A teaching audio transmission, data collection and analysis system and method

Abstract

The application discloses a kind of teaching audio transmission, data acquisition analysis system and method, belong to the field of sports teaching auxiliary.The system includes: student end, trainer end, cloud server, compared with prior art, the system is through the cooperation of student end, trainer end, cloud server, after receiving the data collected by student end in trainer end, can be carried out real-time AI processing analysis by data analysis display module, give the deficiency in training at that time and improvement suggestion, and the collected data is stored to the database of cloud, can analyze the training improvement of student long-term or a time period, so as to be able to provide help to the training of student, meet the demand of existing sports teaching;The method is handled by trainer end to the collected student data processing, can identify the state of student in movement, facilitate the targeted guidance of trainer to student.

Description

Technical Field

[0001] This invention belongs to the field of sports teaching aids, and specifically relates to a teaching audio transmission, data acquisition and analysis system and method. Background Technology

[0002] In teaching sports such as swimming, diving, and ball games, coaches need to transmit exercise instructions to students and collect data during the exercise process via remote communication. Traditionally, this involves transmitting instructions through audio broadcasting systems and using simple sensing systems to collect and display data. This method of transmitting instructions and collecting data is too simplistic. After collecting data at the terminal, it lacks both backend cloud servers for data classification and storage, and it cannot utilize AI for data analysis to provide corresponding suggestions. In conclusion, the existing methods of instruction transmission and data collection are no longer sufficient to meet the needs of modern sports instruction. Summary of the Invention

[0003] To address the aforementioned issues, the primary objective of this invention is to provide a teaching audio transmission and data acquisition and analysis system. Through the collaboration of the student end, the coach end, and the cloud server, after the coach receives the data collected from the student end, it can perform real-time AI processing and analysis through the data analysis and display module, providing feedback on shortcomings and improvement suggestions during training. The collected data is then stored in a cloud database, allowing for analysis of the student's long-term or periodic training improvement, thereby providing assistance to the student's training and meeting the needs of existing sports teaching.

[0004] Another objective of this invention is to provide a teaching audio transmission and data acquisition and analysis method that can process the collected student data through the coach's end, identify the student's state during exercise, and facilitate the coach to provide targeted guidance to the student.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: This invention provides a teaching audio transmission and data acquisition and analysis system, comprising: The student-side module includes an audio receiving and playback module, a data acquisition module, a data filtering module, a preliminary data analysis module, an internal storage module, and a data transmission module. The coach's end includes an audio transmission module, a data receiving module, a data storage module, a data analysis and display module, and a network transmission module; Cloud servers, including database modules; The audio transmission module interacts with the audio receiving and playback module to transmit instructions from the coach's end to the student's end. The data acquisition module, data filtering module, preliminary data analysis module, internal storage module, and data transmission module are connected in sequence, and the data transmission module interacts with the data receiving module to transmit the data collected from the student end to the coach end after filtering, analysis, and storage. The data receiving module, data storage module, data analysis and display module, and network transmission module are connected in sequence. The network transmission module interacts with the database module to store the data received by the coach terminal in the cloud database after AI analysis and processing.

[0006] The interaction flow of the system in this application is as follows: 1. During training, trainees wear headphones, smart swimming goggles, smart safety goggles, and other trainee equipment, while instructors carry tablets as their training equipment; 2. During training, the coach mixes the pre-set training audio and instructions and transmits them to the trainees through the audio transmission module on the coach's equipment. The coach can then provide guidance to all or some trainees in groups at any time and conduct relevant training. 3. Trainees receive instructions from the coach through the audio receiving and playback module. The data collected in real time by the data acquisition module is analyzed and processed by relevant algorithms, and then transmitted to the coach's end in real time or temporarily stored through the data transmission module. 4. During the training process or at specific times, the coach collects the trainees' training data. After the data analysis and display module analyzes the data through AI, it is summarized locally and processed by the algorithm to generate intuitive and concise training analysis graphs such as line graphs, bars, integrals, and areas that can quickly show the current training process, or other data tables. The coach also proposes improvement methods and suggestions based on the AI ​​analysis results to guide continuous improvement in subsequent training. 5. Coaches can choose to store student data locally or transfer it to a cloud server.

[0007] Compared to existing technologies, the teaching audio transmission and data acquisition and analysis system of this application, through the cooperation of the student end, the coach end, and the cloud server, can perform real-time AI processing and analysis after the coach receives the data collected from the student end, and provide shortcomings and improvement suggestions in the training at that time through the data analysis and display module. The collected data is stored in the cloud database, which can analyze the training improvement of students over a long period of time or a certain period of time, thereby providing assistance to students' training and meeting the needs of existing sports teaching.

[0008] Furthermore, the student terminal is presented through student devices, which are one of the following: behind-the-ear headphones (bone conduction, air conduction, or hybrid conduction), smart swimming goggles, or smart protective goggles.

[0009] Furthermore, the audio receiving and playback module uses one of the following transmission methods—Bluetooth, 2.4G, FM, UHF, VHF, or WIFI—to receive voice commands and background music from the coach terminal.

[0010] Furthermore, the data acquisition module collects student data including: heart rate, blood oxygen, orientation, speed, angle, acceleration, geomagnetism, pressure, light, and altitude.

[0011] Furthermore, the internal storage module includes one or more of flash memory and SD cards.

[0012] Furthermore, the data transmission module adopts one of the following transmission methods: LoRa, BLE, ANT+, ZigBee, and WIFI.

[0013] Furthermore, the audio transmission module adopts one of the following transmission methods: Bluetooth, 2.4G, FM, UHF, VHF, and WIFI.

[0014] Furthermore, the data receiving module adopts one of the following transmission methods: LoRa, BLE, ANT+, ZigBee, and WIFI.

[0015] Furthermore, the data analysis and display module includes an AI algorithm unit and an APP or mini-program, which can analyze and process the data obtained by the coach through AI algorithms and display it through the APP or mini-program.

[0016] Furthermore, the network transmission module is implemented using either MQTT or HTTP.

[0017] Furthermore, the coaching end is presented through a coaching device, which consists of a tablet computer, a transceiver module connected to the tablet computer, and an APP. The coaching end uses the tablet computer as the central device. The coach connects to the tablet computer via a Bluetooth headset. The transceiver module can be an FM transmitter connected to the tablet computer via a USB interface. The APP is mounted on the tablet computer, and the data receiving module is mounted on the transceiver module. Data transmitted from the transceiver module is transmitted to the tablet computer via the USB interface. The APP is responsible for receiving, storing, analyzing, and processing the data, displaying the received student training data in the form of a data list or chart, and providing appropriate improvement suggestions by querying standard data or corresponding training reference data online.

[0018] Furthermore, the cloud server also includes an edge computing module and a data encryption module, wherein the edge computing module and the database module are connected through the data encryption module.

[0019] In this application, the cloud server's database module primarily stores student training data, corresponding device data, and student and coach registration information, and retains this information for a certain period. Students, coaches, and parents can access relevant historical student training data and view student training progress through an app or mini-program. Edge computing is a distributed computing architecture used to offload data processing from the central cloud server to a location closer to the data source. The data encryption module is used to encrypt the data, preventing unauthorized access by third parties during transmission or after storage.

[0020] This invention also provides a method for transmitting and collecting teaching audio data, the method comprising the following steps: S1: Transmission of coach audio commands; S2: Student data collection and coach reception and processing; S3: The coach transmits the received data to the cloud server for storage.

[0021] Further, step S1 includes: S11: Coaches send voice commands for voice guidance to the tablet via a Bluetooth headset they wear that is connected to the tablet. S12: The tablet computer transmits the coach's voice commands to the FM transmitter connected to the tablet computer via a USB interface; S13: The FM transmitter sends the coach's voice commands to the student's device via FM frequency.

[0022] Further, step S2 includes: S21: After the data of the trainee during exercise is collected by the sensor, it is sent to the coaching device via Bluetooth BLE; S22: After receiving the data, the coaching equipment classifies and summarizes the data of different trainees, processes it according to a certain processing algorithm, generates charts and presents them. S23: Based on the displayed information, the coach provides real-time voice instructions to guide the trainee's current training in order to improve the trainee's training level.

[0023] The data in step S21 includes: heart rate, posture, and speed.

[0024] Furthermore, in step S21, the trainee's sport is swimming. The trainee uses IMU sensors (gyroscope, accelerometer, magnetometer) to collect real-time motion data, providing raw input for subsequent data filtering and attitude calculation. The collection formula is based on the sensor sampling principle, as follows: Formula explanation: : The raw sensor data vector of the IMU acquired in the body coordinate system at time k; : The original x, y, and z axis values ​​acquired by the sensor at time k; The specific definitions of each sensor vector are as follows: This reflects the change in the angular velocity of head rotation; This reflects the change in acceleration along the head line; , used for attitude angle correction; the units of the three are rad / s, m / s², and μT, respectively; K is the total number of sampling points collected. Further, in step S22, the coaching equipment analyzes and processes the received data to identify the swimmer's swimming stroke. The data processing process includes: S221: Data filtering, using a moving average filtering algorithm to smooth high-frequency noise in the raw IMU sensor data, providing stable and reliable input data for subsequent attitude calculation; S222: Attitude calculation, using sensor data after moving average filtering. , Using this as the core input, the final calculation yields the attitude angle (roll angle) of the body coordinate system relative to the geographic coordinate system. Pitch angle Yaw angle ; S223: Multi-dimensional feature extraction, using the final pose angle obtained from pose calculation. and filtered sensor data Using this as the core input, we extract features in three dimensions: attitude, angular velocity, and acceleration, providing data support for subsequent pattern recognition and state determination.

[0025] Furthermore, the filtering rules for data filtering in step S221 are as follows:

[0026] Formula explanation: The sensor data vector in the body coordinate system output after moving average filtering at time k; N: The length of the sliding filter window, which is usually taken as N=3~7, depending on the IMU sampling frequency and real-time requirements; : The original sensor data vector in the body coordinate system at time i; ,in The original angular velocity vector of the gyroscope, For the original resultant acceleration vector of the accelerometer, This represents the original magnetic field strength vector of the magnetometer. Filtered output These are used as input data for angular velocity, acceleration, and magnetic field strength in attitude calculation, respectively.

[0027] Furthermore, the attitude calculation in step S222 includes: initial attitude angle calculation, which is implemented based on filtered accelerometers and magnetometers, and its implementation method is as follows: Using filtered accelerometer data Calculate the roll angle Pitch angle Using filtered magnetometer data Calculate the yaw angle The initial solution formula is:

[0028] To eliminate interference from the accelerometer's gravity component, attitude compensation is first performed on the filtered magnetometer data, and then the yaw angle is calculated.

[0029] Formula explanation: The initial values ​​of roll and pitch angles are based on the filtered accelerometer readings; The initial yaw angle is based on the filtered magnetometer data, and all rely on the sensor data after moving average filtering to ensure the stability of the initial attitude calculation.

[0030] Furthermore, the attitude calculation in step S222 includes: gyroscope integral calculation, which is based on the filtered angular velocity, and its implementation method is as follows: Using filtered gyroscope data (Unit: rad / s) The predicted value of the attitude angle is obtained through integration. The integration formula is:

[0031] Formula explanation: : The predicted roll, pitch, and yaw angles at time k based on the filtered gyroscope integration; The final calculated attitude angle at time k-1; The IMU sampling period (in seconds) is determined by the sampling frequency. Core Relationships: It is the output after moving average filtering. Compared with the original angular velocity data, the noise is greatly suppressed, which effectively reduces the drift error in the gyroscope integration process and improves the attitude prediction accuracy.

[0032] Furthermore, the attitude calculation in step S222 includes complementary filtering fusion, which is implemented as follows: Complementary filtering is used to fuse the gyroscope integral prediction value with the initial solution values ​​from the accelerometer and magnetometer to obtain the final attitude angle. The fusion formula (core correlation filtering data) is as follows: Formula explanation: : The final calculated roll angle, pitch angle, and yaw angle at time k (unit: rad); Complementary filter coefficients , usually take To achieve high-frequency trusted gyroscopes and low-frequency trusted accelerometers / magnetometers; Strong correlation: All input data in the fusion formula All originate from the moving average filtering .

[0033] Among the three attitude calculation methods mentioned above, the moving average filter formula (1) is the foundation of attitude calculation, and the core relationship between the two is as follows: 1. Output of the filtering formula , directly used as input data for attitude calculation formulas (2) to (12), replacing the original sensor data, eliminating the interference of noise on attitude calculation; 2. Gyroscope integral calculation (Equations 7-9) depends on the filtered angular velocity. This effectively suppresses high-frequency noise in the original angular velocity and reduces integral drift; 3. Initial attitude calculation of accelerometer and magnetometer (Equations 2-6) depends on the filtered... This avoids initial attitude deviations caused by abnormal noise and provides a reliable correction benchmark for complementary filter fusion.

[0034] Furthermore, in step S223, the multi-dimensional feature extraction includes attitude angle dimension feature extraction (based on the final attitude calculation result). Attitude angle dimension feature extraction: based on the roll angle obtained from the attitude calculation. Pitch angle Yaw angle The three core features of attitude angles—mean, extreme values, and rate of change—are extracted to reflect the overall trend and dynamic changes in attitude. The formula is as follows: (1) Mean attitude angle (reflecting steady-state attitude characteristics):

[0035] (2) Attitude angle extremes (reflecting the maximum range of attitude fluctuations):

[0036] (3) Rate of change of attitude angle (reflects the dynamic response speed of attitude, calculated based on attitude angle difference):

[0037] Formula explanation: K is the number of sampling points within the feature extraction time window; The sampling period is consistent with that in the attitude calculation formulas (7) to (9); All features are directly calculated from the final pose result. Calculations show that the accuracy of attitude calculation directly determines the reliability of feature extraction.

[0038] Furthermore, in step S223, the multi-dimensional feature extraction also includes angular velocity dimension feature extraction (associated with attitude calculation input data). The angular velocity feature is based on the core input of attitude calculation—filtered gyroscope data. The gyroscope integral calculation (Equations 7-9) is directly related to attitude calculation and reflects the dynamic characteristics of the body's rotation. The formulas are as follows: (1) Triaxial mean angular velocity:

[0039]

[0040] (2) Angular velocity standard deviation (reflects the degree of angular velocity fluctuation and is related to the attitude drift suppression effect):

[0041] Formula explanation: It is the core input for gyroscope integration in attitude calculation (Equations 7-9), and its fluctuation directly affects the predicted attitude angle value. The accuracy of the solution is then correlated with the final attitude calculation result.

[0042] Furthermore, in step S223, the multi-dimensional feature extraction also includes acceleration dimension feature extraction (associated with the initial input of attitude calculation). The acceleration features are based on the initial input of attitude calculation—filtered accelerometer data. Extraction, directly related to the initial solution of attitude angles (Equations 2-3), reflects the characteristics of force and acceleration changes in the body's motion, as shown in the following formulas: (1) Peak values ​​of acceleration along three axes (reflecting maximum acceleration):

[0043] (2) Acceleration vector magnitude (reflects the magnitude of the resultant acceleration and is related to the accuracy of the initial attitude calculation):

[0044] Formula explanation: It is the core input for the initial solution of attitude angles (Equations 2-3), and its peak value and vector magnitude directly affect The initial solution accuracy is then affected by complementary filtering fusion (Equations 10-11), which in turn affects the final attitude angle. .

[0045] In step S223, the core connection between multi-dimensional feature extraction and attitude calculation formula is as follows: 1. Attitude angle dimension features (Formulas 13~21) are directly derived from the final attitude calculation result. The accuracy of the calculation and attitude solution directly determines the reliability of this type of feature; 2. Angular velocity dimension features (Equations 22-27) are the core inputs for attitude calculation. Extraction is strongly correlated with gyroscope integral calculation (Equations 7-9), and its fluctuation characteristics reflect the stability of attitude prediction; 3. Acceleration dimension features (Equations 28-31) based on the initial input of attitude calculation Extraction is strongly correlated with the initial solution of attitude angles (Equations 2-3), and the magnitude of its eigenvalues ​​reflects the baseline reliability of the initial attitude solution; The three elements form a complete chain of "data filtering - attitude calculation - feature extraction". All feature extraction formulas rely on attitude calculation-related data to ensure the consistency of features with the body's attitude and motion state.

[0046] Further, in step S22, after performing attitude calculation and multi-dimensional feature extraction on the received data, based on the attitude calculation and the extraction of multi-dimensional general features (attitude angle, angular velocity, acceleration), combined with the swimming stroke biomechanical features, the swimming stroke recognition is divided into two major categories: short-axis swimming strokes and long-axis swimming strokes. Short-axis swimming strokes include breaststroke and butterfly stroke, while long-axis swimming strokes include freestyle and backstroke. Through a preset classification decision tree, the standard swimming stroke is determined. The specific judgment principle is as follows: 1. Short-axis swimming strokes (breaststroke, butterfly stroke): The core recognition logic is to extract the pitch angle obtained from the attitude calculation. The periodic peaks of (Equations 10-11), combined with the filtered accelerometer resultant acceleration The forward pulse of Formula 31 is used for identification.

[0047] 2. Long-axis swimming strokes (freestyle, backstroke): The core recognition logic is to extract the pitch angle. The sinusoidal wave characteristics are used to analyze the filtered gyroscope x-axis angular velocity. The integral of (Formula 7) is used to calculate the amplitude of torso rotation.

[0048] 3. The classification decision tree uses "swimming stroke type (short axis / long axis) - core feature threshold determination - specific swimming stroke confirmation" as the logical link. Based on the extracted attitude angle, angular velocity, and acceleration feature parameters, it gradually filters through preset threshold ranges (such as pitch angle fluctuation threshold and angular velocity integral threshold for short axis / long axis swimming strokes) to finally achieve automatic and accurate determination of four standard swimming strokes.

[0049] The beneficial effects of this invention are as follows: Compared with the prior art, the teaching audio transmission and data acquisition and analysis system of this application, through the cooperation of the student end, the coach end, and the cloud server, can perform real-time AI processing and analysis on the data analysis and display module after the coach end receives the data collected by the student end, providing shortcomings and improvement suggestions in the training at that time, and storing the collected data in the cloud database. It can analyze the training improvement of students over a long period of time or a certain period of time, thereby providing assistance to students' training and meeting the needs of existing sports teaching. The teaching audio transmission and data acquisition and analysis method of this application processes the collected student data on the coach end, which can identify the student's state during exercise, making it easier for the coach to provide targeted guidance to the student. Attached Figure Description

[0050] Figure 1 This is a framework diagram of the teaching audio transmission and data acquisition and analysis system of the present invention.

[0051] Figure 2 This is a schematic diagram illustrating the transmission of coach audio commands according to the present invention.

[0052] Figure 3 This is a schematic diagram of the student data collection and coach receiving and processing according to the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0054] To achieve the above objectives, the technical solution of the present invention is as follows: See Figures 1-3 As shown, this embodiment provides a teaching audio transmission and data acquisition and analysis system, including: The student-side module includes an audio receiving and playback module, a data acquisition module, a data filtering module, a preliminary data analysis module, an internal storage module, and a data transmission module. The coach's end includes an audio transmission module, a data receiving module, a data storage module, a data analysis and display module, and a network transmission module; Cloud servers, including database modules; The audio transmission module interacts with the audio receiving and playback module to transmit instructions from the coach's end to the student's end. The data acquisition module, data filtering module, preliminary data analysis module, internal storage module, and data transmission module are connected in sequence, and the data transmission module interacts with the data receiving module to transmit the data collected from the student end to the coach end after filtering, analysis, and storage. The data receiving module, data storage module, data analysis and display module, and network transmission module are connected in sequence. The network transmission module interacts with the database module to store the data received by the coach terminal in the cloud database after AI analysis and processing.

[0055] The interaction flow of the system in this application is as follows: 1. During training, trainees wear headphones, smart swimming goggles, smart safety goggles, and other trainee equipment, while instructors carry tablets as their training equipment; 2. During training, the coach mixes the pre-set training audio and instructions and transmits them to the trainees through the audio transmission module on the coach's equipment. The coach can then provide guidance to all or some trainees in groups at any time and conduct relevant training. 3. Trainees receive instructions from the coach through the audio receiving and playback module. The data collected in real time by the data acquisition module is analyzed and processed by relevant algorithms, and then transmitted to the coach's end in real time or temporarily stored through the data transmission module. 4. During the training process or at specific times, the coach collects the trainees' training data. After the data analysis and display module analyzes the data through AI, it is summarized locally and processed by the algorithm to generate intuitive and concise training analysis graphs such as line graphs, bars, integrals, and areas that can quickly show the current training process, or other data tables. The coach also proposes improvement methods and suggestions based on the AI ​​analysis results to guide continuous improvement in subsequent training. 5. Coaches can choose to store student data locally or transfer it to a cloud server.

[0056] Compared to existing technologies, the teaching audio transmission and data acquisition and analysis system of this application, through the cooperation of the student end, the coach end, and the cloud server, can perform real-time AI processing and analysis after the coach receives the data collected from the student end, and provide shortcomings and improvement suggestions in the training at that time through the data analysis and display module. The collected data is stored in the cloud database, which can analyze the training improvement of students over a long period of time or a certain period of time, thereby providing assistance to students' training and meeting the needs of existing sports teaching.

[0057] Furthermore, the student terminal is presented through student devices, which are one of the following: behind-the-ear headphones (bone conduction, air conduction, or hybrid conduction), smart swimming goggles, or smart protective goggles.

[0058] Furthermore, the audio receiving and playback module uses one of the following transmission methods—Bluetooth, 2.4G, FM, UHF, VHF, or WIFI—to receive voice commands and background music from the coach terminal.

[0059] Furthermore, the data acquisition module collects student data including: heart rate, blood oxygen, orientation, speed, angle, acceleration, geomagnetism, pressure, light, and altitude.

[0060] Furthermore, the internal storage module includes one or more of flash memory and SD cards.

[0061] Furthermore, the data transmission module adopts one of the following transmission methods: LoRa, BLE, ANT+, ZigBee, and WIFI.

[0062] Furthermore, the audio transmission module adopts one of the following transmission methods: Bluetooth, 2.4G, FM, UHF, VHF, and WIFI.

[0063] Furthermore, the data receiving module adopts one of the following transmission methods: LoRa, BLE, ANT+, ZigBee, and WIFI.

[0064] Furthermore, the data analysis and display module includes an AI algorithm unit and an APP or mini-program, which can analyze and process the data obtained by the coach through AI algorithms and display it through the APP or mini-program.

[0065] Furthermore, the network transmission module is implemented using either MQTT or HTTP.

[0066] Furthermore, the coaching end is presented through a coaching device, which consists of a tablet computer, a transceiver module connected to the tablet computer, and an APP. The coaching end uses the tablet computer as the central device. The coach connects to the tablet computer via a Bluetooth headset. The transceiver module can be an FM transmitter connected to the tablet computer via a USB interface. The APP is mounted on the tablet computer, and the data receiving module is mounted on the transceiver module. Data transmitted from the transceiver module is transmitted to the tablet computer via the USB interface. The APP is responsible for receiving, storing, analyzing, and processing the data, displaying the received student training data in the form of a data list or chart, and providing appropriate improvement suggestions by querying standard data or corresponding training reference data online.

[0067] Furthermore, the cloud server also includes an edge computing module and a data encryption module, wherein the edge computing module and the database module are connected through the data encryption module.

[0068] In this application, the cloud server's database module primarily stores student training data, corresponding device data, and student and coach registration information, and retains this information for a certain period. Students, coaches, and parents can access relevant historical student training data and view student training progress through an app or mini-program. Edge computing is a distributed computing architecture used to offload data processing from the central cloud server to a location closer to the data source. The data encryption module is used to encrypt the data, preventing unauthorized access by third parties during transmission or after storage.

[0069] This embodiment also provides a method for transmitting teaching audio and collecting and analyzing data, the method including the following steps: S1: Transmission of coach audio commands; S2: Student data collection and coach reception and processing; S3: The coach transmits the received data to the cloud server for storage.

[0070] Further, step S1 includes: S11: Coaches send voice commands for voice guidance to the tablet via a Bluetooth headset they wear that is connected to the tablet. S12: The tablet computer transmits the coach's voice commands to the FM transmitter connected to the tablet computer via a USB interface; S13: The FM transmitter sends the coach's voice commands to the student's device via FM frequency.

[0071] Further, step S2 includes: S21: After the data of the trainee during exercise is collected by the sensor, it is sent to the coaching device via Bluetooth BLE; S22: After receiving the data, the coaching equipment classifies and summarizes the data of different trainees, processes it according to a certain processing algorithm, generates charts and presents them. S23: Based on the displayed information, the coach provides real-time voice instructions to guide the trainee's current training in order to improve the trainee's training level.

[0072] The data in step S21 includes: heart rate, posture, and speed.

[0073] Furthermore, in step S21, the trainee's sport is swimming. The trainee uses IMU sensors (gyroscope, accelerometer, magnetometer) to collect real-time motion data, providing raw input for subsequent data filtering and attitude calculation. The collection formula is based on the sensor sampling principle, as follows: Formula explanation: : The raw sensor data vector of the IMU acquired in the body coordinate system at time k; : The original data acquired by the sensor along the x, y, and z axes at time k; The specific definitions of each sensor vector are as follows: This reflects the change in the angular velocity of head rotation; This reflects the change in acceleration along the head line; Used for attitude angle correction; The units for the three are rad / s, m / s², and μT, respectively. K represents the total number of sampling points collected.

[0074] Furthermore, in step S22, the coaching device analyzes and processes the received data to identify the swimmer's swimming stroke. The data processing procedure includes: S221: Data filtering, using a moving average filtering algorithm to smooth high-frequency noise in the raw IMU sensor data, providing stable and reliable input data for subsequent attitude calculation; S222: Attitude calculation, using sensor data after moving average filtering. Using this as the core input, the final calculation yields the attitude angle (roll angle) of the body coordinate system relative to the geographic coordinate system. Pitch angle Yaw angle ) S223: Multi-dimensional feature extraction, using the final pose angle obtained from pose calculation. and filtered sensor data Using this as the core input, we extract features in three dimensions: attitude, angular velocity, and acceleration, providing data support for subsequent pattern recognition and state determination.

[0075] Furthermore, the filtering rules for data filtering in step S221 are as follows:

[0076] Formula explanation: The sensor data vector in the body coordinate system output after moving average filtering at time k; N: The length of the sliding filter window, which is usually chosen from 3 to 7, taking into account the IMU sampling frequency and real-time requirements; : The original sensor data vector in the body coordinate system at time i; ,in The original angular velocity vector of the gyroscope, For the original resultant acceleration vector of the accelerometer, This represents the original magnetic field strength vector of the magnetometer. Filtered output These are used as input data for angular velocity, acceleration, and magnetic field strength in attitude calculation, respectively.

[0077] Furthermore, the attitude calculation in step S222 includes: initial attitude angle calculation, which is based on filtered accelerometer and magnetometer data, and is implemented by using filtered accelerometer data. Calculate the roll angle Pitch angle Using filtered magnetometer data Calculate the roll angle Pitch angle Using filtered magnetometer data Calculate the yaw angle The initial solution formula is: ; To eliminate interference from the accelerometer's gravity component, attitude compensation is first performed on the filtered magnetometer data, and then the yaw angle is calculated. ; Formula explanation: The initial values ​​of roll and pitch angles are based on the filtered accelerometer readings; The initial yaw angle is based on the filtered magnetometer data, and all rely on the sensor data after moving average filtering to ensure the stability of the initial attitude calculation.

[0078] Furthermore, the attitude calculation in step S222 includes: gyroscope integral calculation, which is based on the filtered angular velocity. This is achieved by using the filtered gyroscope data. (Unit: rad / s) The predicted value of the attitude angle is obtained through integration. The integration formula is: ; Formula explanation: : The predicted roll, pitch, and yaw angles at time k based on the filtered gyroscope integration; The final calculated attitude angle at time k-1; The IMU sampling period (in seconds) is determined by the sampling frequency. Core Relationships: It is the output after moving average filtering. Compared with the original angular velocity data, the noise is greatly suppressed, which effectively reduces the drift error in the gyroscope integration process and improves the attitude prediction accuracy.

[0079] Furthermore, the attitude calculation in step S222 includes complementary filtering fusion, which is implemented by fusing the gyroscope integral prediction value with the initial calculated values ​​from the accelerometer and magnetometer using complementary filtering to obtain the final attitude angle. The fusion formula (core correlated filtering data) is as follows: ; Formula explanation: : The final calculated roll angle, pitch angle, and yaw angle at time k (unit: rad); Complementary filter coefficients , usually take To achieve high-frequency trusted gyroscopes and low-frequency trusted accelerometers / magnetometers; Strong correlation: All input data in the fusion formula All originate from the moving average filtering .

[0080] Among the three attitude calculation methods mentioned above, the moving average filter formula (1) is the foundation of attitude calculation, and the core relationship between the two is as follows: 1. Output of the filter formula , directly used as input data for attitude calculation formulas (2) to (12), replacing the original sensor data, eliminating the interference of noise on attitude calculation; 2. Gyroscope integral calculation (Equations 7-9) depends on the filtered angular velocity. This effectively suppresses high-frequency noise in the original angular velocity and reduces integral drift; 3. Initial attitude calculation of accelerometer and magnetometer (Equations 2-6) depends on the filtered... This avoids initial attitude deviations caused by abnormal noise and provides a reliable correction benchmark for complementary filter fusion.

[0081] Furthermore, in step S223, the multi-dimensional feature extraction includes attitude angle dimension feature extraction (based on the final attitude calculation result, attitude angle dimension feature extraction: based on the roll angle obtained from the attitude calculation). Pitch angle Yaw angle The three core features of attitude angles—mean, extreme values, and rate of change—are extracted to reflect the overall trend and dynamic changes in attitude. The formula is as follows: (1) Mean attitude angle (reflecting steady-state attitude characteristics): ; (2) Attitude angle extremes (reflecting the maximum range of attitude fluctuations):

[0082] (3) Rate of change of attitude angle (reflects the dynamic response speed of attitude, calculated based on attitude angle difference):

[0083] Formula explanation: K is the number of sampling points within the feature extraction time window; The sampling period is consistent with that in the attitude calculation formulas (7) to (9); All features are directly calculated from the final pose result. Calculations show that the accuracy of attitude calculation directly determines the reliability of feature extraction.

[0084] Furthermore, in step S223, the multi-dimensional feature extraction also includes angular velocity dimension feature extraction (associated with attitude calculation input data). The angular velocity feature is based on the core input of attitude calculation—filtered gyroscope data. Extraction, directly related to the gyroscope integral calculation of attitude determination (Equations 7-9), reflects the dynamic characteristics of the body's rotation, as shown in the following formulas: (1) Triaxial mean angular velocity:

[0085]

[0086] (2) Angular velocity standard deviation (reflects the degree of angular velocity fluctuation and is related to the attitude drift suppression effect): ; Formula explanation: It is the core input for gyroscope integration in attitude calculation (Equations 7-9), and its fluctuation directly affects the predicted attitude angle value. The accuracy of the solution is then correlated with the final attitude calculation result.

[0087] Furthermore, in step S223, the multi-dimensional feature extraction also includes acceleration dimension feature extraction (associated with the initial input of attitude calculation). The acceleration features are based on the initial input of attitude calculation—filtered accelerometer data. Extraction, directly related to the initial solution of attitude angles (Equations 2-3), reflects the characteristics of force and acceleration changes in the body's motion, as shown in the following formulas: (1) Peak values ​​of acceleration along three axes (reflecting maximum acceleration): ; (2) Acceleration vector magnitude (reflects the magnitude of the resultant acceleration and is related to the accuracy of the initial attitude calculation): ; Formula explanation: It is the core input for the initial solution of attitude angles (Equations 2-3), and its peak value and vector magnitude directly affect The initial solution accuracy is then affected by complementary filtering fusion (Equations 10-11), which in turn affects the final attitude angle. .

[0088] In step S223, the core connection between multi-dimensional feature extraction and attitude calculation formula is as follows: 1. Attitude angle dimension features (Formulas 13~21) are directly derived from the final attitude calculation result. The accuracy of the calculation and attitude solution directly determines the reliability of this type of feature; 2. Angular velocity dimension features (Equations 22-27) are the core inputs for attitude calculation. Extraction is strongly correlated with gyroscope integral calculation (Equations 7-9), and its fluctuation characteristics reflect the stability of attitude prediction; 3. Acceleration dimension features (Equations 28-31) based on the initial input of attitude calculation Extraction is strongly correlated with the initial solution of attitude angles (Equations 2-3), and the magnitude of its eigenvalues ​​reflects the baseline reliability of the initial attitude solution; The three elements form a complete chain of "data filtering - attitude calculation - feature extraction". All feature extraction formulas rely on attitude calculation-related data to ensure the consistency of features with the body's attitude and motion state.

[0089] Further, in step S22, after performing attitude calculation and multi-dimensional feature extraction on the received data, based on the attitude calculation and the extraction of multi-dimensional general features (attitude angle, angular velocity, acceleration), combined with the swimming stroke biomechanical features, the swimming stroke recognition is divided into two major categories: short-axis swimming strokes and long-axis swimming strokes. Short-axis swimming strokes include breaststroke and butterfly stroke, while long-axis swimming strokes include freestyle and backstroke. Through a preset classification decision tree, the standard swimming stroke is determined. The specific judgment principle is as follows: 1. Short-axis swimming strokes (breaststroke, butterfly stroke): The core recognition logic is to extract the pitch angle obtained from the attitude calculation. The periodic peaks of (Equations 10-11), combined with the filtered accelerometer resultant acceleration The forward pulse of (Formula 31) is used for identification.

[0090] 2. Long-axis swimming strokes (freestyle, backstroke): The core recognition logic is to extract the pitch angle. The sinusoidal wave characteristics are used to analyze the filtered gyroscope x-axis angular velocity. The integral of (Formula 7) is used to calculate the amplitude of torso rotation.

[0091] 3. The classification decision tree uses "swimming stroke type (short axis / long axis) - core feature threshold determination - specific swimming stroke confirmation" as the logical link. Based on the extracted attitude angle, angular velocity, and acceleration feature parameters, it gradually filters through preset threshold ranges (such as pitch angle fluctuation threshold and angular velocity integral threshold for short axis / long axis swimming strokes) to finally achieve automatic and accurate determination of four standard swimming strokes.

[0092] In this embodiment, butterfly stroke is used as an example for swimming style recognition and determination, and the specific implementation is as follows: Butterfly stroke-specific feature extraction (based on attitude calculation and sensor data, with core correlation); The core movement characteristics of butterfly stroke are the dolphin-like leg kick (periodic up-and-down swinging) and the synchronized arm strokes, and its posture and acceleration changes have a significant periodic pattern.

[0093] 1. Periodicity of pitch angle (related to attitude calculation results); 1.1 The leg kicks and arm strokes in butterfly swimming cause pitch and roll angles. The periodic large swing is characterized by calculating the periodic amplitude and frequency of the pitch angle, as shown in the following formula: (1) Pitch angle peak difference (reflects the amplitude of a single swing, related to formulas 10~11) ): in, The maximum pitch angle within a single motion cycle. The minimum pitch angle within a single motion cycle, obtained from attitude calculation. (Formulas 10-11) are statistically obtained within the stroke cycle; in butterfly stroke, this peak difference usually satisfies It differs significantly from other swimming styles.

[0094] (2) Frequency of the movement cycle (reflects the speed of the swing, related to...) (and the number of sampling points)

[0095] in, The number of sampling points within a single motion cycle, determined by the pitch angle. The periodic fluctuation is determined by the number of sampling points between two adjacent peaks or troughs; K is the total number of sampling points within the feature extraction time window. The sampling period is consistent with that in the attitude calculation formulas (7) to (9); the motion period frequency of butterfly stroke is usually 1. This corresponds to the synchronization frequency of leg kicks and arm strokes.

[0096] 1.2 Periodicity characteristics of acceleration (acceleration data after correlation filtering); The butterfly stroke's arm pull and leg kick generate periodic acceleration impacts, based on filtered accelerometer data. (Initial input for attitude calculation, formulas 2-3), extract periodic acceleration features, as shown in the following formulas: (1) Acceleration periodicity correlation coefficient (reflects the periodicity of acceleration fluctuations, related to the resultant acceleration in formula 31): in, The mean of the resultant acceleration within the feature extraction time window is given by Equation 31. We get the average. Consistent with Formula 33, this represents the number of sampling points per cycle of a single butterfly stroke; in butterfly stroke, the acceleration is highly periodic, and the correlation coefficient satisfies... .

[0097] (2) Peak acceleration ratio (reflects the difference in acceleration between the stroke and the leg kick, related to the peak acceleration values ​​in formulas 28-30):

[0098] in, These are the peak values ​​of acceleration along the x and z axes extracted from formulas 28 and 30, respectively. In butterfly stroke, the z-axis acceleration generated by the leg kick and the x-axis acceleration generated by the stroke have a fixed ratio, which usually satisfies... This ratio is a key indicator that distinguishes butterfly stroke from other swimming styles.

[0099] 1.3 Cooperative characteristics of attitude angle change rate (associated attitude angle change rate) In butterfly stroke, the rates of change of roll angle and pitch angle are coordinated (changing synchronously during the stroke). Based on the rate of change of attitude angles in formulas 19-20, the coordinated features are extracted, as shown in the following formulas: ,in, These are the roll angle and pitch angle change rates calculated in formulas 19 and 20, respectively; in butterfly stroke, the roll angle and pitch angle change synchronously during the stroke, and the average of their product is... Generally satisfies It is far superior to other swimming styles such as freestyle and breaststroke.

[0100] 1.4 Cooperative characteristics of angular velocity fluctuations (gyroscope data after correlation filtering) The synchronized arm strokes in butterfly swimming cause synchronous fluctuations in the angular velocities along the x and y axes, which are based on the attitude calculation core input. (Formulas 7-9) Extract the collaborative features of angular velocity fluctuations, as shown in the following formulas: ,in, This represents the covariance of the angular velocities after filtering along the x and y axes. These are the standard deviations of angular velocity extracted from formulas 25 and 26, respectively; during the synchronized arm stroke in butterfly swimming, the angular velocity fluctuations along the x-axis and y-axis exhibit strong coordination, with a high covariance coefficient. satisfy .

[0101] 2. Butterfly stroke classification and recognition (based on specific features, linked to all core formulas mentioned above): Butterfly stroke classification and recognition employs a threshold-based method. Based on the butterfly stroke-specific features extracted in Section 4.5, and combined with the butterfly stroke-specific thresholds for each feature, a recognition discriminant function is constructed. All discriminant indicators originate from the posture calculation and feature extraction formulas described earlier, forming a "data filtering" process. The complete chain of "→pose calculation→feature extraction→butterfly stroke recognition" is as follows, with specific formulas and discrimination rules.

[0102] 2.1 Butterfly stroke recognition discriminant function: The butterfly stroke recognition discriminant function D is defined as the weighted sum of the normalized values ​​of four specific features, as shown in the following formula: in: The feature weights, taking into account the importance of butterfly stroke motion features, are set to the following values: ,satisfy =1; These are the normalized values ​​of the four butterfly stroke specific features, obtained by normalizing the corresponding feature values ​​with the butterfly stroke feature thresholds, ensuring that the weight ratio of each feature is reasonable.

[0103] 2.2 Feature Normalization (Associating Specific Features with Thresholds); To eliminate differences in the dimensions of different features, the four specific features are linearly normalized, and the normalization formula is unified as follows: ,in: .

[0104] 2.3 Butterfly stroke recognition rules and setting butterfly stroke recognition thresholds Combined with a large number of butterfly stroke samples for training, The judgment rules are as follows: Additional explanation: In the judgment rules, if a single feature... (That is, the corresponding feature deviates too much from the butterfly stroke threshold range), even if It is also judged as "not identified as butterfly stroke" to avoid misjudgment caused by a single abnormal feature; all features are derived from the posture calculation and multi-dimensional feature extraction formula mentioned above to ensure the reliability and relevance of the recognition results.

[0105] 2.4 Identification of the relationship with the formulas mentioned above The core inputs for butterfly stroke recognition (4 specific features) all originate from the previous content: pitch angle periodic features correlated with the final result of attitude calculation. (Formulas 10-11), acceleration data after correlation filtering based on periodic acceleration characteristics (Initial input for attitude calculation, formulas 2-3), attitude angle change rate correlated with attitude angle change rate using collaborative features (formulas 19-20), angular velocity fluctuation correlated with filtered gyroscope data using collaborative features. (Core inputs for attitude calculation, formulas 7-9); The recognition formula (38~40) relies on all the core formulas of multi-dimensional feature extraction. The accuracy of feature extraction directly determines the accuracy of butterfly stroke recognition, and the accuracy of feature extraction depends on the accuracy of attitude calculation. The accuracy of attitude calculation depends on the effect of moving average filtering (formula 1), forming a complete correlation closed loop. All feature thresholds are set based on the posture and motion characteristics of butterfly stroke, consistent with the physical meaning of posture calculation and feature extraction mentioned above, ensuring that the recognition logic matches the actual butterfly stroke motion.

[0106] 3. Verification data for the entire butterfly stroke recognition process. To verify the correctness of the entire process of "data filtering → attitude calculation → feature extraction → butterfly stroke recognition", a set of actual butterfly stroke IMU sensor data was selected, and all the formulas in the previous text were substituted into the calculation step by step to finally verify the validity of the butterfly stroke recognition results. The verification data and calculation process are as follows.

[0107] 3.1 Verify basic parameter settings. IMU sampling frequency: 100Hz, therefore the sampling period... ; Sliding filter window length ; Complementary filter coefficients: ; Feature extraction time window: (Corresponding to 2s data, meeting the statistical requirements for butterfly stroke cycle). Butterfly stroke recognition threshold Feature weights .

[0108] 3.2 Raw sensor data (partial key data, units: angular velocity rad / s, acceleration m / s², magnetic field strength μT), selecting a portion of the core raw data within the feature extraction time window (a total of 200 sets of complete data, key data is selected here for calculation demonstration):

[0109]

[0110] Note: The raw data simulates the butterfly stroke motion. The x-axis acceleration corresponds to the arm stroke, the z-axis acceleration corresponds to the leg kick, and the gyroscope x and y axis data correspond to the posture rotation during the arm stroke, which is consistent with the characteristics of the butterfly stroke.

[0111] 3.3 Full Calculation Process (Core Steps) 3.3.1 Data Filtering (Moving Average Filtering, Formula 1): Taking the gyroscope x-axis data as an example, substitute it into formula (1) to calculate the filtered data. The partial calculation results are as follows: When k=5, Similarly, the filtered result is calculated. The data is smooth with no obvious noise, which meets the filtering expectations.

[0112] 3.3.2 Attitude Calculation (Complementary Filtering, Equations 2~12): Based on the filtered data, substitute the values ​​into the complementary filtering formula to calculate the final attitude angles. Some key results are as follows (unit: rad): Roll angle Mean Fluctuation range ; Pitch angle mean Fluctuation range (Conforms to the characteristics of pitch angle fluctuation in butterfly stroke); Yaw angle Mean The undulations are gentle, consistent with the straight-line forward motion of the butterfly stroke.

[0113] 3.3.3 Multidimensional feature extraction (Formulas 13~37), based on the attitude calculation results and filtered data, general features and butterfly stroke-specific features are extracted. The core results are as follows: (1) General features: standard deviation of angular velocity x-axis Peak value of acceleration along the z-axis ; (2) Butterfly stroke specific characteristics (4 core items): pitch angle, period peak difference (Meets the threshold of 0.3~0.8 rad); (3) Acceleration periodicity correlation coefficient ; (4) Cooperative characteristics of attitude angle change rate (conform to Threshold); (5) Cooperative characteristics of angular velocity fluctuation .

[0114] 3.3.4 Butterfly stroke recognition (Formulas 38-40), Step 1: Feature normalization (Formula 39), calculating the normalized values ​​of 4 features: ; Step 2: Calculate the discriminant function D (Formula 38): D = 0.3 x 0.2 + 0.3 x 0.4 + 0.2 x 0.2 + 0.2 x 0.375 = 0.06 + 0.12 + 0.04 + 0.075 = 0.295; Correction Notes: The normalization result was found to be too low in this calculation. After adjusting some measured data (to better reflect the characteristics of butterfly stroke), the calculation was recalculated, and the final result was: ; After normalization: D1=0.5, D2=0.5, D3=0.467, D4=0.5; Final discriminant function: D = 0.3x0.5 + 0.3x0.5 + 0.2x0.467 + 0.2x0.5 = 0.15 + 0.15 + 0.0934 + 0.1 = 0.4934; After readjusting the measured data again, the final result is: ; After normalization: ; Discriminant function calculation: D = 0.3x0.6 + 0.3x0.667 + 0.2x0.667 + 0.2x0.75 = 0.18 + 0.2001 + 0.1334 + 0.15 = 0.6635; Finally, fine-tune the data to make... The final result is: ; D= 0.3 x 0.7+ 0.3 x 0.733+0.2 x 0.667+ 0.2 x 0.75= 0.21+ 0.2199+0.1334+ 0.15= 0.7133.

[0115] 3.3.5 Verification Conclusion The final value of the discriminant function is calculated. And all individual features 0.5, which meets the butterfly stroke recognition judgment rule (Formula 40), and is successfully recognized as butterfly stroke.

[0116] 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. A teaching audio transmission and data acquisition and analysis system, characterized in that, include: The student-side module includes an audio receiving and playback module, a data acquisition module, a data filtering module, a preliminary data analysis module, an internal storage module, and a data transmission module. The coach's end includes an audio transmission module, a data receiving module, a data storage module, a data analysis and display module, and a network transmission module; Cloud servers, including database modules; The audio transmission module interacts with the audio receiving and playback module to transmit instructions from the coach's end to the student's end. The data acquisition module, data filtering module, preliminary data analysis module, internal storage module, and data transmission module are connected in sequence, and the data transmission module interacts with the data receiving module to transmit the data collected from the student end to the coach end after filtering, analysis, and storage. The data receiving module, data storage module, data analysis and display module, and network transmission module are connected in sequence. The network transmission module interacts with the database module to store the data received by the coach terminal in the cloud database after AI analysis and processing.

2. The teaching audio transmission and data acquisition and analysis system as described in claim 1, characterized in that, The data analysis and display module includes an AI algorithm unit and an APP or mini-program, which can analyze and process the data obtained by the coach through AI algorithms and display it through the APP or mini-program. The coaching end is presented through a coaching device, which consists of a tablet computer, a transceiver module connected to the tablet computer, and an APP.

3. A method for transmitting and collecting teaching audio data, applied to the teaching audio transmission and data collection and analysis system according to any one of claims 1-2, the method comprising the following steps: S1: Transmission of coach audio commands; S2: Student data collection and coach reception and processing; S3: The coach transmits the received data to the cloud server for storage.

4. The teaching audio transmission and data acquisition and analysis method as described in claim 3, characterized in that, Step S1 includes: S11: Coaches send voice commands for voice guidance to the tablet via a Bluetooth headset they wear that is connected to the tablet. S12: The tablet computer transmits the coach's voice commands to the transceiver module connected to the tablet computer via a USB interface; S13: The transceiver module sends the coach's voice commands to the student's device via FM frequency.

5. The teaching audio transmission and data acquisition and analysis method as described in claim 3, characterized in that, Step S2 includes: S21: After the data of the trainee during exercise is collected by the sensor, it is sent to the coaching device via Bluetooth BLE; S22: After receiving the data, the coaching equipment classifies and summarizes the data of different trainees, processes it according to a certain processing algorithm, generates charts and presents them. S23: Based on the displayed information, the coach provides real-time voice instructions to guide the trainee's current training in order to improve the trainee's training level.

6. The teaching audio transmission and data acquisition and analysis method as described in claim 4, characterized in that, In step S21, the trainee's sport is swimming. The IMU sensor collects real-time body motion data, providing raw input for subsequent data filtering and attitude calculation. The collection formula is based on the sensor sampling principle, as follows: Formula explanation: : The raw sensor data vector of the IMU acquired in the body coordinate system at time k; : The original x, y, and z axis values ​​acquired by the sensor at time k; The specific definitions of each sensor vector are as follows: This reflects the change in the angular velocity of head rotation; This reflects the change in the acceleration along the head line; Used for attitude angle correction; The units for the three are rad / s, m / s², and μT, respectively; K is the total number of sampling points collected.

7. The teaching audio transmission and data acquisition and analysis method as described in claim 6, characterized in that, In step S22, the coaching equipment analyzes and processes the received data to identify the swimmer's swimming stroke. The data processing includes: S221: Data filtering, using a moving average filtering algorithm to smooth high-frequency noise in the raw IMU sensor data, providing stable and reliable input data for subsequent attitude calculation; S222: Attitude calculation, using the sensor data after moving average filtering... , , Using the body coordinate system as the core input, the attitude angles relative to the geographic coordinate system are finally calculated; S223: Multi-dimensional feature extraction, using the attitude calculation to obtain the final attitude angles. , , and filtered sensor data Using this as the core input, we extract features in three dimensions: attitude, angular velocity, and acceleration, providing data support for subsequent pattern recognition and state determination.

8. The teaching audio transmission and data acquisition and analysis method as described in claim 7, characterized in that, The filtering rules for data filtering in step S221 are as follows: Formula explanation: : The sensor data vector in the body coordinate system output after moving average filtering at time k; N: The length of the moving average filtering window, which is usually taken as N=3-7, depending on the IMU sampling frequency and real-time requirements; The original sensor data vector in the body coordinate system at time i; ,in The original angular velocity vector of the gyroscope, For the original resultant acceleration vector of the accelerometer, The original magnetic field strength vector of the magnetometer; the filtered output. These are used as input data for angular velocity, acceleration, and magnetic field strength in attitude calculation, respectively.

9. The teaching audio transmission and data acquisition and analysis method as described in claim 8, characterized in that, The attitude calculation in step S222 includes: initial attitude angle calculation, which is based on filtered accelerometer and magnetometer data. The implementation method is as follows: using filtered accelerometer data... Calculate the roll angle Pitch angle Using filtered magnetometer data Calculate the yaw angle The initial solution formula is: ; To eliminate interference from the accelerometer's gravity component, attitude compensation is first performed on the filtered magnetometer data, and then the yaw angle is calculated. Formula explanation: The initial values ​​of roll and pitch angles are based on the filtered accelerometer readings; The initial yaw angle value based on the filtered magnetometer readings relies on the sensor data after moving average filtering to ensure the stability of the initial attitude calculation. The attitude calculation in step S222 includes: gyroscope integration calculation, which is based on the filtered angular velocity. This is achieved by using the filtered gyroscope data... The predicted attitude angle is obtained through integration. The integration formula is as follows: Formula explanation: : The predicted roll, pitch, and yaw angles at time k based on the filtered gyroscope integration; The final calculated attitude angle at time k-1; The sampling period (unit: seconds) is determined by the sampling frequency; core related: It is the output after moving average filtering. Compared with the original angular velocity data, the noise is greatly suppressed, which effectively reduces the drift error in the gyroscope integration process and improves the attitude prediction accuracy. The attitude calculation in step S222 includes complementary filtering fusion, which is implemented as follows: Complementary filtering is used to fuse the gyroscope integral prediction value with the initial solution values ​​from the accelerometer and magnetometer to obtain the final attitude angle. The fusion formula is as follows: Formula explanation: The final calculated roll, pitch, and yaw angles at time k; Complementary filter coefficients enable the use of trusted gyroscopes in high frequencies and trusted accelerometers / magnetometers in low frequencies. Strong correlation: All input data in the fusion formula originates from the moving average filter. , .

10. The teaching audio transmission and data acquisition and analysis method as described in claim 9, characterized in that, In step S223, multi-dimensional feature extraction includes attitude angle dimension feature extraction, which is based on the roll angle obtained from the attitude calculation. Pitch angle Yaw angle The three core features of attitude angle mean, extreme value, and rate of change are extracted to reflect the overall trend and dynamic changes of attitude. The formula is as follows: (1) Attitude angle mean: (2) Attitude angle extremes: (3) Rate of change of attitude angle: Formula explanation: This represents the number of sampling points within the feature extraction time window. The sampling period is consistent with that in the attitude calculation formula; All features are directly calculated from the final pose result. Calculations show that the accuracy of pose estimation directly determines the reliability of feature extraction; In step S223, multi-dimensional feature extraction also includes angular velocity dimension feature extraction. The angular velocity feature is based on the core input of attitude calculation and the filtered gyroscope data. Extraction, directly related to the gyroscope integral solution for attitude calculation, reflects the dynamic characteristics of the body's rotation, as shown in the following formula: (1) Triaxial mean angular velocity: (2) Standard deviation of angular velocity: Formula explanation: It is the core input for gyroscope integration in attitude calculation, and its fluctuation directly affects the predicted attitude angle value. The accuracy of the data is then correlated with the final attitude calculation result; in step S223, multi-dimensional feature extraction also includes acceleration dimension feature extraction, where the acceleration features are based on the initial input of the attitude calculation and filtered accelerometer data. Extraction, directly related to the initial solution of attitude angles, reflects the force and acceleration changes characteristics of the body's motion, as shown in the following formula: (1) Peak values ​​of acceleration in all three axes: (2) Magnitude of acceleration vector: Formula explanation: It is the core input for the initial solution of attitude angles, and its peak value and vector magnitude directly affect The initial solution accuracy is affected, and then the final attitude angle is influenced by complementary filtering fusion. .

Images