Gesture recognition method, terminal device, storage medium and computer program product

Abstract

The invention discloses a gesture recognition method, terminal equipment, a storage medium and a computer program product, and relates to the technical field of signal processing, and the method comprises the steps: obtaining a pre-trained basic gesture recognition model; obtaining user training data of the target user, and adjusting a preset offset calibration module based on the user training data to obtain an individual calibration module; a target gesture recognition model is constructed based on the individual calibration module and the basic gesture recognition model, and when gesture recognition is conducted on the to-be-recognized electromyographic signals of the target user according to the target gesture recognition model, feature vectors corresponding to the to-be-recognized electromyographic signals are adjusted according to the individual calibration module. The individual calibration module is introduced to capture and adapt to the unique features of the target user, so that the target gesture recognition model can better capture and adapt to the unique features of the target user, recognition errors caused by individual differences are reduced, and the accuracy of gesture recognition is improved.

Description

Technical Field

[0001] The present application relates to the field of signal processing technology, and in particular to a gesture recognition method, a terminal device, a storage medium, and a computer program product. Background Art

[0002] Electromyography (EMG) is an electrophysiological signal generated by muscle tissue during contraction. EMG signals can express muscle movements. By capturing and analyzing EMG signals, accurate recognition and classification of gestures can be achieved. At present, gesture recognition algorithms based on EMG signals need to first obtain a general gesture recognition model through data training of a large number of users, and then perform gesture recognition through this gesture recognition model. However, due to differences in muscle structure and skin characteristics of different individuals, the EMG signals generated by the same gesture in different individuals are also different, which makes it difficult for traditional recognition algorithms to adapt to the differences of each individual in practical applications, affecting the accuracy of gesture recognition.

[0003] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0004] The main purpose of this application is to provide a gesture recognition method, terminal device, storage medium and computer program product, aiming to solve the technical problem that gesture recognition based on electromyographic signals is difficult to adapt to individual differences, resulting in the accuracy of gesture recognition being affected.

[0005] To achieve the above objectives, the present application proposes a gesture recognition method, the method comprising:

[0006] Obtain a pre-trained basic gesture recognition model, wherein the basic gesture recognition model is trained using historical electromyographic signals of multiple users as input data and gestures corresponding to the historical electromyographic signals as labels;

[0007] Acquire user training data of a target user, and adjust a preset bias calibration module based on the user training data to obtain an individual calibration module, wherein the user training data includes a user electromyographic signal of the target user and a target gesture corresponding to the user electromyographic signal;

[0008] A target gesture recognition model is constructed based on the individual calibration module and the basic gesture recognition model, wherein when gesture recognition is performed on the target user's electromyographic signal to be recognized according to the target gesture recognition model, a feature vector corresponding to the electromyographic signal to be recognized is adjusted according to the individual calibration module.

[0009] In one embodiment, the step of adjusting the preset bias calibration module based on the user training data to obtain the individual calibration module includes:

[0010] Inputting the user's electromyographic signal into the basic gesture recognition model to obtain a signal feature vector;

[0011] Inputting the signal feature vector into a preset bias calibration module to obtain an adjusted feature vector, and inputting the adjusted feature vector into the basic gesture recognition model to obtain a basic recognition result;

[0012] The bias calibration module is adjusted based on the basic recognition result and the target gesture to obtain an individual calibration module.

[0013] In one embodiment, the basic gesture recognition model includes a feature extraction module and a feature classification module;

[0014] The step of inputting the user's electromyographic signal into the basic gesture recognition model to obtain a signal feature vector includes:

[0015] Inputting the user's electromyographic signal into the feature extraction module to obtain a signal feature vector;

[0016] The step of inputting the adjusted feature vector into the basic gesture recognition model to obtain a basic recognition result includes:

[0017] The adjusted feature vector is input into the feature classification module to obtain a basic recognition result.

[0018] In one embodiment, the bias calibration module includes a first hyperparameter bias vector and a second hyperparameter bias vector;

[0019] The step of inputting the signal feature vector into a preset offset calibration module to obtain an adjusted feature vector comprises:

[0020] Inputting the signal feature vector into a preset bias calibration module, and obtaining an intermediate feature vector by multiplying the signal feature vector by the first hyperparameter bias vector;

[0021] An adjusted feature vector is obtained by adding the intermediate feature vector and the second hyperparameter offset vector.

[0022] In one embodiment, before the step of obtaining user training data of the target user, the method further includes:

[0023] Collecting the periodic electromyographic signals of the target user according to a preset period, and performing gesture recognition on each of the periodic electromyographic signals based on the basic gesture recognition model to obtain a plurality of periodic recognition results;

[0024] Detecting whether to trigger a model calibration operation based on each of the cycle identification results;

[0025] If it is determined that the model calibration operation is triggered, the step of obtaining the user training data of the target user and adjusting the preset bias calibration module based on the user training data to obtain the individual calibration module is performed.

[0026] In one embodiment, the step of detecting whether to trigger a model calibration operation based on each of the cycle identification results further includes:

[0027] Detecting whether there is an invalid recognition result in each of the periodic recognition results, wherein the invalid recognition result indicates that the basic gesture recognition model fails to recognize a valid gesture;

[0028] If the invalid identification result exists in each of the periodic identification results, determining to trigger a model calibration operation;

[0029] If the invalid recognition result does not exist in each of the periodic recognition results, user feedback information of the target user based on each of the periodic recognition results is obtained, and when the user feedback information is that the gesture result is inaccurate, it is determined to trigger the model calibration operation.

[0030] In one embodiment, the step of obtaining user training data of the target user includes:

[0031] Outputting a preset target gesture and prompting the target user to perform the target gesture;

[0032] The user electromyographic signal generated when the target user performs the target gesture within a preset time period is collected, and the user training data is constructed using the user electromyographic signal as training input data and the target gesture as a training label.

[0033] In addition, to achieve the above objectives, the present application also proposes a terminal device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the gesture recognition method described above.

[0034] In addition, to achieve the above objectives, the present application also proposes a storage medium, which is a computer-readable storage medium, and stores a computer program on the storage medium. When the computer program is executed by a processor, the steps of the gesture recognition method described above are implemented.

[0035] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the gesture recognition method described above are implemented.

[0036] In the present application, a pre-trained basic gesture recognition model is obtained, wherein the basic gesture recognition model is trained using historical electromyographic signals of multiple users as input data and gestures corresponding to the historical electromyographic signals as labels. A basic gesture recognition model is obtained by training with historical electromyographic signals of multiple users. The basic gesture recognition model is universal and provides a basic framework for subsequent individual calibration.

[0037] The user training data of the target user is obtained, and the preset bias calibration module is adjusted based on the user training data to obtain an individual calibration module, wherein the user training data includes the user electromyographic signal of the target user and the target gesture corresponding to the user electromyographic signal; a target gesture recognition model is constructed based on the individual calibration module and the basic gesture recognition model, wherein when performing gesture recognition on the electromyographic signal to be recognized of the target user according to the target gesture recognition model, the feature vector corresponding to the electromyographic signal to be recognized is adjusted according to the individual calibration module. The preset bias calibration module is adjusted using the user training data of the target user to obtain an individual calibration module, and in the gesture recognition process, the feature vector corresponding to the electromyographic signal to be recognized of the target user is adjusted through the individual calibration module, so that the adjusted feature vector can reflect the unique characteristics of the target user, thereby enhancing the individual adaptability of the model and improving the accuracy of gesture recognition.

[0038] In the present application, an individual calibration module is introduced to capture and adapt to the unique characteristics of the target user, so that the target gesture recognition model can better capture and adapt to the unique characteristics of the target user, thereby reducing the recognition error caused by individual differences and improving the accuracy of gesture recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 A flowchart of the first embodiment of the gesture recognition method of the present application is provided;

[0042] Figure 2 This is a schematic diagram of the structure of a basic gesture recognition model involved in one embodiment of the present application;

[0043] Figure 3A schematic diagram of the structure of a target gesture recognition model involved in an implementation method of the application;

[0044] Figure 4 A schematic diagram of a brief flow of a gesture recognition method provided in an embodiment of the present application;

[0045] Figure 5 This is a schematic diagram of the module structure of the gesture recognition device according to an embodiment of the present application;

[0046] Figure 6 Schematic diagram of the device structure of the hardware operating environment involved in the gesture recognition method in the embodiment of the present application.

[0047] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0048] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0049] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0050] The main solution of the embodiment of the present application is: obtaining a pre-trained basic gesture recognition model, wherein the basic gesture recognition model is trained using historical electromyographic signals of multiple users as input data and gestures corresponding to the historical electromyographic signals as labels; obtaining user training data of a target user, and adjusting a preset bias calibration module based on the user training data to obtain an individual calibration module, wherein the user training data includes the user electromyographic signal of the target user and a target gesture corresponding to the user electromyographic signal; constructing a target gesture recognition model based on the individual calibration module and the basic gesture recognition model, wherein, when performing gesture recognition on the electromyographic signal to be recognized of the target user according to the target gesture recognition model, adjusting the feature vector corresponding to the electromyographic signal to be recognized according to the individual calibration module.

[0051] In this embodiment, for ease of description, the following description is made with the terminal device as the execution subject.

[0052] With the rapid development of human-computer interaction technology, electromyography gesture recognition technology has attracted much attention due to its wide application in rehabilitation medicine, health monitoring, prosthetic control, game control and virtual reality. The core of EMG gesture recognition is to accurately recognize and classify gestures by capturing and analyzing the electrical signals generated by human muscles. Its application scenarios include but are not limited to: in rehabilitation medicine, by identifying the patient's gesture movements, assisting rehabilitation training and evaluation; in health monitoring, real-time monitoring of the user's movement status and health status; in prosthetic control, by identifying the user's intention, precise control of the prosthesis can be achieved; in games and virtual reality, it provides a natural and intuitive interaction method to enhance the user experience.

[0053] However, in practical applications, EMG gesture recognition technology faces the challenge of insufficient algorithm generalization. Differences in muscle structure, skin characteristics and usage environment between individuals lead to significant performance differences in traditional recognition algorithms under different users or environments. This difference is mainly reflected in the following aspects: First, differences in muscle structure between individuals lead to significant differences in EMG signals generated by the same gesture in different individuals; second, differences in skin characteristics such as thickness and humidity will also affect the collection and transmission of EMG signals; in addition, changes in the usage environment, such as temperature, humidity, sensor position, etc., will also affect the quality and stability of EMG signals. These factors together lead to the unstable performance of traditional recognition algorithms in practical applications, limiting the popularization and application of EMG gesture recognition technology.

[0054] To solve this problem, researchers have proposed a variety of improvement methods, including feature extraction, signal preprocessing, model optimization, etc. However, while these methods improve recognition accuracy, they are accompanied by increased computational complexity, making it difficult to meet the real-time and response speed requirements in practical applications. Therefore, there is an urgent need for a recognition method that can quickly align and calibrate individual feature biases to improve the adaptability and generalization ability of the algorithm, and ensure efficient recognition performance in a variety of practical usage scenarios.

[0055] The present application provides a solution, which uses the user training data of the target user to adjust the preset bias calibration module to obtain an individual calibration module. In the gesture recognition process, the feature vector corresponding to the target user's electromyographic signal to be recognized is adjusted through the individual calibration module, so that the adjusted feature vector can reflect the unique characteristics of the target user, thereby enhancing the individual adaptability of the model and improving the accuracy of gesture recognition. That is, by introducing the individual calibration module to capture and adapt to the unique characteristics of the target user, the target gesture recognition model can better capture and adapt to the unique characteristics of the target user, thereby reducing the recognition error caused by individual differences and improving the accuracy of gesture recognition.

[0056] In addition, compared to improving recognition accuracy through feature extraction, signal preprocessing, model optimization and other methods, the present application introduces a bias calibration module in the model training stage, and uses a small amount of user training data of new users to adjust the bias calibration module to obtain an individual calibration module. There is no need to make secondary adjustments to the basic gesture recognition model, which shortens the training and calibration time of the model, reduces the computational complexity in the model training process, and reduces the time cost and computing resource cost of adapting to individual differences, thereby improving the adaptability and generalization ability of the algorithm, and ensuring efficient recognition performance in a variety of actual usage scenarios.

[0057] It should be noted that the terminal device that executes the present embodiment may be a wearable device, such as a bionic hand, a wristband, or other device, or an electronic device that establishes a communication connection with the wearable device, such as a tablet computer, a personal computer, a mobile phone, etc. The present embodiment and the following embodiments are described below using a terminal device as an example.

[0058] Based on this, the present application embodiment provides a gesture recognition method, referring to Figure 1 , Figure 1 This is a flowchart of the first embodiment of the gesture recognition method of the present application.

[0059] In this embodiment, the gesture recognition method includes steps S10 to S30:

[0060] Step S10, obtaining a pre-trained basic gesture recognition model, wherein the basic gesture recognition model is trained using historical electromyographic signals of multiple users as input data and gestures corresponding to the historical electromyographic signals as labels;

[0061] The electromyographic signal is the temporal and spatial superposition of the action potentials of motor units in numerous muscle fibers. It is a bioelectric signal generated along with muscle contraction. When the muscle contracts, the flow of ions in the muscle fibers will produce a potential difference, which can be captured by electrodes to form an electromyographic signal. Therefore, user actions can be identified through the electromyographic signal.

[0062] It should be noted that the basic gesture recognition model is a model that has been trained with a large amount of data and can recognize gestures from electromyographic signals. The model can preliminarily recognize the electromyographic signals of new users, but due to the existence of individual differences, the recognition accuracy of the basic gesture recognition model is limited. In this embodiment, the historical electromyographic signals of multiple users are used as input data, and the gestures corresponding to the historical electromyographic signals (that is, the electrical signals captured from the user's muscle activities by electromyographic technology) are used as labels for training to obtain a basic gesture recognition model. The training process is not limited and can be set according to actual needs. For example, an end-to-end data streaming training method can be used in the model training process to train the basic gesture recognition model.

[0063] It is understandable that the basic gesture recognition model can be trained in the cloud and then saved in the terminal device, or it can be trained and saved in the terminal device, and there is no limitation here. When saving the basic gesture recognition model, the model can be saved as a whole to simplify the model loading and reasoning process, making the deployment and operation on the terminal device more intuitive and efficient; the model can also be saved in segments, and feature extraction modules and feature classification modules are generated separately to achieve modular design of the model, making the update and maintenance of the model more flexible, and also facilitating targeted optimization and adjustment in different application scenarios. For example, when a new feature extraction method appears, the feature extraction module can be updated without changing the feature classification module; for example, when the classification algorithm is improved, the feature classification module can also be updated.

[0064] Step S20, obtaining user training data of the target user, and adjusting a preset bias calibration module based on the user training data to obtain an individual calibration module, wherein the user training data includes a user electromyographic signal of the target user and a target gesture corresponding to the user electromyographic signal;

[0065] It should be noted that the bias calibration module is a module used to capture and adjust user-specific signal features. Through training, it can adapt to the electromyographic signal pattern of a specific user to reduce recognition errors caused by differences between users. During the training process, the bias calibration module learns the user-specific electromyographic signal features and adjusts these features to better adapt to the input requirements of the model. During the recognition process, the bias calibration module preprocesses or adjusts the user's electromyographic signal based on the learned features to extract a more accurate feature vector. It should be noted that the bias calibration module can be a simple linear model, a neural network layer, or other forms of adjustment mechanisms. There is no specific limitation here and it can be set according to actual needs.

[0066] In this embodiment, the user who needs to perform differentiated adaptation of the basic gesture recognition model is called a target user, and the training data with a label corresponding to the target user is obtained, which is hereinafter referred to as user training data for distinction. The user training data may include the electromyographic signal of the target user (hereinafter referred to as the user electromyographic signal) and its corresponding gesture action (referred to as the target gesture for convenience of description). The preset bias calibration module is adjusted through the user training data to obtain an individual calibration module for the target user.

[0067] It should be noted that the process of obtaining user training data is not limited here and can be set according to actual needs. For example, in a feasible implementation, the user's electromyographic signal can be obtained, and the user's annotation information based on the electromyographic signal can be obtained to obtain the user training data. The specific process can be: collecting the user's electromyographic signal of the target user through the electromyographic sensor device; obtaining the signal annotation of the target user based on the user's electromyographic signal, and determining the target gesture corresponding to the user's electromyographic signal based on the signal annotation. Furthermore, the terminal device can input the user's electromyographic signal into the basic gesture recognition model to obtain a preliminary recognition result, and then obtain the target user's feedback information on the preliminary recognition result, and use the feedback information as the annotation information to correct and enhance the accuracy and relevance of the training data set through direct feedback from the user, which helps to improve the accuracy of subsequent individual calibration modules and user satisfaction. Exemplarily, in another feasible implementation, the terminal device may generate a preset target gesture, and guide the target user to perform the target gesture according to the instructions through a screen or other forms of prompts. During the user's execution of the gesture, the terminal device collects the user's electromyographic signal, and uses the collected electromyographic signal as the user's electromyographic signal, and uses the target gesture as a label to construct a training data set. In this implementation, the consistency and comparability of the training data are ensured through a standardized data collection process. Since the user performs the gesture under clear guidance, data errors caused by user misunderstanding or improper operation are reduced, thereby improving the accuracy and reliability of the training data.

[0068] It should be noted that the process of adjusting the preset bias calibration module based on the user training data to obtain the individual calibration module is not limited here. For example, in a feasible implementation, the bias calibration module can be trained locally on the terminal device. This implementation can process user training data in real time, reducing the delay and cost of data transmission, and because data processing and model training are completed locally, there is no need to upload data to the cloud or other servers, which can effectively avoid the risk of user privacy leakage and improve data security. For example, in another implementation, the bias calibration module can be adjusted using a strategy of collaborative work between the cloud and the terminal. In this implementation, the powerful computing power and storage resources of the cloud can be used to avoid the impact of the training process on the operating performance of the terminal device, ensuring a smooth experience for users during use. The specific process of training in this embodiment may be: inputting the user's electromyographic signal into the basic gesture recognition model to obtain a signal feature vector; sending the signal feature vector and the target gesture to a cloud device, wherein the cloud device optimizes the parameters of the bias calibration module deployed in the cloud based on the first feature vector and the target gesture to obtain updated parameters and returns them; receiving the updated parameters sent by the cloud device, and updating the preset bias calibration module with the updated parameters to obtain an individual calibration module. Furthermore, in order to ensure the security of the cloud transmission process, data encryption processing may be performed during the transmission process. The specific data encryption transmission process is not limited here and can be set according to actual needs.

[0069] In this embodiment, by introducing the individual calibration module, the gesture recognition model can more accurately recognize the gesture movements of the target user, reduce the recognition errors caused by differences between users, and thus improve the overall recognition accuracy.

[0070] Step S30, building a target gesture recognition model based on the individual calibration module and the basic gesture recognition model, wherein when performing gesture recognition on the target user's electromyographic signal to be recognized according to the target gesture recognition model, adjusting the feature vector corresponding to the electromyographic signal to be recognized according to the individual calibration module.

[0071] The individual calibration module is combined with the basic gesture recognition model to construct a gesture recognition model for the target user, and then gesture recognition is performed based on the gesture recognition model, wherein the basic gesture recognition model is used for feature extraction and feature classification, and the individual calibration module is used to perform individual calibration on the feature vector extracted according to the basic gesture recognition model, that is, the feature distribution of the feature vector obtained by the feature extraction module in the target gesture recognition model is adjusted by the individual calibration module based on the electromyographic signal to be recognized, thereby achieving the purpose of adjusting the feature vector of the target user, so that the feature vector used for feature classification is personalized with the target user, and the accuracy of gesture recognition is improved. That is, the bias calibration module is used to adjust the feature distribution of the feature vector after receiving the feature vector extracted by the feature extraction module in the basic gesture recognition model, and the bias calibration module outputs the adjusted feature vector to the feature classification module in the basic gesture recognition model to obtain the gesture recognition result. In the training process of the bias calibration module, the user electromyographic signal is input to obtain the feature vector, the feature vector is input to the bias calibration module to obtain the adjusted feature vector, and the adjusted feature vector is input to the feature classification module to obtain the model training output result. The bias calibration module is tuned based on the model training output result and the target gesture to obtain the individual calibration module.

[0072] In this embodiment, the preset bias calibration module is adjusted using the user training data of the target user to obtain an individual calibration module. In the gesture recognition process, the feature vector corresponding to the target user's electromyographic signal to be recognized is adjusted by the individual calibration module, so that the adjusted feature vector can reflect the unique characteristics of the target user, thereby enhancing the individual adaptability of the model and improving the accuracy of gesture recognition. That is, by introducing the individual calibration module to capture and adapt to the unique characteristics of the target user, the target gesture recognition model can better capture and adapt to the unique characteristics of the target user, thereby reducing the recognition error caused by individual differences and improving the accuracy of gesture recognition.

[0073] In addition, compared with improving recognition accuracy through feature extraction, signal preprocessing, model optimization and other methods, this embodiment introduces a bias calibration module in the model training stage, and uses a small amount of user training data of new users to adjust the bias calibration module to obtain an individual calibration module. There is no need to perform secondary adjustment on the basic gesture recognition model, which shortens the training and calibration time of the model, reduces the computational complexity in the model training process, and reduces the time cost and computing resource cost of adapting to individual differences, thereby improving the adaptability and generalization ability of the algorithm and ensuring efficient recognition performance in a variety of actual usage scenarios.

[0074] In one feasible implementation, the step S20: the step of obtaining user training data of the target user, comprises:

[0075] Step S201, outputting a preset target gesture, and prompting the target user to perform the target gesture;

[0076] It should be noted that different gestures can be pre-stored in the terminal device, and then the gesture performed by the target user (ie, the target gesture) is determined from different preset gestures. The process of determining the target gesture is not restricted here, for example, it can be randomly selected, and can be set according to actual needs.

[0077] After determining the target gesture that the target user needs to perform, the user is conveyed the instruction to perform the target gesture through a preset prompt method (such as visual prompt, voice guidance, etc.). In a specific implementation, the most appropriate prompt method can be selected according to the physiological characteristics of the target user (such as vision condition, hearing condition, etc.). For example, for users with normal vision, visual prompts (such as animations or pictures on the screen) can be used; for users with normal hearing, voice guidance can be used.

[0078] It is understandable that through clear instructions and prompts, it can be ensured that the collected user electromyographic signal data corresponds to a specific gesture, thereby improving the accuracy and reliability of the data. Choosing the right prompt method can increase user participation and comfort, making the data collection process smoother.

[0079] Step S202, collecting the user electromyographic signal generated when the target user performs the target gesture within a preset time period, and constructing user training data using the user electromyographic signal as training input data and the target gesture as a training label.

[0080] Use an electromyographic signal acquisition device to capture electrical signals from the muscle activity of the target user, and associate the user's electromyographic signal with the target gesture to form a training data set. The specific process of constructing user training data can be: while the target user performs the target gesture, start the electromyographic signal acquisition device to record the electromyographic signal data within a preset time length, wherein the preset time length can be reasonably set according to the complexity and duration of the gesture to ensure that the complete gesture action can be captured, and there is no specific restriction here; associate the collected electromyographic signal data with the target gesture to form a training data pair, each data pair includes an electromyographic signal feature vector (usually data after preprocessing and feature extraction) and a corresponding gesture label.

[0081] It is understandable that through precise data collection and labeling processes, a high-quality training data set can be generated for training using the specific data of the target user, thereby generating a personalized gesture recognition model for the user and improving the accuracy and robustness of recognition.

[0082] In one feasible implementation, the step S20: adjusting the preset bias calibration module based on the user training data to obtain the individual calibration module, comprises:

[0083] Step S203, inputting the user's electromyographic signal into the basic gesture recognition model to obtain a signal feature vector;

[0084] The target user's electromyographic signal is input into the feature extraction module of the trained basic gesture recognition model to extract the feature vector related to gesture recognition. The specific process can be: select one or more segments of data from the collected user's electromyographic signal as the original data input to the basic gesture recognition model; feature extraction: the basic gesture recognition model processes the input electromyographic signal step by step, and finally outputs one or more feature vectors, which are referred to as signal feature vectors for distinction. The signal feature vector is the key information related to gesture recognition extracted by the model from the original signal.

[0085] It can be understood that through the feature extraction capability of the basic gesture recognition model, complex electromyographic signals can be converted into concise and efficient feature vectors to facilitate subsequent processing and recognition, maintaining consistency with the model's internal processing flow for adjustment of the bias calibration module in subsequent steps.

[0086] Step S204, inputting the signal feature vector into a preset bias calibration module to obtain an adjusted feature vector, and inputting the adjusted feature vector into the basic gesture recognition model to obtain a basic recognition result;

[0087] The extracted signal feature vector is input into a preset bias calibration module to adjust the distribution or weight of the feature vector, and then the adjusted feature vector is input into the feature classification module in the basic gesture recognition model to obtain the basic recognition result.

[0088] It can be understood that by adjusting the bias calibration module, the feature vector can better reflect the specific gesture characteristics of the target user, thereby improving the accuracy of gesture recognition. The introduction of the bias calibration module increases the flexibility of the model, enabling it to adapt to the differences in gesture characteristics of different users.

[0089] Step S205: adjusting the parameters of the bias calibration module based on the basic recognition result and the target gesture to obtain an individual calibration module.

[0090] Compare the difference between the basic recognition result and the target gesture to calculate the error (such as classification error rate, cross entropy loss, etc.); according to the error calculation result, use the optimization algorithm (such as gradient descent, stochastic gradient descent, etc.) to adjust the parameters of the bias calibration module. The goal of the adjustment is to minimize the error and thus improve the recognition accuracy; perform multiple iterative optimizations until the predetermined convergence conditions are reached to obtain the individual calibration module (such as the error is less than a certain threshold, the number of iterations reaches an upper limit, etc.).

[0091] It can be understood that, through the iterative optimization process, the parameters of the bias calibration module can be continuously adjusted to better adapt to the gesture characteristics of the target user, and personalized optimization of the basic gesture recognition model can be achieved to improve the accuracy and adaptability of gesture recognition.

[0092] In one feasible implementation, the bias calibration module includes a first hyperparameter bias vector and a second hyperparameter bias vector;

[0093] The step S204: inputting the signal feature vector into a preset offset calibration module to obtain an adjusted feature vector, comprises:

[0094] Step S2041, inputting the signal feature vector into a preset bias calibration module, and obtaining an intermediate feature vector by multiplying the first hyperparameter bias vector and the signal feature vector;

[0095] The bias calibration module consists of two key hyperparameters: the first hyperparameter bias vector and the second hyperparameter bias vector. The first hyperparameter bias vector and the second hyperparameter bias vector adjust the input signal feature vector to optimize the accuracy of gesture recognition. It should be noted that the first hyperparameter bias vector and the second hyperparameter bias vector have the same shape.

[0096] The signal feature vector is obtained from the basic gesture recognition model, and the intermediate feature vector is obtained by element-wise multiplication of the signal feature vector and the first hyperparameter bias vector. By multiplying the first hyperparameter bias vector and the signal feature vector, weighted processing of the features is achieved, which can highlight the features related to the target gesture recognition, while weakening or ignoring irrelevant features.

[0097] Step S2042: Obtain an adjusted feature vector by adding the intermediate feature vector to the second hyperparameter offset vector.

[0098] The intermediate feature vector is added to the second hyperparameter bias vector to obtain an adjusted feature vector, which is used for subsequent gesture recognition. By adding the intermediate feature vector to the second hyperparameter bias vector, the feature offset processing is realized, and the feature distribution can be further adjusted to better adapt to the gesture characteristics of the target user.

[0099] The first hyperparameter bias vector and the second hyperparameter bias vector are learnable parameters and can be adjusted according to the recognition results during the training process, thereby realizing personalized adjustment and optimization of the signal feature vector, thereby increasing the flexibility and adaptability of the model and providing more accurate and reliable feature information for subsequent gesture recognition.

[0100] In one feasible implementation, the basic gesture recognition model includes a feature extraction module and a feature classification module; the step S203: inputting the user's electromyographic signal into the basic gesture recognition model to obtain a signal feature vector includes:

[0101] Step S2031, inputting the user's electromyographic signal into the feature extraction module to obtain a signal feature vector;

[0102] In this embodiment, the basic gesture recognition model is stored in segments, specifically divided into a feature extraction module and a feature classification module. The two modules work together to extract features from the input user electromyographic signals, and perform gesture classification based on these features.

[0103] The feature extraction module may contain multiple processing layers, such as convolutional layers, pooling layers, and fully connected layers. The user's electromyographic signal is used as the input of the feature extraction module. Each layer in the feature extraction module processes the input electromyographic signal step by step to extract gesture-related features (i.e., signal feature vectors). The specific architecture of the feature extraction module is not limited here and can be set according to actual needs.

[0104] It can be understood that the feature extraction module can convert complex electromyographic signals into concise and efficient feature vectors. Through the processing of the feature extraction module, it can ensure that the data format and dimension input to the feature classification module are unified, which is conducive to model training and reasoning.

[0105] In this implementation, the step S204: inputting the adjusted feature vector into the basic gesture recognition model to obtain a basic recognition result includes:

[0106] Step S2043, inputting the adjusted feature vector into the feature classification module to obtain a basic recognition result.

[0107] The adjusted feature vector obtained from the bias calibration module has been optimized and adjusted according to the gesture characteristics of the target user. The adjusted feature vector is input into the feature classification module. The feature classification module may contain multiple classifiers or classification layers. Each classifier (or classification layer) classifies the input adjusted feature vector, determines which gesture category the adjusted feature vector belongs to, and obtains the basic recognition result.

[0108] For example, please refer to Figure 2The basic gesture recognition model can include a feature extraction module and a feature classification module. The feature extraction module is composed of a convolution layer (conv), a pooling layer (Pooling), and an activation function (Relu). Different stacking layers can be selected according to needs. The result of the last layer of convolution is the feature of the data. In order to quickly calibrate the training phase, after obtaining the output of the feature extraction model, the feature is normalized by the mean variance. The formula is: feature′=(feature-Mean)*Var. The feature classification module includes a fully connected layer (fc, Fully Connected layer) and an activation function (softmax). The feature classification module classifies the input features. During the training process, this module can be trained simultaneously with the feature extraction module because this part is defined as a classification problem. Figure 2 As shown, the training process of the basic gesture recognition model can be: input the basic training data (that is, the historical electromyographic signals and the gestures corresponding to the historical electromyographic signals) into the feature extraction module, extract the features through the multi-layer feature extraction blocks to obtain the signal features, calculate the average value (mean) of the signal features to reduce noise or extract global features, then subtract the average value from the signal features to normalize them to obtain the normalized features, perform dot product operation on the normalized features and the variance (var) of the signal features to obtain the feature vector (feature), input the feature vector into the fully connected layer, and then input the activation function to obtain the category result, and adjust the model parameters based on the category result to obtain the basic gesture recognition model. Figure 3 , the target gesture recognition model introduces a fast calibration module (also known as a bias calibration module) based on the basic gesture recognition model. The fast calibration module receives the feature vector extracted by the feature extraction model, processes the feature vector through the gamma vector (the first hyperparameter bias vector) and the beta vector (the second hyperparameter bias vector) to obtain the processed feature, and inputs the processed feature into the feature classification module. The process of the fast calibration module processing the feature vector can be: feature′=feature*gamma+beta, where gamma can be initialized to a vector of all 1s, beta can be initialized to a vector of all 0s, and the size of the hyperparameter bias vector is the same as the feature size output by the feature extraction module.

[0109] It can be understood that through the processing of the feature classification module, the adjusted feature vector can be accurately classified, thereby improving the accuracy of gesture recognition. As a part of the model, the feature classification module can work together with the feature extraction module and the bias calibration module to adapt to the differences in gesture features of different users and achieve accurate recognition and classification of user electromyographic signals.

[0110] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated later. On this basis, please refer to the step S20: before the step of obtaining the user training data of the target user, it also includes:

[0111] Step S01, collecting periodic electromyographic signals of the target user according to a preset period, and performing gesture recognition on each of the periodic electromyographic signals based on the basic gesture recognition model to obtain multiple periodic recognition results;

[0112] It should be noted that the preset period can be set according to the actual application scenario and the needs of the target user, and is not limited here. In this embodiment, the target user's electromyographic signal is collected according to the preset period, hereinafter referred to as the periodic electromyographic signal for distinction, and the basic gesture recognition model is used to perform gesture recognition on the periodic electromyographic signal. The purpose is to collect gesture recognition results over a period of time to facilitate subsequent analysis of whether the model needs to be calibrated.

[0113] It is understandable that through regular collection and identification, a large amount of recognition result data can be accumulated, providing a basis for subsequent analysis of model performance. By continuously monitoring the recognition results, possible problems with the model can be discovered in a timely manner, providing early warning for model calibration.

[0114] Step S02, detecting whether a model calibration operation is triggered based on each of the cycle identification results;

[0115] The accumulated periodic recognition results are analyzed to determine whether the model calibration operation needs to be triggered. Specifically, the judgment can be made through evaluation indicators such as accuracy, recall rate, F1 score, etc., which are not limited here. Exemplarily, the process of detecting whether to trigger the model calibration operation can be: according to the accumulated periodic recognition results, the accuracy of the basic gesture recognition model is calculated; according to the actual application requirements, the accuracy threshold is set, and the accuracy threshold is used to determine whether the model needs to be calibrated; if the accuracy is lower than the accuracy threshold, the model calibration operation is triggered; otherwise, it is not triggered.

[0116] It is understandable that by regularly testing the recognition results and comparing the evaluation indicators, the degradation of model performance can be discovered in time, and the calibration operation can be triggered to ensure the accuracy and stability of the model. In addition, evaluating whether to perform the calibration operation can avoid frequent triggering of the calibration operation due to slight fluctuations, reducing unnecessary resource consumption.

[0117] Step S03, if it is determined that the model calibration operation is triggered, the step of obtaining the user training data of the target user and adjusting the preset bias calibration module based on the user training data to obtain an individual calibration module is executed.

[0118] After determining that the model calibration operation needs to be triggered, the model calibration operation is performed, that is, the step of obtaining the user training data of the target user is performed, and the subsequent model calibration process is performed based on the user training data. The specific process is not described in detail here, and the first embodiment can be referred to for details. In this embodiment, through the preprocessing process before the model calibration, possible problems of the model can be discovered in time, and necessary training data can be collected for the subsequent model calibration process.

[0119] In a feasible implementation manner, the step S02: the step of detecting whether to trigger a model calibration operation based on each of the cycle identification results, further includes:

[0120] Step S021, detecting whether there is an invalid recognition result in each of the periodic recognition results, wherein the invalid recognition result indicates that the basic gesture recognition model fails to recognize a valid gesture;

[0121] The periodic recognition result may be a valid recognition result or an invalid recognition result, wherein a valid recognition result refers to a result that can characterize a valid gesture action, such as a gesture category label or probability distribution, and an invalid recognition result refers to a result that characterizes that the basic gesture recognition model fails to recognize a valid gesture. When the basic gesture recognition model cannot recognize any valid gesture from the input electromyographic signal, an invalid recognition result is output. The invalid recognition result may be caused by the low quality of the input electromyographic signal, high noise, or a large difference from the data used in model training.

[0122] The recognition results in each cycle are checked to identify those results that failed to successfully identify a valid gesture, i.e., invalid recognition results. Exemplarily, the process of detecting whether there are invalid recognition results in each of the cycle recognition results can be: traversing each cycle recognition result, checking whether the cycle recognition result is a valid gesture label, if the check result is empty (i.e., no gesture is recognized) or a preset invalid label (such as "unknown" or "unrecognizable"), it is regarded as an invalid recognition result, and the number of invalid recognition results that appear in each cycle is recorded.

[0123] By detecting invalid recognition results, potential problems in the model during the recognition process can be discovered in a timely manner, providing a direct basis for model calibration.

[0124] Step S022, if there is an invalid recognition result in each of the period recognition results, determining to trigger a model calibration operation;

[0125] If invalid recognition results are detected in each cycle recognition result, it indicates that there may be problems with the current basic gesture recognition model and a calibration operation needs to be triggered to improve its performance. It can be understood that by directly responding to invalid recognition results, the model calibration operation can be quickly triggered, and reducing invalid recognition results can improve the overall accuracy and reliability of the system, thereby improving the user experience.

[0126] Step S023: if the invalid recognition result does not exist in each of the periodic recognition results, user feedback information of the target user based on each of the periodic recognition results is obtained, and when the user feedback information is that the gesture result is inaccurate, it is determined to trigger the model calibration operation.

[0127] If no invalid recognition results are detected in the recognition results of each cycle, there may also be inaccurate recognition results. Therefore, direct feedback from the target user on the gesture recognition results is obtained, and based on these feedbacks, it is determined whether the model calibration operation needs to be triggered. Specifically, in the case where there are no invalid recognition results in the recognition results, the target user's feedback on the gesture recognition results is collected through the user interface or other means; the user feedback is analyzed to determine whether there is an inaccurate recognition result; if the recognition result is determined to be inaccurate based on the user feedback, it is determined to trigger the model calibration operation. It can be understood that calibration based on user feedback can more accurately identify and solve problems in the model during the recognition process, thereby improving the overall accuracy of the system.

[0128] In this embodiment, whether to trigger the model calibration operation is detected based on periodic recognition results and user feedback. By comprehensively considering invalid recognition results and user feedback, it can be more comprehensively and accurately determined whether the model needs to be calibrated, thereby improving system performance and user experience.

[0129] Exemplarily, in order to help understand the implementation process of the gesture recognition method obtained by combining this embodiment with the above-mentioned embodiment one, the third embodiment of the present application is proposed. In this embodiment, a basic gesture recognition model is designed for EMG gesture data. By standardizing the gesture feature layer, the stability and consistency of the model in different users and environments are ensured. The standardization helps to reduce differences between individuals and improve the generalization ability of the model. It should be noted that the calibration process is performed entirely on the terminal device and does not involve uploading and cloud processing of user data, thereby ensuring the protection of user privacy. This feature is particularly suitable for application scenarios with high requirements for data privacy, such as medical health and personal devices, and enhances user trust and acceptance of technology. Specifically, please refer to Figure 4 , the specific process of performing gesture recognition in this embodiment may be:

[0130] EMG gesture data segment interception, by intercepting EMG gesture data segments to obtain gesture data that can be used for training. The main purpose of this step is to collect enough high-quality data to ensure the effectiveness of subsequent model training. The process of intercepting EMG gesture data segments includes extracting signals of specific time periods from the user's electromyographic signals, which correspond to the electromyographic activities when the user performs different gestures.

[0131] Train the EMG gesture recognition base model (also known as the basic gesture recognition model). Input the intercepted gesture data into the constructed gesture recognition model for training. During the model training process, an end-to-end data streaming training method is adopted. The training process includes data preprocessing, feature extraction, model training and verification, which will not be described here. Through repeated iterations and optimization, the model can learn and recognize the features of different gestures. After the training is completed, a basic model is generated that can recognize various EMG gesture data.

[0132] The model is stored in segments. After the training is completed, the basic model is saved separately, the gesture feature extraction model and the feature classification model. The main function of the feature extraction model is to analyze the input data, extract its features, and represent these features in an abstract form. The feature classification model is responsible for classifying the extracted features. By analyzing and comparing the features, the model can output the results of gesture recognition. The feature classification process includes the input of feature vectors, the application of classifiers, and the output of results. The segmented saving mechanism not only improves the modular design of the model, making the update and maintenance of the model more flexible, but also facilitates targeted optimization and adjustment in different application scenarios.

[0133] The model usage process is divided into the case without calibration and the case with calibration:

[0134] In the absence of calibration, the basic recognition model can be used directly, and the electromyographic signal can be input into the feature extraction module to obtain the features of the EMG data, and then the feature classification model will classify these features, ensuring that the model can still provide reliable recognition performance in the absence of individualized calibration data, which is suitable for rapid deployment and testing stages.

[0135] When a new user starts to use the model, it enters the calibration usage. A small amount of new user gesture data with labels will be obtained. The gesture feature extraction model will be used to extract features from these new data to generate feature vectors, and then fast calibration training will be performed on the embedded chip. Since the gesture feature extraction model design contains standardized operations, only a small number of iterations are required to complete the calibration. The fast calibration training process includes feature extraction of new data, input of feature vectors, adjustment of model parameters, and verification of calibration results. Specifically, the model introduces two hyperparameter bias vectors, gamma (also known as the first hyperparameter bias vector) and beta (also known as the second hyperparameter bias vector). These vectors directly act on the output feature vector of the feature extraction module to adjust the distribution of features, thereby affecting the final classification results of the feature classification module. In this way, the model can dynamically adapt to the feature differences of different users. Since the dimension of the hyperparameter bias vector is the same as that of the feature vector and the number of parameters is small, only a small amount of user gesture data is needed to complete the training on the embedded device. This fast adaptability enables the gesture recognition model to quickly adjust to new users or environmental changes, improving the user experience and the practicality of the model. When a new user uses the model, the model can be adapted to the new data by applying the trained bias, thereby achieving rapid calibration and improving the user experience. When a new user uses the model, the system automatically applies the calibrated parameters to ensure the accuracy and reliability of the recognition results. In this way, the present embodiment achieves rapid calibration for new users, enabling the model to adapt to the individual differences of different users and provide more accurate gesture recognition services. It is understandable that because the parameters of the feature extraction module and the feature classification module are fixed, they cannot be updated through back propagation during the training of gamma and beta, so the present embodiment uses random sampling descent to update, which can also achieve parameter learning of gamma and beta. After the rapid training is completed, the classification of the calibration user needs to be corrected by the rapid calibration module to achieve the final classification effect. The entire model does not affect the recognition effect of the base model, and supports parameter reset and multiple rapid calibrations.

[0136] This embodiment proposes a method of feature transformation and standardization processing to address the problem of signal changes caused by differences in muscle structure and skin characteristics between individuals in EMG gesture recognition. By standardizing the gesture feature layer, the model can effectively adjust and compensate for these differences, so that the signals of different users appear more consistent in the feature space. This processing not only improves the generalization ability of the model, but also enhances its stability and consistency in different environments. The standardized features are easier to process by the classification module, thereby improving the accuracy of recognition and ensuring that the model maintains efficient recognition performance in a variety of practical usage scenarios. By designing a model inference structure with bias calibration and a fast calibration method for individual feature bias, the generalization ability and adaptability of the EMG gesture recognition algorithm are improved. Specifically, the model inference structure with bias calibration can effectively adjust and compensate for differences between individuals by introducing a feature transformation mechanism, so that the model can maintain a high recognition accuracy under different users and environments. At the same time, the fast calibration method for individual feature bias further improves the adaptability and response speed of the model by quickly capturing and adjusting the bias of individual features. This innovative design not only improves the accuracy and robustness of EMG gesture recognition, but also significantly shortens the training and calibration time of the model, making it more suitable for practical application scenarios. By designing a fast adaptation mechanism, the model can be quickly trained and deployed on embedded devices, and training can be completed with a small amount of user gesture data. By using the low parameter count characteristics of the hyperparameter bias vector, the model can be quickly adjusted to adapt to new users or environmental changes. This rapid adaptability not only improves the user experience, but also makes the model suitable for deployment in embedded environments such as mobile devices and wearable devices, meeting the needs of low-power and low-computing smart wearable devices.

[0137] In addition, compared with the traditional method, the recognition accuracy of this embodiment under different users and environments is significantly improved, and the calibration time is greatly shortened. Therefore, this embodiment not only improves the adaptability, accuracy and robustness of EMG gesture recognition in technology, but also provides strong privacy protection measures at the application level, which significantly improves its feasibility and security in actual application scenarios. These innovations provide new solutions and development directions for the widespread application of EMG gesture recognition technology.

[0138] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the gesture recognition method of the present application. More simple transformations based on this technical concept are all within the protection scope of the present application.

[0139] This application also provides a gesture recognition device, please refer to Figure 5 , the gesture recognition device comprises:

[0140] An acquisition module 10 is used to acquire a pre-trained basic gesture recognition model, wherein the basic gesture recognition model is trained using historical electromyographic signals of multiple users as input data and gestures corresponding to the historical electromyographic signals as labels;

[0141] An adjustment module 20, configured to obtain user training data of a target user, and adjust a preset bias calibration module based on the user training data to obtain an individual calibration module, wherein the user training data includes a user electromyographic signal of the target user and a target gesture corresponding to the user electromyographic signal;

[0142] A construction module 30 is used to construct a target gesture recognition model based on the individual calibration module and the basic gesture recognition model, wherein, when gesture recognition is performed on the target user's electromyographic signal to be recognized according to the target gesture recognition model, the feature vector corresponding to the electromyographic signal to be recognized is adjusted according to the individual calibration module.

[0143] Optionally, the adjustment module 20 is used to:

[0144] Inputting the user's electromyographic signal into the basic gesture recognition model to obtain a signal feature vector;

[0145] Inputting the signal feature vector into a preset bias calibration module to obtain an adjusted feature vector, and inputting the adjusted feature vector into the basic gesture recognition model to obtain a basic recognition result;

[0146] The bias calibration module is adjusted based on the basic recognition result and the target gesture to obtain an individual calibration module.

[0147] Optionally, the basic gesture recognition model includes a feature extraction module and a feature classification module; the adjustment module 20 is used to:

[0148] Inputting the user's electromyographic signal into the feature extraction module to obtain a signal feature vector;

[0149] The step of inputting the adjusted feature vector into the basic gesture recognition model to obtain a basic recognition result includes:

[0150] The adjusted feature vector is input into the feature classification module to obtain a basic recognition result.

[0151] Optionally, the bias calibration module includes a first hyperparameter bias vector and a second hyperparameter bias vector; the adjustment module 20 is used to:

[0152] Inputting the signal feature vector into a preset bias calibration module, and obtaining an intermediate feature vector by multiplying the signal feature vector by the first hyperparameter bias vector;

[0153] An adjusted feature vector is obtained by adding the intermediate feature vector and the second hyperparameter offset vector.

[0154] Optionally, the device further comprises a detection module, configured to:

[0155] Collecting the periodic electromyographic signals of the target user according to a preset period, and performing gesture recognition on each of the periodic electromyographic signals based on the basic gesture recognition model to obtain a plurality of periodic recognition results;

[0156] Detecting whether to trigger a model calibration operation based on each of the cycle identification results;

[0157] If it is determined that the model calibration operation is triggered, the step of obtaining the user training data of the target user and adjusting the preset bias calibration module based on the user training data to obtain the individual calibration module is performed.

[0158] Optionally, the detection module is used to:

[0159] Detecting whether there is an invalid recognition result in each of the periodic recognition results, wherein the invalid recognition result indicates that the basic gesture recognition model fails to recognize a valid gesture;

[0160] If the invalid identification result exists in each of the periodic identification results, determining to trigger a model calibration operation;

[0161] If the invalid recognition result does not exist in each of the periodic recognition results, user feedback information of the target user based on each of the periodic recognition results is obtained, and when the user feedback information is that the gesture result is inaccurate, it is determined to trigger the model calibration operation.

[0162] Optionally, the adjustment module 20 is used to:

[0163] Outputting a preset target gesture and prompting the target user to perform the target gesture;

[0164] The user electromyographic signal generated when the target user performs the target gesture within a preset time period is collected, and the user training data is constructed using the user electromyographic signal as training input data and the target gesture as a training label.

[0165] The gesture recognition device provided by the present application adopts the gesture recognition method in the above embodiment, which can solve the technical problem that gesture recognition based on electromyographic signals is difficult to adapt to individual differences, resulting in the accuracy of gesture recognition being affected. Compared with the prior art, the beneficial effects of the gesture recognition device provided by the present application are the same as those of the gesture recognition method provided by the above embodiment, and other technical features in the gesture recognition device are the same as those disclosed in the above embodiment method, which will not be described in detail here.

[0166] The present application provides a terminal device, which includes: at least one processor; and a memory that is communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the gesture recognition method in the above-mentioned embodiment.

[0167] Reference below Figure 6 , which shows a schematic diagram of the structure of a terminal device suitable for implementing the embodiment of the present application. The terminal device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The terminal device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0168] like Figure 6As shown, the terminal device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the terminal device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the terminal device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a terminal device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or provided instead.

[0169] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0170] The terminal device provided by the present application adopts the gesture recognition method in the above embodiment, which can solve the technical problem that gesture recognition based on electromyographic signals is difficult to adapt to individual differences, resulting in the accuracy of gesture recognition being affected. Compared with the prior art, the beneficial effects of the terminal device provided by the present application are the same as the beneficial effects of the gesture recognition method provided by the above embodiment, and the other technical features in the terminal device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0171] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0172] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0173] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the gesture recognition method in the above-mentioned embodiment.

[0174] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0175] The computer-readable storage medium may be included in the terminal device; or may exist independently without being installed in the terminal device.

[0176] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by a terminal device, the terminal device implements the methods of the above embodiments.

[0177] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0178] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0179] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0180] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned gesture recognition method, and can solve the technical problem that gesture recognition based on electromyographic signals is difficult to adapt to individual differences, resulting in the accuracy of gesture recognition being affected. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the gesture recognition method provided in the above-mentioned embodiment, and will not be elaborated here.

[0181] The present application also provides a computer program product, including a computer program, which implements the steps of the gesture recognition method as described above when executed by a processor.

[0182] The computer program product provided by the present application can solve the technical problem that gesture recognition based on electromyographic signals is difficult to adapt to individual differences, resulting in the accuracy of gesture recognition being affected. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as those of the gesture recognition method provided by the above embodiment, and will not be elaborated here.

[0183] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A gesture recognition method, characterized in that: The gesture recognition method comprises: Obtain a pre-trained basic gesture recognition model, wherein the basic gesture recognition model is trained using historical electromyographic signals of multiple users as input data and gestures corresponding to the historical electromyographic signals as labels; Acquire user training data of a target user, and adjust a preset bias calibration module based on the user training data to obtain an individual calibration module, wherein the user training data includes a user electromyographic signal of the target user and a target gesture corresponding to the user electromyographic signal; A target gesture recognition model is constructed based on the individual calibration module and the basic gesture recognition model, wherein when gesture recognition is performed on the target user's electromyographic signal to be recognized according to the target gesture recognition model, a feature vector corresponding to the electromyographic signal to be recognized is adjusted according to the individual calibration module.

2. The gesture recognition method according to claim 1, characterized in that: The step of adjusting the preset bias calibration module based on the user training data to obtain the individual calibration module includes: Inputting the user's electromyographic signal into the basic gesture recognition model to obtain a signal feature vector; Inputting the signal feature vector into a preset bias calibration module to obtain an adjusted feature vector, and inputting the adjusted feature vector into the basic gesture recognition model to obtain a basic recognition result; The bias calibration module is adjusted based on the basic recognition result and the target gesture to obtain an individual calibration module.

3. The gesture recognition method according to claim 2, characterized in that: The basic gesture recognition model includes a feature extraction module and a feature classification module; The step of inputting the user's electromyographic signal into the basic gesture recognition model to obtain a signal feature vector includes: Inputting the user's electromyographic signal into the feature extraction module to obtain a signal feature vector; The step of inputting the adjusted feature vector into the basic gesture recognition model to obtain a basic recognition result includes: The adjusted feature vector is input into the feature classification module to obtain a basic recognition result.

4. The gesture recognition method according to claim 2, characterized in that: The bias calibration module includes a first hyperparameter bias vector and a second hyperparameter bias vector; The step of inputting the signal feature vector into a preset offset calibration module to obtain an adjusted feature vector comprises: Inputting the signal feature vector into a preset bias calibration module, and obtaining an intermediate feature vector by multiplying the signal feature vector by the first hyperparameter bias vector; An adjusted feature vector is obtained by adding the intermediate feature vector and the second hyperparameter offset vector.

5. The gesture recognition method according to claim 1, characterized in that: Before the step of obtaining the user training data of the target user, the method further includes: Collecting the periodic electromyographic signals of the target user according to a preset period, and performing gesture recognition on each of the periodic electromyographic signals based on the basic gesture recognition model to obtain a plurality of periodic recognition results; Detecting whether to trigger a model calibration operation based on each of the cycle identification results; If it is determined that the model calibration operation is triggered, the step of obtaining the user training data of the target user and adjusting the preset bias calibration module based on the user training data to obtain the individual calibration module is performed.

6. The gesture recognition method according to claim 5, characterized in that: The step of detecting whether to trigger a model calibration operation based on each of the cycle identification results further includes: Detecting whether there is an invalid recognition result in each of the periodic recognition results, wherein the invalid recognition result indicates that the basic gesture recognition model fails to recognize a valid gesture; If the invalid identification result exists in each of the periodic identification results, determining to trigger a model calibration operation; If the invalid recognition result does not exist in each of the periodic recognition results, user feedback information of the target user based on each of the periodic recognition results is obtained, and when the user feedback information is that the gesture result is inaccurate, it is determined to trigger the model calibration operation.

7. The gesture recognition method according to any one of claims 1 to 6, characterized in that: The step of obtaining user training data of the target user includes: Outputting a preset target gesture and prompting the target user to perform the target gesture; The user electromyographic signal generated when the target user performs the target gesture within a preset time period is collected, and the user training data is constructed using the user electromyographic signal as training input data and the target gesture as a training label.

8. A terminal device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the gesture recognition method according to any one of claims 1 to 7.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the gesture recognition method according to any one of claims 1 to 7 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the gesture recognition method according to any one of claims 1 to 7 are implemented.

Images