AI intelligent correction fitness training method and smith training device thereof

Abstract

The application discloses an AI intelligent correction fitness training method and a Smith training device, and comprises the following steps: S1, voice instruction awakening, receiving user voice instructions and identifying, controlling the fitness device and correcting the fitness posture according to the voice instructions; S2, visual perception coverage, collecting front and side images of the user, obtaining the user's body key point position, key point confidence and multi-view motion characteristics. The application solves the problem that the existing fitness device is inconvenient for assisting users to correct the fitness posture during use, and the user is prone to sports injuries due to non-standard movements, and the incorrect force mode can cause the target muscle group to be unable to be accurately activated, but promote other irrelevant muscles to compensate for the force, thereby reducing the training effect.

Description

Technical Field

[0001] This invention relates to the field of fitness training technology, specifically to an AI-powered intelligent correction fitness training method and its Smith training device. Background Technology

[0002] AI intelligent correction refers to a training method that uses computer vision and artificial intelligence algorithms to track the user's skeletal joints in real time and correct their posture. When a user exercises, the AI ​​system uses a camera to locate the user's key body parts and tracks the changes in the joint angles of these parts throughout the movement. The AI ​​compares the user's actual movement pattern with the standard ideal posture, and once a deviation is detected, it will immediately provide voice prompts or screen demonstrations to guide the correction.

[0003] This corrective method is applied to large fitness equipment, such as the Smith machine, a comprehensive fitness machine. Its basic principle is that the barbell moves up and down on fixed steel rails, equipped with intermittent locking points and safety devices. During exercise, improper posture can put unnecessary pressure on joints and ligaments. For example, knee buckling during squats can easily damage the knees; timely correction can prevent muscle strains or chronic injuries.

[0004] Existing fitness equipment does not readily assist users in correcting their posture during training. Due to the lack of a real-time correction mechanism, users are highly susceptible to sports injuries caused by improper movements. Furthermore, incorrect force exertion patterns can prevent the target muscle groups from being precisely activated, instead prompting other unrelated muscles to compensate and reducing the effectiveness of training. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-powered intelligent corrective fitness training method and its Smith training device. This invention solves the problems of existing fitness equipment, which is not convenient for assisting users in correcting their fitness postures during use. Due to the lack of a real-time correction mechanism, users are prone to sports injuries due to non-standard movements. Furthermore, incorrect force exertion patterns can prevent the target muscle groups from being accurately activated, instead prompting other unrelated muscles to compensate and reducing the training effect.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an AI-powered intelligent corrective fitness training method, comprising the following steps: S1. Voice command wake-up: Receives and recognizes user voice commands, controls the fitness equipment and corrects fitness postures according to the voice commands. S2, Visual perception coverage, collects frontal and side images of the user, and obtains the location of key points of the user's human body, key point confidence and multi-view motion features; S3. Motion recognition and analysis: Match feature points in frontal and side images to generate multi-view motion features or quasi-3D pose estimation results for the user, and simultaneously identify the motion trajectory of the training actuators inside the fitness equipment. S4. Determination of movement standard: Detect the movement trajectory and determine whether the exercise movement is standard. S5 features intelligent feedback and correction, providing voice and visual feedback based on the results of exercise movements to guide users in performing standard posture exercises.

[0007] In a preferred embodiment of the AI-powered intelligent correction fitness training method of the present invention, during the speech recognition process in step S1, the acoustic signals within the central training space are continuously monitored. The analog audio signals are converted into text data through a speech recognition engine, and the semantics of the instructions are parsed through a natural language understanding algorithm. When a wake-up word is detected, the wake-up mechanism is triggered, and the system enters the instruction receiving state.

[0008] In a preferred embodiment of the AI-powered intelligent correction fitness training method of the present invention, in step S1, a correction signal or a control signal is generated based on the parsing result of the voice command. The control signal adjusts the training parameters of the training actuators inside the fitness equipment, and the correction signal controls the visual acquisition module to capture the user's current action image.

[0009] In a preferred embodiment of the AI-powered intelligent correction fitness training method of the present invention, when acquiring user images in step S2, the first visual acquisition module acquires the user's frontal image and two-dimensional plane projection information in the central training space, and the second visual acquisition module acquires the user's lateral posture contour, lateral projection of key human body points, and corresponding confidence levels.

[0010] In a preferred embodiment of the AI-powered intelligent correction fitness training method of the present invention, when processing the image in step S2, the data obtained by the first visual acquisition module and the second visual acquisition module are fused based on user orientation judgment, multi-view key point recognition and device coordinate system calibration to generate user multi-view motion features or quasi-three-dimensional posture estimation results.

[0011] In a preferred embodiment of the AI ​​intelligent correction fitness training method of the present invention, during feature point matching in S3, the displacement of the training actuator in three-dimensional space is calculated through visual recognition, and the motion trajectory and running state of the training actuator are constructed by combining the running data of the training actuator.

[0012] In a preferred embodiment of the AI-powered intelligent correction fitness training method of the present invention, when making standard judgments on the exercise movements in step S4, the movement judgment rules or movement feature configuration tables corresponding to the target training movements are retrieved from the cloud or local database. In the spatial dimension, the distance deviation between the user's 3D motion trajectory and the standard trajectory is calculated, and the angles of key joints are measured in real time. In the temporal dimension, the user's actual movement sequence is aligned and matched with the rhythm of the standard movements using a dynamic time warping algorithm, and the smoothness and continuity of the force exertion phase are evaluated to generate time and spatial deviation values.

[0013] As a preferred embodiment of the AI-powered intelligent correction fitness training method of the present invention, the time and space deviation values ​​are substituted into the comprehensive scoring formula to generate a quantified movement standard score. The state machine is judged according to the preset safety threshold. If the standard score is higher than the qualified line and there is no dangerous posture, it is judged as a standard movement. If there is a slight deviation, it is judged as needing correction, and the specific error feature points are extracted. If a violent shaking of the trajectory or a joint angle exceeding the safety limit is detected, it is immediately judged as a dangerous instability, thereby providing accurate data support for subsequent triggering of voice prompts, screen displays, or emergency locking of equipment.

[0014] In a preferred embodiment of the AI-powered intelligent correction fitness training method of the present invention, when providing voice and visual feedback in step S5, different feedback is triggered based on the judgment result of step S4. If the movement is standard, the system provides voice encouragement through a speaker. If a deviation is detected, a natural language error correction instruction is immediately generated and broadcast in real time. At the same time, the display feedback module outputs visual guidance, superimposing a comparison between the user's posture and the standard posture on the screen, and marking the joints that need adjustment with highlighted arrows. Through the dual closed-loop feedback of hearing and vision, the system accurately guides the user to correct the force exertion pattern, reduces the training risk caused by movement deviation, and improves the consistency between the training movements and the preset specifications.

[0015] A Smith training device includes: The framework module has a central training space inside, and the training execution component is installed inside the central training space. A voice acquisition module, which is installed on the frame module, is used to acquire user voice. The first visual acquisition module is located in at least two different viewing directions in the front, side, top or column area of ​​the frame; The second vision acquisition module is located on the side of the frame module. In conjunction with the first vision module, it identifies the user's posture changes, joint angles, and motion amplitude characteristics in the lateral or frontal viewing direction based on the user's different standing positions and orientations. The central processing module receives the collected video and audio data, processes and analyzes the data, and determines whether the exercise movements are standard. The display feedback module provides voice and visual feedback and guidance to the user based on the judgment result.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By combining frontal and lateral dual visual acquisition modules, this invention can not only acquire two-dimensional projection information and generate multi-view motion features or quasi-three-dimensional posture estimation results for users, but also, with the dynamic time warping algorithm and spatiotemporal dimension comparison with standard models, can keenly capture joint angle deviations and force exertion rhythm problems. This effectively solves the problem of motion assessment in traditional fitness relying on subjective experience and lacking quantitative standards, and greatly reduces the risk of sports injuries.

[0017] 2. This invention combines voice command wake-up and multimodal feedback interaction modes, allowing users to control the device or trigger corrections through natural language without manual operation. During the correction process, dual auditory and visual feedback is used to generate specific error correction instructions through natural language. On the other hand, highlighted arrows on the screen visually indicate the joints that need adjustment. This audiovisual guidance method allows users to instantly understand complex movement deviations, quickly adjust the force exertion mode, and improve the efficiency of individual training. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] Example 1 Please see Figure 1 An AI-powered intelligent corrective fitness training method includes the following steps: S1. Voice command wake-up: Receives and recognizes user voice commands, controls the fitness equipment and corrects fitness postures according to the voice commands. S2, Visual perception coverage, collects frontal and side images of the user, and obtains the location of key points of the user's human body, key point confidence and multi-view motion features; S3. Motion recognition and analysis: Match feature points in frontal and side images to generate multi-view motion features or quasi-3D pose estimation results for the user, and simultaneously identify the motion trajectory of the training actuators inside the fitness equipment. S4. Determination of movement standard: Detect the movement trajectory and determine whether the exercise movement is standard. S5 features intelligent feedback and correction, providing voice and visual feedback based on the results of exercise movements to guide users in performing standard posture exercises.

[0020] In S1, when recognizing speech, the acoustic signal in the central training space is continuously monitored. The speech recognition engine converts the analog audio signal into text data and the natural language understanding algorithm parses the semantics of the instruction. When a wake word is detected, the wake-up mechanism is triggered and the instruction receiving state is entered.

[0021] S1 generates correction or control signals based on the parsing results of voice commands. The control signals adjust the exercise parameters of the training actuators inside the fitness equipment, while the correction signals control the visual acquisition module to capture the user's current movement.

[0022] When acquiring user images in S2, the first vision acquisition module obtains the user's frontal image and two-dimensional plane projection information in the central training space, while the second vision acquisition module obtains the user's lateral posture contour, lateral projection of human body key points, and corresponding confidence levels.

[0023] When processing images in S2, the data acquired by the first vision acquisition module and the second vision acquisition module are fused based on user orientation judgment, multi-view key point recognition and device coordinate system calibration to generate user multi-view motion features or quasi-3D pose estimation results.

[0024] In S3, when matching feature points, the displacement of the training executor in three-dimensional space is calculated through visual recognition, and the motion trajectory and running state of the training executor are constructed by combining the running data of the training executor.

[0025] The operational data of the training device includes equipment data such as barbell slider position, wire encoder, tension sensor, and motor encoder. The equipment data includes the wire movement length, wire movement speed, wire force data, and motor resistance.

[0026] When S4 makes standard judgments on exercise movements, it retrieves the movement judgment rules or movement feature configuration table corresponding to the target training movement from the cloud or local database. In the spatial dimension, it calculates the distance deviation between the user's 3D motion trajectory and the standard trajectory, and measures the angles of key joints in real time. In the temporal dimension, it uses a dynamic time warping algorithm to align and match the user's actual movement sequence with the rhythm of the standard movement, evaluates whether the force exertion phase is smooth and continuous, and generates time and spatial deviation values.

[0027] Substituting the time and space deviation values ​​into the comprehensive scoring formula generates a quantified motion standard score. Based on the preset safety threshold, a state machine is used for judgment. If the standard score is higher than the passing line and there is no dangerous posture, it is judged as a standard motion. If there is a slight deviation, it is judged as needing correction, and specific error feature points are extracted. If severe trajectory shaking or joint angle exceeding the safety limit is detected, it is immediately judged as dangerous instability, thus providing accurate data support for subsequent triggering of voice prompts, screen displays, or emergency locking of the device.

[0028] When S5 provides voice and visual feedback, it triggers different feedback based on the judgment result of S4. If the movement is standard, the system provides voice encouragement through the speaker. If a deviation is detected, it immediately generates natural language error correction instructions for real-time broadcast. At the same time, the display feedback module outputs visual guidance, overlaying the comparison between the user's posture and the standard posture on the screen and marking the joints that need adjustment with highlighted arrows. Through the dual closed-loop feedback of hearing and vision, it accurately guides the user to correct the force application pattern, reduces the training risk caused by movement deviation, and improves the consistency between training movements and preset specifications.

[0029] By combining frontal and lateral dual visual acquisition modules, it can not only acquire two-dimensional projection information and generate multi-view motion features or quasi-three-dimensional posture estimation results for users, but also, with the dynamic time warping algorithm and spatiotemporal dimension comparison with standard models, it can keenly capture joint angle deviations and force exertion rhythm problems. This effectively solves the problem of motion assessment relying on subjective experience and lacking quantitative standards in traditional fitness, and greatly reduces the risk of sports injuries.

[0030] By combining voice command wake-up and multimodal feedback interaction modes, users can control the device or trigger corrections through natural language without manual operation. During the correction process, dual auditory and visual feedback is used to generate specific error correction instructions through natural language. On the other hand, highlighted arrows on the screen visually indicate the joints that need adjustment. This audiovisual guidance method allows users to instantly understand complex movement deviations, quickly adjust their force application patterns, and improve the efficiency of individual training.

[0031] Example 2 A Smith training device includes: The framework module has a central training space inside, and the training execution component is installed inside the central training space. A voice acquisition module, which is installed on the frame module, is used to acquire user voice. The first visual acquisition module is located in at least two different viewing directions in the front, side, top or column area of ​​the frame; The second vision acquisition module is located on the side of the frame module. In conjunction with the first vision module, it identifies the user's posture changes, joint angles, and motion amplitude characteristics in the lateral or frontal viewing direction based on the user's different standing positions and orientations. The central processing module receives the collected video and audio data, processes and analyzes the data, and determines whether the exercise movements are standard. The display feedback module provides voice and visual feedback and guidance to the user based on the judgment result.

[0032] Furthermore, the training components include a slide barbell assembly and a sliding steel wire assembly.

[0033] Furthermore, the slide barbell assembly includes a barbell bar, a counterweight assembly, a safety locking mechanism, and an electronically controlled resistance adjustment module.

[0034] Furthermore, the sliding steel wire assembly includes a first steel wire assembly and a second steel wire assembly disposed on the left and right sides of the frame module. The first steel wire assembly and the second steel wire assembly have sliding outlet ends and can slide along the height direction or the width direction.

[0035] Furthermore, the central training space has a width of 2.2-2.8m and a height of 1.8-2.2m, and the frame module is a steel structure or an aluminum alloy structure.

[0036] Furthermore, the voice acquisition module consists of one or more microphones.

[0037] Furthermore, the height of the first visual acquisition module is 1.5-2.0m, the height of the second visual acquisition module is 1.2-1.8m, and the installation angle is 30°-60°.

[0038] Furthermore, the display feedback module can be one or more of a display screen, a speaker, an indicator light, and a vibration device.

[0039] By constructing a multi-dimensional visual and voice acquisition system and combining it with real-time data analysis from the central processing module, intelligent judgment and accurate feedback of training action standards are achieved. The top first visual acquisition module and the side second visual acquisition module work together to generate multi-view motion features or quasi-three-dimensional posture estimation results for users. With the auxiliary analysis of voice data, the comprehensiveness and accuracy of action posture recognition are ensured.

[0040] The feedback module provides dual guidance through voice and vision based on the judgment results, transforming the traditional post-training debriefing into real-time data-driven error correction. This closed-loop design not only effectively enhances the value of a single training session but also helps users establish correct movement patterns, significantly reducing the risk of injury caused by movement distortion or imbalance.

[0041] In terms of hardware structure and spatial adaptation, it integrates a sliding barbell assembly and a multi-directional sliding steel wire assembly, and is equipped with an electronically controlled resistance adjustment and safety locking mechanism, taking into account both the diversity and safety of strength training. The cable outlet of the sliding steel wire assembly can be flexibly adjusted along the height and width directions to achieve isolation and compound training of multiple muscle groups throughout the body.

[0042] The steel or aluminum alloy frame ensures the stability of the equipment under high-intensity use, while the camera installation height and angle ensure no blind spots in monitoring and minimize visual interference and physical interference during training.

[0043] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-powered intelligent corrective fitness training method, characterized in that, Includes the following steps: S1. Voice command wake-up: Receives and recognizes user voice commands, controls the fitness equipment and corrects fitness postures according to the voice commands. S2, Visual perception coverage, collects frontal and side images of the user, and obtains the location of key points of the user's human body, key point confidence and multi-view motion features; S3. Motion recognition and analysis: Match feature points in frontal and side images to generate multi-view motion features or quasi-3D pose estimation results for the user, and simultaneously identify the motion trajectory of the training actuators inside the fitness equipment. S4. Determination of movement standard: Detect the movement trajectory and determine whether the exercise movement is standard. S5 features intelligent feedback and correction, providing voice and visual feedback based on the results of exercise movements to guide users in performing standard posture exercises.

2. The AI-powered intelligent corrective fitness training method according to claim 1, characterized in that: In S1, when recognizing speech, the acoustic signal in the central training space is continuously monitored. The speech recognition engine converts the analog audio signal into text data and the natural language understanding algorithm parses the semantics of the instruction. When a wake word is detected, the wake-up mechanism is triggered and the instruction receiving state is entered.

3. The AI-powered intelligent corrective fitness training method according to claim 2, characterized in that: In step S1, a correction signal or a control signal is generated based on the parsing result of the voice command. The control signal adjusts the exercise parameters of the training actuators inside the fitness equipment, and the correction signal controls the visual acquisition module to capture the user's current action image.

4. The AI-powered intelligent corrective fitness training method according to claim 1, characterized in that: When acquiring user images in S2, the first visual acquisition module obtains the user's frontal image and two-dimensional plane projection information in the central training space, and the second visual acquisition module obtains the user's lateral posture contour, the lateral projection of human body key points, and the corresponding confidence level.

5. The AI-powered intelligent corrective fitness training method according to claim 4, characterized in that: In S2, when processing the image, the data obtained by the first vision acquisition module and the second vision acquisition module are fused based on user orientation judgment, multi-view key point recognition and device coordinate system calibration to generate user multi-view motion features or quasi-three-dimensional pose estimation results.

6. The AI-powered intelligent corrective fitness training method according to claim 1, characterized in that: In step S3, when matching feature points, the displacement of the training executor in three-dimensional space is calculated through visual recognition, and the motion trajectory and running state of the training executor are constructed by combining the running data of the training executor.

7. The AI-powered intelligent corrective fitness training method according to claim 1, characterized in that: In step S4, when making a standard judgment on the exercise movement, the movement judgment rules or movement feature configuration table corresponding to the target training movement are retrieved from the cloud or local database. In the spatial dimension, the distance deviation between the user's 3D motion trajectory and the standard trajectory is calculated, and the angles of key joints are measured in real time. In the temporal dimension, the dynamic time warping algorithm is used to align and match the user's actual movement sequence with the rhythm of the standard movement, evaluate whether the force exertion phase is smooth and continuous, and generate time and spatial deviation values.

8. The AI-powered intelligent corrective fitness training method according to claim 7, characterized in that: Substituting the time and space deviation values ​​into the comprehensive scoring formula generates a quantified motion standard score. Based on the preset safety threshold, a state machine is used for judgment. If the standard score is higher than the passing line and there is no dangerous posture, it is judged as a standard motion. If there is a slight deviation, it is judged as needing correction, and specific error feature points are extracted. If severe trajectory shaking or joint angle exceeding the safety limit is detected, it is immediately judged as dangerous instability, thus providing accurate data support for subsequent triggering of voice prompts, screen displays, or emergency locking of the device.

9. The AI-powered intelligent corrective fitness training method according to claim 1, characterized in that: When S5 provides voice and visual feedback, different feedback is triggered based on the judgment result of S4. If the action is standard, the system provides voice encouragement through the speaker. If a deviation is detected, a natural language error correction instruction is immediately generated and broadcast in real time. At the same time, the display feedback module outputs visual guidance, superimposing a comparison between the user's posture and the standard posture on the screen, and marking the joints that need adjustment with highlighted arrows. Through the dual closed-loop feedback of hearing and vision, the system accurately guides the user to correct the force application pattern, reduces the training risk caused by movement deviation, and improves the consistency between the training movements and the preset specifications.

10. A Smith training device, characterized in that, include: The framework module has a central training space inside, and the training execution component is installed inside the central training space. A voice acquisition module, which is installed on the frame module, is used to acquire user voice. The first visual acquisition module is located in at least two different viewing directions in the front, side, top or column area of ​​the frame; The second vision acquisition module is located on the side of the frame module. In conjunction with the first vision module, it identifies the user's posture changes, joint angles, and motion amplitude characteristics in the lateral or frontal viewing direction based on the user's different standing positions and orientations. The central processing module receives the collected video and audio data, processes and analyzes the data, and determines whether the exercise movements are standard. The display feedback module provides voice and visual feedback and guidance to the user based on the judgment result.

Images