Skiing posture recognition system

Abstract

The invention relates to the technical field of somatosensory interaction, and discloses a skiing posture recognition system, which comprises a data acquisition module and an inertial measurement unit sensor network deployed in a plurality of limb links of a skier and used for acquiring movement data of limbs; the positioning module is used for collecting the position and trajectory data of the skier; and the data processing and synchronizing module is used for receiving and processing the data from the data acquisition module and carrying out time synchronization on the motion data and the position and track data. According to the skiing posture recognition system, multi-source abstract data of an IMU sensor, a plantar pressure sensor and GNSS positioning are converted into three-dimensional dynamic playback consistent with real actions through three-dimensional posture reconstruction and ski track scene copying, and quantitative data labels such as key joint angles and vertical blade angles are matched, so that the problems that data are difficult to understand and postures are difficult to distinguish in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of motion-sensing interaction technology, and in particular to a skiing posture recognition system. Background Technology

[0002] With the rapid development of motion-sensing interaction and wearable computing technologies, human motion capture and posture recognition based on inertial sensors has become a research hotspot in the field of human-computer interaction. This technology, through the deployment of miniature sensor nodes on key parts of the human body, captures kinematic data of the limbs in real time and reconstructs the human's posture in three-dimensional space using algorithmic models. It has been widely applied in virtual reality, human-computer interaction, and sports science analysis. Applying this technology to skiing, enabling accurate identification and quantitative analysis of skiers' postures, is of great significance for scientific training, skill improvement, and sports injury prevention. This is the core driving force behind the development of skiing posture recognition technology.

[0003] However, existing ski posture recognition solutions still have significant shortcomings. First, most systems focus on recording basic motion data or simple action counting based on threshold judgments, lacking refined deconstruction and visual reproduction of continuous and complex ski postures. This results in unintuitive feedback, making it difficult for skiers to associate abstract data with their specific posture deficiencies. Second, existing solutions typically separate motion recognition from spatial location information, failing to accurately bind identified posture anomalies to specific geographical locations on the slope for post-flight visualization and review. Furthermore, they fail to utilize this binding relationship to provide skiers with location-based, predictive, real-time corrective prompts during subsequent skiing, leading to severe feedback lag and low training efficiency. Summary of the Invention

[0004] The technical problem to be solved by the present invention is that the existing technology has the disadvantage of poor predictability and correction. To address this, we propose a skiing posture recognition system.

[0005] To achieve the above objectives, this application adopts the following technical solution: a skiing posture recognition system, comprising:

[0006] The data acquisition module consists of an inertial measurement unit sensor network deployed in multiple limb segments of the skier to collect limb motion data; and the positioning module is used to collect the skier's position and trajectory data.

[0007] The data processing and synchronization module is used to receive and process data from the data acquisition module, and to synchronize the motion data with the position and trajectory data in time.

[0008] The 3D pose reconstruction module is used to drive a predefined 3D skeletal model based on synchronized data and reconstruct the corresponding ski slope scene in the virtual environment to generate a 3D dynamic playback that is consistent with the skier's real movements.

[0009] The attitude analysis module has a built-in standard attitude knowledge base, which is used to compare the attitude in 3D dynamic playback with the standard attitude and identify attitude anomalies.

[0010] The visualization and interaction module is used to display the results of the 3D dynamic playback and attitude analysis module, and to receive user marking instructions for attitude anomalies and their locations.

[0011] The closed-loop feedback control module stores user-marked posture anomalies, corresponding standard posture data, and geographical location. During subsequent skiing, when it is predicted that the skier will reach the marked position, the current motion data is compared with the standard data in real time. If a posture anomaly is detected, the tactile feedback device located on the corresponding limb is triggered to provide a prompt.

[0012] Preferably, the inertial measurement unit sensor network includes multiple 9-axis IMU sensor nodes, which are deployed on the main parts of the skier's upper and lower limbs via elastic straps or fixed bases.

[0013] Preferably, the positioning module is a GNSS receiver integrated into a mobile terminal or a standalone host.

[0014] Preferably, the data processing and synchronization module achieves data synchronization by adding timestamps from the same clock source to all data sources and using an interpolation algorithm to time-align data with inconsistent sampling periods.

[0015] Preferably, the three-dimensional posture reconstruction module uses an inverse kinematics algorithm to convert the calculated absolute limb posture information into rotational drive signals for each joint in the three-dimensional skeletal model.

[0016] Preferably, the visualization interaction module supports users to observe the 3D dynamic playback from multiple perspectives, slow down, and pause, and overlays angle data labels of key joints on the model.

[0017] Preferably, the closed-loop feedback control module includes:

[0018] The rule management unit is used to store correction rules that include posture anomaly descriptions, standard posture characteristics, triggering geographical locations, and fault tolerance thresholds.

[0019] The real-time judgment unit is used to predict the time of arrival at the triggered geographical location based on the current positioning information, and to initiate real-time attitude comparison within the warning window;

[0020] The tactile actuation unit is used to control the tactile feedback device of the target limb to generate vibration when the comparison result exceeds the fault tolerance threshold.

[0021] Preferably, the tactile feedback device is a linear resonant actuator integrated into the sensor node of the inertial measurement unit.

[0022] Preferably, the data acquisition module further includes a foot pressure sensing unit integrated into the ski shoe insole, used to collect data on the distribution of the skier's center of gravity; the posture analysis module is also used to assess the skier's balance state based on the center of gravity distribution data.

[0023] Preferably, the posture analysis module further includes a motion classifier, which employs a machine learning model based on a one-dimensional convolutional neural network or a long short-term memory network to automatically identify the type of skiing motion being performed by the skier based on temporal motion data.

[0024] The technical effects and advantages of this invention are as follows:

[0025] In this invention, multi-source abstract data from IMU sensors, plantar pressure sensors, and GNSS positioning are transformed into a three-dimensional dynamic playback consistent with real movements through three-dimensional posture reconstruction and ski slope scene replication. Combined with quantitative data tags such as key joint angles and edge angles, this solves the problems of difficult-to-understand data and difficult-to-identify postures in traditional technologies. This allows skiers to intuitively perceive the relationship between their own movements and the ski slope position. In addition, user-marked posture anomalies are bound to geographical locations, and arrival time is predicted by sliding average speed. Posture data is compared in real time within the warning window, and targeted tactile feedback is provided by a linear resonant actuator integrated into the IMU node, effectively helping skiers form correct muscle memory. Attached Figure Description

[0026] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0027] Figure 1 This is a flowchart of the process of the present invention;

[0028] Figure 2 This is a system module logical architecture diagram of the present invention. Detailed Implementation

[0029] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0030] Reference Figures 1-2 As shown, the present invention provides a technical solution: a skiing posture recognition system, comprising:

[0031] The data acquisition module is an inertial measurement unit (IMU) sensor network deployed across multiple limb segments of the skier to collect limb motion data. This IMU sensor network includes multiple 9-axis IMU sensor nodes, which are deployed on the skier's upper and lower limbs via elastic straps or fixed bases. Seven 9-axis IMU sensors (model: MPU-9250) are selected and fixed to the skier's key limbs via elastic straps. These straps are positioned 5cm above the elbow joints of the left and right upper arms, 5cm above the wrist joints of the left and right forearms, level with the sacrum in the lower back, and 8cm above the ankle joints of the left and right lower legs. The straps are tightened to ensure a snug fit between the sensor and clothing or ski boots. The sensors perform a self-test upon power-up and initialize the accelerometer, gyroscope, and magnetometer via the I2C interface. The sampling frequency is set to 200Hz.

[0032] The user maintains a standard standing posture for 30 seconds. During this period, the system collects static data from each IMU sensor. For the accelerometer, the measurement should only reflect the acceleration due to gravity; for the gyroscope, the measurement should be approximately zero. The system calculates the average value of the data for each axis over these 30 seconds and stores this as zero bias, which is subtracted in real time from subsequent dynamic data.

[0033] To prevent interference from iron structures (such as cable car supports) in the ski resort on geomagnetic measurements, the system guides users to wave their handheld devices or wearable sensors in a figure-eight pattern. By collecting a large amount of magnetometer data in different orientations, an ellipsoidal fitting algorithm is used to calibrate the raw magnetometer readings, which are affected by interference from hard and soft iron, onto a standard spherical surface, thus obtaining an accurate heading reference. Due to slight differences in wearing position, the body coordinate system of each sensor is not completely consistent with the human skeletal coordinate system. The system guides users to perform a series of standard movements, such as arm swings in place and squats, and uses algorithms to automatically estimate a fixed rotation matrix from each IMU sensor to its corresponding limb skeletal segment. This matrix will be used in subsequent attitude calculations to unify all sensor data into a human-centered biomechanical coordinate system.

[0034] After deployment, start the sensor self-calibration program, collect 30 seconds of static data, and complete zero-drift calibration and coordinate system initialization.

[0035] The positioning module is a GNSS receiver integrated into a mobile terminal or a standalone host, used to collect the skier's location and trajectory data. It uses a dual-mode GNSS receiver, model UbloxF9P, which is integrated into the user's smartphone or a dedicated portable host and enables multi-band positioning mode.

[0036] During initialization, ephemeris data download and base station differential signal synchronization are completed. The positioning sampling frequency is set to 10Hz, and the dynamic positioning accuracy is calibrated to ≤2 meters. During gliding, the three-dimensional position coordinates are collected in real time and set as X, Y, Z, altitude as H and movement speed as V. Each frame of data is accompanied by a high-precision timestamp from the mobile phone system (accuracy ≤1ms).

[0037] The data acquisition module also includes a plantar pressure sensing unit integrated into the ski boot insole, used to collect data on the distribution of the skier's center of gravity. The posture analysis module is further used to assess the skier's balance based on the center of gravity distribution data. The sensing insole is placed inside the ski boot, ensuring the sensing points are aligned with key pressure areas on the sole of the foot. Communication with the main terminal is established via Bluetooth Low Energy, with the sampling frequency set to 100Hz. During the data acquisition phase, the pressure values ​​at each sensing point are acquired in real time. And calculate the coordinates of the pressure center point as the core data for the center of gravity distribution.

[0038] A timed interrupt-triggered sampling mechanism is adopted, with the IMU sensor generating a sampling interrupt every 5ms to synchronously collect acceleration, angular velocity, and magnetometer data. The GNSS receiver and plantar pressure sensor collect data synchronously according to their respective sampling frequencies. All data is transmitted to the terminal via Bluetooth protocol, with transmission delay controlled to ≤50ms. The raw data is initially denoised using a sliding window mean filter.

[0039] Taking the X-axis acceleration data from an IMU sensor as an example: ,in, This is the filtered X-axis acceleration data for the k-th frame. For the original data, such as The original values ​​up to frame k are 1.2g, 1.3g, 1.25g, 1.32g, and 1.28g, respectively. Calculate... It is used to suppress high-frequency noise.

[0040] The data processing and synchronization module is used to receive and process data from the data acquisition module, and to synchronize the motion data with the position and trajectory data in time. The data processing and synchronization module achieves data synchronization by adding timestamps from the same clock source to all data sources and using an interpolation algorithm to time-align data with inconsistent sampling periods.

[0041] It receives raw data from the IMU sensor, GNSS receiver, and plantar pressure sensor, and then... The criteria for removing outliers are as follows: if the data deviates from the mean by more than three times the standard deviation, it is replaced with valid data from the previous frame. LMS adaptive filtering is applied to the IMU sensor data for further noise reduction and to improve data stability. If the filter order is set to 8, the step size factor is... .

[0042] Add an absolute timestamp from the mobile phone system clock to all data sources to establish a unified time reference. Identify the sampling period of each data source and associate the synchronized IMU sensor attitude data, GNSS receiver position trajectory data, and plantar pressure center of gravity data to generate a dataset. Each data record contains a timestamp t, limb movement parameters, position parameters, and center of gravity parameters, in JSON format for easy calling by subsequent modules.

[0043] If an improved Madgwick gradient descent algorithm is used to fuse multi-source data, the absolute limb pose quaternion can be solved. Core update formula step size factor :

[0044] ,in The pose quaternion of the previous frame, and the error vector. Its model Substitute Attitude calculation error .

[0045] The 3D pose reconstruction module is used to drive a predefined 3D skeletal model based on synchronized data and reconstruct the corresponding ski slope scene in the virtual environment to generate a 3D dynamic playback that is consistent with the skier's real movements.

[0046] Extracting GNSS receiver trajectory point clouds from the dataset ,like , The system removes duplicate and outlier points, retains valid trajectory points, and uses the Delaunay triangulation algorithm to model the terrain of the trajectory point cloud area, generating a triangular mesh surface. Based on elevation data, it identifies key nodes of the ski slope, including steep slopes ≥25°, gentle slopes ≤15°, and turning points, and automatically labels them.

[0047] The system constructs a 17-joint human skeleton model, including the head, cervical spine, shoulders, elbows, wrists, hips, knees, ankles, and pelvis. It establishes the correspondence between IMU sensors and skeletal segments. Through inverse kinematics algorithms, it converts the posture quaternions calculated by the IMU receiver into local rotation angles of the skeletal joints, driving the model to reproduce skiing movements. It uses Bezier curve interpolation to smooth the joint angles of adjacent frames, ensuring natural and smooth movements without stuttering or abrupt changes.

[0048] The absolute attitude calculated by the IMU sensor is converted into skeletal joint angles. A hybrid inverse kinematics solution strategy combining analytical and optimization methods is employed. For example, when solving for the knee joint angle, the thigh and lower leg are considered as a two-link system. Given the positions of the pelvis (hip joint) and ankle (end effector) in the global coordinate system, as well as the bone lengths of the thigh and lower leg, the flexion angle of the knee joint can be directly calculated analytically using geometric relationships such as the law of cosines.

[0049] By integrating the ski slope scene with the skeletal model, PBR rendering technology is used to achieve real-time rendering with a rendering frame rate of ≥30fps. Data labels, including joint angles and edge angles, are superimposed on key joints of the model to generate a complete 3D dynamic playback video.

[0050] For example, in Delaunay triangulation, three trajectory points are selected:

[0051] A(1200.5, 350.2, 1800.3);

[0052] B(1205.8,352.1,1798.6);

[0053] C(1203.2,351.5,1799.4);

[0054] Verify whether any point D(1202.8, 350.9, 1799.8) in the plane satisfies:

[0055] ;

[0056] Determine that A, B, and C form a Delaunay triangle to ensure the rationality and smoothness of the snow track terrain model, with an undulation restoration error of ≤0.5 meters.

[0057] The attitude analysis module has a built-in standard attitude knowledge base, which is used to compare the attitude in 3D dynamic playback with the standard attitude and identify attitude anomalies.

[0058] First, time-series data from the IMU receiver is extracted within a 2-second time window to form a 14×400 multidimensional input dataset, including data from 7 sensors and acceleration or angular velocity. This input data is then fed into a 1D-CNN-LSTM action classifier, an end-to-end deep learning model that automatically learns features from the raw sensor data and identifies actions, outputting the probability distributions for snowplow turns, parallel turns, and carabiner turns.

[0059] First, a time window, such as 2 seconds, is received, corresponding to 400 time points. The acceleration and angular velocity data of all 7 IMU sensors are used to form a 14-channel × 400 time point two-dimensional tensor, which contains 2-3 one-dimensional convolutional layers and uses small convolutional kernels.

[0060] Each convolutional layer is followed by a ReLU activation function and a one-dimensional max pooling layer. The one-dimensional convolution slides along the time dimension, enabling the extraction of feature patterns within local time segments, such as "a specific angular velocity pulse pattern at the ankle during the start of a turn." The pooling layer provides a certain degree of time invariance, inputting the feature sequence extracted by the CNN into a bidirectional LSTM layer.

[0061] LSTM is designed specifically for sequential data. Its internal gating mechanism, including the input gate, forget gate, and output gate, enables it to learn long-distance dependencies. For example, it can identify that a "parallel turn" is a complete and coherent process from pressing down the center of gravity, raising the blade, gliding, to getting up and releasing, rather than just a posture at a certain moment. It connects a fully connected layer and a Softmax activation function to output the probability of each action category.

[0062] For advanced skiers, distinguishing between carbines and slalom turns is a key technical point. These two movements may be similar in local posture, but their rhythm and sequence patterns are different.

[0063] Carbine turns: large turning radius, full curves, sustained high pressure in the turn, and long period of attitude stability. LSTM will learn this timing pattern of "long-term stable edge standing".

[0064] Short turns: small turning radius, fast pace, frequent weight transfer, and rapid alternation of edge-setting and decompression actions. LSTM can learn this timing pattern of "high frequency, short-duration edge-setting".

[0065] By capturing these temporal dynamic features, the classifier can distinguish between the two with high accuracy, thereby calling upon different standard knowledge bases for more precise analysis.

[0066] For example, the probability of a snowplow turn is 0.96 and a parallel turn is 0.03. The category with the highest probability is the current action type. Key parameters are extracted from the three-dimensional skeletal model, such as the knee flexion angle of 45°, the hip extension angle of 30°, the left board edge angle of 12°, the right board edge angle of 17°, the body tilt angle of 8°, the center of gravity position in the front-to-back direction of the skis, and the left-to-right direction of the skis.

[0067] Calculate the symmetry of parameters between the left and right limbs: Difference in edge angle = 17° - 12° = 5;

[0068] Based on the action classification results, the standard parameter range of the corresponding action is called from the standard posture knowledge base. For example, for the basic snowplow turn, the knee flexion angle is controlled between 30° and 50°, the edge angle is between 5° and 15°, and the difference between the left and right edge angles is ≤5°. The deviation between the actual parameters and the standard parameters is calculated. The deviation of the right edge angle is 17° - 15° = 2°, and the difference between the left and right edge angles is 5°, which is equal to the fault tolerance threshold. Therefore, it is determined that the edge angle is asymmetrical.

[0069] Based on plantar pressure data, pressure distribution uniformity and center of gravity stability were calculated. A weighted scoring method was used to generate a balance score, with pressure distribution uniformity accounting for 40% and center of gravity stability accounting for 60%.

[0070] In the calculation of equilibrium evaluation parameters, the uniformity of pressure distribution is calculated using the following formula:

[0071] Where Uniformity is the pressure distribution uniformity coefficient, ranging from [0,1]. The closer it is to 1, the more uniform the pressure distribution. The total pressure on the left foot, Total pressure on the right foot; The pressure is evenly distributed between the left and right feet, that is ;

[0072] In the calculation of center of gravity stability, Stability is the centroid stability coefficient, which ranges from [0,1], with values ​​closer to 1 indicating greater centroid stability.

[0073] The length of the trajectory of the pressure center point;

[0074] The maximum permissible trajectory length for the center of gravity shift;

[0075] In the balanced scoring formula:

[0076] ;

[0077] Where A is the balanced total score, with a value range of [1, 100] points;

[0078] The weight for pressure distribution uniformity is set to 0.4, which is equivalent to 40%.

[0079] This is the weight for the stability of the center of gravity, with a value of 0.6, which means it accounts for 60%.

[0080] The visualization and interaction module is used to display the results of the 3D dynamic playback and posture analysis module, and to receive user marking instructions for posture anomalies and their locations. The visualization and interaction module supports users to observe the 3D dynamic playback from multiple perspectives, slow down, and pause, and to overlay and display the angle data labels of key joints on the model.

[0081] By analyzing 3D dynamic playback data, it provides basic controls such as play, pause, stop, and speed adjustment. Based on the timeline and ski slope map, it provides a quick jump function, allowing users to click on any node to jump directly to the playback screen of the corresponding time period.

[0082] To maximize the effectiveness of the review, the system offers powerful comparative analysis capabilities. Users can choose to display their own 3D model alongside or overlay the system's built-in standard motion model in the same snow scene. The system renders the two models using different colors (e.g., blue for the user and green for the standard), making the differences immediately apparent. In overlay mode, the system can draw deviation vectors between the joints of the user's model. For example, if the user's knee buckles inward, a red arrow will be drawn from the standard knee joint position to the user's knee joint position. The direction and length of the arrow visually indicate the direction and severity of the error.

[0083] A closed-loop feedback control module is used to store user-marked posture anomalies and corresponding standard posture data and geographical locations. During subsequent skiing, when it is predicted that the skier will reach the marked location, the current motion data is compared with the standard data in real time. If a posture anomaly is detected, a tactile feedback device located on the corresponding limb is triggered to provide a prompt. The closed-loop feedback control module includes:

[0084] The rule management unit is used to store correction rules that include posture anomaly descriptions, standard posture characteristics, triggering geographical locations, and fault tolerance thresholds.

[0085] The real-time judgment unit is used to predict the time of arrival at the triggered geographical location based on the current positioning information, and to initiate real-time attitude comparison within the warning window;

[0086] A tactile actuation unit is used to control the tactile feedback device of the target limb to generate vibration when the comparison result exceeds the fault tolerance threshold. The tactile feedback device is a linear resonant actuator integrated in the sensor node of the inertial measurement unit.

[0087] By encapsulating the abnormal information marked by users into correction rules, each rule includes: abnormality type, right plate edge angle too large, standard parameter edge angle, triggering geographical location, fault tolerance threshold, feedback mode, and associated limb.

[0088] During subsequent gliding, the positioning data and gliding speed of the GNSS receiver are acquired in real time, and the straight-line distance between the current position and the trigger position is calculated.

[0089] When the predicted arrival time is less than or equal to 5 seconds, an early warning window is activated, and the corresponding correction rule is activated. Within the early warning window, the sampling frequency of the IMU sensor is increased, and the right plate vertical blade angle parameter is extracted in real time.

[0090] Calculate the deviation between real-time parameters and standard parameters. If the attitude is normal, no vibration feedback is triggered. If the real-time edge angle deviates during subsequent gliding, vibration is triggered.

[0091] The posture analysis module also includes a motion classifier, which uses a machine learning model based on a one-dimensional convolutional neural network or a long short-term memory network to automatically identify the type of skiing motion that the skier is performing based on time-series motion data.

[0092] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A skiing posture recognition system, characterized in that, include: The data acquisition module is an inertial measurement unit sensor network deployed in multiple limb segments of the skier to collect limb motion data; And a positioning module, used to collect the skier's location and trajectory data; The data processing and synchronization module is used to receive and process data from the data acquisition module and synchronize the motion data with the position and trajectory data in time; the three-dimensional posture reconstruction module is used to drive a predefined three-dimensional skeleton model based on the synchronized data and reconstruct the corresponding ski slope scene in the virtual environment to generate a three-dimensional dynamic playback consistent with the skier's real movements. The attitude analysis module has a built-in standard attitude knowledge base, which is used to compare the attitude in 3D dynamic playback with the standard attitude and identify attitude anomalies. The visualization and interaction module displays the results of the 3D dynamic playback and posture analysis module, and receives user marking instructions for posture abnormalities and their locations. The closed-loop feedback control module stores the posture abnormalities marked by the user, along with the corresponding standard posture data and geographical location. During subsequent skiing, when it is predicted that the skier will reach the marked location, the current motion data is compared with the standard data in real time. If a posture abnormality is detected, the tactile feedback device located on the corresponding limb is triggered to provide a prompt.

2. The skiing posture recognition system according to claim 1, characterized in that: The inertial measurement unit sensor network includes multiple 9-axis IMU sensor nodes, which are deployed on the main parts of the skier's upper and lower limbs via elastic straps or fixed bases.

3. The skiing posture recognition system according to claim 1, characterized in that: The positioning module is a GNSS receiver integrated into a mobile terminal or a standalone host.

4. The skiing posture recognition system according to claim 1, characterized in that: The data processing and synchronization module achieves data synchronization by adding timestamps from the same clock source to all data sources and using an interpolation algorithm to align data with inconsistent sampling periods.

5. The skiing posture recognition system according to claim 1, characterized in that: The three-dimensional posture reconstruction module uses an inverse kinematics algorithm to convert the calculated absolute limb posture information into rotational drive signals for each joint in the three-dimensional skeletal model.

6. The skiing posture recognition system according to claim 1, characterized in that: The visualization and interaction module allows users to observe the 3D dynamic playback from multiple perspectives, slow down, and pause, and overlay angle data labels of key joints on the model.

7. The skiing posture recognition system according to claim 1, characterized in that: The closed-loop feedback control module includes: The rule management unit is used to store correction rules that include posture anomaly descriptions, standard posture characteristics, triggering geographical locations, and fault tolerance thresholds. The real-time judgment unit is used to predict the time of arrival at the triggered geographical location based on the current positioning information, and to initiate real-time attitude comparison within the warning window; The tactile actuation unit is used to control the tactile feedback device of the target limb to generate vibration when the comparison result exceeds the fault tolerance threshold.

8. The skiing posture recognition system according to claim 1, characterized in that: The tactile feedback device is a linear resonant actuator integrated into the sensor node of the inertial measurement unit.

9. The skiing posture recognition system according to claim 1, characterized in that: The data acquisition module also includes a foot pressure sensing unit integrated into the ski shoe insole, used to collect data on the distribution of the skier's center of gravity; the posture analysis module is also used to assess the skier's balance based on the center of gravity distribution data.

10. The skiing posture recognition system according to claim 1, characterized in that: The posture analysis module also includes a motion classifier, which uses a machine learning model based on a one-dimensional convolutional neural network or a long short-term memory network to automatically identify the type of skiing motion that the skier is performing based on time-series motion data.

Images