Intelligent physical education data management method and system

Abstract

The application relates to the technical field of data management, in particular to an intelligent sports teaching data management method and system. The method comprises the following steps: collecting motion video stream data and inertial motion data of students; mapping a camera coordinate system of the video stream data and a body coordinate system of the inertial motion data to a world coordinate system; constructing a state space model based on an extended Kalman filter, and iteratively calculating to obtain a corrected student three-dimensional skeleton posture sequence; calculating an optimal matching path and a similarity distance between the student three-dimensional skeleton posture sequence and a standard action template sequence; and generating a quantitative action score according to the similarity distance. The application eliminates integral drift while ensuring the anti-occlusion capability by using an extended Kalman filter algorithm; meanwhile, an improved dynamic time warping algorithm is introduced, the standardization and objectivity of action quality scoring are realized, and the scientificity, safety and individualization level of sports teaching are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of data management technology, specifically to an intelligent method and system for managing sports teaching data. Background Technology

[0002] In traditional physical education teaching, teachers rely on visual observation and subjective experience to guide and correct students' movements. This model has inherent limitations: First, the human eye struggles to capture the instantaneous technical details of high-speed movements (such as the wrist pronation speed during a badminton swing), resulting in a lack of precision in instruction; second, in large-class teaching environments, teachers cannot simultaneously ensure the proper form of movement for all students, making it difficult to implement individualized instruction.

[0003] Furthermore, the level of informatization in existing physical education teaching management varies greatly. Although some schools have introduced academic affairs management systems, these are mostly limited to administrative tasks such as course scheduling and grade entry, lacking in-depth involvement in the core aspects of "teaching, learning, practice, testing, and evaluation." The core asset in the physical education teaching process—sports data—has long been in a state of loss or fragmentation, failing to form effective "digital assets" to support the improvement of teaching quality.

[0004] To address the aforementioned issues, some auxiliary training systems based on single technologies have emerged in the existing technology field. However, these systems all suffer from significant technical bottlenecks in practical applications and are unable to meet the needs of complex and ever-changing sports teaching scenarios.

[0005] Current visual motion capture technology primarily relies on RGB cameras to capture key points on the human body. While this non-contact solution minimizes user disruption, it faces significant challenges in sports teaching scenarios:

[0006] Occlusion problem: In multi-player competitive sports (such as basketball and football) or sports with complex limb movements (such as yoga and martial arts), key points are easily lost due to occlusion by one's own limbs, sports equipment, or other people. Once key points are lost, vision-based pose estimation algorithms often produce severe jitter or incorrect skeletal connections, leading to scoring failure.

[0007] Environmental sensitivity: Changes in lighting conditions (such as direct sunlight outdoors or insufficient lighting in indoor venues) significantly affect the signal-to-noise ratio of image sensors, thereby reducing the confidence of key point detection.

[0008] Wearable devices (such as smart bracelets and smart insoles) use accelerometers, gyroscopes, and magnetometers to monitor movement. Their advantages include high sampling rates and immunity to obstruction and lighting conditions, but they also have the following inherent drawbacks:

[0009] Integral drift: Inertial navigation systems calculate displacement by integrating acceleration twice and attitude by integrating angular velocity once. Due to sensor noise (white noise, random walk) and initial alignment errors, the integration calculation accumulates errors over time. Without external calibration, the calculated position and attitude will deviate significantly from the true values ​​after long-term operation.

[0010] Lack of absolute position reference: Inertial sensors excel at measuring relative motion, but struggle to determine a person's absolute coordinates within a given area. For example, the system can track how many steps a student has taken, but it cannot determine if the student has veered off course.

[0011] Existing automatic scoring algorithms mostly employ simple Euclidean distance or cosine similarity comparisons. This method ignores individual student differences (height, wingspan) and variations in movement rhythm. For example, a student might complete a standard shooting motion slower than a professional athlete, and directly aligning the comparisons along the timeline would lead to significant errors. The lack of a universal evaluation algorithm capable of adapting to time distortion is a key challenge hindering the widespread adoption of intelligent teaching systems.

[0012] In conclusion, there is an urgent need for an intelligent sports teaching data management method and system that can integrate the advantages of visual and inertial data and adopt adaptive evaluation algorithms to address the shortcomings of current technologies. Summary of the Invention

[0013] To address the shortcomings of existing technologies, this invention provides an intelligent method and system for managing physical education teaching data.

[0014] An intelligent method for managing physical education teaching data includes the following steps:

[0015] By using visual acquisition devices deployed at the teaching site and inertial sensors worn by students, the system simultaneously collects students' motion video stream data and inertial motion data.

[0016] Through static initial alignment and T-Pose calibration, the rotation offset matrix between the inertial coordinate system and the visual world coordinate system is calculated, and the camera coordinate system of the video stream data and the body coordinate system of the inertial motion data are uniformly mapped to the world coordinate system.

[0017] A state-space model based on extended Kalman filter is constructed. The kinematic parameters calculated by the inertial sensor are used as the state prediction values, and the coordinates of human key points extracted by vision are used as the observation update values. The corrected three-dimensional skeletal posture sequence of the student is obtained through iterative calculation.

[0018] The system retrieves a preset standard motion template sequence from the cloud database and uses a dynamic time warping algorithm with a constraint window to calculate the optimal matching path and similarity distance between the student's 3D skeletal pose sequence and the standard motion template sequence.

[0019] A quantified action score is generated based on the similarity distance.

[0020] Preferably, the sampling frequency of the inertial sensor is adjusted in real time according to the student's motion state, specifically including: real-time monitoring of the acceleration variance of the inertial sensing device; when the variance is lower than a preset static threshold, switching the sampling frequency to a low-power mode; when the variance is higher than a preset motion threshold or a specific high-dynamic motion feature is detected, switching the sampling frequency to a high-performance mode.

[0021] Preferably, the state vector of the extended Kalman filter model includes at least the position, linear velocity, attitude quaternion, angular velocity of the human skeletal segment, and the zero bias of the inertial sensor's acceleration and the zero bias of the gyroscope; the coordinates of key human points extracted by vision are used as observation update values ​​to estimate and eliminate the zero bias of the acceleration and the zero bias of the gyroscope, so as to correct the integral drift of the inertial sensor.

[0022] Preferably, the noise covariance matrix of the observed update value is dynamically adjusted according to the confidence level output by the visual algorithm: when the confidence level of the visual key point is lower than a preset threshold, the value of the corresponding item in the noise covariance matrix is ​​increased, and the weight of the visual data in the fusion calculation is reduced, thereby suppressing the pose estimation jump caused by occlusion.

[0023] Preferably, the dynamic time warping algorithm with constraint windows uses Sakoe-Chiba strip windows or Itakura parallelogram windows to limit the degree to which the matching path deviates from the diagonal, preventing the forced matching of action segments with excessively large time span differences.

[0024] An intelligent sports teaching data management system includes a data acquisition module, a data processing module, and a data analysis module;

[0025] The data acquisition module is used to simultaneously acquire students' motion video stream data and inertial motion data through visual acquisition devices deployed at the teaching site and inertial sensors worn by students.

[0026] The data processing module calculates the rotation offset matrix between the inertial coordinate system and the visual world coordinate system through static initial alignment and T-Pose calibration, and maps the camera coordinate system of the video stream data and the body coordinate system of the inertial motion data to the world coordinate system.

[0027] The data analysis module is used to quantify students' action scores based on the collected motion video stream data and inertial motion data.

[0028] Preferably, the data analysis module includes an attitude correction unit, a distance calculation unit, and a score generation unit;

[0029] The correction unit is used to construct a state-space model based on extended Kalman filter, use the kinematic parameters calculated by the inertial sensor as the state prediction value, use the coordinates of human key points extracted by vision as the observation update value, and obtain the corrected user three-dimensional skeleton posture sequence through iterative calculation.

[0030] The distance calculation unit is used to calculate the optimal matching path and similarity distance between the user's 3D skeletal pose sequence and the standard motion template sequence using a dynamic time warping algorithm with a constraint window.

[0031] The scoring generation unit is used to generate a quantified action score based on the similarity distance.

[0032] Compared with the prior art, the advantages of this invention are:

[0033] By fusing visual and inertial sensing data and utilizing the Kalman filter system to achieve complementary advantages, high-precision 3D pose estimation is maintained even in occluded and high-dynamic scenes.

[0034] An improved Dynamic Time Warping (DTW) algorithm is introduced to eliminate interference caused by differences in motion rates, thereby achieving standardized and objective scoring of motion quality. Attached Figure Description

[0035] Figure 1 This is a flowchart of an intelligent sports teaching data management method proposed in this invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0037] Reference Figure 1 This invention provides an intelligent method for managing physical education teaching data.

[0038] S1. Simultaneously collect students' motion video stream data and inertial motion data through visual acquisition equipment deployed at the teaching site and inertial sensing equipment worn by students.

[0039] The visual acquisition equipment uses an industrial-grade RGB-D depth camera to ensure that the resolution of the video stream is no less than 1920×1080@60fps and the depth stream is no less than 1280×720@60fps. It also uses a global shutter to avoid the rolling shutter effect when shooting high-speed motion (such as swinging a racket) and to ensure the geometric realism of each frame.

[0040] In one specific embodiment, such as deployment in a standard basketball or badminton court, 4-6 units can be deployed around the perimeter of the court using a "ring-shaped redundant coverage" method. The installation height is 2.5 meters to 3.5 meters, with a downward tilt angle of 15° to 30° to minimize mutual obstruction between players.

[0041] The inertial sensing device uses a low-power Bluetooth SoC, supports BLE 5.0 / 5.2 protocols, and has a floating-point unit (FPU) to support simple local quaternion calculations. The standard configuration is a 5-point system (both wrists, both ankles, and the center of the waist). The waist node serves as the root node, responsible for aggregating limb data or serving as a reference for the body's center of mass.

[0042] To balance data accuracy and battery life, the device incorporates an adaptive sampling rate algorithm.

[0043] Low power mode (25Hz-50Hz): Reduces the sampling rate when the acceleration variance remains below the threshold (e.g., when stationary or walking slowly).

[0044] High Dynamic Range (100Hz-200Hz): Automatically increases the sampling rate when drastic motion features are detected.

[0045] In addition, before data acquisition, the visual acquisition device and the inertial sensing device are synchronized using the Network Time Protocol to ensure that the timestamp error between the video frame and the inertial sensing data packet is less than 1ms.

[0046] S2. Through static initial alignment and T-Pose calibration, the rotation offset matrix between the inertial coordinate system and the visual world coordinate system is calculated, and the camera coordinate system of the visual data and the body coordinate system of the inertial data are uniformly mapped to the world coordinate system.

[0047] in:

[0048] World coordinate system: Defined by the visual system. It is usually based on the center of the field as the origin, with the Z-axis pointing vertically upwards.

[0049] Navigation coordinate system: defined by inertial sensing equipment. The Z-axis is parallel to the gravity vector (pointing upwards), and the Y-axis points to the geomagnetic north pole (or true north after magnetic declination correction).

[0050] Body coordinate system: fixed to the sensor chip, rotating with the movement of the human body.

[0051] In a preferred embodiment, the specific mapping method is as follows:

[0052] Static gravity alignment:

[0053] During the system initialization phase, the wearer should remain standing still.

[0054] The inertial sensing device measures the component of the gravity vector in the sensor coordinate system as follows.

[0055] The vision system uses the PnP algorithm to calculate the attitude rotation matrix of the sensor (or the body part to which it is attached) in the world coordinate system.

[0056] Calculate the rotation matrix so that the gravity vector of the inertial sensor is parallel to the Z-axis of the visual coordinate system:

[0057] ;

[0058] In the formula, The Z-axis vector of the visual coordinate system. The gravity vector of the inertial sensing device. It is a rotation matrix.

[0059] Heading angle alignment:

[0060] Because the magnetometer is affected by interference from indoor ferromagnetic materials, relying solely on the magnetic north pole is unreliable. The system employs "visual-assisted heading correction".

[0061] The user stands facing the positive X-axis of the visual coordinate system (T-Pose ready posture).

[0062] The heading angle output by the inertial sensor is The body orientation angle measured by the visual system is (Reference value).

[0063] Calculate heading deviation .

[0064] Final unified rotation matrix: This matrix will store the data in the inertial navigation coordinate system. Mapped to the visual world coordinate system, i.e. .

[0065] Sensor-to-bone calibration:

[0066] The user performs a T-Pose (arms raised horizontally).

[0067] The system records the attitude of each sensor at this time. .

[0068] Assuming that the skeletal coordinate system is aligned with the world coordinate system in T-Pose, then the calibration matrix is ​​calculated. .

[0069] In subsequent movements, the actual skeletal posture This eliminates errors caused by the arbitrary angle at which the sensor is worn.

[0070] S3. Construct a state-space model based on extended Kalman filter, use the kinematic parameters calculated by the inertial sensor as the state prediction value, use the coordinates of human key points extracted by vision as the observation update value, and obtain the corrected user 3D skeletal pose sequence through iterative calculation.

[0071] S301. Define the state vector: The system state vector includes the motion states of each skeletal segment of the human body.

[0072] ;

[0073] In the formula, Let be the system state vector. This represents the position of the bone segment in the world coordinate system. Linear velocity, Let be the quaternion of the pose. Angular velocity, and This is for the zero bias of the accelerometer and gyroscope.

[0074] S302, State Prediction: Using the kinematic parameters calculated by the inertial sensor, the state is recursively predicted to determine the state at the next moment.

[0075] ;

[0076] In the formula, and These are the measurements from the accelerometer and gyroscope, respectively. The vector of gravitational acceleration. This is the rotation matrix corresponding to the quaternion. This is quaternion multiplication.

[0077] This step not only updates the position and attitude, but also transmits the error covariance matrix. :

[0078] ;

[0079] In the formula, The Jacobian matrix of the state transition matrix. Let be the process noise covariance.

[0080] S303 Measurement Update: When the visual acquisition device captures a new frame of image, the captured data is used to correct the predicted value of the inertial sensor.

[0081] Visual algorithms (such as YOLO-Pose) output 2D pixel coordinates. By combining these with depth values ​​from a depth camera, 3D observation coordinates can be obtained through backprojection. .

[0082] Update equation:

[0083] ;

[0084] In the formula, This is the corrected system state vector. The joint position as observed visually. This is the observation matrix (i.e., the direct observation locations). To observe the noise covariance matrix, its value is dynamically adjusted: when the confidence level of the visual algorithm output is low (e.g., due to occlusion), it is increased. The value of makes the filter trust the IMU's predictions more; conversely, it trusts visual observations more.

[0085] Through the above process, the high-frequency characteristics of the inertial sensor ensure smooth movement and rapid response, while the absolute position information from vision continuously corrects the zero bias of the inertial sensor. and This eliminates integral drift and enables stable tracking over a long period of time.

[0086] S4. Retrieve a preset standard motion template sequence from the cloud database, and use a dynamic time warping algorithm with a constraint window to calculate the optimal matching path and similarity distance between the user's 3D skeletal pose sequence and the standard motion template sequence.

[0087] The system stores a library of standard movement templates covering various sports in its cloud database. These templates were recorded by professional athletes and have undergone data cleaning and smoothing processes.

[0088] It is understandable that the lengths of student action sequences and standard sequences are usually different (due to varying action speeds). Direct frame-by-frame comparison will lead to errors. This system uses the DTW algorithm to find the optimal nonlinear mapping between the two sequences. The specific steps are as follows:

[0089] S401. Define the characteristic sequence:

[0090] The student action feature sequence is represented as follows: ,in Indicates the first The feature vector of the frame (such as key joint angles or quaternions).

[0091] The standard template feature sequence is represented as: ,in Indicates the first The feature vector of a frame.

[0092] in, and These represent the frame numbers of the two sequences, typically... .

[0093] S402. Construct and populate the cumulative distance matrix:

[0094] To prevent unreasonable excessive time distortion, a constraint window size is set. (For example, 10% of the sequence length). Only when Time calculation Otherwise, set it to infinity.

[0095] The formula for calculating the value of each element in the matrix is:

[0096] ;

[0097] In the formula, The th in the cumulative distance matrix Line number The element values ​​of the column, For local distance metrics, Euclidean distance is used for joint angles and inner product distance is used for quaternions.

[0098] S403, Backtracking to find the optimal matching path:

[0099] The optimal matching path is a sequence of indices in the cumulative distance matrix, specifically:

[0100] From the finish line Start backtracking until the starting point. .

[0101] At every step ,choose The node with the smallest cumulative distance value is taken as the previous node of the path.

[0102] The final determined sequence is the optimal matching path, which can intuitively reflect when the student's actions are too fast or too slow.

[0103] S404. Calculate the normalized similarity distance:

[0104] The original distance is That is, the th in the cumulative distance matrix Line number To eliminate the influence of sequence length on the score, the element values ​​of the column are normalized based on the original distance:

[0105] ;

[0106] S5. Generate a quantified action score based on the similarity distance.

[0107] The specific formula for calculating the action score is as follows:

[0108] ;

[0109] In the formula, Rate the action. The base score, For similarity distance, This is the attenuation coefficient, used to control the strictness of the scoring:

[0110] For beginners, a lower setting can be used. Even with a certain margin of error, a value (such as 0.05) will result in a slower decline in the score, thus serving as an encouragement.

[0111] For professionals, a higher setting can be made. Even a small difference in movement (such as 0.2) can cause the score to drop sharply, requiring extremely high precision.

[0112] An intelligent sports teaching data management system includes a data acquisition module, a data processing module, and a data analysis module.

[0113] The data acquisition module is used to simultaneously collect students' motion video stream data and inertial motion data through visual acquisition devices deployed at the teaching site and inertial sensors worn by students.

[0114] The data processing module calculates the rotation offset matrix between the inertial coordinate system and the visual world coordinate system through static initial alignment and T-Pose calibration, and maps the camera coordinate system of the video stream data and the body coordinate system of the inertial motion data to the world coordinate system.

[0115] The data analysis module is used to quantify students' action scores based on the collected motion video stream data and inertial motion data.

[0116] The data analysis module includes an attitude correction unit, a distance calculation unit, and a score generation unit.

[0117] The correction unit is used to construct a state-space model based on extended Kalman filtering, use the kinematic parameters calculated by the inertial sensor as the state prediction value, use the coordinates of human key points extracted by vision as the observation update value, and obtain the corrected user three-dimensional skeletal posture sequence through iterative calculation.

[0118] The distance calculation unit is used to calculate the optimal matching path and similarity distance between the user's 3D skeletal pose sequence and the standard motion template sequence using a dynamic time warping algorithm with a constraint window.

[0119] The scoring generation unit is used to generate a quantified action score based on the similarity distance.

[0120] In this solution, a hybrid database architecture is adopted to support the storage and efficient querying of massive amounts of data.

[0121] In a preferred embodiment, the specific database schema definition includes:

[0122] Table 1: Basic Information Table

[0123]

[0124] Table 2: Course and Training Session Schedule

[0125]

[0126] Table 3: Action Evaluation Results Table

[0127]

[0128] Table 1 records students' basic information such as height, weight, and other relevant data. Table 2 is the course and training session table, which records specific information about a single course or training session, such as course type, exercise type, and environment data. Table 3 is the movement evaluation result table, which links the movement evaluation results of the training to specific students and courses through foreign keys to facilitate data storage and efficient querying.

[0129] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0130] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent method for managing physical education teaching data, characterized in that, Includes the following steps: By using visual acquisition devices deployed at the teaching site and inertial sensors worn by students, the system simultaneously collects students' motion video stream data and inertial motion data. Through static initial alignment and T-Pose calibration, the rotation offset matrix between the inertial coordinate system and the visual world coordinate system is calculated, and the camera coordinate system of the video stream data and the body coordinate system of the inertial motion data are uniformly mapped to the world coordinate system. A state-space model based on extended Kalman filter is constructed. The kinematic parameters calculated by the inertial sensor are used as the state prediction values, and the coordinates of human key points extracted by vision are used as the observation update values. The corrected three-dimensional skeletal posture sequence of the student is obtained through iterative calculation. The state vector of the extended Kalman filter model includes at least the position, linear velocity, attitude quaternion, angular velocity of the human skeletal segment, and the zero bias of the inertial sensor's acceleration and the zero bias of the gyroscope. The coordinates of key human points extracted by vision are used as observation update values ​​to estimate and eliminate the zero bias of the acceleration and the zero bias of the gyroscope, so as to correct the integral drift of the inertial sensor. The noise covariance matrix of the observed update value is dynamically adjusted according to the confidence level output by the visual algorithm: when the confidence level of the visual key point is lower than the preset threshold, the value of the corresponding item in the noise covariance matrix is ​​increased, and the weight of the visual data in the fusion calculation is reduced, thereby suppressing the pose estimation jump caused by occlusion. The system retrieves a preset standard motion template sequence from the cloud database and uses a dynamic time warping algorithm with a constraint window to calculate the optimal matching path and similarity distance between the student's 3D skeletal pose sequence and the standard motion template sequence. A quantified action score is generated based on the similarity distance.

2. The intelligent physical education teaching data management method according to claim 1, characterized in that, The sampling frequency of the inertial sensor is adjusted in real time according to the student's motion state. Specifically, this includes: real-time monitoring of the acceleration variance of the inertial sensor; when the variance is lower than a preset static threshold, switching the sampling frequency to a low-power mode; and when the variance is higher than a preset motion threshold, switching the sampling frequency to a high-performance mode.

3. The intelligent physical education teaching data management method according to claim 1, characterized in that, The dynamic time warping algorithm with constraint windows uses Sakoe-Chiba strip windows or Itakura parallelogram windows to limit the degree to which the matching path deviates from the diagonal, preventing the forced matching of action segments with excessively large time span differences.

4. An intelligent physical education teaching data management system, used to implement the intelligent physical education teaching data management method according to any one of claims 1-3, characterized in that, It includes a data acquisition module, a data processing module, and a data analysis module; The data acquisition module is used to simultaneously acquire students' motion video stream data and inertial motion data through visual acquisition devices deployed at the teaching site and inertial sensors worn by students. The data processing module calculates the rotation offset matrix between the inertial coordinate system and the visual world coordinate system through static initial alignment and T-Pose calibration, and maps the camera coordinate system of the video stream data and the body coordinate system of the inertial motion data to the world coordinate system. The data analysis module is used to quantify students' action scores based on the collected motion video stream data and inertial motion data.

5. The intelligent physical education teaching data management system according to claim 4, characterized in that, The data analysis module includes an attitude correction unit, a distance calculation unit, and a score generation unit; The posture correction unit is used to construct a state-space model based on extended Kalman filter, use the kinematic parameters calculated by the inertial sensor as the state prediction value, use the coordinates of human key points extracted by vision as the observation update value, and obtain the corrected user three-dimensional skeleton posture sequence through iterative calculation. The distance calculation unit is used to calculate the optimal matching path and similarity distance between the user's 3D skeletal pose sequence and the standard motion template sequence using a dynamic time warping algorithm with a constraint window. The scoring generation unit is used to generate a quantified action score based on the similarity distance.

Images