle. Then, the blind spot body nodes are identified in the standard image in the movement simulation results. This process involves extracting the specific stage a student is currently in, determined through stage analysis, from multiple pre-defined aerobics stages. The corresponding instructional movement sequence and standard image for that stage are then obtained. The extracted instructional movements for that stage are combined with the previously calculated student deviation angle (i.e., the difference between the student's current facial orientation and the standard orientation). The posture of the instructor in the standard image is adjusted to match the instructor's orientation to the student's deviation angle, simulating the standard instructional movement performed at that angle. After the movement simulation, the standard image in the simulation results is analyzed and identified based on the full set of body nodes contained in the standard body data. These missing body nodes are marked as blind spots in the standard image, as shown in the following formula: ; Where β is the actual orientation angle of the teaching figure after simulation, and β0 is the standard orientation angle of the original teaching figure in the teaching action at this stage.

[0028] The body node simulation unit is used to compare blind zone body nodes by combining real-time learning images with standard images. After the comparison is consistent, the relevant blind zone body nodes of the aerobics stage are extracted, and the relevant blind zone body nodes are simulated and analyzed based on the body nodes identified by the real-time learning images. By comparing the consistency of blind zone nodes, non-standard movements caused by abnormal occlusion positions can be quickly eliminated. Based on non-blind zone nodes, the blind zone nodes are simulated and completed, realizing a complete analysis of all body nodes and avoiding recognition omissions caused by occlusion. At the same time, the position of blind zone nodes is inferred by combining the movement patterns of non-blind zone nodes (such as joint linkage and limb trajectory), ensuring that the simulation results conform to the logic of human movement and improving the reliability of the completed data.

[0029] The body node simulation unit includes a blind zone body node comparison module and a node simulation module; The blind spot body node comparison module compares the blind spot body nodes of the real-time learning image with the blind spot body nodes of the standard image. When the comparison result shows that the blind spot body nodes are consistent, the module proceeds to the node simulation module. Conversely, when the comparison result shows that the blind spot body nodes are inconsistent, the trainee's movement is determined to be non-standard, and the trainee is reminded. The steps are as follows: From the real-time learning images, the body nodes of the trainees' blind spots identified in the previous stage are extracted and compiled into a real-time blind spot node list. At the same time, from the standard images corresponding to the current aerobics stage, the body nodes of the marked standard blind spots are extracted and compiled into a standard blind spot node list. The core dimensions of the comparison are clarified, including the type of blind spot node and the relative positional relationship of the nodes (such as the distance and angle between the blind spot node and the adjacent non-blind spot node), to ensure that the comparison standard is consistent. Then, the real-time blind spot node list is compared with the standard blind spot node list one by one according to the set dimensions to determine whether the blind spot nodes in the two lists are completely matched in terms of type and relative position. If the comparison results show that the two sets of blind spot nodes are completely consistent, the system enters the subsequent node simulation module; if the comparison results show that there are differences (such as the real-time blind spot node is the right elbow and the standard blind spot node is the left elbow), the system directly determines that the student's current movement is not standard and issues a reminder to the student through voice, text or screen annotation.

[0030] The node simulation module is used to extract blind spot body nodes related to the movements in the current aerobics stage. Then, it combines the non-blind spot body nodes identified by the real-time learning image with the relevant blind spot body nodes for simulation analysis. Based on the non-blind spot body nodes, it performs action logic analysis on the relevant blind spot body nodes, thereby simulating the relevant blind spot body nodes. If the current stage is a lateral extension movement, then the obscured lateral waist joint, ipsilateral elbow joint, and other blind spot nodes closely related to the extension movement are the relevant nodes. Then, the relationship between non-blind spot body nodes and relevant blind spot body nodes in human movement is analyzed (e.g., the linkage between the elbow and shoulder joints, the synergistic movement of the knee and hip joints). The logical connection between the two in the movement is clarified. Based on the position and angle characteristics of the non-blind spot body nodes, combined with the laws of human kinematics (e.g., limb movement trajectory, joint range of motion), the reasonable position and posture of the relevant blind spot body nodes are inferred, simulating their specific state in the current movement, thus completing the blind spot node information. The formula is as follows: ; Among them, a b To simulate the angles of relevant blind zone body nodes, k1 and k2 are the influence coefficients of the non-blind zone node angles on the blind zone node angles, pre-set based on the human kinematics model, reflecting the linkage weight between the two, α nb1 and α nb2 For real-time identification of non-blind zone body nodes, such as the shoulder and wrist joint angles related to blind zone nodes, c is a correction coefficient used to compensate for individual movement differences and make the simulation results more consistent with the actual movement logic.

[0031] The action recognition and analysis unit is used to input simulated relevant blind zone body nodes into the real-time learning image, and compare the body nodes by combining the input real-time learning image with the standard image, and perform action standard recognition based on the comparison results.

[0032] Based on the discrepancy data, targeted prompts are made to trainees to adjust their movements (such as insufficient left arm raising), directly guiding the learning process and improving the practicality and efficiency of the aids.

[0033] The action recognition and analysis unit includes a simulated input module and a final judgment module; The simulated input module is used to input the relevant blind zone body nodes simulated by the node simulation module into the real-time learning image, and then compare the body nodes of the real-time learning image with the standard image to obtain the difference data of body nodes between the real-time learning image and the standard image. The steps are as follows: The relevant blind spot body nodes generated by the node simulation module are added to the real-time learning image, so that the real-time image contains complete body node information of the trainee (non-blind spot nodes + simulated blind spot nodes). Then, the real-time learning image with complete nodes is compared with the standard image corresponding to the current aerobics stage. The position and angle of each body node are compared one by one. Based on the comparison results, the difference information of body nodes in the real-time image and the standard image is statistically recorded, including the type of difference node (left elbow, right knee joint), the degree of difference (position offset distance, angle deviation degree), etc., to form body node difference data.

[0034] The final judgment module is used to perform posture deviation analysis based on body node difference data. It analyzes the overall posture deviation reflected by the difference data, judges whether the deviation is within a reasonable range, clarifies the specific parts and directions that need to be adjusted in the student's movements, and then provides auxiliary reminders to the student through posture deviation analysis. If the body node difference data shows no difference, the student's movement is judged to be standard, and no reminder is given.

[0035] It solves problems such as differences in body proportions, angle deviations, and limb obstruction, and achieves dynamic recognition and precise assistance for aerobics movements, thereby improving students' learning efficiency.

[0036] The second objective of this invention is to provide a motion recognition method for aerobics-assisted learning, including any of the above-mentioned motion recognition systems for aerobics-assisted learning, comprising the following steps: S1. Obtain the trainees' standard body data, adjust the human posture of the standard aerobics image based on the standard body data, then collect the trainees' real-time learning images, analyze the offset angle by combining the real-time learning images with the standard body data, and simultaneously identify body nodes in the real-time learning images using a skeleton tracking algorithm, and scale the teaching figure in the standard image based on the identified body nodes. S2. Analyze the aerobics stage based on the real-time learning images, then simulate the movements by combining the aerobics stage with the offset angle. In the simulation results, identify blind-zone body nodes in the standard image. Compare the blind-zone body nodes between the real-time learning images and the standard image. If a match is found, extract the relevant blind-zone body nodes for the current aerobics stage, and perform simulation analysis on the relevant blind-zone body nodes based on the body nodes identified in the real-time learning images. S3. Input the simulated relevant blind zone body nodes into the real-time learning image, and compare the body nodes by combining the real-time learning image with the standard image. Based on the comparison results, perform action standard recognition.

[0037] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A motion recognition system for assisting learning aerobics, characterized in that: It includes an aerobics image collection unit, an image adjustment unit, a motion simulation unit, a body node simulation unit, and a motion recognition and analysis unit; The aerobics image collection unit is used to acquire the trainees' standard body data and adjust the human posture of the aerobics standard images based on the standard body data. The image adjustment unit is used to acquire real-time learning images of trainees, combine the real-time learning images with standard body data for offset angle analysis, and simultaneously use a skeleton tracking algorithm to identify body nodes in the real-time learning images and scale the teaching figures in the standard images according to the identified body nodes. The motion simulation unit is used to perform aerobics stage analysis based on real-time learning images, then combine the aerobics stage with the offset angle to perform motion simulation, and identify blind zone body nodes in the standard image in the motion simulation results. The body node simulation unit is used to compare blind zone body nodes by combining real-time learning images with standard images. After the comparison is consistent, the relevant blind zone body nodes of the aerobics stage are extracted, and the relevant blind zone body nodes are simulated and analyzed based on the body nodes identified by the real-time learning images. The action recognition and analysis unit is used to input simulated relevant blind zone body nodes into the real-time learning image, and compare the body nodes by combining the input real-time learning image with the standard image, and perform action standard recognition based on the comparison results.

2. The motion recognition system for aerobics-assisted learning according to claim 1, characterized in that: The aerobics image collection unit acquires standard body data. Before the aerobics training begins, the trainee stands parallel to the camera, and the camera collects images of the trainee's limbs to determine the trainee's body proportions. Then, the limb images are compiled as standard body data.

3. The motion recognition system for aerobics-assisted learning according to claim 1, characterized in that: The aerobics image collection unit extracts standard images from the aerobics instructional video, making the instructional video consist of multiple standard images. It then simulates and extracts the teaching figures from the standard images and adjusts the body proportions of the teaching figures in the standard images according to standard body data, so that the body proportions of the teaching figures in the aerobics instructional video are consistent with the body proportions of the students.

4. The motion recognition system for aerobics-assisted learning according to claim 1, characterized in that: The image adjustment unit includes a real-time acquisition module, an offset analysis module, and an image scaling module; The real-time acquisition module is used to capture the process of students learning aerobics in real time through a camera, and obtain real-time learning images of the students; The offset analysis module is used to acquire facial data of real-time learning images, and then perform difference angle analysis by combining the facial orientation of real-time learning images with the facial orientation of standard body data, and use the difference angle obtained from the analysis as the deviation angle between the student and the standard body data. The image scaling module is used to identify body nodes in real-time learning images using a skeleton tracking algorithm, obtain the body nodes of students in real-time learning images, and then scale the teaching figures in the standard image proportionally based on the obtained body nodes, so that the teaching figures and students in the real-time learning images can fit together perfectly.

5. The motion recognition system for aerobics-assisted learning according to claim 4, characterized in that: The image scaling module acquires the trainee's body nodes in the real-time learning image, including blind spot body nodes and non-blind spot body nodes. First, all body nodes are determined using standard body data. Then, the real-time learning image is analyzed in conjunction with all the determined body nodes. Body nodes not displayed in the real-time learning image are designated as blind spot body nodes, while body nodes displayed in the real-time learning image are designated as non-blind spot body nodes.

6. The motion recognition system for aerobics-assisted learning according to claim 1, characterized in that: The motion simulation unit includes a phase analysis module and a blind spot body node extraction module; The stage analysis module is used to divide the aerobics instructional video into stages based on the continuity of the movements, so that the instructional video is divided into multiple stages, and then the real-time learning images are combined with the instructional video to determine the aerobics stage. The blind spot body node extraction module is used to extract the aerobics stage determined in the stage analysis stage, and then combine the teaching movements of the aerobics stage with the offset angle to simulate the movements, so that the teaching figure simulates the completion of the teaching movements after the offset angle, and then the blind spot body nodes are identified in the standard image in the movement simulation results.

7. The motion recognition system for aerobics-assisted learning according to claim 1, characterized in that: The body node simulation unit includes a blind zone body node comparison module and a node simulation module; The blind spot body node comparison module is used to compare the blind spot body nodes of the real-time learning image with the blind spot body nodes of the standard image. When the comparison result shows that the blind spot body nodes are consistent, the node simulation module is entered. Otherwise, when the comparison result shows that the blind spot body nodes are inconsistent, the student's action is judged to be non-standard and the student is reminded. The node simulation module is used to extract blind spot body nodes related to the movements in the current aerobics stage, and then combine the non-blind spot body nodes identified by the real-time learning image with the relevant blind spot body nodes for simulation analysis. Based on the non-blind spot body nodes, the module performs movement logic analysis on the relevant blind spot body nodes, thereby simulating the relevant blind spot body nodes.

8. The motion recognition system for aerobics-assisted learning according to claim 1, characterized in that: The action recognition and analysis unit includes a simulated input module and a final determination module; The simulation input module is used to input the relevant blind zone body nodes simulated by the node simulation module into the real-time learning image, and to compare the body nodes of the real-time learning image with the standard image after input, so as to obtain the body node difference data between the real-time learning image and the standard image. The final judgment module is used to perform posture deviation analysis based on body node difference data, and then provide auxiliary reminders to trainees through posture deviation analysis; If the body node difference data shows no difference, the student's movement is judged to be standard, and no reminder is given.

9. A motion recognition method for aerobics-assisted learning, comprising the motion recognition system for aerobics-assisted learning as described in any one of claims 1-8, characterized in that: Includes the following steps: S1. Obtain the trainees' standard body data, adjust the human posture of the standard aerobics image based on the standard body data, then collect the trainees' real-time learning images, analyze the offset angle by combining the real-time learning images with the standard body data, and simultaneously identify body nodes in the real-time learning images using a skeleton tracking algorithm, and scale the teaching figure in the standard image based on the identified body nodes. S2. Analyze the aerobics stage based on the real-time learning images, then simulate the movements by combining the aerobics stage with the offset angle. In the simulation results, identify blind-zone body nodes in the standard image. Compare the blind-zone body nodes between the real-time learning images and the standard image. If a match is found, extract the relevant blind-zone body nodes for the current aerobics stage, and perform simulation analysis on the relevant blind-zone body nodes based on the body nodes identified in the real-time learning images. S3. Input the simulated relevant blind zone body nodes into the real-time learning image, and compare the body nodes by combining the real-time learning image with the standard image. Based on the comparison results, perform action standard recognition.

Images