A motion rectification method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application provide a motion correction method and device, electronic equipment and storage medium, applied to the technical field of intelligent sports, wherein the method comprises: acquiring real-time motion data of user motion, determining motion posture features of the user according to the real-time motion data; determining abnormal motion indicators in each of the motion posture features based on a standard action model; determining root cause motion indicators in each of the abnormal motion indicators according to a pre-trained root cause prediction model, wherein the root cause prediction model is generated by training based on real coach correction history data; determining motion hardware device control parameters matched with the root cause motion indicators, and controlling motion auxiliary equipment of the user based on the motion hardware device control parameters. Embodiments of the present application identify root cause actions of motion action errors in abnormal motion indicators, correct user motion states based on root cause actions, cut off error muscle memory, and improve motion correction effect.

Description

Technical Field

[0001] This invention relates to the field of intelligent sports technology, and in particular to a motion correction method, device, electronic device, and storage medium. Background Technology

[0002] With the rapid development of artificial intelligence, computer vision, and big data technologies, intelligent sports have become an important direction for competitive training and mass fitness. Traditional sports teaching relies heavily on manual observation and experience-based judgment, making it difficult to accurately capture athletes' movement defects. Currently, the application of artificial intelligence and machine vision technologies in the sports field is deepening, gradually evolving from post-match statistics to training assistance. In intelligent training equipment for racket-based sports (such as intelligent ball-serving machines), voice prompts and posture recognition have been introduced, but the two are often disconnected. Existing sports ball-driving systems can include pre-announcement, post-match analysis, and threshold alarm types. Taking racket-based sports as an example, existing devices can usually only provide preset static voice prompts before the movement occurs, and cannot trigger real-time intervention guidance based on the user's actual movement deformation during training, leading to users easily forming incorrect muscle memory. In addition, when a user makes an incorrect movement, it often causes multiple body part indicators (such as footwork, center of gravity, and swing trajectory) to deviate from the standard threshold simultaneously. Existing motion evaluation systems often simply list and pile up all error indicators, causing users to experience "cognitive overload" within a very short reaction time, unsure of where to correct their mistakes first; the system also fails to pinpoint the "root cause" of motion distortion. Therefore, there is an urgent need for a method that corrects errors based on their root causes. Summary of the Invention

[0003] This invention provides a motion correction method, device, electronic device, and storage medium to solve the problem that existing motion correction systems cannot detect the root cause of erroneous movements, resulting in the inability to trigger accurate intervention and guidance.

[0004] According to one aspect of the present invention, a motion correction method is provided, wherein the method includes: Acquire real-time motion data of the user's movement, and determine the user's motion posture characteristics based on the real-time motion data; Abnormal motion indicators in each of the aforementioned motion posture features are determined based on the standard motion model; The root cause motion index within each of the abnormal motion indexes is determined according to a pre-trained root cause prediction model, wherein the root cause prediction model is trained and generated based on the historical correction data of real coaches. Determine the control parameters of the motion hardware device that match the root cause motion index, and operate the user's motion assistive device based on the control parameters of the motion hardware device.

[0005] According to another aspect of the present invention, a motion correction device is provided, wherein the device comprises: The feature extraction module is used to acquire real-time motion data of the user's movement and determine the user's motion posture features based on the real-time motion data. Anomaly detection module is used to determine abnormal motion indicators in each of the motion posture features based on a standard motion model; The root cause determination module is used to determine the root cause motion index among each of the abnormal motion indexes according to a pre-trained root cause prediction model, wherein the root cause prediction model is trained and generated based on the historical correction data of real coaches. The device coordination module is used to determine the control parameters of the motion hardware device that match the root cause motion index, and to control the user's motion assistive device based on the control parameters of the motion hardware device.

[0006] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the motion correction method according to any embodiment of the present invention.

[0007] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the motion correction method according to any embodiment of the present invention.

[0008] The technical solution of this invention collects real-time motion data of the user's movement process, obtains motion posture features within the real-time motion data, determines abnormal motion indicators based on the motion posture features and standard movement models, obtains root cause motion indicators within each abnormal motion indicator through a root cause prediction model, acquires control parameters for motion hardware devices matching the root cause motion indicators, and controls the user's motion assistive devices according to the motion hardware control parameters. This invention can identify the root cause of erroneous motion movements within multiple abnormal motion indicators through abnormal root cause prediction, correct the user's motion state based on the root cause movements, cut off erroneous muscle memory, and improve the motion correction effect.

[0009] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart of a motion correction method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of another motion correction method provided in Embodiment 2 of the present invention; Figure 3 This is a flowchart of another motion correction method provided in Embodiment 3 of the present invention; Figure 4 This is a flowchart of another motion correction method provided in Embodiment 4 of the present invention; Figure 5 This is an example diagram of a motion correction method provided in Embodiment 5 of the present invention; Figure 6 This is a schematic diagram of a motion correction device according to Embodiment Six of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device that implements the motion correction method of the present invention. Detailed Implementation

[0012] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0013] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0014] Example 1 Figure 1 This is a flowchart of a motion correction method according to Embodiment 1 of the present invention. This embodiment is applicable to correcting erroneous motion movements. The method can be executed by a motion correction device, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes: Step 110: Obtain real-time motion data of the user's movement and determine the user's motion posture characteristics based on the real-time motion data.

[0015] The sports activities can include racket-based sports, such as table tennis and tennis. Real-time motion data can be various data streams containing information about the user's movement, and can be collected by sensors such as depth cameras, binocular cameras, pressure sensors, and inertial sensors. Motion posture features can be characteristic indicators reflecting the user's motion state, and can include, but are not limited to, the user's posture, position, and movement trajectory.

[0016] In this embodiment of the invention, real-time motion data of a user in motion can be collected. The user's motion posture features can be extracted from the real-time motion data based on visual recognition technology, or the user's motion position or speed can be collected using inertial sensors. For example, taking real-time motion data including video data as an example, a top-down target detection algorithm can be used. This algorithm first locates the human body regions in each frame of the real-time motion data, and then independently detects the coordinates of key joints for each human body region to obtain motion posture features. Alternatively, a bottom-up target detection algorithm can be used. This algorithm first detects all key joints in each frame of the real-time motion data, and then groups and assigns each key joint to different human body parts using limb management rules.

[0017] Step 120: Determine the abnormal motion indicators in each motion posture feature based on the standard motion model.

[0018] Among them, the standard motion model can be a motion model used to identify whether a user's movement is standard. The standard motion model can correspond to a single action or a set of actions. For example, in tennis, a standard motion model can correspond to a standard action or a class of standard actions. The standard motion model can correspond to the sliding step, cross step, lunge, shuffle step, or starting step in footwork movement, etc. Or, the standard motion model can correspond to footwork movement, backswing and swing, hitting action, follow-through and recovery action, etc.

[0019] In this embodiment of the invention, a pre-configured standard motion model can be obtained. The obtained motion posture features can be compared with the standard motion features in the standard motion model to determine abnormal motion indicators that do not match the standard motion model. It is understood that in some embodiments, user actions can be first coarsely identified according to motion posture features, then the corresponding standard motion model can be selected based on the action type identified by coarse granularity, and then each motion posture feature can be judged based on the standard motion model to obtain abnormal motion indicators.

[0020] Step 130: Determine the root cause motion index within each abnormal motion index based on the pre-trained root cause prediction model, wherein the root cause prediction model is trained and generated based on the historical correction data of real coaches.

[0021] The root cause prediction model can identify the root cause motion indicators leading to abnormalities in other motion indicators based on the inherent correlations between various abnormal motion indicators. For example, abnormal motion indicators include a backward stance, a high center of gravity, and a delayed hitting point. When a high center of gravity leads to a backward stance and a delayed hitting point, the high center of gravity can be the root cause of both. The root cause prediction model can be trained using historical correction data from real coaches. This model can output root cause motion indicators that match the real coach's selections. The aforementioned historical correction data can include the coach's set of abnormal motion indicators and the root cause motion indicators within each set.

[0022] In this embodiment of the invention, when multiple abnormal motion indicators are obtained, a pre-trained root cause prediction model can be invoked, and one of the root cause motion indicators among the multiple abnormal motion indicators can be predicted according to the root cause prediction model.

[0023] Step 140: Determine the control parameters of the motion hardware device that match the root cause of motion index, and operate the user's motion assistive device based on the control parameters of the motion hardware device.

[0024] Among them, the control parameters of sports hardware devices can be the control parameters of devices that assist users in their movements. The control parameters of sports hardware devices can control sports assistive devices to perform different functions. Sports assistive devices may include visual monitoring systems, ball machines, wearable devices, audio devices, etc.

[0025] In this embodiment of the invention, the control parameters of the motion hardware device that assists in correcting user movements can be determined according to the root cause motion index. The process of determining the control parameters of the motion hardware device can be implemented through pre-configured rules or intelligence. The user's motion assistive device can be controlled to perform corresponding control functions according to the determined control parameters of the motion hardware device, thereby correcting the user's erroneous motion movements.

[0026] For example, the acquired root cause motion index can be selected from a pre-configured rule set to match its target rule. Different types of root cause motion indices can correspond to different target rules. Control parameters for the motion hardware device can be generated according to the target rule to control the motion assistive device. For instance, if the root cause motion index is a hitting point error, the ball machine can be paused, or the ball machine can be re-served according to the serving speed corresponding to the root cause motion index. As another example, if the motion assistive device is a voice player, for the acquired root cause motion index, the voice player can be controlled to play a reminder voice corresponding to that root cause motion index, so that the voice playback device outputs a unique and crucial voice-based error correction command, such as "Attention! Take a step to find the ball first."

[0027] This invention, in its embodiments, collects real-time motion data from a user's movement process, acquires motion posture features within the real-time motion data, determines abnormal motion indicators based on these features and a standard movement model, obtains root cause motion indicators within each abnormal motion indicator using a root cause prediction model, acquires control parameters for motion hardware devices matching the root cause motion indicators, and controls the user's motion assistive devices based on these parameters. This invention, through abnormal root cause prediction, identifies the root cause of erroneous motion movements within multiple abnormal motion indicators, corrects the user's movement state based on the root cause movement, cuts off erroneous muscle memory, and improves the effectiveness of motion correction.

[0028] In some embodiments, controlling a user's motion assistive device based on motion hardware device control parameters includes: when there are at least two abnormal motion indicators, controlling the voice playback module in the motion assistive device to play only a single corrective voice command matching the root cause motion indicator according to the motion hardware device control parameters.

[0029] In this embodiment of the invention, when at least two abnormal motion indicators are obtained, and there are multiple abnormal motion indicators, motion hardware device control parameters can be generated. These motion hardware device control parameters can control the voice playback module in the motion assistive device, so that the voice playback module only plays a single correction voice command for the root cause motion indicator among each abnormal motion indicator, avoiding the confusion caused by playing correction voice commands for multiple abnormal motion indicators at the same time. This makes the voice correction process conform to the cognitive limits of information processing in human movement, and can improve the teaching effect of motion correction.

[0030] Example 2 Figure 2 This is a flowchart of another motion correction method provided in Embodiment 2 of the present invention. This embodiment of the present invention describes the process of recognizing motion posture features. See [link to flowchart]. Figure 2 The method provided in this embodiment of the invention specifically includes the following steps: Step 210: Collect real-time video data of the user while in motion using a high-precision vision sensor as real-time motion data.

[0031] In this embodiment of the invention, one or more high-precision visual sensors can be pre-deployed in the user's sports field. The user's movements can be collected through the high-precision visual sensors during the user's movement, thereby forming real-time video data as real-time motion data.

[0032] Step 220: Based on the artificial intelligence skeletal key point detection algorithm, identify the real-time video data to obtain the user's three-dimensional posture features, wherein the three-dimensional posture features include at least the three-dimensional coordinate information of the key skeletal joints of the human body.

[0033] Among them, the artificial intelligence skeletal key point detection algorithm can be an algorithm that detects key points of the human skeleton based on video or image data. The artificial intelligence skeletal key point detection algorithm can include HRNet, SimpleBaseline, CPN, HigherHRNet, MediaPipe Pose, BlazePose, etc.

[0034] In this embodiment of the invention, real-time video data can be input into an artificial intelligence skeletal key point detection algorithm for processing. The algorithm can extract the three-dimensional coordinate information of different artificial key skeletal joints in the real-time video data. These human skeletal joints may include, but are not limited to, shoulder joints, elbow joints, wrist joints, hip joints, knee joints, and ankle joints used to determine the torso and core; neck, shoulder, elbow, wrist, and finger roots used to determine the swing, backswing, and hitting point; hip, knee, ankle, and toe / heel joints used to analyze stance, footwork, and center of gravity movement; and spine and pelvis used to determine body rotation and side-body force exertion posture.

[0035] Step 230: Determine the spatial position features based on the positional relationship of the first three-dimensional coordinate information of different key bone joints of the foot within the three-dimensional posture features in the same time period.

[0036] The first three-dimensional coordinate information can be the three-dimensional coordinate information belonging to the key bone joints of the foot. The first three-dimensional coordinate information can include the three-dimensional coordinate information of each key bone joint of the foot. The key bone joints of the foot can include, but are not limited to, the left ankle, right ankle, toe, and heel.

[0037] Specifically, the first three-dimensional coordinate information belonging to the key bone joints of the foot can be extracted from the three-dimensional pose features. The horizontal distance, relative orientation, and support area between the left and right feet can be determined by the first three-dimensional coordinate information of different key bone joints of the foot in the same frame. Based on the above information, the foot width, orientation, and center of gravity support range can be determined as spatial position features.

[0038] Step 240: Determine the dynamic features based on the second three-dimensional coordinate information of at least one key point of the upper limb within the three-dimensional posture features and the continuous movement trajectory at different times.

[0039] The second three-dimensional coordinate information can be the three-dimensional coordinate information belonging to the key points of the upper limb, which may include the wrist, hip, trunk, etc.

[0040] In this embodiment of the invention, second three-dimensional coordinate information belonging to key points of the upper limbs such as the wrist, hip, and torso can be extracted from the three-dimensional posture features. For each key point of the upper limb, the second three-dimensional coordinate information at different times can be arranged into a continuous movement trajectory. The continuous movement trajectory corresponding to each key point of the upper limb can be used as a motion feature.

[0041] Step 250: Use the three-dimensional pose features, spatial position features, and dynamic features as motion pose features.

[0042] Specifically, the acquired 3D posture features, spatial position features, and dynamic features can be used as the user's motion posture features.

[0043] Step 260: Determine the abnormal motion indicators in each motion posture feature based on the standard motion model.

[0044] Step 270: Determine the root cause motion index within each abnormal motion index based on the pre-trained root cause prediction model, wherein the root cause prediction model is generated based on the historical correction data of real coaches.

[0045] Step 280: Determine the control parameters of the motion hardware device that match the root cause of motion index, and operate the user's motion assistive device based on the control parameters of the motion hardware device.

[0046] This invention employs an artificial intelligence skeletal keypoint detection algorithm to identify user 3D posture features in real-time video data. It extracts first 3D coordinate information of different key foot joints within the 3D posture features, determines spatial position features based on the positional relationships of these first 3D coordinates within the same time frame, and uses the continuous movement trajectories formed by these first 3D coordinates at different times as dynamic features. These 3D posture features, spatial position features, and dynamic features are then used as motion posture features. Abnormal motion indicators are determined within these motion posture features according to a standard action model. A root cause prediction model is then invoked to determine the root cause motion indicators within each abnormal motion indicator. Control parameters for the motion hardware device matching the root cause motion indicators are obtained, and the user's motion assistive device is controlled based on these parameters. This invention can identify abnormal root causes within various abnormal motion indicators and correct user errors based on these root causes, thus severing erroneous muscle memory and improving motion correction efficiency.

[0047] In other embodiments of the invention, the method further includes: collecting pressure data of the user during exercise based on a foot-worn smart wearable device; collecting dynamic motion data of the user during exercise based on a swing inertial sensor, wherein the dynamic motion data includes basic swing data, action posture data, force data, and exercise volume data, and the swing inertial sensor is worn by the user or the user's exercise equipment; and fusing the pressure data and dynamic motion data into real-time motion data.

[0048] Among them, the foot-worn smart wearable device can be a wearable device for the user's feet. This device can include, but is not limited to, smart insoles, smart socks, and smart footwear built-in modules. It can integrate pressure sensors and communication modules to collect real-time pressure-related data between the user's feet and the contact surface during exercise. This pressure data, collected by the foot-worn smart wearable device, can include pressure values, pressure distribution, pressure change rate, and corresponding plantar areas. The pressure data can reflect the user's center of gravity position, lower limb force balance, and stance stability during exercise. The racket swing inertial sensor can be worn on the user's body (e.g., wrist, arm) or on sports equipment (e.g., the handle / frame of a badminton racket, tennis racket, or table tennis racket). It collects the dynamic motion state of the user during the swing through an integrated three-axis accelerometer or three-axis gyroscope. This dynamic motion data, collected by the swing inertial sensor, can be processed to include basic swing data, posture data, force data, and exercise volume data.

[0049] In this embodiment of the invention, basic swing data may include maximum swing speed, average swing speed, instantaneous ball speed, swing duration, backswing time, and follow-through buffer time, etc., which can be core indicators for measuring the basic characteristics of the swing action; action posture data includes racket pitch angle, roll angle, yaw angle, racket face angle, swing trajectory (planar / three-dimensional arc), as well as the user's wrist pronation / external rotation angle, arm swing amplitude, etc., which can be used to judge the standardization of the swing action. Force data may include instantaneous explosive force of the swing, peak force, peak acceleration, force continuity, braking force, force release efficiency, etc., used to determine the user's power transfer and force generation techniques during the swing; exercise volume data includes total number of swings, effective number of hits, frequency of consecutive hits, exercise load, exercise intensity, calorie consumption, etc., used to quantify the user's total exercise volume and exercise intensity.

[0050] Specifically, when acquiring pressure data and dynamic motion data collected by foot-worn smart wearable devices and racket swing inertial sensors, a data fusion algorithm can be activated to process and fuse these data. This can involve time synchronization calibration of the two types of data, matching the swing dynamic motion data with the corresponding plantar pressure data, for example, correlating the force data at the moment of impact with the peak plantar pressure data at that moment. Alternatively, algorithms can eliminate errors from single data points, such as using pressure data to correct swing posture judgments (e.g., swing deviations caused by unstable user stance can be identified using plantar pressure distribution data), while using dynamic motion data to supplement the rationality analysis of lower limb force exertion (e.g., insufficient swing force can be determined by combining plantar pressure data to determine if insufficient push-off force is the cause). Alternatively, algorithms can integrate the two types of data, extract core correlation information, eliminate redundant data, and form comprehensive real-time motion data covering user lower limb stance, upper limb swing, force efficiency, and exercise intensity.

[0051] Furthermore, based on the above embodiments of the invention, the acquired pressure data and dynamic motion data can be fused with the real-time video data collected in the above embodiments to further improve the feature dimension of the real-time motion data and increase the reliability of motion correction.

[0052] Example 3 Figure 3 This is a flowchart of another motion correction method provided in Embodiment 3 of the present invention. This embodiment of the present invention describes the training and usage process of the root cause prediction model. See [link to documentation]. Figure 3 The method provided in this embodiment of the invention specifically includes the following steps: Step 310: Construct a training dataset based on the number of historical correction data based on a threshold, wherein each historical correction data includes an anomaly index sequence and root cause indexes labeled by the real coach.

[0053] The threshold number can be the amount of data used to determine the training effect of the root cause prediction model, and the threshold number can be configured based on experience.

[0054] In this embodiment of the invention, a threshold number of historical error correction data from real coaches can be collected, and a training dataset can be constructed based on this historical error correction data. This historical error correction data can include anomaly indicator sequences and root cause indicators labeled by real coaches. It is understood that the root cause indicators labeled by real coaches can be manually labeled by the real coaches, or automatically labeled by an image recognition model that identifies the motion data of real coaches. Each piece of historical error correction data can contain two core components: first, an anomaly indicator time-series sequence composed of multiple anomaly motion indicators, which can include various errors that occur during the execution of motion actions and their timing; second, real root cause indicators manually labeled by experienced sports coaches or automatically labeled by experienced sports coaches, i.e., the core root cause errors corresponding to the simultaneous occurrence of multiple anomalies. The training set constructed in this way allows the model to fully learn the temporal correlations, causal relationships, and priority judgment rules between different anomalies, providing accurate and reliable supervision signals for subsequent deep learning-based root cause prediction models.

[0055] Step 320: Set the loss function of the root cause prediction model, wherein the loss function includes at least the output result of the root cause prediction model and the true result of the root cause index.

[0056] Specifically, the loss function can be set for the root cause prediction model. This loss function can be constructed based on the difference between the model output and the true label. The loss function can be expressed as the error formula between the output of the root cause prediction model and the true result of the root cause index. For example, the loss function can be the mean square error or arithmetic square error between the output of the root cause prediction model and the true result of the root cause index.

[0057] For example, in one embodiment, the loss function can be the cross-entropy loss function, such as: ,in, The total number of categories of root cause indicators. Root cause outcome labels for real coaches This represents the predicted probability of the root cause index category output by the root cause prediction model.

[0058] Step 330: Train the temporal graph neural network model or the long short-term memory network model with added attention mechanism into a root cause prediction model based on the training dataset and loss function.

[0059] In this embodiment of the invention, the root cause prediction model can be a temporal graph neural network model or a long short-term memory network model with added attention mechanism. The temporal graph neural network model or the long short-term memory network model with added attention mechanism can be trained according to the constructed training dataset and the configured loss function to obtain the root cause prediction model.

[0060] Step 340: Obtain real-time motion data of the user's movement and determine the user's motion posture characteristics based on the real-time motion data.

[0061] Step 350: Determine the abnormal motion indicators in each motion posture feature based on the standard motion model.

[0062] Step 360: Input each abnormal motion index into the root cause prediction model to obtain the root cause motion index within each abnormal motion index; wherein, the root cause prediction model includes at least a time-series graph neural network model trained based on historical data for error correction or a long short-term memory network model with added attention mechanism.

[0063] Among them, the temporal graph neural network model is a deep learning model that processes graph structure data in the order of temporal evolution. The temporal graph neural network model can determine the spatial structure relationship and temporal dependence between various abnormal motion indicators, thereby obtaining the root cause motion indicators within each abnormal motion indicator. The temporal graph neural network model may include a spatial graph convolution model, a temporal evolution module, and an output prediction model.

[0064] In this embodiment of the invention, abnormal motion indicators can be input into a time-series graph neural network model trained as a root cause prediction model or a long short-term memory network model with an added attention mechanism. The implicit correlation and causal weight between each abnormal motion indicator can be obtained through the time-series graph neural network model or the long short-term memory network model with an added attention mechanism, thereby selecting the root cause motion indicator from among the abnormal motion indicators.

[0065] Step 370: Determine the control parameters of the motion hardware device that match the root cause of motion index, and operate the user's motion assistive device based on the control parameters of the motion hardware device.

[0066] In this embodiment of the invention, a training dataset is constructed by collecting a threshold number of historical data for error correction. A loss function is set based on the output items of the root cause prediction model and the actual results of the root cause indicators. A time-series graph neural network model or a long short-term memory network model with added attention mechanism is trained as a root cause prediction model according to the training dataset and the loss function. Real-time motion data of the user's movements is collected, and motion posture features are extracted from the real-time motion data. Abnormal motion indicators in each motion posture feature are identified according to a standard action model. The root cause motion indicators in the abnormal motion indicators are identified according to the root cause prediction model, and the control parameters of the motion hardware device matching the root cause motion indicators are determined. The user's motion assistive device is then controlled based on the control parameters of the motion hardware device. This embodiment of the invention can train the root cause prediction model based on a dataset containing real coach data, which can improve the accuracy of root cause identification. By identifying the implicit causal relationship between abnormal movements through the root cause prediction model, the root cause of the user's erroneous movements can be discovered in a timely manner, filtering out superficial erroneous movements and enhancing the user's motion correction effect.

[0067] In some embodiments of the invention, abnormal motion indicators in each motion posture feature are determined based on a standard motion model, including: According to the user's sport type, a matching standard motion model is determined within the preset motion model; standard three-dimensional posture features, standard spatial position features, and standard dynamic features of the standard motion model are extracted; the motion posture features are compared with the standard three-dimensional posture features, standard spatial position features, and standard dynamic features respectively; the feature indicators whose differences from the standard three-dimensional posture features, standard spatial position features, and standard dynamic features are greater than the fault tolerance threshold are determined as abnormal motion indicators; wherein, the feature indicators include at least one of the three-dimensional posture features, spatial position features, and dynamic features.

[0068] The sports type can be the type of sport the user is currently playing, which can include tennis, table tennis, or badminton, etc. The preset motion models can be feature sets of various standard movements corresponding to different sports. These preset motion models can include, but are not limited to, ready stance movements, footwork movements, backswing and swing movements, hitting movements, follow-through and recovery movements; ready stance movements include open stance, semi-open stance, and closed stance; footwork movements include sliding steps, crossover steps, lunges, shuffle steps, and starting steps; backswing and swing movements include forehand backswing, backhand backswing, and one-handed / two-handed backhand grip swings; and hitting movements include forehand hits, backhand hits, serves, volleys, overhead smashes, and slices, etc.

[0069] In this embodiment of the invention, the type of sport currently selected by the user or determined by scene recognition can be obtained first. The standard action model corresponding to the type of sport can be accurately matched in the preset action model. The standard action model can cover the complete feature benchmarks of various core movements in the sport (such as forehand hitting, backhand hitting, serving, footwork movement, etc.). From the matched standard action model, standard three-dimensional posture features (including three-dimensional angles of each key joint, relative position of limbs, etc.), standard spatial position features (including standard stance width, footwork trajectory, spatial coordinates of hitting point, etc.) and standard dynamic features (including standard swing speed, center of gravity movement trajectory, action timing rhythm, etc.) can be extracted in batches as the core benchmarks for subsequent action comparison and evaluation. Then, the user's motion posture features, which are collected and processed in real time by artificial intelligence skeletal key point detection algorithm, are compared with the three types of extracted standard features by performing dimension-by-dimensional difference calculation or similarity comparison. Based on a preset reasonable fault tolerance threshold, the feature indicators in the user's motion posture features that differ from any one of the standard three-dimensional posture features, standard spatial position features, and standard dynamic features by more than the fault tolerance threshold are uniformly determined as abnormal motion indicators. The feature indicators can be one of the three-dimensional posture features, spatial position features, and dynamic features, or they can include multiple features at the same time.

[0070] Based on the above embodiments of the invention, the method further includes: if the root cause prediction model is not enabled, then the root cause motion index within each abnormal motion index is determined based on a preset dynamic chain causal matrix or a priority decision tree; wherein the preset dynamic chain causal matrix includes at least one dynamic causal chain indicating the relationship between different motion posture indicators, or the priority decision tree includes at least one decision branch indicating the priority order of actions between different motion posture indicators.

[0071] The preset kinetic chain causal matrix can be a pre-constructed standardized relational data table containing multiple causal relationships between human movement postures and force indicators. It clarifies the sequence of motion force transmission, how defects in preceding movements chain to trigger subsequent abnormal movements, and quantifies the causal mapping relationship of the whole-body kinetic chain. The preset kinetic chain causal matrix can include multiple kinetic causal chains, each reflecting the force linkage relationship between a series of movement posture indicators. The priority decision tree can be a hierarchical tree-like decision rule model, which can filter abnormal movement indicators step by step according to the logic of movement formation, the order of movements, and the division of influence weights, quickly identifying the dominant root cause movement indicator. Each decision branch in the priority decision tree can correspond to the standard sequential execution logic of a series of movement posture indicators. For example, the abnormal movement indicator matched by the decision branch node closest to the root of the priority decision tree within each abnormal movement indicator can be the root cause movement indicator. It is understandable that the kinetic causal chains in the preset kinetic chain causal matrix and the decision branches for the action priority order in the priority decision tree can be determined based on the principles of sports biomechanics. The preset kinetic chain causal matrix and the priority decision tree are determined by exhaustively enumerating common erroneous movement combinations.

[0072] In this embodiment of the invention, after the system detects and confirms that the root cause prediction model is not enabled, it can automatically switch to the rule-based judgment mode. It then calls a pre-configured preset dynamic chain causal matrix or priority judgment tree to analyze the identified abnormal motion indicators. For example, it can match each abnormal motion indicator based on the dynamic causal transmission relationship of multiple sets of motion posture indicators recorded in the dynamic chain causal matrix, determine the dynamic causal chain satisfied by the combination of each abnormal motion indicator, and determine the abnormal motion indicator corresponding to the source node in the dynamic causal chain as the root cause motion indicator. Alternatively, it can judge each abnormal motion indicator based on multiple judgment branches designed according to the action sequence logic within the priority judgment tree, determine the judgment branch node matching each abnormal motion indicator, and determine the abnormal motion indicator closest to the root node of the priority judgment tree among the matched judgment branch nodes as the root cause motion indicator. In this embodiment of the invention, it can compare and filter layer by layer according to the motion dynamic transmission law and the action execution priority order to accurately locate the source root cause motion indicator corresponding to each abnormal motion indicator, completing the motion action anomaly tracing and judgment in the absence of a root cause prediction model.

[0073] Example 4 Figure 4 This is a flowchart of another motion correction method provided in Embodiment 4 of the present invention. This embodiment of the present invention describes the closed-loop control process of a motion assist device. See [link to documentation]. Figure 4 The method provided in this embodiment of the invention specifically includes the following steps: Step 410: Obtain real-time motion data of the user's movement and determine the user's motion posture characteristics based on the real-time motion data.

[0074] Step 420: Determine the abnormal motion indicators in each motion posture feature based on the standard motion model.

[0075] Step 430: Determine the root cause motion index within each abnormal motion index based on the pre-trained root cause prediction model, wherein the root cause prediction model is generated based on the historical correction data of real coaches.

[0076] Step 440: In response to the root cause motion index, generate a lock-up command for the ball-serving machine within the motion assist device as a control parameter for the motion hardware device.

[0077] The lock-up command can be a control command that triggers the ball machine to pause the serve. The lock-up command can be determined based on the ball machine's communication protocol. The ball machine can be an automated ball training device, consisting of a ball supply system, a launching mechanism, and an intelligent control system. It can automatically launch tennis balls at a specified speed, spin, angle, and frequency to a predetermined area on the court according to set parameters, simulating the effects of a real serve and shot, and is used to assist players in conducting independent training without a sparring partner.

[0078] In this embodiment of the invention, when the root cause motion index is obtained, a lock-up command can be generated. The lock-up command can be transmitted to the ball-serving machine included in the motion assist device through a pre-configured communication protocol. When the ball-serving machine receives the lock-up command, it can pause the ball-serving.

[0079] Step 450: Generate monitoring instructions for the visual monitoring system within the motion assistive device according to the root cause motion index, which will then be used as control parameters for the motion hardware device.

[0080] Among them, the visual monitoring system can be the core perception module of intelligent training. It consists of a multi-view high-definition camera, an image acquisition card, a real-time processing unit, and an artificial intelligence analysis module. It is specifically used to capture the user's movement posture, tennis ball trajectory, and training scene information in real time during tennis training, providing accurate and real-time raw data support for subsequent action analysis, anomaly detection, and root cause prediction.

[0081] Specifically, the motion actions corresponding to the root cause motion indicators can be extracted, and monitoring instructions for the visual monitoring system can be generated according to these motion actions, thereby controlling the visual monitoring system to monitor the user's aforementioned motion actions.

[0082] Step 460: Transmit a lock command to the ball machine to control the ball machine to pause the ball serving operation.

[0083] In this embodiment of the invention, a lock command can be sent to the ball-serving machine to control the ball-serving machine to pause the serve, thereby giving the user time to correct any erroneous actions.

[0084] Step 470: Transmit monitoring instructions to the visual monitoring system to trigger the visual monitoring system to monitor and collect current visual data of root cause motion indicators.

[0085] Specifically, the monitoring instructions for the corresponding root cause motion index can be transmitted to the visual monitoring system in real time via wired or wireless communication, driving the visual monitoring system to collect the current visual data of the user's motion based on the key action parts, motion time periods, and feature dimensions corresponding to the root cause motion index.

[0086] Step 480: If the user's motion state is determined to meet the unlocking requirements based on the current visual data, an unlocking command is generated, and the ball-launching machine is unlocked according to the unlocking command.

[0087] The unlocking requirement can be a condition for controlling the ball-serving machine to serve again. The unlocking condition can include the current visual data meeting the standard motion model or the difference between the current visual data and the standard motion model meeting a specified condition.

[0088] In this embodiment of the invention, the current visual data can be extracted to obtain the user's motion state. The motion state can be compared with the unlocking requirements. If the current visual data meets the standard motion model or the difference between the current visual data and the standard motion model meets the specified conditions, it is determined that the user's motion state meets the unlocking requirements. Then, an unlocking command can be generated and sent to the ball-serving machine, so that the ball-serving machine can continue to serve.

[0089] This invention, in its embodiments, collects real-time motion data of the user, extracts motion posture features from the real-time motion data, identifies abnormal motion indicators in each motion posture feature according to a standard action model, calls a root cause prediction model to determine the root cause motion indicator within each abnormal motion indicator, and, in response to the root cause motion indicator, generates a locking command for the ball-serving machine within the motion assistance device, and generates a monitoring command for the visual monitoring system within the motion assistance device. Based on the locking command, the ball-serving machine is paused; based on the monitoring command, the visual monitoring system is controlled to collect current visual data; if the user's motion state is determined to meet the unlocking requirements based on the current visual data, an unlocking command is generated, and the ball-serving machine is unlocked according to the unlocking command. This invention, in its embodiments, trains the root cause prediction model using a dataset of real coaching data, which can improve the accuracy of root cause identification. By identifying the implicit causal relationships between abnormal movements through the root cause prediction model, the underlying causes of user errors can be discovered in a timely manner, filtering out superficial errors and enhancing the user's motion correction effect.

[0090] Furthermore, based on the above embodiments of the invention, determining that the user's motion state meets the unlocking requirements based on the current visual data includes: extracting motion features of the corresponding root cause motion index according to the current visual data; determining that the motion features meet the standard action model, and thus determining that the user's motion state meets the unlocking requirements.

[0091] In this embodiment of the invention, feature extraction can be performed on the current visual data to obtain motion features that match the root cause motion index. The motion features extracted from the user's visual data can be compared with a standard action model. If the motion features meet the standard action model, it is determined that the user's motion state meets the unlocking requirements. Specifically, when extracting motion features from the user's visual data, targeted motion feature extraction can be performed according to the action type corresponding to the root cause motion index, which can improve feature extraction efficiency. For example, if the root cause motion feature is a low center of gravity, then the center of gravity feature of the user's center of gravity can be extracted from the user's visual data.

[0092] Example 5 Figure 5This is an example diagram of a motion correction method provided in Embodiment 5 of the present invention. This embodiment provides a correction method that can be implemented based on a closed-loop voice teaching system with real-time visual feedback. This system can not only "see" motion deviations in real time and "speak" correction instructions, but more importantly, it establishes a "kinetic chain root cause diagnosis mechanism." When multiple motion indicators are abnormal simultaneously, the system can intelligently filter out apparent errors, extract the most serious problem or the root cause of the problem, and output a unique and crucial voice correction instruction, achieving efficient, focused, accompanying coach-level guidance. This closed-loop voice teaching system may include a high-definition camera, a ball-serving machine, and a voice playback system, etc. The motion correction method executed by this system may include the following steps: Step 1: Real-time visual data acquisition High-definition cameras are used to capture users' motion video streams and depth data in real time at high frame rates.

[0093] Step 2: 3D State Recognition and Feature Extraction By using an AI-powered skeletal keypoint detection algorithm, features are extracted from motion video streams and depth data, thereby extracting the user's 3D posture features (coordinates of each joint), spatial position (footwork and stance), and dynamic features (swing speed, center of gravity trajectory) in real time.

[0094] Step 3: Real-time comparison of multi-dimensional thresholds The extracted features are compared with the "standard dynamic model" of the current training subject to filter out all abnormal indicators that exceed the fault tolerance threshold. For example, three abnormal indicators, "stance too far back", "center of gravity too high" and "hitting point too late", are detected at the same time.

[0095] Step 4: Root Cause Extraction and Priority Arbitration The system has a built-in root cause prediction model based on temporal features and deep learning. This root cause prediction model may include a temporal graph neural network temporal action evaluation model or a long short-term memory (LSTM) model.

[0096] The processing logic is as follows: When multiple errors exist in the set of abnormal indicators, the system does not directly output all the audio. Instead, it inputs these abnormal feature vector sets containing temporal information into a pre-trained deep learning model. This model is trained using massive amounts of historical correction data from professional coaches (including real coaches' correction choices when multiple errors coexist), and can automatically calculate the implicit correlations and causal weights between multiple error features in complex action sequences.

[0097] For example, if both [A: late hitting point] and [B: footwork not in place] are detected simultaneously, the model, based on the attention mechanism in its network parameters, identifies the anomaly of feature B in time sequence as a strongly correlated antecedent factor leading to the anomaly of feature A. The model then isolates the apparent error A and outputs B with the highest confidence as the highest priority root cause for correction.

[0098] Step 5: Trigger voice intervention and closed-loop control Precise voice broadcast: For the single root cause (B) extracted in step four, the corresponding corrective voice template is called ("Attention! Take a step to find the ball first!") to ensure that the instruction is single and clear.

[0099] Training status intervention (lock and unlock): Send a command to pause the ball-serving machine. The vision system continuously monitors the user's footwork and ready posture. Only when it recognizes that the user has lowered their center of gravity and adjusted their footwork as required (reaching the unlock threshold) will the system announce "Good, ready" and resume serving.

[0100] In the embodiments of the present invention, when multiple action errors deviate from the threshold, graph neural networks (GNN), long short-term memory networks (LSTM) or other deep learning algorithms are used to reason about multi-dimensional temporal abnormal feature vectors, automatically identify the implicit causal relationship between action errors, determine the root cause error action within the abnormal feature vector, and output a unique and most critical voice correction instruction to achieve efficient and focused accompanying coaching-level guidance.

[0101] Example 6 Figure 6 This is a schematic diagram of a motion correction device according to Embodiment Six of the present invention, as shown below. Figure 6 As shown, the device includes: The feature extraction module 510 is used to acquire real-time motion data of the user's movement and determine the user's motion posture features based on the real-time motion data.

[0102] The anomaly identification module 520 is used to determine abnormal motion indicators in each motion posture feature based on the standard motion model.

[0103] The root cause determination module 530 is used to determine the root cause motion index among each abnormal motion index according to the pre-trained root cause prediction model, wherein the root cause prediction model is trained and generated based on the historical correction data of real coaches.

[0104] The device coordination module 540 is used to determine the control parameters of the motion hardware device that match the root cause motion index, and to control the user's motion assistive device based on the control parameters of the motion hardware device.

[0105] In this embodiment of the invention, a feature extraction module collects real-time motion data of the user's movement process to obtain motion posture features within the real-time motion data. An anomaly identification module determines abnormal motion indicators based on the motion posture features and a standard action model. A root cause determination module obtains the root cause motion indicator within each abnormal motion indicator using a root cause prediction model. A device coordination module obtains the control parameters of the motion hardware device that match the root cause motion indicator and controls the user's motion assistive device according to the motion hardware control parameters. This embodiment of the invention can identify the root cause of the erroneous motion movement within multiple abnormal motion indicators through anomaly root cause prediction. Based on the root cause movement, the user's motion state is corrected, which can cut off erroneous muscle memory and improve the motion correction effect.

[0106] In some embodiments of the invention, the feature extraction module 510 includes: The data acquisition unit is used to acquire real-time video data of the user while they are in motion using a high-precision vision sensor as real-time motion data.

[0107] The pose extraction unit is used to identify real-time video data based on an artificial intelligence skeletal key point detection algorithm to obtain the user's three-dimensional pose features, which include at least the three-dimensional coordinate information of the key skeletal joints of the human body.

[0108] Spatial position unit, used to determine spatial position features based on the positional relationship of the first three-dimensional coordinate information of different key bone joints of the foot within the three-dimensional posture features in the same time period.

[0109] The dynamic feature unit is used to determine the dynamic features based on the second three-dimensional coordinate information of at least one upper limb key point in the three-dimensional posture features and the continuous movement trajectory at different times.

[0110] The feature determination unit is used to treat three-dimensional pose features, spatial position features, and dynamic features as motion pose features.

[0111] Based on the above embodiments of the invention, the feature extraction module 510 further includes: a feature fusion unit, used to collect pressure data of the user during exercise based on the foot smart wearable device; collect dynamic motion data of the user during exercise based on the swing inertial sensor, wherein the dynamic motion data includes basic swing data, action posture data, force data and exercise volume data, and the swing inertial sensor is worn by the user or the user's exercise equipment; and fuse the pressure data and the dynamic motion data into the real-time motion data.

[0112] Based on the above embodiments of the invention, the anomaly detection module 520 includes: The model selection unit is used to determine the matching standard motion model within the preset motion models according to the user's sport type.

[0113] Standard feature units are used to extract standard 3D pose features, standard spatial position features, and standard dynamic features of standard motion models.

[0114] The feature matching unit is used to compare motion posture features with standard 3D posture features, standard spatial position features, and standard dynamic features, respectively.

[0115] Anomaly determination unit is used to identify anomalous motion indicators as feature indices whose differences from standard three-dimensional posture features, standard spatial position features, and standard dynamic features are greater than the fault tolerance threshold. The feature indices include at least one of the three-dimensional posture features, spatial position features, and dynamic features.

[0116] In some embodiments of the invention, the root cause determination module 530 is specifically used to: input each abnormal motion index into the root cause prediction model to obtain the root cause motion index within each abnormal motion index; wherein, the root cause prediction model includes at least a time-series graph neural network model trained based on historical data for error correction or a long short-term memory network model with added attention mechanism.

[0117] Based on the above embodiments of the invention, the root cause determination module 530 is used to: determine the root cause motion index within each of the abnormal motion indexes based on a preset dynamic chain causal matrix or a priority decision tree if the root cause prediction model is not enabled; wherein the preset dynamic chain causal matrix includes at least one dynamic causal chain indicating the relationship between different motion posture indicators, or the priority decision tree includes at least one decision branch indicating the priority order of actions between different motion posture indicators.

[0118] In some embodiments of the invention, the module further includes: a root cause model training module for constructing a training dataset based on a threshold number of historical data for correction, wherein each piece of historical data for correction includes an anomaly index sequence and a root cause index labeled by the real coach; setting a loss function for the root cause prediction model, wherein the loss function includes at least the output result of the root cause prediction model and the true result of the root cause index; and training a temporal graph neural network model or a long short-term memory network model with added attention mechanism as a root cause prediction model based on the training dataset and the loss function.

[0119] In some embodiments of the invention, the device coordination module 540 is specifically used to: generate a locking command for the ball-serving machine within the motion assistance device as a control parameter for the motion hardware device in response to the root cause motion index; generate a monitoring command for the visual monitoring system within the motion assistance device as a control parameter for the motion hardware device according to the root cause motion index; transmit the locking command to the ball-serving machine to control the ball-serving machine to pause its serving operation; transmit the monitoring command to the visual monitoring system to trigger the visual monitoring system to monitor and collect the current visual data of the root cause motion index; if the user's motion state is determined to meet the unlocking requirements based on the current visual data, an unlocking command is generated, and the ball-serving machine is unlocked according to the unlocking command.

[0120] Based on the above embodiments of the invention, the device collaboration module 540 determines that the user's motion state meets the unlocking requirements based on the current visual data, including: extracting motion features of the corresponding root cause motion index according to the current visual data; determining that the motion features meet the standard action model, and then determining that the user's motion state meets the unlocking requirements.

[0121] In some embodiments of the invention, the device coordination module 540 is further configured to: when there are at least two abnormal motion indicators, control the voice playback module in the motion assistive device to play only the single corrective voice command matching the root cause motion indicator according to the motion hardware device control parameters.

[0122] The motion correction device provided in the embodiments of the present invention can execute the motion correction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0123] Example 7 Figure 7 This is a schematic diagram of the structure of an electronic device implementing the motion correction method of an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0124] like Figure 7As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0125] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0126] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as motion correction methods.

[0127] In some embodiments, the motion correction method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the motion correction method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the motion correction method by any other suitable means (e.g., by means of firmware).

[0128] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0129] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0130] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0131] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0132] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0133] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0134] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0135] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A motion correction method, characterized in that, The method includes: Acquire real-time motion data of the user's movement, and determine the user's motion posture characteristics based on the real-time motion data; Abnormal motion indicators in each of the aforementioned motion posture features are determined based on the standard motion model; The root cause motion index within each of the abnormal motion indexes is determined according to a pre-trained root cause prediction model, wherein the root cause prediction model is trained and generated based on the historical correction data of real coaches. Determine the control parameters of the motion hardware device that match the root cause motion index, and operate the user's motion assistive device based on the control parameters of the motion hardware device.

2. The method according to claim 1, characterized in that, The step of acquiring real-time motion data of the user's movement and determining the user's motion posture characteristics based on the real-time motion data includes: The real-time motion data is obtained by collecting real-time video data of the user while he is in motion using a high-precision vision sensor. The real-time video data is identified based on an artificial intelligence skeletal key point detection algorithm to obtain the user's three-dimensional posture features, wherein the three-dimensional posture features include at least the three-dimensional coordinate information of key human skeletal joints. The spatial position features are determined based on the positional relationship of the first three-dimensional coordinate information of different key bone joints of the foot within the three-dimensional posture features at the same time. The dynamic features are determined based on the second three-dimensional coordinate information of at least one key point of the upper limb within the three-dimensional posture features and the continuous movement trajectory at different times. The three-dimensional posture features, the spatial position features, and the dynamic features are used as the motion posture features.

3. The method according to claim 1 or 2, characterized in that, Also includes: The user's pressure data during exercise is collected by a smart wearable foot device; The user's dynamic motion data during exercise is collected based on a swing inertial sensor. The dynamic motion data includes basic swing data, action posture data, force data, and exercise volume data. The swing inertial sensor is worn by the user or the user's exercise equipment. The pressure data and the dynamic motion data are fused into the real-time motion data.

4. The method according to claim 1, characterized in that, The determination of abnormal motion indicators in each of the motion posture features based on the standard motion model includes: The standard motion model is determined within the preset motion model according to the user's type of sport. Extract the standard three-dimensional posture features, standard spatial position features, and standard dynamic features of the standard motion model; The motion posture features are compared with the standard three-dimensional posture features, the standard spatial position features, and the standard dynamic features, respectively; The abnormal motion index is defined as the feature index whose difference from the standard three-dimensional posture feature, the standard spatial position feature, and the standard dynamic feature is greater than the fault tolerance threshold. The feature indicators include at least one of three-dimensional pose features, spatial position features, and dynamic features.

5. The method according to claim 1, characterized in that, The step of determining the root cause motion index within each of the abnormal motion indices based on a pre-trained root cause prediction model includes: Each of the abnormal motion indicators is input into the root cause prediction model to obtain the root cause motion indicator within each of the abnormal motion indicators. The root cause prediction model includes at least a time-series graph neural network model trained based on the historical data of the correction or a long short-term memory network model with an added attention mechanism.

6. The method according to claim 1, characterized in that, Also includes: If the root cause prediction model is determined to be disabled, the root cause motion index within each of the abnormal motion indexes is determined based on a preset dynamic chain causal matrix or priority decision tree. The preset kinetic chain causal matrix includes at least one kinetic causal chain indicating the relationship between different motion posture indicators, or the priority decision tree includes at least one decision branch indicating the priority order of actions between different motion posture indicators.

7. The method according to claim 1 or 5, characterized in that, Also includes: A training dataset is constructed based on the number of historical correction data points that meet a threshold, wherein each piece of historical correction data includes a sequence of outlier indicators and root cause indicators labeled by the real coach. Set the loss function of the root cause prediction model, wherein the loss function includes at least the output result item of the root cause prediction model and the true result item of the root cause index; The time-series graph neural network model or the long short-term memory network model with added attention mechanism is trained into the root cause prediction model based on the training dataset and the loss function.

8. The method according to claim 1, characterized in that, The step of determining the control parameters of the motion hardware device that match the root cause of the motion index, and controlling the user's motion assistive device based on the control parameters of the motion hardware device, includes: In response to the root cause motion index, a lock command for the ball-serving machine within the motion assist device is generated as a control parameter for the motion hardware device; The monitoring instructions of the visual monitoring system within the motion assistive device are generated according to the root cause motion index and used as the control parameters of the motion hardware device. The locking command is transmitted to the ball-serving machine to control the ball-serving machine to pause its serving operation; The monitoring command is transmitted to the visual monitoring system to trigger the visual monitoring system to monitor and collect the current visual data of the root cause motion index; If the user's motion state is determined to meet the unlocking requirements based on the current visual data, an unlocking command is generated, and the ball-launching machine is unlocked according to the unlocking command.

9. The method according to claim 8, characterized in that, The step of determining whether the user's motion state meets the unlocking requirements based on the current visual data includes: Extract motion features corresponding to the root cause motion index based on the current visual data; If the motion characteristics are determined to satisfy the standard action model, then the user's motion state is determined to satisfy the unlocking requirements.

10. The method according to claim 1, characterized in that, The method of controlling the user's motion assistive device based on the control parameters of the motion hardware device includes: When at least two of the abnormal motion indicators exist, the voice playback module in the motion assistive device is controlled according to the motion hardware device control parameters to play only the single corrective voice command matching the root cause motion indicator.

11. A motion correction device, characterized in that, The device includes: The feature extraction module is used to acquire real-time motion data of the user's movement and determine the user's motion posture features based on the real-time motion data. Anomaly detection module is used to determine abnormal motion indicators in each of the motion posture features based on a standard motion model; The root cause determination module is used to determine the root cause motion index among each of the abnormal motion indexes according to a pre-trained root cause prediction model, wherein the root cause prediction model is trained and generated based on the historical correction data of real coaches. The device coordination module is used to determine the control parameters of the motion hardware device that match the root cause motion index, and to control the user's motion assistive device based on the control parameters of the motion hardware device.

12. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the motion correction method according to any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the motion correction method according to any one of claims 1-10.

Images