Upper limb movement abnormality detection method and system based on neural network

Abstract

The embodiment of the present application provides a method and system for detecting upper limb movement abnormalities based on a neural network. The method and system achieve comprehensive capture of the subject's upper limb movements by acquiring action behavior data of a first subject in an upper limb training space scene when performing an upper limb training task, namely, upper limb movement capture data. The method then performs abnormal state detection on these data under multiple abnormal characteristic states, and generates respective corresponding abnormal state degree values. The method generates abnormal training parameter values ​​of the first subject by performing weighted calculation on the abnormal state degree values ​​corresponding to each abnormal characteristic state. The neural network is trained using the acquired upper limb movement capture data and the calculated abnormal training parameter values, and a target neural network is generated. The method thus enables abnormal state detection based on the input upper limb movement capture data. The method has strong generalization ability and real-time performance, and improves the accuracy and efficiency of upper limb movement abnormality detection.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and system for detecting abnormal upper limb movement based on a neural network. Background Art

[0002] In existing technologies, upper limb movement abnormality detection usually relies on professional observation and evaluation, or preliminary judgment through simple sensor data. However, these methods often have problems such as high subjectivity, insufficient accuracy, and limited real-time feedback capabilities. Especially in complex upper limb training scenarios, accurate identification of subtle abnormalities in movement and timely intervention are crucial to improving training effectiveness and safety. Summary of the invention

[0003] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the present application is to provide a method and system for detecting upper limb movement abnormalities based on a neural network.

[0004] In a first aspect, the present application provides a method for detecting upper limb movement abnormalities based on a neural network, which is applied to an upper limb training monitoring system, and the method comprises:

[0005] Acquire upper limb motion capture data to be analyzed, wherein the upper limb motion capture data is motion behavior data captured when a first subject in an upper limb training space scene performs an upper limb training task; the upper limb training task is used to guide the motion area of ​​the upper limb motion to be locked in a target defined area during the training process in which the first subject continuously performs the upper limb motion;

[0006] Perform abnormal state detection on the upper limb motion capture data under multiple abnormal characteristic states, and generate an abnormal state degree value corresponding to each abnormal characteristic state; wherein the abnormal state degree value corresponding to each abnormal characteristic state represents the abnormal movement score of the first subject under the corresponding abnormal characteristic state;

[0007] Performing weighted calculation on the abnormal state degree value corresponding to each of the abnormal characteristic states to generate an abnormal training parameter value of the first subject;

[0008] Each of the upper limb motion capture data is used as a training sample, and the abnormal training parameter value of the first subject is used as a training label to train the neural network and generate a target neural network to detect abnormal conditions of the input upper limb motion capture data based on the target neural network.

[0009] In a second aspect, an embodiment of the present application further provides an upper limb training monitoring system, which includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores a computer program, and the computer program is loaded and executed according to the processor to implement the neural network-based upper limb movement abnormality detection method of the above first aspect.

[0010] Based on any one of the above aspects, the beneficial effects of this application are as follows:

[0011] The embodiment of the present application realizes the comprehensive capture of the subject's upper limb movements by acquiring the action behavior data of the first subject in the upper limb training space scene when performing the upper limb training task, that is, the upper limb motion capture data. Abnormal state detection is performed on these data under multiple abnormal characteristic states, and the corresponding abnormal state degree values ​​are generated, which effectively reflect the abnormal degree of the subject's movement under different abnormal characteristic states. Further, by weighted calculation of the abnormal state degree value corresponding to each abnormal characteristic state, the abnormal training parameter value of the first subject is generated, which comprehensively expresses the overall abnormal degree of the subject's movement when performing the upper limb training task, and provides accurate training labels for subsequent neural network training. In addition, the neural network is trained using the acquired upper limb motion capture data and the calculated abnormal training parameter value, and a target neural network is generated, so that abnormal state detection can be performed based on the input upper limb motion capture data, which has strong generalization ability and real-time performance, can be widely used in the field of upper limb movement abnormality detection, and provides strong technical support for upper limb movement rehabilitation training and abnormal diagnosis. Thus, the accuracy and efficiency of upper limb movement abnormality detection are improved, and a new solution is provided for upper limb movement rehabilitation training and abnormal diagnosis.

[0012] That is, the present application obtains the upper limb motion capture data to be analyzed, and performs abnormal state detection on the upper limb motion capture data under multiple abnormal characteristic states. After generating the abnormal state degree value corresponding to each of the abnormal characteristic states, a weighted calculation is performed on it, so as to obtain the abnormal training parameter value of the first subject. Using these data as training samples and labels, the neural network is trained to generate the target neural network, and then the input upper limb motion capture data can be accurately detected in abnormal state. Thus, by performing deep learning on the upper limb motion data through the neural network, the abnormal features in the upper limb movements can be effectively identified, and the accuracy and real-time performance of abnormal detection are improved; on the other hand, since the abnormal state degree value is used as a training label, the training process is more in line with actual needs, and the training effect of the neural network and its application effect in upper limb motion abnormality detection are further improved. In addition, the abnormal state degree value is weightedly calculated according to different abnormal characteristic states, so as to more accurately evaluate the abnormal training parameter value of the subject, which provides a basis for subsequent rehabilitation training. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of a flow chart of a method for detecting abnormal upper limb movements based on a neural network provided in an embodiment of the present application;

[0014] Figure 2 A block diagram showing the functional structure of an upper limb training monitoring system for implementing the above-mentioned neural network-based upper limb movement abnormality detection method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] See also Figure 1 As shown, the present application provides a method for detecting upper limb movement abnormalities based on a neural network, comprising the following steps.

[0016] Step S110, obtaining upper limb motion capture data to be analyzed, wherein the upper limb motion capture data is action behavior data captured when a first subject in an upper limb training space scene performs an upper limb training task. The upper limb training task is used to guide the action area of ​​the upper limb action to be locked in a target definition area during the training process in which the first subject continuously performs the upper limb action.

[0017] In this embodiment, the upper limb training monitoring system can receive data from the motion capture devices connected thereto. These motion capture devices are set in a space scene specifically for upper limb training, surrounding a first subject. The first subject is performing a series of upper limb training tasks, such as swinging arms, throwing and other actions. The motion capture device accurately records the upper limb movements of the first subject through sensors and cameras, and transmits these motion capture data to the upper limb training monitoring system in real time. These motion capture data are called upper limb motion capture data, which record in detail the movement behaviors of the subject when performing the training tasks.

[0018] That is, in detail, the upper limb motion capture data refers to the data about the movement of the upper limbs (such as arms, hands, etc.) captured by a specific device or system. These upper limb motion capture data may include a variety of information such as position, speed, acceleration, direction, etc., which are used to record and analyze the movement state of the upper limbs in detail. For example, in a sports training, an athlete wears an armband equipped with sensors. When he swings his arms, the sensors on the armband will capture his arm movement trajectory, speed changes and other data, and transmit these data to the analysis system.

[0019] The upper limb training space scene refers to the specific environment or scene for upper limb training. This upper limb training space scene may include multiple elements such as training equipment, space layout, training task settings, etc., aiming to provide an environment suitable for upper limb training. For example, a certain area in a gym is specially set up as an upper limb training area, equipped with various upper limb training equipment, such as dumbbells, barbells, pullers, etc. This area is an upper limb training space scene.

[0020] The upper limb training tasks refer to specific upper limb training movements or movement combinations set for the subjects. These tasks are designed to achieve the effect of improving upper limb strength, flexibility or other training goals through specific training movements. For example, a coach designs a set of upper limb strength training tasks for basketball players, including push-ups, dumbbell curls, pull-ups and other movements, and requires the athletes to complete these movements in a specified number of times and order.

[0021] The target definition area refers to an ideal action execution area set for the subject when performing upper limb training. The subject is guided to complete the training action in this area to ensure the correctness of the action and maximize the training effect. For example, when performing shooting training, the coach will set an ideal shooting area in front of the basket (such as an area near the free throw line) and ask the athlete to try to shoot the ball into this area to improve the accuracy and hit rate of shooting. This set shooting area is the target definition area.

[0022] It is worth noting that when the first subject performs the upper limb training task, the action needs to be performed continuously, rather than intermittently or stopping halfway. This helps to maintain the continuity and rhythm of the training and improve the training effect. For example, when doing push-up training, the coach requires the athlete to do 10 push-ups in a row without pausing or resting in between. This continuous action execution method is to perform upper limb movements uninterruptedly.

[0023] Step S120, performing abnormal state detection on the upper limb motion capture data under multiple abnormal characteristic states, and generating an abnormal state degree value corresponding to each abnormal characteristic state, wherein the abnormal state degree value corresponding to each abnormal characteristic state represents the movement abnormality score of the first subject under the corresponding abnormal characteristic state.

[0024] After receiving the upper limb motion capture data, the upper limb training monitoring system begins to analyze it. For example, the upper limb training monitoring system has a built-in set of complex algorithms that can identify a variety of abnormal characteristic states, such as unsmooth movements, movement ranges beyond predetermined limits, abnormal movement speeds, etc. For each abnormal characteristic state, the upper limb training monitoring system will detect the upper limb motion capture data and evaluate the extent to which the subject's movements show this abnormal characteristic. The result of this evaluation is quantified into an abnormal state degree value, which represents the subject's movement abnormality score under a specific abnormal characteristic state.

[0025] Step S130, performing weighted calculation on the abnormal state degree value corresponding to each of the abnormal characteristic states to generate an abnormal training parameter value of the first subject.

[0026] In this embodiment, after obtaining the abnormal state degree value under each abnormal characteristic state, the upper limb training monitoring system will perform further calculations. These abnormal state degree values ​​are not equally weighted, because different abnormal characteristic states may have different degrees of influence on the training effect. Therefore, the upper limb training monitoring system will perform weighted calculations on each abnormal state degree value according to a preset weight value, and finally obtain a comprehensive abnormal training parameter value. This value reflects the overall abnormality of the first subject's movements.

[0027] Step S140, using each of the upper limb motion capture data as a training sample and the abnormal training parameter value of the first subject as a training label, training the neural network and generating a target neural network to perform abnormal state detection on the input upper limb motion capture data based on the target neural network.

[0028] In this embodiment, the upper limb training monitoring system regards the upper limb motion capture data that has been collected and processed before as a better training sample. Each training sample is associated with an abnormal training parameter value, which serves as a training label and indicates the degree of abnormality of the action in the training sample. The upper limb training monitoring system uses these training samples with training labels to train a neural network. During the training process, the neural network learns how to predict the corresponding abnormal training parameter values ​​based on the input upper limb motion capture data. Over time, the prediction ability of the neural network gradually improves, and eventually becomes a target neural network that can accurately detect abnormal states of upper limb movements.

[0029] Based on the above steps, the embodiment of the present application realizes the comprehensive capture of the subject's upper limb movements by acquiring the action behavior data of the first subject in the upper limb training space scene when performing the upper limb training task, that is, the upper limb motion capture data. Abnormal state detection is performed on these data under multiple abnormal characteristic states, and the corresponding abnormal state degree values ​​are generated, which effectively reflect the abnormal degree of the subject's movement under different abnormal characteristic states. Further, by weighted calculation of the abnormal state degree value corresponding to each abnormal characteristic state, the abnormal training parameter value of the first subject is generated, which comprehensively expresses the overall abnormal degree of the subject's movement when performing the upper limb training task, and provides accurate training labels for subsequent neural network training. In addition, the neural network is trained using the acquired upper limb motion capture data and the calculated abnormal training parameter value, and a target neural network is generated, so that abnormal state detection can be performed based on the input upper limb motion capture data, which has strong generalization ability and real-time performance, can be widely used in the field of upper limb movement abnormality detection, and provides strong technical support for upper limb movement rehabilitation training and abnormal diagnosis. Thus, the accuracy and efficiency of upper limb movement abnormality detection are improved, and a new solution is provided for upper limb movement rehabilitation training and abnormal diagnosis.

[0030] That is, the present application obtains the upper limb motion capture data to be analyzed, and performs abnormal state detection on the upper limb motion capture data under multiple abnormal characteristic states. After generating the abnormal state degree value corresponding to each of the abnormal characteristic states, a weighted calculation is performed on it, so as to obtain the abnormal training parameter value of the first subject. Using these data as training samples and labels, the neural network is trained to generate the target neural network, and then the input upper limb motion capture data can be accurately detected in abnormal state. Thus, by performing deep learning on the upper limb motion data through the neural network, the abnormal features in the upper limb movements can be effectively identified, and the accuracy and real-time performance of abnormal detection are improved; on the other hand, since the abnormal state degree value is used as a training label, the training process is more in line with actual needs, and the training effect of the neural network and its application effect in upper limb motion abnormality detection are further improved. In addition, the abnormal state degree value is weightedly calculated according to different abnormal characteristic states, so as to more accurately evaluate the abnormal training parameter value of the subject, which provides a basis for subsequent rehabilitation training.

[0031] In a possible implementation, the abnormal characteristic state includes at least one of an abnormal motion extension characteristic state, an abnormal motion stability characteristic state, and an abnormal motion distortion characteristic state.

[0032] In this embodiment, the abnormal characteristic state of movement extension refers to the abnormality of the amplitude or range of movement compared with the standard or expectation during the execution of the upper limb movement. For example, the amplitude of the arm swing movement is too small or too large. The abnormal characteristic state of movement stability refers to the degree of instability of the upper limb movement during execution, such as trembling, shaking, etc. This may be caused by muscle fatigue, lack of strength or coordination problems. The abnormal characteristic state of movement distortion involves the degree of distortion or deformation of the upper limb movement during execution. This may be caused by joint flexibility problems, muscle tension or insufficient movement skills.

[0033] The abnormal state detection is performed on the upper limb motion capture data under multiple abnormal characteristic states to generate an abnormal state degree value corresponding to each abnormal characteristic state, including at least one of the following:

[0034] A. If the abnormal characteristic state includes an abnormal characteristic state of motion extension, abnormal state detection is performed on the upper limb motion capture data under the abnormal characteristic state of motion extension to generate a first abnormal state degree value corresponding to the abnormal characteristic state of motion extension. The first abnormal state degree value represents the motion abnormality score of the first subject under the abnormal characteristic state of motion extension.

[0035] For example, an upper limb training monitoring system receives upper limb motion capture data of a first subject when performing an arm swinging motion. The upper limb training monitoring system analyzes these data through an algorithm and compares the amplitude and range of the arm swing with a preset standard value. If it is found that the arm swing amplitude is significantly smaller or larger than the standard range, the upper limb training monitoring system will determine that there is an abnormal extension of the movement. Thus, a first abnormal state degree value is generated, and the first abnormal state degree value quantifies the degree of abnormal extension of the arm swinging motion. For example, a higher value may indicate that the arm swing amplitude is too small, while a lower value may indicate that the arm swing amplitude is too large.

[0036] B. If the abnormal characteristic state includes an abnormal characteristic state of motion stability, abnormal state detection is performed on the upper limb motion capture data under the abnormal characteristic state of motion stability to generate a second abnormal state degree value corresponding to the abnormal characteristic state of motion stability. The second abnormal state degree value represents the motion abnormality score of the first subject under the abnormal characteristic state of motion stability.

[0037] For example, in another training, the first subject is asked to keep the arm stable and motionless for a long time. The upper limb training monitoring system analyzes the captured data to monitor the slight movement and tremor of the arm. If it is detected that the arm stability is significantly lower than the preset threshold, it is judged as abnormal movement stability. As a result, the upper limb training monitoring system generates a second abnormal state degree value, which reflects the abnormal degree of arm stability. A higher value indicates that the arm is more unstable, while a lower value indicates relatively good stability.

[0038] C. If the abnormal characteristic state includes an abnormal characteristic state of motion distortion, abnormal state detection is performed on the upper limb motion capture data under the abnormal characteristic state of motion distortion to generate a third abnormal state degree value corresponding to the abnormal characteristic state of motion distortion. The third abnormal state degree value represents the motion abnormality score of the first subject under the abnormal characteristic state of motion distortion.

[0039] For example, in the training of throwing action, the upper limb training monitoring system analyzes the upper limb movement data of the first subject, especially the movement trajectory of the arm and wrist. If it is found that the movement trajectory is obviously distorted or deformed compared with the standard throwing movement trajectory, it is judged that the movement distortion is abnormal. As a result, the upper limb training monitoring system generates a third abnormal state degree value, which quantifies the degree of abnormal distortion of the throwing action. A higher value indicates that the movement distortion is greater, while a lower value indicates that the movement is closer to the standard.

[0040] After generating the abnormal state degree value corresponding to each abnormal characteristic state, the upper limb training monitoring system can process these values ​​comprehensively according to the preset weights or algorithms to obtain a comprehensive abnormal training parameter value. This comprehensive value can provide more comprehensive information for coaches or researchers to help them evaluate the upper limb training effect and problems of the first subject.

[0041] In a possible implementation, the upper limb training task includes K groups of training action sequences, and the upper limb motion capture data includes: the movement amplitudes corresponding to the K groups of training action sequences.

[0042] The abnormal state detection is performed on the upper limb motion capture data under the abnormal state of motion extension to generate a first abnormal state degree value corresponding to the abnormal state of motion extension, including:

[0043] Step A110, performing motion extension analysis based on the motion amplitude corresponding to each group of the training motion sequence, and generating a motion extension abnormality corresponding to each group of the training motion sequence. The motion extension abnormality is a mapping parameter value of each group of the training motion sequence for resisting the motion abnormality resistance suffered by the first subject when performing the upper limb motion, and the motion extension abnormality is positively correlated with the motion abnormality resistance.

[0044] For example, in a professional sports training center, a first subject is performing an upper limb training task. This upper limb training task includes K sets of training action sequences, each of which is designed to train a specific muscle group or sports skill. In order to accurately evaluate the subject's performance, advanced motion capture equipment is used to collect data.

[0045] When the first subject performs each training sequence, the motion capture device records the range of motion in detail. These data are transmitted to the upper limb training monitoring system in real time for subsequent analysis and processing.

[0046] In this embodiment, after receiving the upper limb motion capture data of the first subject, the upper limb training monitoring system starts to perform motion extension analysis. For each set of training action sequences, the upper limb training monitoring system compares its motion amplitude with a predefined standard amplitude. These standard amplitudes are formulated based on the experience of professional coaches and the principles of sports science, and represent normal and efficient ways of performing actions.

[0047] For example, the upper limb training monitoring system can analyze the difference in the range of motion, taking into account a variety of factors, such as the speed, acceleration, direction, etc. of the motion, to ensure the accuracy and comprehensiveness of the analysis. If the range of a certain training movement sequence is significantly different from the standard range, the upper limb training monitoring system will consider that the movement sequence has an extension abnormality.

[0048] After analysis, the upper limb training monitoring system generates the degree of abnormal extension of movement corresponding to each set of training action sequences. This degree of abnormal extension of movement is a specific value that quantifies the degree of deviation of the subject from the standard amplitude when performing the set of movements. The higher the degree of abnormal extension of movement, the more abnormal the subject's movement extension is, and the greater the possible problem is.

[0049] The first subject encounters a resistance or difficulty when performing an upper limb movement, which causes the movement amplitude to deviate from the standard amplitude, i.e., movement extension abnormality. The movement extension abnormality is a quantitative representation of this deviation, which is proportional to the resistance to movement abnormality suffered by the subject.

[0050] Specifically, abnormal resistance to movement refers to various internal or external factors that the subject may encounter when performing upper limb movements, which causes the movement to be performed unsmoothly or deviate from expectations. These factors may include muscle fatigue, lack of strength, coordination problems, unskilled skills, external interference, etc. These resistances will cause the subject to have an additional burden during the execution of the movement, causing the movement amplitude, speed or direction to deviate from the standard or expectation. Therefore, the abnormal extension of movement is a quantitative indicator used to measure the degree of deviation of the movement amplitude of the subject when performing upper limb movements. It is calculated by comparing the difference between the actual movement amplitude and the standard amplitude. If there is a large difference between the actual movement amplitude and the standard amplitude, it indicates that the subject suffered a large abnormal resistance to movement when performing the movement, resulting in a high abnormal extension of movement. On the contrary, if the actual movement amplitude is close to the standard amplitude, it indicates that the subject suffered less abnormal resistance to movement when performing the movement, and the abnormal extension of movement is low.

[0051] Therefore, the degree of abnormal extension of movement is positively correlated with abnormal resistance of movement, which means that as the resistance of abnormal movement increases, the degree of abnormal extension of movement will also increase accordingly. This positive correlation provides an important basis for evaluating the subject's action execution and identifying potential problems. By monitoring and analyzing the changes in the degree of abnormal extension of movement, coaches or researchers can understand the difficulties or challenges that subjects may encounter when performing upper limb movements, and accordingly develop targeted training plans or provide necessary guidance and support.

[0052] Step A120: weighted calculation is performed on the generated K action extension abnormalities to generate an action extension abnormality parameter. The action extension abnormality parameter represents the weighted action extension abnormality obtained by weighted calculation.

[0053] After obtaining the abnormal degree of movement extension of each training action sequence, the upper limb training monitoring system starts weighted calculation. The purpose of weighted calculation is to combine the abnormal degrees of different training action sequences to obtain an overall evaluation index, which helps coaches or researchers to fully understand the abnormality of the subject's movement extension.

[0054] In detail, the upper limb training monitoring system assigns different weights to each set of training action sequences according to factors such as importance and difficulty. Then, these weights are used to weightedly calculate the abnormality of movement extension to obtain a comprehensive movement extension abnormality parameter. This movement extension abnormality parameter is a weighted average that reflects the overall movement extension abnormality of the subject when performing all training action sequences.

[0055] Through weighted calculation, the upper limb training monitoring system obtains a specific movement extension abnormality parameter value. The higher the movement extension abnormality parameter value, the more serious the subject's overall movement extension abnormality. Coaches or researchers can evaluate the subject's performance based on this value and develop corresponding training plans or provide targeted guidance.

[0056] Step A130: output the motion extension abnormality parameter as a first abnormal state degree value corresponding to the motion extension abnormality characteristic state.

[0057] The movement abnormality score of the first subject in the movement extension abnormality characteristic state includes a movement range deviation, and the movement range deviation is positively correlated with the weighted movement extension abnormality.

[0058] After completing the calculation of the abnormal extension parameter, the upper limb training monitoring system outputs the abnormal extension parameter value as the first abnormal state degree value corresponding to the abnormal extension characteristic state. This first abnormal state degree value is a quantitative indicator used to evaluate the degree of abnormal movement of the subject under the abnormal extension characteristic state. It includes information such as the deviation of the range of motion, providing a more comprehensive evaluation basis for coaches or researchers.

[0059] Therefore, the coach or researcher can identify which action sequences of the subject have extension abnormalities according to the first abnormal state degree value, and adjust the training plan or provide targeted guidance accordingly. For example, if the abnormality degree of a training action sequence is high, the coach can focus on the execution of the training action sequence to help the subject correct the action posture or improve the action skills. In this way, the training effect of the subject can be effectively improved.

[0060] In a possible implementation, the upper limb training task includes K groups of training action sequences, and the upper limb motion capture data includes the movement amplitudes corresponding to the K groups of training action sequences.

[0061] In this embodiment, the training action sequence is designed to train the strength, stability, coordination, etc. of the upper limbs. The upper limb motion capture data received by the upper limb training monitoring system includes the motion amplitudes corresponding to the K groups of training action sequences. These upper limb motion capture data are obtained through motion capture technology and can accurately reflect the amplitude changes of the subject when performing each group of actions.

[0062] For example, the upper limb training task may be: the subject is asked to complete an upper limb training task including 5 sets (K=5) of training action sequences. The 5 sets of training action sequences are: push-ups, dumbbell curls, pull-ups, dumbbell presses, and bent-over rows.

[0063] The motion capture system then recorded the range of motion of the subjects when they completed these five sets of movements. For example, in push-ups, the motion capture system recorded the range of the subjects' sinking to the lowest point and pushing up to the highest point each time; in dumbbell curls, the range of the dumbbells from the starting position to the highest point was recorded, etc.

[0064] The abnormal state detection is performed on the upper limb motion capture data under the abnormal characteristic state of motion stability to generate a second abnormal state degree value corresponding to the abnormal characteristic state of motion stability, including:

[0065] Step B110, walking through the movement amplitudes corresponding to the K groups of training movement sequences.

[0066] For example, in this embodiment, the upper limb training monitoring system first processes the data of push-ups, then dumbbell curls, and so on, until the data of bent-over rowing is processed.

[0067] Step B120, when walking to the motion amplitude corresponding to the x-th group of training motion sequences, if x is greater than 1, obtaining the motion amplitude corresponding to the x-1-th group of training motion sequences.

[0068] For example, when processing dumbbell curl data, the upper limb training monitoring system will also refer to the previously recorded push-up data. Similarly, when processing pull-up data, it will refer to the dumbbell curl data, and so on.

[0069] Step B130, based on the movement amplitude corresponding to the xth group of training movement sequences and the movement amplitude corresponding to the x-1th group of training movement sequences, determine the movement extension change parameter. The movement extension change parameter is used to represent the change value between the movement amplitude corresponding to the x-1th group of training movement sequences and the movement amplitude corresponding to the xth group of training movement sequences.

[0070] For example, an upper body training monitoring system compares the range of motion variation between push-ups and dumbbell curls. If the subject is able to maintain a stable range of motion in push-ups, but significantly decreases or increases the range of motion in dumbbell curls, then the extension variation parameter between the two exercises will be high.

[0071] Step B140: After obtaining K-1 action extension and variation parameters, weighted calculation is performed on the K-1 action extension and variation parameters to generate an action abnormal fluctuation parameter. The action abnormal fluctuation parameter represents the weighted action extension and variation parameter generated by weighted calculation.

[0072] For example, the upper limb training monitoring system comprehensively considers all the extension change parameters from push-ups to dumbbell curls, dumbbell curls to pull-ups, pull-ups to dumbbell presses, and dumbbell presses to bent-over rows. Since the difficulty and importance of each set of movements may be different, the upper limb training monitoring system assigns different weights to each set of changes, and then performs weighted calculations to obtain an overall abnormal movement fluctuation parameter.

[0073] Step B150, outputting the abnormal motion fluctuation parameter as a second abnormal state degree value corresponding to the abnormal characteristic state of motion stability.

[0074] The abnormal motion score of the first subject in the abnormal motion stability characteristic state includes abnormal motion fluctuation. The abnormal motion fluctuation is positively correlated with the weighted motion extension variation parameter. x∈[1, K].

[0075] After weighted calculation, the upper limb training monitoring system obtains a specific value, which is the second abnormal state degree value corresponding to the abnormal characteristic state of the subject's movement stability in this upper limb training task. The higher this value is, the worse the subject's movement stability is.

[0076] If the second abnormal state degree value output by the upper limb training monitoring system is high, the coach can know that the subject may have stability problems when performing certain training movements. For example, the subject may have unstable range of motion in dumbbell curls due to insufficient strength or unskilled skills. Based on this information, the coach can provide more targeted guidance and training plan adjustments to the subject to help him improve his movement stability and enhance training results.

[0077] In a possible implementation manner, the motion amplitude includes a horizontal motion amplitude and a vertical motion amplitude.

[0078] Step B130 may include:

[0079] Step B131, performing difference calculation based on the horizontal movement amplitude corresponding to the xth group of training movement sequence and the horizontal movement amplitude corresponding to the x-1th group of training movement sequence, to generate a change in the horizontal movement amplitude of the xth group of training movement sequence relative to the x-1th group of training movement sequence.

[0080] Step B132, performing difference calculation based on the vertical movement amplitude corresponding to the xth group of training action sequences and the vertical movement amplitude corresponding to the x-1th group of training action sequences, to generate a vertical movement amplitude change of the xth group of training action sequences relative to the x-1th group of training action sequences.

[0081] Step B133: generating a motion extension change parameter based on the horizontal motion amplitude change and the vertical motion amplitude change of the xth group of training motion sequences relative to the x-1th group of training motion sequences.

[0082] For example, in the training area of ​​a fitness center, the subject is performing a series of upper limb training movements. These movements include horizontal and vertical movements, such as pushing, pulling, lifting, etc. In order to fully evaluate the subject's performance, the upper limb training monitoring system needs to consider both the horizontal and vertical range of motion.

[0083] The upper limb training monitoring system receives detailed data on the range of motion of the subject when performing each set of training movements, including horizontal and vertical range of motion. These data are obtained through a high-precision motion capture system and can accurately reflect the subject's movement execution.

[0084] When the upper limb training monitoring system processes the xth group of training action sequences, the following steps are performed to determine the action extension change parameters:

[0085] The upper limb training monitoring system first extracts the horizontal movement amplitude data of the xth group of training action sequences and the x-1th group of training action sequences.

[0086] Then, it calculates the difference between the two groups of data, that is, the change in the horizontal movement amplitude of group x relative to group x-1. This change reflects the stability of the subject's horizontal movement amplitude.

[0087] Next, the upper limb training monitoring system extracts the corresponding vertical movement amplitude data. Similarly, the difference in vertical movement amplitude between the xth group and the x-1th group is calculated to obtain the vertical movement amplitude change. This change reflects the stability of the subject's vertical movement amplitude.

[0088] The upper limb training monitoring system combines the calculated horizontal movement amplitude change and vertical movement amplitude change to generate a comprehensive movement extension change parameter through a specific algorithm (such as weighted average, Euclidean distance, etc.). This movement extension change parameter not only takes into account the movement change in the horizontal direction, but also the movement change in the vertical direction, so as to more comprehensively evaluate the movement stability of the subject.

[0089] In this way, the upper limb training monitoring system can generate a movement extension change parameter for each training action sequence, which is then used for further analysis and evaluation. Coaches can use these parameters to identify possible problems when subjects perform actions, such as uncoordinated movements and uneven force distribution, so as to develop more targeted training plans to help subjects improve their movements and enhance training results.

[0090] Therefore, the upper limb training monitoring system can more accurately evaluate the upper limb training performance of the subjects through comprehensive analysis of the horizontal and vertical range of motion, and provide more valuable feedback and guidance to coaches and subjects.

[0091] In a possible implementation, performing abnormal state detection on the upper limb motion capture data under the abnormal characteristic state of the motion distortion degree to generate a third abnormal state degree value corresponding to the abnormal characteristic state of the motion distortion degree includes:

[0092] Step C110, performing continuous path fitting on the upper limb motion capture data to generate a continuous action path.

[0093] For example, in the training area of ​​a fitness center, a subject is performing a series of upper limb training movements. The upper limb training monitoring system receives the subject's upper limb motion capture data, which records in detail the subject's motion trajectory when performing each movement. Now, the upper limb training monitoring system needs to perform abnormal state detection on these data under the abnormal characteristic state of motion distortion to evaluate whether the subject's movement has abnormal distortion.

[0094] The upper limb training monitoring system first performs continuous path fitting on the received upper limb motion capture data. This means that the upper limb training monitoring system will connect each movement trajectory point of the subject to form a continuous movement path. This path can clearly reflect the movement trajectory of the subject when performing the action.

[0095] Step C120 , segmenting the action continuous path based on the multiple deformation points of the action continuous path to generate multiple action continuous sub-paths, each action continuous sub-path corresponding to a variation range.

[0096] Next, the upper limb training monitoring system will segment the action continuous path based on multiple deformation points. Deformation points are points on the path where the curvature changes greatly, usually indicating important turning points or change points of the action. Through segmentation, the upper limb training monitoring system divides the continuous path into multiple action continuous sub-paths, each of which corresponds to a specific range of change. This makes it easier to analyze and process each sub-path separately.

[0097] Step C130 , identifying significant deformation intervals for the variation ranges corresponding to the plurality of continuous motion sub-paths, and generating a plurality of significant deformation intervals.

[0098] For example, the upper limb training monitoring system then identifies the significant deformation intervals for the range of change corresponding to each continuous sub-path of the movement. The significant deformation interval refers to the interval where the curvature change in the sub-path is particularly obvious or abnormal, which may indicate that the subject has distortion or abnormality when performing this part of the movement. The upper limb training monitoring system will automatically identify and mark these intervals using algorithms.

[0099] Step C140: determining a third abnormal state degree value corresponding to the abnormal characteristic state of the motion distortion degree based on the number of the significant deformation intervals and the variation ranges corresponding to the plurality of motion continuous sub-paths.

[0100] Finally, the upper limb training monitoring system determines the third abnormal state degree value corresponding to the abnormal characteristic state of the motion distortion based on the number of multiple significant deformation intervals identified and the range of change corresponding to each continuous sub-path of the motion. This value is a quantitative indicator used to evaluate the degree of abnormality of the subject in terms of motion distortion. Specifically, the upper limb training monitoring system may calculate this value by comprehensively considering factors such as the number, length, and curvature change amplitude of the significant deformation intervals.

[0101] As a result, the upper limb training monitoring system can generate a third abnormal state degree value that reflects the abnormal degree of distortion of the subject's upper limb movements. The coach can use this value to evaluate whether the subject has abnormal distortion when performing upper limb training movements, and then determine whether the movements are standard and coordinated. If abnormalities are found, the coach can provide timely guidance and corrections to help the subject improve the quality of the movement and improve the training effect. At the same time, this data-driven evaluation method is more objective and accurate, and can provide coaches with more powerful decision-making support.

[0102] In a possible implementation, step C110 may include:

[0103] Step C111, deconstructing the upper limb motion capture data to generate horizontal motion behavior data and vertical motion behavior data.

[0104] For example, the upper limb training monitoring system first deconstructs the received upper limb motion capture data. This means that the upper limb training monitoring system decomposes the raw data into two parts: horizontal action behavior data and vertical action behavior data. The horizontal action behavior data mainly describes the subject's motion trajectory in the horizontal direction, while the vertical action behavior data describes the subject's motion trajectory in the vertical direction. Such deconstruction helps the upper limb training monitoring system to more accurately analyze the subject's motion performance in different directions.

[0105] Step C112, adaptively updating the horizontal motion behavior data using a continuous function model to generate horizontal motion path data, and adaptively updating the vertical motion behavior data based on the continuous function model to generate vertical motion path data.

[0106] Next, the upper limb training monitoring system uses the continuous function model to adaptively update the horizontal motion behavior data and the vertical motion behavior data respectively. The continuous function model is a mathematical tool that can be used to describe the trajectory of continuous changes. The upper limb training monitoring system will adjust the parameters of the continuous function model according to the actual motion trajectory of the subject to make it better fit the subject's motion path. Through adaptive updating, the upper limb training monitoring system generates horizontal motion path data and vertical motion path data, which respectively reflect the subject's continuous motion trajectory in the horizontal and vertical directions.

[0107] Step C113, integrating the horizontal motion path data and the vertical motion path data to generate a motion continuous path.

[0108] Finally, the upper limb training monitoring system integrates the horizontal motion path data and the vertical motion path data. The integration process is to merge these two parts of data into a complete continuous motion path. This continuous motion path combines the subject's motion trajectory in the horizontal and vertical directions, and can fully reflect the subject's upper limb motion performance. Through such integrated processing, the upper limb training monitoring system obtains a continuous and smooth motion path, which provides a basis for subsequent analysis and evaluation.

[0109] As a result, the upper limb training monitoring system successfully generated a continuous path for the subject's upper limb movements. This continuous path for upper limb movements provides valuable feedback information for coaches and subjects. Coaches can evaluate whether the subject's movements are smooth and coordinated, and whether there are distortions or abnormalities based on the continuous path for movements. At the same time, subjects can also understand their own movement performance by observing their own continuous path for movements, so as to make targeted improvements and enhancements. This data-driven analysis method not only improves the accuracy and objectivity of the evaluation, but also provides coaches and subjects with more intuitive and comprehensive feedback on movement performance.

[0110] In a possible implementation, step C120 may include:

[0111] Step C121, selecting a target deformation point from a plurality of deformation points of the action continuous path.

[0112] Step C122, segmenting the action continuous path based on the target deformation point to generate multiple action continuous sub-paths, wherein the variation ranges corresponding to the respective action continuous sub-paths are arranged in sequence, and except for the variation range corresponding to the first action continuous sub-path and the variation range corresponding to the last action continuous sub-path, the deformation points corresponding to the critical points of each variation range are all the target deformation points.

[0113] In this embodiment, the upper limb training monitoring system first selects target deformation points from multiple deformation points of the continuous path of the action. These target deformation points are automatically selected by the upper limb training monitoring system according to specific algorithms or standards (such as curvature change threshold, distance threshold, etc.), and they represent the most significant or most critical change points on the path. The process of selecting target deformation points may involve operations such as sorting, screening or clustering the deformation points to ensure that the selected points can effectively reflect the changing characteristics of the subject's movements.

[0114] Once the target deformation points are selected, the upper limb training monitoring system will segment the action continuity path based on these points. The segmentation process is to cut the entire action continuity path into multiple smaller parts, each of which is called an action continuity sub-path. Each action continuity sub-path corresponds to a specific range of variation, and these ranges of variation are arranged in sequence on the action continuity path. Except for the range of variation corresponding to the first action continuity sub-path and the range of variation corresponding to the last action continuity sub-path, the deformation points corresponding to the critical points of each range of variation (i.e., the starting point and end point of the sub-path) are all the target deformation points selected previously.

[0115] As a result, the upper limb training monitoring system decomposes the complex action path into multiple action sub-paths that are easier to analyze and understand. Each sub-path focuses on a specific stage or change interval in the subject's action trajectory, allowing coaches and subjects to more clearly identify and understand the action characteristics and problems in each stage. This segmented analysis method helps to improve the accuracy and pertinence of action evaluation, and provides strong support for subsequent guidance and improvement.

[0116] In a possible implementation, step C130 may include:

[0117] Step C131 , walking through the variation ranges corresponding to the plurality of action continuous sub-paths, each of the variation ranges including a first critical point and a second critical point.

[0118] Step C132, obtaining the amount of change in the motion direction of each variation range visited, wherein the amount of change in the motion direction is formed by the direction vector of the first critical point of the variation range and the direction vector of the second critical point of the same variation range.

[0119] Step C133: outputting the change range of the movement direction change amount not less than the preset change amount as a significant deformation interval.

[0120] In this embodiment, the upper limb training monitoring system starts to walk through the ranges of change corresponding to the multiple continuous sub-paths of the action one by one. Each range of change is defined by a first critical point and a second critical point, which are the starting point and the end point of the sub-path, respectively. The upper limb training monitoring system processes the range of change corresponding to each sub-path in sequence.

[0121] For each range of change, the upper limb training monitoring system calculates the change in direction of the movement. This change is obtained by comparing the direction vector of the first critical point of the range of change with the direction vector of the second critical point. The direction vector is a mathematical quantity that describes the direction and speed of the movement path at a specific point. The upper limb training monitoring system uses these direction vectors to calculate the magnitude of the direction change.

[0122] The upper limb training monitoring system compares the calculated change in the direction of movement with the preset change threshold. If the change in the direction of movement in a certain range of change is not less than the preset change, then this range of change is considered to be a significant deformation interval. The upper limb training monitoring system outputs these significant deformation intervals for subsequent analysis and processing.

[0123] Through the above steps, the upper limb training monitoring system can accurately identify the significant deformation intervals in the continuous sub-path of the movement. These significant deformation intervals usually correspond to important turning points or abnormal changes in the subject's movements, and are the parts that coaches and subjects need to pay special attention to. Coaches can evaluate the subject's movement quality based on the location and characteristics of these significant deformation intervals, identify possible problems, and provide targeted guidance and corrections. At the same time, subjects can also understand their own movement performance by observing these significant deformation intervals, so as to make targeted improvements and improvements. This data-driven analysis method not only improves the accuracy and objectivity of the evaluation, but also provides coaches and subjects with more intuitive and comprehensive feedback on movement performance.

[0124] In a possible implementation, the upper limb training task includes K sets of training action sequences, and step C140 may include:

[0125] Step C141, connect the first critical point and the second critical point of the variation range corresponding to each action continuous sub-path to generate an action route corresponding to each action continuous path. Each action continuous sub-path has multiple action path positions.

[0126] In this embodiment, the upper limb training monitoring system first connects the first critical point and the second critical point of the variation range corresponding to each action continuous sub-path to generate an action route corresponding to each action continuous sub-path. This action route can be regarded as the trajectory of the subject's arm movement when performing a certain action. Each action continuous sub-path has multiple action path positions, which represent different points on the trajectory.

[0127] Step C142, determining the deviation between each action path position on each action continuous sub-path and the action route corresponding to each action continuous sub-path. The deviation corresponding to each action path position represents the distortion measurement value of the action continuous path at the corresponding action path position.

[0128] Next, the upper limb training monitoring system determines the deviation between each action path position and the corresponding action route on each continuous sub-path of the action. Deviation is a measure of the difference between the action path position and the ideal route, reflecting the degree of distortion of the subject's arm trajectory when performing the action. The upper limb training monitoring system uses an algorithm to calculate the distance or angle difference between each action path position and the action route to obtain the corresponding deviation value.

[0129] Step C143, determining the distortion measure value of the continuous motion path based on the deviation corresponding to each motion path position, the number of motion path positions and the training number K of the training motion sequence.

[0130] Based on the deviations corresponding to each position in the action path, the number of positions in the action path, and the number of training sequences K, the upper limb training monitoring system determines the distortion metric of the action path. This distortion metric is a comprehensive assessment of the deviations of all positions in the entire action path, taking into account the number and complexity of the movements, as well as the consistency of the subjects' performance over multiple training sessions.

[0131] In detail, in one example, first, the upper limb training monitoring system has determined the deviation corresponding to each motion path position, and the number of motion path positions represents how many data points there are in the entire motion continuous path. This number is intuitively important for calculating the distortion metric value and determines the density and coverage of the data points. Generally, the more the number of motion path positions, the more accurate the calculated distortion metric value. The number of training (K) represents the number of times the subject performs the same training action sequence. If the subject shows a similar distortion pattern in multiple training sessions, then this distortion may be more representative. Therefore, the number of training sessions can be used as a weight factor when calculating the distortion metric value.

[0132] In order to comprehensively consider the deviations of all motion path positions and the number of training, the upper limb training monitoring system will traverse all motion path positions, add up the deviations at each motion path position, and obtain a total deviation. This total deviation reflects the overall difference between the subject's movement and the ideal route throughout the continuous path of the movement. The number of motion path positions represents the number of data points used to calculate the total deviation. In order to take into account the number of motion path positions, the upper limb training monitoring system can divide the total deviation by the number of motion path positions to obtain an average deviation, thereby ensuring that different numbers of position data are comparable when calculating the distortion metric.

[0133] Next, the upper limb training monitoring system needs to consider the effect of the number of training sessions ( K ) on the distortion metric. If the subject shows a similar distortion pattern in multiple training sessions, then the distortion may be more significant. To reflect this, the upper limb training monitoring system can multiply the average deviation by a weight factor related to the number of training sessions. This weight factor can be designed according to the actual situation, such as a function of the number of training sessions or a fixed value.

[0134] Finally, the weighted average deviation is used as the distortion measure of the action continuous path. This distortion measure combines the deviation of all action path positions, the number of positions, and the influence of the number of trainings, and can more comprehensively reflect the degree of distortion of the subject when performing the action.

[0135] In practical applications, the calculated distortion measure values ​​may need to be further processed and analyzed, such as normalization to eliminate dimensional differences, or application of statistical methods for significance testing and reliability assessment.

[0136] In addition, in other examples, other calculation methods may also be used, and reference may be made to the relevant description sections of the subsequent embodiments.

[0137] Step C144, determining a motion distortion parameter based on the distortion measure value of the motion continuous path and the number of the significant deformation intervals, and outputting the motion distortion parameter as a third abnormal state degree value corresponding to the motion distortion abnormal characteristic state.

[0138] Finally, the upper limb training monitoring system determines the action distortion parameter based on the distortion measure of the action continuous path and the number of multiple significant deformation intervals previously identified. For example, the preset mapping weight corresponding to the number of significant deformation intervals can be obtained, and then the distortion measure is multiplied by the preset mapping weight to determine the action distortion parameter. This action distortion parameter is a quantitative indicator used to indicate the degree of distortion of the subject's action when performing an upper limb training task. The upper limb training monitoring system may perform weighted calculations or other mathematical processing on the distortion measure and the number of significant deformation intervals to obtain a comprehensive evaluation value as the action distortion parameter.

[0139] Through the above steps, the upper limb training monitoring system can accurately evaluate the degree of distortion of the subject's movements when performing upper limb training tasks, and output the corresponding third abnormal state degree value. This value can provide valuable feedback information for coaches, helping them understand whether there is distortion or abnormality in the subject's movement performance, and formulate targeted training plans and corrective measures accordingly. At the same time, the subject can also use this value to understand the quality of his or her movements, so as to make targeted improvements and improvements. This data-driven analysis method not only improves the accuracy and objectivity of the evaluation, but also provides coaches and subjects with more intuitive and comprehensive feedback on movement performance.

[0140] In a possible implementation, step C143 may include:

[0141] Step C1431, based on the spatial relationship between each motion path position and the corresponding motion route, determine the distortion vector weight of each motion path position. The distortion vector weight represents the abnormal training parameter value determined due to the motion distortion trajectory of the motion continuous path.

[0142] For each motion path position, the upper limb training monitoring system calculates the spatial relationship between the motion path and the corresponding motion route, which includes factors such as distance, angle, and direction. Based on these spatial relationships, the upper limb training monitoring system assigns a distortion vector weight to each motion path position. This distortion vector weight reflects the importance of the motion path position in the overall motion path and its contribution to the motion distortion.

[0143] For example, suppose that in a certain movement, the subject's hand should move along a straight line from point A to point B. But in fact, the subject's hand deviates from this straight line during the movement, especially at a certain position C in the middle. The upper limb training monitoring system will identify position C and assign it a higher distortion vector weight because it contributes the most to the overall movement distortion.

[0144] In one example, suppose a throwing motion is being analyzed, and the corresponding motion path is a straight line from the shoulder to the target. The subject may deviate from this motion path when actually performing the motion. Then, the distortion vector weight for each position can be determined based on the spatial relationship between the subject's motion path position and the ideal path.

[0145] First, a motion capture system is used to collect the subject's motion path data when performing the throwing action. This data includes the hand position (x, y, z coordinates) at each time point. At the same time, the ideal throwing path is defined, which is a straight line from the starting point (shoulder position) to the target point (throwing target).

[0146] For each position point on the motion path, calculate the shortest distance between it and the ideal route. This can be achieved by calculating the perpendicular distance from the point to the line. In addition to distance, direction deviation and angle deviation are also considered. Direction deviation can be obtained by calculating the angle between the velocity vector of the position point and the direction vector of the ideal route. Angle deviation is the angle between the line connecting the position point and the nearest point on the ideal route and the ideal route.

[0147] The above-calculated spatial relationship parameters (distance, direction deviation, angle deviation) are used to define the distortion vector. This distortion vector reflects the distortion of the position point in three-dimensional space. The distortion vector weight can be defined according to the needs of the actual application. One implementation method is to normalize the distance, direction deviation and angle deviation, and add them weightedly to obtain a comprehensive weight value. For example, the following formula can be used:

[0148] Warp vector weight = w1 * (normalized distance) + w2 * (normalized direction deviation) + w3 * (normalized angle deviation)

[0149] Among them, w1, w2 and w3 are weight factors, which can be determined based on experience or experimental data. Normalization is to ensure that different parameters have the same dimension and range in the weight calculation.

[0150] Once the distortion vector weights for each motion path position are calculated, these weights can be used to further analyze the degree of motion distortion, such as multiplying the deviation of each position by its weight when calculating the deviation from the target.

[0151] For example, suppose that at a certain time point t, the subject's hand position deviates from the ideal route by 5 cm, 10 degrees in direction, and 5 degrees in angle. These values ​​can be normalized to the range of 0 to 1, and the weight factors w1=0.5, w2=0.3, and w3=0.2 are assumed. Then, the distortion vector weight of the position point is calculated as follows:

[0152] Normalized distance = 5cm / (maximum possible distance)

[0153] Normalized heading deviation = 10 degrees / (maximum possible heading deviation)

[0154] Normalized angular deviation = 5 degrees / (maximum possible angular deviation)

[0155] Distortion vector weight = 0.5 * normalized distance + 0.3 * normalized direction deviation + 0.2 * normalized angle deviation

[0156] It is worth noting that the "maximum possible distance", "maximum possible direction deviation" and "maximum possible angle deviation" here are defined according to the specific application scenario and action range. They represent the maximum values ​​that may appear on these parameters and are used to normalize the actual measurement values ​​to the same range. In actual applications, these maximum values ​​may need to be determined based on experience, experimental data or expert advice.

[0157] Step C1432, performing calculations based on the deviation corresponding to each motion path position, the distortion vector weight of the corresponding motion path position, and the number of motion path positions to generate a target deviation.

[0158] The upper limb training monitoring system then iterates through all the motion path positions and multiplies the deviation of each motion path position by its corresponding distortion vector weight. This is done to quantify the contribution of each motion path position to the overall distortion. The upper limb training monitoring system then adds up all the weighted deviations and divides them by the number of motion path positions to obtain an average weighted deviation, which is the target deviation.

[0159] Continuing with the above example, if the deviation of position C is 0.5 (assuming the deviation range is 0 to 1), and its distortion vector weight is 2 (assuming the weights of other positions are all 1), then the contribution of position C to the target deviation is 0.5* 2 = 1. The upper limb training monitoring system will similarly calculate the contributions of all positions, add them together, and divide by the number of positions to get the final target deviation.

[0160] Step C1433: determining the distortion measurement value of the action continuous path based on the training quantity K of the training action sequence and the target deviation.

[0161] After determining the target deviation of the subject in a single training session, in order to more comprehensively evaluate the subject's movement distortion, the upper limb training monitoring system also needs to consider the number of training sessions K. The upper limb training monitoring system can multiply the target deviation by the number of training sessions K to obtain a weighted average distortion measurement value. This weighted average distortion measurement value comprehensively considers the distortion degree of a single training session and the consistency of multiple training sessions.

[0162] For example, if a subject performs 5 training sessions (K=5), and the target deviations for each session are 0.2, 0.15, 0.22, 0.18, and 0.21, then the upper limb training monitoring system can first calculate the average target deviation for these 5 sessions (e.g., using a simple average or a weighted average), and then multiply this average by 5 to obtain the final distortion metric value. The higher the distortion metric value, the more consistent the subject's performance of greater movement distortions across multiple sessions.

[0163] Therefore, the upper limb training monitoring system can determine the distortion measurement value of the continuous action path based on the deviation of the action path position, the distortion vector weight and the training quantity K, thereby providing an important basis for subsequent action quality analysis and anomaly detection.

[0164] In a possible implementation, the method further includes:

[0165] Step D110, obtaining abnormal training parameter values ​​of a plurality of second subjects who have completed upper limb training in the upper limb training space scene, wherein the upper limb training category of the plurality of second subjects is the same as the upper limb training category of the first subject.

[0166] Exemplarily, in an upper limb rehabilitation training system, an upper limb training monitoring system is responsible for monitoring, analyzing and evaluating the upper limb training movements of the subject. The upper limb rehabilitation training system has collected abnormal training parameter values ​​of multiple second subjects who have completed the same upper limb training category as the first subject. Now, the upper limb training monitoring system is to evaluate the training movements of the first subject and decide whether to update its movement evaluation features based on the comparison results of its abnormal training parameter values ​​with those of the second subject.

[0167] For example, the upper limb training monitoring system first accesses its database to retrieve abnormal training parameter values ​​of multiple second subjects with the same upper limb training category as the first subject. These abnormal training parameter values ​​are previously obtained by analyzing the training movements of the second subjects, reflecting the degree of distortion or other abnormal performance of their movements during training.

[0168] Assume that 10 abnormal training parameter values ​​of the second subject are stored in the upper limb training monitoring system database, and these abnormal training parameter values ​​are all collected when the second subject performs the same upper limb lifting training as the first subject.

[0169] Step D120, based on the abnormal training parameter values ​​of the plurality of second subjects, determining the abnormal training parameter value interval corresponding to the upper limb training category.

[0170] The upper limb training monitoring system performs statistical analysis on the abnormal training parameter values ​​of the retrieved multiple second subjects, such as calculating the mean value, standard deviation, etc., to determine a reasonable abnormal training parameter value interval. This abnormal training parameter value interval represents the distribution range of the abnormal training parameter values ​​of most second subjects under the upper limb training category.

[0171] For example, the upper limb training monitoring system calculates that the average value of the abnormal training parameter values ​​of the 10 second subjects is 50, and the standard deviation is 10. Therefore, an abnormal training parameter value range can be determined, such as the average value plus or minus one standard deviation, that is, between 40 and 60.

[0172] Step D130: If the abnormal training parameter value of the first subject is within the abnormal training parameter value interval, the action evaluation feature of the first subject is locked unchanged. The action evaluation feature includes abnormal action resistance.

[0173] The upper limb training monitoring system then compares the abnormal training parameter value of the first subject with the determined abnormal training parameter value interval. If the value of the first subject falls within the interval, it means that its performance is similar to that of most of the second subjects and there is no significant abnormality. Therefore, the upper limb training monitoring system decides to lock the action evaluation characteristics of the first subject unchanged, including abnormal resistance to action, etc. For example, if the abnormal training parameter value of the first subject is 45, then this value is within the interval of 40 to 60, so the upper limb training monitoring system will not update its action evaluation characteristics.

[0174] Step D140: If the abnormal training parameter value of the first subject is greater than the abnormal training parameter value interval, a candidate subject whose abnormal training parameter value matches the abnormal training parameter value of the first subject is selected from the plurality of second subjects.

[0175] If the abnormal training parameter value of the first subject exceeds the determined interval, it means that his performance is different from that of most second subjects, and there may be more significant abnormalities. In this case, the upper limb training monitoring system will select candidate subjects whose abnormal training parameter values ​​are similar to those of the first subject from multiple second subjects for more detailed comparison and analysis. For example, if the abnormal training parameter value of the first subject is 70, which exceeds the interval of 40 to 60, the upper limb training monitoring system will search for candidate subjects whose abnormal training parameter values ​​are also higher (for example, above 65) among the 10 second subjects.

[0176] Step D150, obtaining a relative abnormal training parameter value between the abnormal training parameter value of the candidate subject and the abnormal training parameter value of the first subject, and if the relative abnormal training parameter value is greater than a relative abnormal threshold value, updating the motion evaluation feature of the first subject.

[0177] Finally, the upper limb training monitoring system calculates the relative abnormal training parameter value between the first subject and the selected candidate subject. This relative abnormal training parameter value reflects the similarity or difference between the abnormal performance of the first subject and the candidate subject. If the relative abnormal training parameter value exceeds the preset relative abnormal threshold value, it means that the abnormal performance of the first subject is more significant than that of the candidate subject, and the upper limb training monitoring system will decide to update its action evaluation features to reflect this difference.

[0178] For example, assume that the upper limb training monitoring system selects three candidate subjects, and their abnormal training parameter values ​​are 68, 72, and 75, respectively. The upper limb training monitoring system calculates the relative abnormal training parameter values ​​between the first subject and the three candidate subjects, such as using a difference or ratio method. If the calculated relative value exceeds a preset threshold value (such as a difference of more than 10), the upper limb training monitoring system updates the action evaluation features of the first subject to better reflect its unique abnormal performance.

[0179] Figure 2 An upper limb training monitoring system 100 is provided in an embodiment of the present application, including a processor 1001 and a memory 1003 and a program code stored in the memory 1003. The processor 1001 executes the above program code to implement the steps of the upper limb movement abnormality detection method based on a neural network.

[0180] Figure 2 The upper limb training monitoring system 100 shown includes: a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, such as through a bus 1002. Optionally, the upper limb training monitoring system 100 may also include a transceiver 1004, which can be used for data interaction between the upper limb training monitoring system 100 and other upper limb training monitoring systems 100, such as data transmission and / or data reception. It should be noted that in actual scheduling, the transceiver 1004 is not limited to one, and the structure of the upper limb training monitoring system 100 does not constitute a limitation on the embodiments of the present application.

[0181] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0182] The bus 1002 may include a path to transmit information between the above components. The bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 1002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0183] The memory 1003 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), magnetic disk storage medium, other magnetic storage devices, or any other medium that can be used to carry or store program code and can be read by a computer, without limitation herein.

[0184] The memory 1003 is used to store program codes for executing the embodiments of the present application, and the execution is controlled by the processor 1001. The processor 1001 is used to execute the program codes stored in the memory 1003 to implement the steps shown in the above method embodiments.

[0185] An embodiment of the present application provides a computer-readable storage medium having program code stored thereon. When the program code is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0186] The above is only an optional implementation method for some implementation scenarios of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the scheme of the present application, other similar implementation methods based on the technical ideas of the present application also belong to the protection scope of the embodiments of the present application.

Claims

1. A method for detecting abnormal upper limb movement based on a neural network, characterized in that: Applied to an upper limb training monitoring system, the method comprises: Acquire upper limb motion capture data to be analyzed, wherein the upper limb motion capture data is motion behavior data captured when a first subject in an upper limb training space scene performs an upper limb training task; the upper limb training task is used to guide the motion area of ​​the upper limb motion to be locked in a target defined area during the training process in which the first subject continuously performs the upper limb motion; Perform abnormal state detection on the upper limb motion capture data under multiple abnormal characteristic states, and generate an abnormal state degree value corresponding to each abnormal characteristic state; wherein the abnormal state degree value corresponding to each abnormal characteristic state represents the abnormal movement score of the first subject under the corresponding abnormal characteristic state; Performing weighted calculation on the abnormal state degree value corresponding to each of the abnormal characteristic states to generate an abnormal training parameter value of the first subject; Each of the upper limb motion capture data is used as a training sample, and the abnormal training parameter value of the first subject is used as a training label to train the neural network and generate a target neural network to detect abnormal conditions of the input upper limb motion capture data based on the target neural network.

2. The upper limb movement abnormality detection method based on neural network according to claim 1 is characterized in that: The abnormal characteristic state includes: at least one of: an abnormal characteristic state of motion extension, an abnormal characteristic state of motion stability, and an abnormal characteristic state of motion distortion; the abnormal state detection is performed on the upper limb motion capture data under multiple abnormal characteristic states to generate an abnormal state degree value corresponding to each abnormal characteristic state, including at least one of the following: If the abnormal characteristic state includes an abnormal characteristic state of motion extension, abnormal state detection is performed on the upper limb motion capture data under the abnormal characteristic state of motion extension to generate a first abnormal state degree value corresponding to the abnormal characteristic state of motion extension; the first abnormal state degree value represents the motion abnormality score of the first subject under the abnormal characteristic state of motion extension; If the abnormal characteristic state includes an abnormal characteristic state of movement stability, abnormal state detection is performed on the upper limb motion capture data under the abnormal characteristic state of movement stability to generate a second abnormal state degree value corresponding to the abnormal characteristic state of movement stability; the second abnormal state degree value represents the abnormal movement score of the first subject under the abnormal characteristic state of movement stability; If the abnormal characteristic state includes an abnormal characteristic state of movement distortion, abnormal state detection is performed on the upper limb motion capture data under the abnormal characteristic state of movement distortion, and a third abnormal state degree value corresponding to the abnormal characteristic state of movement distortion is generated; the third abnormal state degree value represents the movement abnormality score of the first subject under the abnormal characteristic state of movement distortion.

3. The upper limb movement abnormality detection method based on neural network according to claim 2 is characterized in that: The upper limb training task includes K groups of training action sequences, and the upper limb motion capture data includes: the action amplitudes corresponding to the K groups of training action sequences; The abnormal state detection is performed on the upper limb motion capture data under the abnormal state of motion extension to generate a first abnormal state degree value corresponding to the abnormal state of motion extension, including: Based on the movement amplitude corresponding to each group of the training movement sequence, the movement extension analysis is performed respectively to generate the movement extension abnormality corresponding to each group of the training movement sequence; the movement extension abnormality is a mapping parameter value of each group of the training movement sequence for resisting the movement abnormality resistance suffered by the first subject when performing the upper limb movement, the movement extension abnormality is positively correlated with the movement abnormality resistance, and the movement extension abnormality quantifies the degree of deviation of the subject from the standard amplitude when performing the training movement sequence, and the movement abnormality resistance refers to various internal or external factors encountered by the subject when performing the upper limb movement, which cause the movement to be performed unsmoothly or deviate from expectations, resulting in additional burden on the subject during the movement execution process, so that the movement amplitude, speed or direction deviates from the standard or expectation; Performing weighted calculation on the generated K action extension abnormalities to generate an action extension abnormality parameter; the action extension abnormality parameter represents the weighted action extension abnormality obtained by weighted calculation; Outputting the action extension abnormality parameter as a first abnormal state degree value corresponding to the action extension abnormality characteristic state; The movement abnormality score of the first subject in the movement extension abnormal characteristic state includes a movement range deviation; and the movement range deviation is positively correlated with the weighted movement extension abnormality.

4. The upper limb movement abnormality detection method based on neural network according to claim 2 is characterized in that: The upper limb training task includes K groups of training action sequences, and the upper limb motion capture data includes the action amplitudes corresponding to the K groups of training action sequences; The abnormal state detection is performed on the upper limb motion capture data under the abnormal characteristic state of motion stability to generate a second abnormal state degree value corresponding to the abnormal characteristic state of motion stability, including: Walk the K groups of training action sequences corresponding to each action amplitude; When walking to the action amplitude corresponding to the x-th group of training action sequences, if x is greater than 1, the action amplitude corresponding to the x-1-th group of training action sequences is obtained; Determine a movement extension variation parameter based on the movement amplitude corresponding to the xth group of training movement sequences and the movement amplitude corresponding to the x-1th group of training movement sequences; the movement extension variation parameter is used to represent the change value between the movement amplitude corresponding to the x-1th group of training movement sequences and the movement amplitude corresponding to the xth group of training movement sequences; After obtaining K-1 action extension change parameters, weighted calculation is performed on the K-1 action extension change parameters to generate action abnormal fluctuation parameters; the action abnormal fluctuation parameters represent the weighted action extension change parameters generated by weighted calculation; Outputting the abnormal motion fluctuation parameter as a second abnormal state degree value corresponding to the abnormal characteristic state of motion stability; Among them, the movement abnormality score of the first subject in the abnormal movement stability characteristic state includes movement abnormality volatility; the movement abnormality volatility is positively correlated with the weighted movement extension change parameter; x∈[1, K].

5. The method for detecting abnormal upper limb movement based on a neural network according to claim 4, characterized in that: The movement range includes a horizontal movement range and a vertical movement range; The step of determining the movement extension change parameter based on the movement amplitude corresponding to the xth group of training movement sequences and the movement amplitude corresponding to the x-1th group of training movement sequences includes: Based on the difference calculation of the horizontal movement amplitude corresponding to the xth group of training movement sequences and the horizontal movement amplitude corresponding to the x-1th group of training movement sequences, a change amount of the horizontal movement amplitude of the xth group of training movement sequences relative to the x-1th group of training movement sequences is generated; Based on the vertical movement amplitude corresponding to the xth group of training action sequences and the vertical movement amplitude corresponding to the x-1th group of training action sequences, a difference calculation is performed to generate a vertical movement amplitude change amount of the xth group of training action sequences relative to the x-1th group of training action sequences; Based on the horizontal movement amplitude change amount and the vertical movement amplitude change amount of the xth group of training movement sequences relative to the x-1th group of training movement sequences, a movement extension change parameter is generated.

6. The method for detecting abnormal upper limb movement based on a neural network according to claim 2, characterized in that: The abnormal state detection is performed on the upper limb motion capture data under the abnormal characteristic state of the motion distortion degree to generate a third abnormal state degree value corresponding to the abnormal characteristic state of the motion distortion degree, including: Deconstructing the upper limb motion capture data to generate horizontal motion behavior data and vertical motion behavior data; Adaptively updating the horizontal motion behavior data using a continuous function model to generate horizontal motion path data, and adaptively updating the vertical motion behavior data according to the continuous function model to generate vertical motion path data; Integrating the horizontal motion path data and the vertical motion path data to generate a motion continuous path; Segmenting the action continuous path based on a plurality of deformation points of the action continuous path to generate a plurality of action continuous sub-paths, each action continuous sub-path corresponding to a variation range; Walking through the variation ranges corresponding to the plurality of action continuous sub-paths, each variation range comprising a first critical point and a second critical point; Obtaining a change amount of the motion direction of each variation range wandered to, wherein the change amount of the motion direction is formed by a direction vector of a first critical point of the variation range and a direction vector of a second critical point of the same variation range; Outputting a range of movement direction changes that is not less than a preset change as a significant deformation interval; Based on the number of the significant deformation intervals and the variation ranges corresponding to the plurality of action continuous sub-paths, a third abnormal state degree value corresponding to the abnormal characteristic state of the action distortion degree is determined.

7. The upper limb movement abnormality detection method based on neural network according to claim 6 is characterized in that: The step of segmenting the action continuous path based on the plurality of deformation points of the action continuous path to generate a plurality of action continuous sub-paths includes: Selecting a target deformation point from a plurality of deformation points of the continuous path of the motion; The action continuous path is segmented based on the target deformation point to generate multiple action continuous sub-paths; wherein, the change ranges corresponding to the respective action continuous sub-paths are arranged in sequence, except for the change ranges corresponding to the first action continuous sub-path and the change ranges corresponding to the last action continuous sub-path, the deformation points corresponding to the critical points of each change range are all the target deformation points.

8. The method for detecting abnormal upper limb movement based on a neural network according to claim 6, characterized in that: The upper limb training task includes K groups of training action sequences, and the third abnormal state degree value corresponding to the abnormal characteristic state of the action distortion degree is determined based on the number of the significant deformation intervals and the change ranges corresponding to the multiple action continuous sub-paths, including: Connecting the first critical point and the second critical point of the variation range corresponding to each action continuous sub-path to generate an action route corresponding to each action continuous path; each action continuous sub-path has multiple action path positions; Determine the deviation between each action path position on each action continuous sub-path and the action route corresponding to each action continuous sub-path; the deviation corresponding to each action path position represents the distortion measurement value of the action continuous path at the corresponding action path position; Determine a distortion vector weight of each motion path position based on a spatial relationship between each motion path position and a corresponding motion route; the distortion vector weight represents an abnormal training parameter value determined due to a motion distortion trajectory of the motion continuous path; Calculation is performed based on the deviation corresponding to each motion path position, the distortion vector weight of the corresponding motion path position, and the number of motion path positions to generate a target deviation; Determining a distortion measure value of the action continuous path based on the training quantity K of the training action sequence and the target deviation; Based on the distortion measure value of the motion continuous path and the number of the significant deformation intervals, a motion distortion parameter is determined, and the motion distortion parameter is output as a third abnormal state degree value corresponding to the motion distortion abnormal characteristic state.

9. The upper limb movement abnormality detection method based on neural network according to claim 3, characterized in that: The method further comprises: Acquiring abnormal training parameter values ​​of a plurality of second subjects who have completed upper limb training in the upper limb training space scene; the upper limb training categories of the plurality of second subjects are the same as the upper limb training category of the first subject; Determining the abnormal training parameter value interval corresponding to the upper limb training category based on the abnormal training parameter values ​​of the plurality of second subjects; If the abnormal training parameter value of the first subject is within the abnormal training parameter value interval, locking the action evaluation feature of the first subject unchanged; wherein the action evaluation feature includes abnormal action resistance; If the abnormal training parameter value of the first subject is greater than the abnormal training parameter value interval, selecting a candidate subject whose abnormal training parameter value matches the abnormal training parameter value of the first subject from the plurality of second subjects; A relative abnormal training parameter value between the abnormal training parameter value of the candidate subject and the abnormal training parameter value of the first subject is obtained, and if the relative abnormal training parameter value is greater than a relative abnormal threshold value, the action evaluation feature of the first subject is updated.

10. An upper limb training monitoring system, characterized in that: The upper limb training monitoring system includes a processor and a machine-readable storage medium, in which machine-executable instructions are stored. The machine-executable instructions are loaded and executed by the processor to implement the upper limb movement abnormality detection method based on neural network as described in any one of claims 1-9.

Images