WO2021052045A1

WO2021052045A1 - Body movement recognition method and apparatus, computer device and storage medium

Info

Publication number: WO2021052045A1
Application number: PCT/CN2020/106654
Authority: WO
Inventors: 田彦秀; 韩久琦
Original assignee: 北京海益同展信息科技有限公司
Priority date: 2019-09-17
Filing date: 2020-08-03
Publication date: 2021-03-25
Also published as: CN110710970B; CN110710970A

Abstract

A body movement recognition method, comprising: (11) establishing a body movement classification model, wherein the classification model is trained on the basis of a training set of surface electromyography signals of sample objects; (12) applying the classification model to the recognition of the body movements of a target object; (13) acquiring the state of recognizing the target object during the recognition of the body movement of the target object; (14) if the state of recognizing the target object satisfies update conditions of the classification model, updating parameters of the classification model to correct the deviation of the classification model in recognizing the body movement of the target object.

Description

Recognition method, device, computer equipment and storage medium of body movement

Cross-references to related applications

This application is based on a Chinese patent application with an application number of 201910876338.6 and an application date of September 17, 2019, and claims the priority of the Chinese patent application. The entire content of the Chinese patent application is hereby incorporated into this application by reference.

Technical field

The present disclosure relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment, and storage medium for recognizing body movements.

Background technique

Surface Electromyography (SEMG) is a bioelectric signal related to neuromuscular activity. When the movement command is transmitted to the relevant muscle fibers through the central nervous system, it will cause the potential changes on the muscle fibers and contraction of the muscle fibers. This potential change is superimposed on the surface of the skin in time and space to form a surface EMG signal, which can be collected by the surface EMG electrode. The surface EMG signal contains information such as the pattern of muscle contraction and contraction strength. Different body movements correspond to different EMG signals. By analyzing the surface EMG signal, the specific action mode corresponding to the signal can be identified. Therefore, the surface EMG signal is widely used in medical diagnosis, sports rehabilitation and other fields. Especially in the control of human-like EMG prosthetic hands, the surface EMG signal as the control source drives the prosthetic hand to make various grasping handshake gestures, which has received extensive research and attention.

The surface EMG signal is very specific. In the gesture pattern recognition, the first aspect is that the performance of the gesture recognition system will be affected by the individual differences in surface EMG signals. Such individual differences are often caused by the electrode displacements of different collected subjects or the muscle shapes between different subjects. Differences in size, strength, fatigue, and skin impedance are caused. This individual difference often leads to different distributions of training data and test data from different subjects, making it difficult to effectively expand the classifier model learned from the current individual And apply to other individuals. On the other hand, the surface EMG signal is time-varying. With the passage of time, the subject is prone to fatigue due to long-term dynamic changes in muscle contraction force. The shape and size of the muscles, the magnitude of the force, the degree of fatigue, and the skin impedance of the same subject Changes will occur, affecting the recognition effect of the classifier model.

Summary of the invention

In view of this, the embodiments of the present disclosure provide a method, device, computer equipment, and storage medium for recognizing body movements.

The technical solutions of the embodiments of the present disclosure are implemented as follows:

The embodiments of the present disclosure provide a method for recognizing body movements, including:

Establishing a classification model of body movements, wherein the classification model is trained based on the training set of the surface EMG signal of the sample object;

Applying the classification model to the recognition of the body movements of the target object;

In the process of recognizing the limb movements of the target object, acquiring the state of recognizing the target object;

When the state of identifying the target object satisfies the update condition of the classification model, the parameters of the classification model are updated to repair the deviation of the limb movement of the classification model in identifying the target object.

Wherein, the establishment of a classification model of body movements includes:

Perform preprocessing and feature vector extraction on the surface EMG signal of the sample object to obtain a feature vector sample set;

Randomly dividing the feature vector sample set into a training set and a test set;

Training a classification model of body movements based on the training set and the labels of the sample objects in the training set, and using the test set to test the recognition accuracy of the classification model;

When the recognition accuracy of the classification model is greater than the first set threshold, the trained classification model is obtained.

Wherein, applying the classification model to the recognition of the body movements of the target object includes:

Acquiring a feature vector of the surface EMG signal of the target object to be recognized;

The dimensionality reduction processing and classification recognition processing are performed on the feature vector through the classification model to obtain a classification and recognition result of the body movement of the target object.

Wherein, the acquiring and identifying the status of the target object includes:

Acquiring the status of whether the target object to be recognized by the classification model is transformed;

And/or, obtain the duration of the classification model identifying the same target object;

And/or, obtaining the number of samples of the surface EMG signal of the same target object identified by the classification model.

Wherein, when the state of identifying the target object satisfies the update condition of the classification model, updating the parameters of the classification model includes:

When at least one of the following update conditions is met, the parameters of the classification model are updated:

The target object to be recognized by the classification model is changed;

The duration of the classification model identifying the same target object is greater than the duration threshold;

The classification model identifies the number of samples of the surface EMG signal of the same target object, which is greater than the sample number threshold.

Wherein, said updating the parameters of the classification model includes:

When the target object to be recognized by the classification model is changed, update the parameters of the classification model through a training method of supervised learning;

When the duration for the classification model to identify the same target object is greater than the duration threshold, or when the number of samples of the surface EMG signal of the same target to be identified by the classification model is greater than the second set threshold, unsupervised learning The training method updates the parameters of the classification model.

Wherein, said updating the parameters of the classification model includes:

When the classification model is a linear discriminant analysis (Linear Discriminant Analysis, LDA) classifier, training the LDA classifier by using labeled surface EMG signal samples of the target object;

In the training process of the classification model, the intra-class divergence matrix and the inter-class divergence matrix of the LDA classifier are adjusted until the following conditions are met: the same type after dimensionality reduction and classification processing is performed by the LDA classifier The distance between the feature vectors of the target object sample meets a first distance threshold, and the distance between the feature vectors of the target object samples of different types meets a second distance threshold;

The adjusted intra-class divergence matrix and the inter-class divergence matrix satisfying the condition are used as parameters of the updated LDA classifier.

The embodiment of the present disclosure provides a device for recognizing body movements. The device includes a building module and a processing module, wherein

The construction module is configured to establish a classification model of body movements, wherein the classification model is trained based on the training set of the surface EMG signal of the sample object;

The processing module is configured to apply the classification model to the recognition of the limb movements of the target object; in the recognition process of the limb movements of the target object, obtain the state of recognizing the target object; when recognizing the target object When the status of satisfies the update condition of the classification model, the parameters of the classification model are updated to repair the deviation of the limb movement of the classification model in identifying the target object.

The embodiment of the present disclosure provides a computer device, including:

Memory, configured to store executable instructions;

The processor is configured to implement the method for recognizing a limb movement in the embodiment of the present disclosure when the executable instruction is executed. Among them, the executable instructions can be installation packages, programs, codes, plug-ins, and libraries (dynamic/static libraries).

The embodiment of the present disclosure provides a limb movement recognition device, including: a processor and a memory configured to store a computer program that can run on the processor;

Wherein, the processor is configured to implement the limb movement recognition method of the embodiment of the present disclosure when running the computer program.

The embodiment of the present disclosure provides a storage medium that stores executable instructions, and when the executable instructions are executed, the method for recognizing body movements in the embodiments of the present disclosure is realized.

Applying the above-mentioned embodiments of the present disclosure has the following beneficial technical effects:

In the process of recognizing the limb movements of the target object, the state of recognizing the target object can be obtained. When the state of recognizing the target object satisfies the update condition of the classification model, the parameters of the classification model are updated to repair all the objects. The classification model recognizes the deviation of the body movement of the target object. In this way, the classification model can be updated in time based on the state of the target object, repairs the deviation of the classification model's recognition of the body movement of the target object, and reduces the impact of the change in the state of the target object on the classification model. The recognition accuracy of the classification model is ensured.

Description of the drawings

FIG. 1 is a schematic flowchart of a method for recognizing body movements according to an embodiment of the disclosure;

2 is a schematic flow chart of another method for recognizing body movements provided by an embodiment of the present disclosure;

FIG. 3 is a schematic flowchart of another method for recognizing body movements according to an embodiment of the present disclosure;

FIG. 4 is a schematic flowchart of another method for recognizing body movements according to an embodiment of the disclosure;

FIG. 5 is a schematic diagram of a model applied by a method for recognizing body movements according to an embodiment of the disclosure;

FIG. 6 is a schematic flowchart of another method for recognizing body movements according to an embodiment of the present disclosure;

FIG. 7 is a schematic diagram of a device for recognizing body movements according to an embodiment of the disclosure;

FIG. 8 is a schematic structural diagram of a computer device provided by an embodiment of the disclosure.

detailed description

Hereinafter, the present disclosure will be further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the embodiments provided here are only used to explain the present disclosure, but not used to limit the present disclosure. In addition, the embodiments provided below are part of the embodiments for implementing the present disclosure, rather than providing all the embodiments for implementing the present disclosure. In the case of no conflict, the technical solutions described in the embodiments of the present disclosure can be combined in any manner. Implement.

It should be noted that in the embodiments of the present disclosure, the terms "including", "including" or any other variations thereof are intended to cover non-exclusive inclusion, so that a method or device including a series of elements not only includes the explicitly stated Elements, and also include other elements not explicitly listed, or elements inherent to the implementation of the method or device. Without more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other related elements in the method or device that includes the element (such as steps in the method or units in the device). For example, the unit may be a part of a circuit, a part of a processor, a part of a program or software, etc.).

For example, the method for recognizing a body movement provided by the embodiment of the present disclosure includes a series of steps, but the method for recognizing a body movement provided by the embodiment of the present disclosure is not limited to the recorded steps. Similarly, the embodiment of the present disclosure provides The recognition device based on a kind of limb movement includes a series of modules, but the device provided in the embodiments of the present disclosure is not limited to include the clearly recorded modules, and may also include settings that need to be set for obtaining relevant information or processing based on information Module.

Before describing the embodiments of the present disclosure in detail, the terms and terms involved in the embodiments of the present disclosure are described. The terms and terms involved in the embodiments of the present disclosure are applicable to the following explanations.

1) Classification model refers to an algorithm model or function used for classification, which can map data in a database or sample set to a category in a given category. For example, the present disclosure is used for a classification model for classifying body movements.

2) Limb movement refers to the corresponding posture when the limb moves. For example, three-finger pinching, side pinching, hooking, powerful grasping, cylindrical grasping, center grasping, palm extension, and wrist bending postures related to human arms.

In order to facilitate the understanding of the embodiments of the present disclosure, the following uses the recognition of limb movements of the prosthesis in the field of bionics as an example to illustrate the application scenarios of the present disclosure: when the user of the prosthesis controls the movement of the prosthesis, it is necessary to control the prosthesis to make limb movements by issuing control instructions. . Therefore, the EMG signal of the human limb can be collected by setting sensors on the human body, because when the user of the prosthesis intends to make a certain action, there will be feedback on the EMG signal. After collecting the EMG signal, pass Analyzing the EMG signal can determine the specific posture action type that the prosthesis user intends to make, so as to control the prosthesis to make corresponding actions. In this way, the user of the prosthesis does not need to control the movement of the prosthesis through the cumbersome way of mechanical switch or manual selection, which improves the experience of the prosthesis user. The embodiments of the present disclosure can be applied to the control of the movement of the prosthesis.

Next, the embodiments of the present disclosure will be described in detail. In some embodiments, the method for recognizing body movements provided by the embodiments of the present disclosure may be implemented by a terminal or a server alone, or implemented by the server and the terminal in cooperation. Please refer to FIG. 1, the present disclosure An embodiment provides a method for recognizing body movements, including:

Step 11: Establish a classification model of body movements, wherein the classification model is trained based on the training set of the surface EMG signal of the sample object;

Here, the execution subject of the limb movement may refer to a target object capable of generating electromyographic signals and capable of performing limb movement, such as humans, animals, and the like. Taking a person as an example, when the limb movement is a movement of a human arm, the electromyographic signal may be collected from the forearm muscles by using two modular bipolar differential electrodes. Here, the sample object may be a sample of the electromyographic signal collected when the execution subject makes a different posture action, for example, a sample of the electromyographic signal collected when the arm makes a movement such as bending, inversion, and eversion. . Here, the sample object can be collected in different time periods, the collection of the sample object can correspond to different execution subjects, and the collection method can be collection according to a set collection frequency. Here, the classification model may be a classification model that has been trained so that the classification accuracy rate reaches an expected value, for example, an LDA classifier. Here, the classifier may be a combination or cascade of multiple sub-classifiers.

Step 12: Apply the classification model to the recognition of body movements of the target object;

Here, it may be that the feature vector of the electromyographic signal generated by the target object is input into the classification model, and the type of the limb movement of the target object is determined through dimensionality reduction processing and classification processing of the classification model. For example, when the EMG signal sample of the current arm is collected, the first feature vector corresponding to the EMG signal is input into the classification model. If the output result of the classification model corresponds to the first category, the first The body action posture corresponding to the category is bending, then it can be determined that the current arm is making a bent body action.

Step 13, in the process of recognizing the limb movements of the target object, acquiring the status of recognizing the target object;

Here, the state of the target object may be the state of whether the target object to be identified by the classification model has been changed. Due to individual differences between the target objects, such individual differences are often caused by the electrode displacements of different collected subjects, or The differences in muscle shape and size, exertion strength, fatigue, and skin impedance between different subjects are caused. Such individual differences often result in different distributions of sample data from different subjects. The state of the target object may also be the state of the duration of the classification model identifying the same target object. As time goes by, the long-term dynamic change of muscle contraction force makes the subject easily fatigued, which may also result in different sample data. The distribution of the target object; the state of the target object can also be the state of the number of samples of the surface EMG signal of the same target object that the classification model recognizes, because the more body movements the target object makes, the more frequent movements will change the contraction of the muscles The subject is prone to fatigue, which will also cause the sample data to have different distributions.

Step 14: When the state of identifying the target object meets the update condition of the classification model, update the parameters of the classification model to repair the deviation of the classification model in identifying the body movement of the target object.

Here, the update condition may be set in advance, for example, the target object to be recognized by the classification model is changed; or, the duration for the classification model to identify the same target object is greater than the duration threshold; or, it is set as The classification model identifies the number of samples of the surface EMG signal of the same target object, which is greater than the sample number threshold.

Here, the parameters of the classification model may be configuration parameters that affect the classification accuracy of the classification model. Taking the LDA classifier as an example, the parameters of the classification model may be values corresponding to the intra-class divergence matrix and/or the inter-class divergence matrix. Here, the deviation of the limb movement may refer to the deviation between the classified body movement and the actual corresponding limb movement, which can be quantified and embodied in the way of accuracy.

In the embodiment of the present disclosure, in the process of recognizing the body movement of the target object, the state of recognizing the target object can be obtained, and when the state of recognizing the target object meets the update condition of the classification model, the state of recognizing the target object is updated. The parameters of the classification model are used to repair the deviation of the body movements of the target object identified by the classification model. In this way, the classification model can be updated in time based on the state of the target object, repair the deviation of the classification model to recognize the body movement of the target object, reduce the impact of the state change of the target object on the classification model, and ensure The recognition accuracy rate of the classification model is calculated.

Please refer to FIG. 2. In some embodiments, in the step 11, the establishment of a classification model of a body movement includes:

Step 21: Perform preprocessing and feature vector extraction on the surface EMG signal of the sample object to obtain a feature vector sample set;

Here, taking arm movements as an example, before collecting the sample object, first clean the arm skin, scrape off the hair on the epidermis of the selected muscle, wash it with water and wipe the skin with a cotton swab dipped in medical alcohol; then, collect the subject EMG signal on the surface of the forearm, the subjects made cylindrical grasp, hook grasp, side pinch, pointing, spherical grasp, three-finger pinch, precise pinch, relaxed posture, varus, valgus, grasping, etc. Gestures, using two EMG electrodes to obtain the EMG sample objects of each gesture and store them.

The preprocessing of the surface EMG signal of the sample object may be filtering or denoising the surface EMG signal; the preprocessing and feature vector extraction of the surface EMG signal can be implemented by a neural network algorithm. , Such as BP neural network algorithm and so on. Here, the feature vector sample set may include a set of multiple EMG signal samples corresponding to the feature vector or the feature vector corresponding matrix.

Step 22: randomly divide the feature vector sample set into a training set and a test set;

Here, the random division may be to randomly select part of the feature vectors from the feature vector samples as the test set, and the remaining feature vectors as the training set. Here, the number of samples in the test set can be configured to any number according to requirements.

Step 23: Train a classification model of body movements based on the training set and the labels of the sample objects in the training set, and use the test set to test the recognition accuracy of the classification model;

Here, the label of the sample object may be used to characterize cylindrical grasping, hooking, side pinching, pointing, spherical grasping, three-finger grasping, precise pinching, relaxed posture, inversion, eversion, and grasping. Mark label for holding and waiting posture. Each sample object corresponds to one of the marking labels. Here, the recognition accuracy of the classification model may refer to the accuracy of comparing the classification result with the corresponding labeled label by inputting any sample in the test set into the training classification model.

Step 24: Obtain the trained classification model when the recognition accuracy of the classification model is greater than the first set threshold.

Here, the first setting threshold may be set in advance, and the size of the first setting threshold may be flexibly set according to actual needs.

Referring to FIG. 3, in some embodiments, in the step 12, applying the classification model to the recognition of the body movements of the target object includes:

Step 31: Obtain the feature vector of the surface EMG signal of the target object to be recognized;

Here, the feature of the surface EMG signal of the target object may be a time domain feature, for example, an integrated EMG value feature, where the integrated EMG value is a value obtained by integrating the surface EMG signal of the target object; the number of zero crossing points Features, the number of zero crossing points refers to the number of times that the surface EMG signal of the target object passes through the zero point. Since the EMG signal originates from the electrical pulse sent by the central nerve, the strength of the EMG signal is also related to the frequency of the electrical pulse, so the number of zero crossing points It can be used as a feature of EMG signal. The feature may be a vector corresponding to the above one feature or multiple features.

Step 32: Perform dimensionality reduction processing and classification recognition processing on the feature vector through the classification model to obtain a classification and recognition result of the body movement of the target object.

Here, the classification model can project the feature points corresponding to the high-dimensional feature vector to a low-dimensional space, so that the features of the EMG signal are projected from the high-dimensional feature space to the low-dimensional space, and belong to the feature points of the same body movement. It is more clustered, and the feature points belonging to different body movements are more separated, so as to obtain classification and recognition results.

In some embodiments, in the step 13, the acquiring and recognizing the status of the target object includes:

Here, the change of the target object can be a limb change, for example, from the subject’s left arm to the right arm; it can also be a subject’s change, for example, from the subject A’s arm to the subject. Try B's arm. Here, the duration of identifying the same target object may be the duration from the input of the feature vector of the first sample to the classification model and the duration to the current time, such as 0.5 hour, 1 hour, or 2 hours. Here, the number of samples of the surface EMG signal for identifying the same target object may be N numbers from the input of the first sample feature vector to the classification model to the current input of the Nth sample number. Here, to obtain the status of the target object, on the one hand, when the target object changes individuals, the classification model can be updated in real time to adapt to the new individual; on the other hand, it can dynamically change muscles based on long-term changes over time. For contractile force, when the subject is prone to fatigue, update the classification model in time.

In some embodiments, in the step 14, when the state of identifying the target object satisfies the update condition of the classification model, updating the parameters of the classification model includes:

The target object to be recognized by the classification model is changed;

Here, the duration threshold and the sample number threshold can be flexibly set according to the recognition accuracy requirements of classification recognition. For example, when the recognition accuracy requirements are high, set a smaller duration threshold and a smaller number threshold; when the recognition accuracy requirements are low, set a larger duration threshold and a larger number threshold;

In some embodiments, in the step 14, the updating the parameters of the classification model includes:

Here, the target object is changed, and the changed target object is clear. This situation is applicable to the situation where there are labeled data samples. By using the labeled calibration data of the changed target object, the classification model parameters are determined by the supervised learning method. Update.

Here, the sample feature changes caused by the passage of time are uncertain, and it is inconvenient to label the sample data, and it is suitable for updating the classification model parameters through an unsupervised learning method. Here, the duration threshold may be statistically obtained based on the big data of the tested individual, such as 0.5 hour, 1 hour, and so on. The duration threshold can be flexibly set.

In some embodiments, referring to FIG. 4, in the step 14, the updating the parameters of the classification model includes:

Step 41: When the classification model is an LDA classifier, train the LDA classifier by using labeled surface EMG signal samples of the target object;

Here, the labeled surface EMG signal sample of the target object may be the surface EMG signal sample recognized in the classification model recognition process.

Step 42: During the training process of the classification model, adjust the intra-class divergence matrix and the inter-class divergence matrix of the LDA classifier until the following conditions are met: dimensionality reduction processing and classification processing are performed by the LDA classifier The distance between the feature vectors of the target object samples of the same type later satisfies the first distance threshold, and the distance between the feature vectors of the target object samples of different types satisfies the second distance threshold;

Here, adjusting the intra-class divergence matrix and the inter-class divergence matrix of the LDA classifier may be adjusted according to the gradient. The first distance threshold and the second distance threshold can be flexibly set according to identification requirements. Here, the same type may mean that the target object sample corresponds to the same body movement, and the different type may mean that the target object sample corresponds to different body movements.

Step 43: Use the adjusted intra-class divergence matrix and the class divergence matrix that satisfy the condition as parameters of the updated LDA classifier.

Here, after the parameters of the LDA classifier are updated, the target object is identified according to the updated parameters of the LDA classifier.

In order to make it easier to understand the implementation process of the method for recognizing body movements provided by the embodiments of the present disclosure, the following is described by one embodiment:

Example 1: Please refer to FIG. 5. In the example, body motion recognition corresponds to gesture recognition. The gesture recognition is divided into a training phase and a recognition phase. The recognition model 52 includes an LDA classification recognition model 51. The method for recognizing body movements in the embodiments of the present disclosure can be applied to the LDA classification recognition model in FIG. 5.

Please refer to Figure 6, the method for recognizing body movements includes the following steps:

Step 61: Perform preprocessing and feature vector extraction on the surface EMG signal of the sample object to obtain a feature vector sample set;

Step 62: Randomly divide the feature vector sample set into a training set and a test set;

Step 63: Train an LDA classification model of body movements based on the training set and the labels of the sample objects in the training set, and use the test set to test the recognition accuracy of the LDA classification model;

Step 64: When the recognition accuracy of the LDA classification model is greater than the first set threshold, obtain the LDA classification model after training;

Step 65: Obtain the feature vector of the surface EMG signal of the target object to be recognized;

Step 66: Perform dimensionality reduction processing and classification and recognition processing on the feature vector through the LDA classification model to obtain a classification and recognition result of the body movement of the target object. Wherein, when the target object to be recognized by the LDA classification model is changed, the parameters of the LDA classification model are updated through a training method of supervised learning; when the duration of the LDA classification model identifying the same target object is greater than the duration threshold , Or when the number of samples of the surface EMG signal of the same target to be recognized by the LDA classification model is greater than the second set threshold, the parameters of the LDA classification model are updated through an unsupervised learning training method. The updating the parameters of the LDA classification model includes: when the LDA classification model is a linear discriminant analysis LDA classifier, training the LDA classifier through the surface EMG signal samples of the target object with labels In the training process of the LDA classification model, adjust the intra-class divergence matrix and the inter-class divergence matrix of the LDA classifier until the following conditions are met: after the dimensionality reduction processing and classification processing are performed by the LDA classifier The distance between the feature vectors of the target object samples of the same type meets the first distance threshold, and the distance between the feature vectors of the target object samples of different types meets the second distance threshold; the adjusted ones that meet the conditions The intra-class divergence matrix and the class divergence matrix are used as parameters of the updated LDA classifier.

In this embodiment, in the process of recognizing the limb movements of the target object, the state of recognizing the target object can be obtained, and when the state of recognizing the target object satisfies the update condition of the LDA classification model, the state of recognizing the target object is updated. The intra-class divergence matrix and the class divergence matrix of the classification model are used to repair the deviation of the body movement of the target object recognized by the classification model. In this way, the LDA classification model can be updated in time based on the status of the target object, repairs the deviation of the LDA classification model to recognize the body movement of the target object, and reduces the impact of the status change of the target object on the classification model. , To ensure the recognition accuracy of the LDA classification model.

The embodiments of the present disclosure also provide a device for recognizing body movements. Please refer to FIG. 7. The device includes a building module and a processing module, where:

The construction module 71 is configured to establish a classification model of body movements, wherein the classification model is trained based on the training set of the surface EMG signal of the sample object;

The processing module 72 is configured to apply the classification model to the recognition of the limb movements of the target object; in the recognition process of the limb movements of the target object, obtain the state of recognizing the target object; when recognizing the target When the state of the object satisfies the update condition of the classification model, the parameters of the classification model are updated to repair the deviation of the limb movement of the classification model in identifying the target object.

The embodiment of the present disclosure also provides a computer device. Please refer to FIG. 8, which is a schematic diagram of the structure of the computer device provided by the embodiment of the present disclosure, including:

The memory 82 is configured to store executable instructions;

When the processor 81 is configured to execute the executable instructions stored in the memory 82, when the processor 81 executes the computer program, it includes the following steps: establishing a classification model of body movements, wherein the classification model is based on samples The training set of the surface EMG signal of the object is trained; the classification model is applied to the recognition of the limb movement of the target object; in the recognition process of the limb movement of the target object, the state of recognizing the target object is obtained; when When the state of identifying the target object satisfies the update condition of the classification model, the parameters of the classification model are updated to repair the deviation of the limb movement of the classification model in identifying the target object.

Here, when the processor 81 executes the computer program, it is also used to: perform preprocessing and feature vector extraction on the surface EMG signal of the sample object to obtain a feature vector sample set; randomly divide the feature vector sample set into Training set and test set; training a classification model of body movements based on the training set and the labels of the sample objects in the training set, and using the test set to test the recognition accuracy of the classification model; when the classification model is When the recognition accuracy is greater than the first set threshold, the trained classification model is obtained.

Here, when the processor 81 executes the computer program, it is also used to achieve: obtain the feature vector of the surface EMG signal of the target object to be recognized; perform dimensionality reduction processing and classification on the feature vector through the classification model The recognition processing obtains the classification and recognition result of the limb movement of the target object.

Here, when the processor 81 executes the computer program, it is also used to: obtain the status of whether the target object to be recognized by the classification model has changed; and/or obtain the duration of the classification model identifying the same target object And/or, obtaining the number of samples of the surface EMG signal of the same target object identified by the classification model.

Here, when the processor 81 executes the computer program, it is also used to implement: when at least one of the following update conditions is met, the parameters of the classification model are updated: the target object to be identified by the classification model is changed; the classification The duration of the model identifying the same target object is greater than the duration threshold; the classification model identifying the number of samples of the surface EMG signal of the same target object is greater than the sample number threshold.

Here, when the processor 81 executes the computer program, it is also used to implement: when the target object to be recognized by the classification model is changed, the parameters of the classification model are updated through a training method of supervised learning; When the duration of the classification model identifying the same target object is greater than the duration threshold, or when the number of samples of the surface EMG signal of the same target to be identified by the classification model is greater than the second set threshold, the training method of unsupervised learning is adopted Update the parameters of the classification model.

Here, when the processor 81 executes the computer program, it is also used to implement: when the classification model is a linear discriminant analysis LDA classifier, train the surface EMG signal samples of the target object with labels. The LDA classifier; in the training process of the classification model, adjust the intra-class divergence matrix and the inter-class divergence matrix of the LDA classifier until the following conditions are met: dimensionality reduction processing and After classification, the distance between the feature vectors of the target object samples of the same type meets the first distance threshold, and the distance between the feature vectors of the target object samples of different classes meets the second distance threshold; The intra-class divergence matrix and the class divergence matrix of the conditions are used as parameters of the updated LDA classifier.

The embodiment of the present disclosure also provides a limb movement recognition device, including: a processor and a memory configured to store a computer program that can run on the processor;

Wherein, the processor is configured to implement the limb movement recognition method provided in the embodiment of the present disclosure when running the computer program.

The embodiment of the present disclosure also provides a storage medium that stores executable instructions, and when the executable instructions are executed, they are used to execute the method for recognizing the body movements provided by the embodiments of the present disclosure.

The above are only specific implementations of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present disclosure. It should be covered within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Industrial applicability

In the embodiments of the present disclosure, a classification model of limb movements is established, wherein the classification model is trained based on the training set of the surface EMG signal of the sample object; the classification model is applied to the recognition of the limb movements of the target object; In the process of recognizing the limb movements of the target object, the state of recognizing the target object is obtained; when the state of recognizing the target object satisfies the update condition of the classification model, the parameters of the classification model are updated to repair the classification The model recognizes the deviation of the limb movement of the target object; in this way, the classification model can be updated in time based on the state of the target object, repairing the deviation of the classification model identifying the limb movement of the target object, and reducing the deviation caused by the target object The impact of the state change on the classification model ensures the recognition accuracy of the classification model.

Claims

A method for recognizing body movements, including:

Establishing a classification model of body movements, wherein the classification model is trained based on the training set of the surface EMG signal of the sample object;

Applying the classification model to the recognition of the body movements of the target object;

In the process of recognizing the limb movements of the target object, acquiring the state of recognizing the target object;

When the state of identifying the target object satisfies the update condition of the classification model, the parameters of the classification model are updated to repair the deviation of the limb movement of the classification model in identifying the target object.
The method according to claim 1, wherein said establishing a classification model of body movements comprises:

Perform preprocessing and feature vector extraction on the surface EMG signal of the sample object to obtain a feature vector sample set;

Randomly dividing the feature vector sample set into a training set and a test set;

Training a classification model of body movements based on the training set and the labels of the sample objects in the training set, and using the test set to test the recognition accuracy of the classification model;

When the recognition accuracy of the classification model is greater than the first set threshold, the trained classification model is obtained.
The method according to claim 1, wherein the applying the classification model to the recognition of body movements of a target object comprises:

Acquiring a feature vector of the surface EMG signal of the target object to be recognized;

The dimensionality reduction processing and classification recognition processing are performed on the feature vector through the classification model to obtain a classification and recognition result of the body movement of the target object.
The method according to claim 1, wherein said acquiring and identifying the status of said target object comprises:

Acquiring the status of whether the target object to be recognized by the classification model is transformed;

And/or, obtain the duration of the classification model identifying the same target object;

And/or, obtaining the number of samples of the surface EMG signal of the same target object identified by the classification model.
The method according to claim 1, wherein the updating the parameters of the classification model when the state of identifying the target object satisfies the update condition of the classification model comprises:

When at least one of the following update conditions is met, the parameters of the classification model are updated:

The target object to be recognized by the classification model is changed;

The duration of the classification model identifying the same target object is greater than the duration threshold;

The classification model identifies the number of samples of the surface EMG signal of the same target object, which is greater than the sample number threshold.
The method according to claim 1, wherein the updating the parameters of the classification model comprises:

When the target object to be recognized by the classification model is changed, update the parameters of the classification model through a training method of supervised learning;

When the duration for the classification model to identify the same target object is greater than the duration threshold, or when the number of samples of the surface EMG signal of the same target to be identified by the classification model is greater than the second set threshold, unsupervised learning The training method updates the parameters of the classification model.
The method according to any one of claims 1 to 6, wherein the updating the parameters of the classification model comprises:

When the classification model is a linear discriminant analysis LDA classifier, training the LDA classifier by using labeled surface EMG signal samples of the target object;

In the training process of the classification model, the intra-class divergence matrix and the inter-class divergence matrix of the LDA classifier are adjusted until the following conditions are met: the same type after dimensionality reduction and classification processing is performed by the LDA classifier The distance between the feature vectors of the target object sample meets a first distance threshold, and the distance between the feature vectors of the target object samples of different types meets a second distance threshold;

The adjusted intra-class divergence matrix and the inter-class divergence matrix satisfying the condition are used as parameters of the updated LDA classifier.
A device for recognizing body movements, said device comprising a building module and a processing module, wherein,

The construction module is configured to establish a classification model of body movements, wherein the classification model is trained based on the training set of the surface EMG signal of the sample object;

The processing module is configured to apply the classification model to the recognition of the limb movements of the target object; in the recognition process of the limb movements of the target object, obtain the state of recognizing the target object; when recognizing the target object When the status of satisfies the update condition of the classification model, the parameters of the classification model are updated to repair the deviation of the limb movement of the classification model in identifying the target object.
The device according to claim 8, wherein:

The construction module is further configured to perform preprocessing and feature vector extraction on the surface EMG signal of the sample object to obtain a feature vector sample set;

Randomly dividing the feature vector sample set into a training set and a test set;

Training a classification model of body movements based on the training set and the labels of the sample objects in the training set, and using the test set to test the recognition accuracy of the classification model;

When the recognition accuracy of the classification model is greater than the first set threshold, the trained classification model is obtained.
The device according to claim 8, wherein:

The processing module is further configured to obtain the feature vector of the surface EMG signal of the target object to be recognized;

The dimensionality reduction processing and classification recognition processing are performed on the feature vector through the classification model to obtain a classification and recognition result of the body movement of the target object.
The device according to claim 8, wherein:

The processing module is further configured to obtain the status of whether the target object to be recognized by the classification model is transformed;

And/or, obtain the duration of the classification model identifying the same target object;

And/or, obtaining the number of samples of the surface EMG signal of the same target object identified by the classification model.
The device according to claim 8, wherein:

The processing module is further configured to update the parameters of the classification model when at least one of the following update conditions is met:

The target object to be recognized by the classification model is changed;

The duration of the classification model identifying the same target object is greater than the duration threshold;

The classification model identifies the number of samples of the surface EMG signal of the same target object, which is greater than the sample number threshold.
The device according to claim 8, wherein:

The processing module is further configured to update the parameters of the classification model through a training method of supervised learning when the target object to be recognized by the classification model is changed;

When the duration for the classification model to identify the same target object is greater than the duration threshold, or when the number of samples of the surface EMG signal of the same target to be identified by the classification model is greater than the second set threshold, unsupervised learning The training method updates the parameters of the classification model.
The device according to any one of claims 8 to 13, wherein:

The processing module is further configured to train the LDA classifier through labeled surface EMG signal samples of the target object when the classification model is a linear discriminant analysis LDA classifier;

In the training process of the classification model, the intra-class divergence matrix and the inter-class divergence matrix of the LDA classifier are adjusted until the following conditions are met: the same type after dimensionality reduction and classification processing is performed by the LDA classifier The distance between the feature vectors of the target object sample meets a first distance threshold, and the distance between the feature vectors of the target object samples of different types meets a second distance threshold;

The adjusted intra-class divergence matrix and the inter-class divergence matrix satisfying the condition are used as parameters of the updated LDA classifier.
A computer device including:

Memory, configured to store executable instructions;

The processor is configured to implement the method for body movement according to any one of claims 1 to 7 when executing the executable instructions stored in the memory.
An apparatus for recognizing body movements, including: a processor and a memory configured to store a computer program that can run on the processor;

Wherein, the processor is configured to implement the method for recognizing body movements according to any one of claims 1 to 7 when running the computer program.
A computer storage medium in which a computer program is stored, and when the computer program is executed by a processor, the method for recognizing a limb movement according to any one of claims 1 to 7 is realized.