Intelligent teaching swimming goggles based on multi-sensor cooperation and AI posture recognition

Abstract

The invention provides intelligent teaching swimming goggles based on multi-sensor cooperation and AI posture recognition, and relates to the technical field of intelligent sports equipment. According to the system, multi-dimensional swimming data, including swimming video streams, motion states, physiological state data and swimming tracks, of students are collected through a multi-sensor module; the data processing and AI analysis unit extracts swimming posture features and generates action correction suggestions and training prompts; a voice interaction module of the communication module broadcasts in a voice form and performs action correction suggestions and training prompts in a mirror front projection display form; and the central control module is deployed on a shore client, so that a coach can check data and carry out teaching and health monitoring management. The problems that in the prior art, a coach cannot capture the underwater posture and physiological state of a student in real time, the timeliness of guidance and interaction is poor, and the student cannot conveniently receive an action correction instruction underwater are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent sports equipment technology, and in particular to an intelligent teaching swimming goggle based on multi-sensor collaboration and AI posture recognition. Background Technology

[0002] In the field of swimming instruction, the traditional teaching model mainly relies on coaches visually observing students' swimming postures and then providing guidance through verbal commands or gestures. Limited by factors such as visual obstructions in the underwater environment, the coach's focus, and the angle of observation, coaches struggle to accurately capture subtle deviations in students' movements and cannot monitor students' physiological states in real time. This can easily lead to problems such as untimely correction of movements and a lack of targeted guidance.

[0003] Meanwhile, during underwater training, trainees cannot easily receive guidance from instructors and usually need to interrupt training to surface to get feedback, which not only affects the continuity of training but also reduces teaching efficiency. Summary of the Invention

[0004] The embodiments of the present invention provide an intelligent teaching swimming goggle based on multi-sensor collaboration and AI posture recognition, which aims to solve the problems of existing technologies where coaches cannot capture students' underwater posture and physiological state in real time, resulting in poor timeliness of guidance and interaction, and students cannot easily receive movement correction instructions underwater.

[0005] To achieve the above objectives, this invention provides an intelligent teaching swimming goggle based on multi-sensor collaboration and AI posture recognition, comprising a multi-sensor module with communication connectivity, a data processing and AI analysis unit, a communication module, a power supply module, a central control module, and a projection module; wherein: The multi-sensor module includes a miniature camera, an inertial sensor array, and a health monitoring sensor, used to collect multi-dimensional swimming data from the trainees; the multi-dimensional swimming data includes swimming video stream data, motion state data, physiological state data, and swimming trajectory. The data processing and AI analysis unit is used to extract swimming posture features of trainees based on the multidimensional swimming data, and generate targeted movement correction suggestions and training tips; The communication module includes a voice interaction module, which is integrated into the strap of the swimming goggles and is used to broadcast the movement correction suggestions and training prompts in voice form. The projection module is integrated into the surface of the eyeglass lens and is used to project the movement correction suggestions and training prompts onto the surface of the eyeglass lens. The central control module, deployed on the shore client, is used to receive the multidimensional swimming data and display it to the swimming coaches on shore; the swimming coaches use the received data for teaching management and / or health monitoring management.

[0006] Furthermore, the communication module also includes an SOS distress module, which is used to control the voice interaction module to broadcast an alarm when the physiological state data exceeds a preset safety threshold, and transmit the alarm to the central control module through a preset wireless communication module.

[0007] Furthermore, the miniature camera is positioned on the left or right front edge of the beam or frame of the swimming goggles, with its optical axis at a certain downward angle to the direction of the swimmer's swimming movement, and is used to collect video data of the swimmer's upper body, including arm strokes and breathing and lateral turning movements.

[0008] Furthermore, the inertial sensor group includes a gyroscope and an accelerometer, which are encapsulated in a sealed cavity within the swimming goggle body. It is used to collect three-dimensional angular velocity and three-dimensional acceleration data of the swimming goggle body during swimming, as the motion state data.

[0009] Furthermore, the physiological data includes heart rate data and blood oxygen data, which are acquired through a wrist-worn device worn by the student and transmitted to the smart teaching swimming goggles via wireless communication.

[0010] Furthermore, the step of extracting swimming posture features from the multidimensional swimming data and generating targeted movement correction suggestions and training prompts includes: The swimming video stream data is input into a predetermined key point detection model to extract the movement trajectory of key points on the trainee's body parts in real time. The motion state data and the motion trajectory of the key points are spatiotemporally aligned and fused to obtain the student's posture feature data; The student's posture feature data is input into a predetermined swimming motion classification and evaluation model, which outputs the current student's swimming motion category and compares it with the standard posture template of the swimming motion category to obtain the degree of posture deviation. The degree of posture deviation, physiological state data, and swimming trajectory are input into a preset correction strategy library, and targeted movement correction suggestions and training prompts are generated.

[0011] Furthermore, the key point detection model is configured as a human pose estimation model based on a convolutional neural network, used to extract two-dimensional or three-dimensional coordinates of predetermined key points of the upper body and arm parts, including the head, shoulders, elbows and wrists, from a single frame video image.

[0012] Furthermore, the spatiotemporal alignment and data fusion include: Synchronize the motion state data with the motion trajectory of the key points using timestamps; The extended Kalman filter is used to fuse the timestamp-synchronized data to obtain smooth and spatiotemporally consistent student posture feature data.

[0013] Furthermore, the swimming motion classification and evaluation model is a motion recognition model based on a temporal neural network. Its input is the time series of the student's posture feature data, and its output is a predetermined swimming motion category and a confidence score for that category. The degree of posture deviation is calculated by measuring the difference between the student's posture feature data and the corresponding standard posture motion template in terms of key point coordinates, angles, or trajectories.

[0014] Furthermore, the correction strategy library, based on the input degree of posture deviation, physiological state data, and swimming trajectory, outputs movement correction suggestions and training prompts including voice prompts, demonstrations of corrective movements, and / or adjustments to training intensity.

[0015] The above technical solution has the following technical effects: The smart swimming goggles of this invention include a multi-sensor module, a data processing and AI analysis unit, a communication module, a power supply module, and a central control module. The multi-sensor module collects multi-dimensional swimming data from the swimmer, including swimming video streams, motion state, and physiological state data. The data processing and AI analysis unit extracts swimming posture features and generates targeted movement correction suggestions and training prompts. The voice interaction module of the communication module is integrated into the temple end, broadcasting movement correction suggestions and training prompts in voice format and displaying them for viewing. The central control module is deployed on a shore-based client, allowing coaches to view data and conduct teaching and health monitoring management. This invention solves the problems of existing technologies where coaches cannot capture swimmers' underwater posture and physiological state in real time, resulting in poor timeliness of guidance and interaction, and swimmers cannot easily receive movement correction instructions underwater. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of an intelligent teaching swimming goggle based on multi-sensor collaboration and AI posture recognition according to an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the various embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, primarily used to illustrate the embodiments, and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementations and the advantages of the present invention. Components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0018] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.

[0019] Figure 1 This is a schematic diagram of the structure of an intelligent teaching swimming goggle based on multi-sensor collaboration and AI posture recognition according to an embodiment of the present invention, as shown below. Figure 1As shown, the system includes a multi-sensor module with communication connectivity, a data processing and AI analysis unit, a communication module, a power supply module, a central control module, and a projection display module; wherein: The multi-sensor module includes a miniature camera, an inertial sensor array, and a health monitoring sensor to collect multi-dimensional swimming data from the swimmer. In one specific implementation, the multi-dimensional swimming data includes swimming video stream data, motion state data, physiological state data, and swimming trajectory. The miniature camera is positioned on the central beam or the left and right edges of the goggles, with its optical axis at a downward angle to the swimmer's swimming direction. It is primarily used to collect video data of the swimmer's upper body, including arm strokes and breathing / lateral turning movements. Since the swimmer's head is typically tilted slightly downwards during swimming, the downward-facing optical axis can more directly align with the arm stroke area, reducing water reflection interference and improving the usability of the arm movement images. The camera primarily collects visual data of the swimmer's upper body, including arm stroke trajectories, shoulder rotation, and head lateral turning during breathing. The inertial sensor array, including a gyroscope and accelerometer, is encapsulated within a sealed cavity of the goggles. It collects three-dimensional angular velocity and three-dimensional acceleration data of the goggles during swimming, serving as motion state data. The gyroscope measures the angular velocity of the goggles' rotation around three axes in real time, reflecting head rotation posture. The accelerometer measures the three-axis linear acceleration, capturing head movement and impact changes. In swimming motion analysis, angular velocity data quantifies the amplitude and frequency of head rotation, used to analyze breathing rhythm and body roll coordination. Acceleration data, combined with time integration, can calculate the head movement trajectory, which, along with video data, verifies the continuity of the movement. Because the sensors are rigidly connected to the goggles, their data directly represent the head movement state and can serve as a benchmark reference point for body posture estimation. Health monitoring sensors are implemented through external linkage. The system does not directly integrate optical or electrical physiological sensors inside the goggles; instead, it receives data collected by an independent smart wearable device worn by the swimmer via a wireless communication module. In practice, the fitness tracker on the student's wrist continuously collects physiological parameters such as heart rate and blood oxygen, and establishes a data transmission connection with the swimming goggles via Bluetooth Low Energy protocol.

[0020] The communication module includes a voice interaction module, integrated into the strap of the swimming goggles, used to broadcast movement correction suggestions and training prompts in voice form; In one specific implementation, the communication module also includes an SOS distress module, used to control the voice interaction module to broadcast an alarm when physiological status data exceeds a preset safety threshold, and to transmit the alarm to the central control module via a pre-set wireless communication module. This module continuously compares real-time heart rate, blood oxygen, and other indicators with pre-set safety ranges. When a persistent abnormality is detected, such as excessively high heart rate or excessively low blood oxygen, the alarm mechanism is triggered. After the alarm is activated, two parallel actions are performed: first, a pre-recorded warning message is immediately broadcast to the student via the voice interaction module, indicating the abnormal state and suggesting stopping exercise; second, a distress signal transmission program is initiated via the wireless communication module. This program sends a standardized distress data packet containing the student's unique identification code, abnormal physiological data, and a timestamp to a preset emergency contact terminal or cloud monitoring platform via the wireless communication module. This protection mechanism achieves a two-layer response from local alarm to remote assistance, ensuring timely external rescue when the student is unable to effectively cope with a dangerous situation. The system supports manual triggering; the student can force the entire alarm and distress process to start by long-pressing a specific button.

[0021] The projection module is integrated into the surface of the eyeglass lens and is used to project movement correction suggestions and training prompts onto the surface of the eyeglass lens; In this embodiment, the projection module is an augmented reality (AR) display module integrated into the surface of the eyeglass lens. Its core function is to overlay virtual information onto the user's natural field of vision to provide real-time motion correction guidance and training prompts. Specifically, the module uses miniaturized projection units or optical waveguide technology to directly project algorithmically generated graphics, text, or symbol information onto the lens surface, forming intuitive visual feedback. Once a motion deviation is detected, the system immediately presents correction suggestions through the AR display module on the lens in a non-invasive manner, such as highlighted arrows, dynamic trajectory lines, floating text descriptions, or graphic overlays, at an appropriate location in the user's field of vision (usually at the edge of the field of vision or next to the target object).

[0022] The central control module is deployed on the shore client to receive multi-dimensional swimming data and display it to the swimming coaches on shore; the swimming coaches use the received data for teaching management and / or health monitoring management.

[0023] In one specific implementation, the central control module is deployed on the shore client and serves as the core hub connecting students' underwater training with instructors' shore guidance. Its core function is to achieve centralized reception, visualization, and implementation of teaching management of training data.

[0024] This module acquires multi-dimensional data, including video streams, movement status, and physiological status, from a multi-sensor module on the swimmer's goggles via wireless communication. This data is then presented to the coach in an intuitive way, overcoming the limitations of traditional coach-based visual observation. Coaches can view details of the swimmer's swimming posture and the trajectory of their movements in real time via a client application, while also monitoring physiological indicators such as heart rate. This allows for accurate assessment of movement deviations and timely monitoring of the swimmer's physical condition, preventing sports injuries.

[0025] Based on this data, coaches can carry out dual management: at the teaching management level, they can develop personalized training plans for trainees by combining AI-generated movement correction suggestions; at the health monitoring management level, they can adjust training intensity and manage emergency alarms based on changes in physiological data.

[0026] In one specific implementation, this module supports the synchronous management of data from multiple students, thereby improving the teaching efficiency of coaches.

[0027] The data processing and AI analysis unit is used to extract swimming posture features from multidimensional swimming data and generate movement correction suggestions and training tips, including: Swimming video stream data is input into a predetermined keypoint detection model to extract the motion trajectories of key points on the swimmer's body in real time. In one specific implementation, the keypoint detection model is configured as a human pose estimation model based on a convolutional neural network, used to extract the two-dimensional or three-dimensional coordinates of predetermined key points of the upper body and arms, including the head, shoulders, elbows, and wrists, from a single frame of video image. In another specific implementation, the model structure is optimized for the swimming scenario, focusing on identifying key anatomical points of the upper body and arms. After processing each frame, a set of coordinate data is output, corresponding to the positions of the head, both shoulder joints, elbow joints, and wrist joints in the image coordinate system. Through the coordinate sequence of consecutive frames, the system constructs the motion trajectory of the key points in the time dimension. This trajectory reflects the spatial motion path and velocity changes of each joint of the upper limbs during swimming. The two-dimensional image coordinates output by the model can be fused with camera calibration parameters and inertial sensor data to partially achieve three-dimensional spatial position estimation. This processing stage transforms the raw video pixel data into structured motion representation data, providing quantitative input for subsequent spatiotemporal alignment, action classification, and deviation analysis.

[0028] Motion state data and keypoint motion trajectories are spatiotemporally aligned and fused to obtain trainee posture feature data; in one specific implementation, spatiotemporal alignment and data fusion include: Timestamp synchronization: The system marks the angular velocity and acceleration data collected by the inertial sensor and the video frame data collected by the camera with precise timestamps, and synchronizes the two types of data streams through time interpolation algorithm to ensure that the motion state and visual features correspond strictly in the time dimension; An extended Kalman filter (EPF) is employed for data fusion: the filter uses the 2D coordinates of visual keypoints as observations and the attitude changes calculated from inertial sensor data using a kinematic model as state predictions. During the filtering iteration process, the system continuously corrects the attitude estimation and outputs a smoothed 3D keypoint motion trajectory. This fusion method effectively addresses the limitations of a single data source, namely, the potential loss or jumps in visual data during rapid movement or water splash occlusion, and the cumulative errors in inertial data. Through Kalman filter fusion, the system obtains spatiotemporally consistent, smooth, and continuous student attitude feature data, providing reliable input for subsequent action recognition. The nonlinear processing capability of the EPF is particularly suitable for complex 3D movements such as swimming.

[0029] The model inputs the student's posture feature data into a predefined swimming motion classification and evaluation model, outputs the current student's swimming motion category, and compares it with the standard posture template of the swimming motion category to obtain the degree of posture deviation. In one specific implementation, the swimming motion classification and evaluation model is a motion recognition model based on a temporal neural network. Its input is time-series student posture feature data, and its output is a predetermined swimming motion category and a confidence score for that category. The degree of posture deviation is calculated by measuring the difference between the student's posture feature data and the corresponding standard posture motion template in terms of key point coordinates, angles, or trajectories.

[0030] This action recognition algorithm, based on a temporal neural network, is specifically designed to handle multi-frame posture data in time-series format. The model's workflow consists of two stages: The first stage performs action classification. The temporal neural network analyzes the evolution patterns of posture features over time, identifying the corresponding swimming action categories, such as the freestyle arm stroke, and outputs a confidence score for each recognition result, reflecting the reliability of the judgment. The second stage performs posture evaluation. The system calls upon a standard posture template corresponding to the identified action category. This template is typically based on data from professional athletes and includes standardized data such as the trajectory of key points and the range of joint angle changes under ideal conditions. By comparing the trainee's actual posture feature data frame-by-frame with the standard template, the system calculates the specific differences in coordinate positions, joint angles, or trajectories of key points. These differences are normalized and weighted, ultimately quantified into comparable posture deviation indicators, such as an elbow lift angle of less than 15 degrees or an outward deviation of more than 10 centimeters in the stroke trajectory. This quantitative evaluation process enables the system to go beyond simple action recognition and achieve precise diagnosis of technical details. The deviation data will serve as the direct basis for generating specific corrective suggestions. Combined with the trainee's real-time physiological data, it will determine whether the deviation is due to improper technique or insufficient physical fitness, thus providing more targeted teaching guidance. The application of temporal neural networks enables the model to capture the complete dynamic process of swimming movements, rather than focusing only on single-frame postures. This is crucial for assessing time-related characteristics such as movement rhythm and coordination.

[0031] The system inputs the degree of posture deviation, physiological state data, and swimming trajectory into a pre-defined correction strategy library, and then matches and generates targeted movement correction suggestions and training prompts. In one specific implementation, the correction strategy library outputs movement correction suggestions and training prompts based on the input degree of posture deviation, physiological state data, and swimming trajectory, including voice prompts, demonstrations of corrective movements, and / or adjustments to training intensity.

[0032] In this embodiment, the quantified posture deviation and real-time physiological state data are input into a preset correction strategy library. This strategy library is essentially a structured decision rule database or a lightweight inference model, storing various common incorrect postures, their corresponding correction methods, and training adjustment suggestions under different physiological states.

[0033] The system first retrieves corresponding correction entries based on the specific type of postural deviation, such as elbow drop or improper breathing timing. Then, it combines real-time heart rate or blood oxygen data to determine the trainee's current exercise load and adjust the training accordingly. For example, if the heart rate is within the safe aerobic range, the system will match voice prompts and breakdown demonstrations focusing on technical details. If the heart rate is close to or exceeds a preset threshold, the strategy library prioritizes outputting safety guidance suggestions such as reducing training intensity and adjusting breathing rhythm. The generated correction suggestions and training prompts are ultimately output in structured data format, including three optional feedback dimensions: first, the specific text content of the voice prompts, used to drive the voice synthesis module; second, standard movement demonstration animations or illustrations that can be accessed on the trainee's interface; and third, intensity and frequency adjustment parameters for subsequent training plans. The system supports dynamic optimization of the strategy library based on the trainee's historical correction effects. For example, if a correction suggestion has been adopted multiple times without improvement, it will automatically try to match alternative correction schemes.

[0034] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A smart teaching swimming goggle based on multi-sensor collaboration and AI posture recognition, characterized in that... It includes a multi-sensor module for communication connectivity, a data processing and AI analysis unit, a communication module, a power supply module, a central control module, and a projection display module; among which: The multi-sensor module includes a miniature camera, an inertial sensor array, and a health monitoring sensor, used to collect multi-dimensional swimming data from the trainees; the multi-dimensional swimming data includes swimming video stream data, motion state data, physiological state data, and swimming trajectory. The data processing and AI analysis unit is used to extract swimming posture features of trainees based on the multidimensional swimming data, and generate targeted movement correction suggestions and training tips; The communication module includes a voice interaction module, which is integrated into the strap of the swimming goggles and is used to broadcast the movement correction suggestions and training prompts in voice form. The projection module is integrated into the surface of the eyeglass lens and is used to project the movement correction suggestions and training prompts onto the surface of the eyeglass lens. The central control module, deployed on the shore client, is used to receive the multidimensional swimming data and display it to the swimming coaches on shore; the swimming coaches use the received data for teaching management and / or health monitoring management.

2. The intelligent teaching swimming goggles according to claim 1, characterized in that... The communication module also includes an SOS distress module, which is used to control the voice interaction module to broadcast an alarm when the physiological state data exceeds a preset safety threshold, and to transmit the alarm to the central control module through a preset wireless communication module.

3. The intelligent teaching swimming goggles according to claim 1, characterized in that... The miniature camera is located on the left or right front edge of the beam or frame of the swimming goggles, with its optical axis at a certain downward angle to the direction of the swimmer's swimming movement. It is used to collect video data of the swimmer's upper body, including arm strokes and breathing and lateral turning movements.

4. The intelligent teaching swimming goggles according to claim 1, characterized in that... The inertial sensor group includes a gyroscope and an accelerometer, which are encapsulated in a sealed cavity within the swimming goggle body. It is used to collect the three-dimensional angular velocity and three-dimensional acceleration data of the swimming goggle body during swimming, as the motion state data.

5. The intelligent teaching swimming goggles according to claim 1, characterized in that... The physiological data, including heart rate and blood oxygen data, is acquired through a wrist-worn device worn by the student and transmitted wirelessly to the smart teaching swimming goggles.

6. The intelligent teaching swimming goggles according to claim 1, characterized in that... The step of extracting swimming posture features from the multidimensional swimming data and generating targeted movement correction suggestions and training tips includes: The swimming video stream data is input into a predetermined key point detection model to extract the movement trajectory of key points on the trainee's body parts in real time. The motion state data and the motion trajectory of the key points are spatiotemporally aligned and fused to obtain the student's posture feature data; The student's posture feature data is input into a predetermined swimming motion classification and evaluation model, which outputs the current student's swimming motion category and compares it with the standard posture template of the swimming motion category to obtain the degree of posture deviation. The degree of posture deviation, physiological state data, and swimming trajectory are input into a preset correction strategy library, and targeted movement correction suggestions and training prompts are generated.

7. The intelligent teaching swimming goggles according to claim 6, characterized in that... The key point detection model is configured as a human pose estimation model based on a convolutional neural network, used to extract two-dimensional or three-dimensional coordinates of predetermined key points of the upper body and arms, including the head, shoulders, elbows and wrists, from a single frame video image.

8. The intelligent teaching swimming goggles according to claim 6, characterized in that... The spatiotemporal alignment and data fusion include: Synchronize the motion state data with the motion trajectory of the key points using timestamps; The extended Kalman filter is used to fuse the timestamp-synchronized data to obtain smooth and spatiotemporally consistent student posture feature data.

9. The intelligent teaching swimming goggles according to claim 6, characterized in that... The swimming motion classification and evaluation model is a motion recognition model based on a temporal neural network. Its input is the time series of the student's posture feature data, and its output is a predetermined swimming motion category and a confidence score for that category. The degree of posture deviation is calculated by measuring the difference between the student's posture feature data and the corresponding standard posture motion template in terms of key point coordinates, angles, or trajectories.

10. The intelligent teaching swimming goggles according to claim 6, characterized in that... The correction strategy library, based on the input degree of posture deviation, physiological state data, and swimming trajectory, outputs movement correction suggestions and training prompts, including voice prompts, demonstrations of corrective movements, and / or adjustments to training intensity.

Images