Method, device and equipment for publishing and sharing human body action data and medium

Abstract

The invention discloses a method, a device, equipment and a medium for publishing and sharing human body action data, and relates to the technical field of data processing, and the method comprises the steps: obtaining multivariate time series data of human body actions, and carrying out the standardization processing, data segmentation and feature extraction of the multivariate time series data, obtaining processed data including action categories and corresponding labels; constructing and training a preset auto-encoder for each action category, and performing data reconstruction on the processed data by using the trained preset auto-encoder to obtain reconstructed data; the trained preset auto-encoder comprises an encoder and a decoder; and releasing and sharing the reconstructed data. According to the technical scheme, information loss can be avoided, the learnability of the human body behavior action data is protected, unauthorized users are prevented from abusing the data, and meanwhile it is ensured that the data still have analyzability conforming to original data in legal sharing and display scenes.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device, equipment and medium for publishing and sharing human motion data. Background Art

[0002] HAR (Human Activity Recognition) is an important research direction in machine learning and sensing technology, and is widely used in medical health monitoring, smart home, security monitoring and other fields. Currently, HAR mainly relies on wearable sensors or computer vision technology to collect data and classify and identify it through machine learning or deep learning models. However, with the widespread application of HAR data, user privacy protection has become a key challenge. Existing related technologies mainly include the following categories: (1) Anonymization; (2) Differential Privacy; (3) Adversarial Learning; (4) Obfuscation. Existing methods (such as anonymization and differential privacy) usually reduce HAR recognition accuracy and make it difficult to balance data availability and security. Even if the data is anonymized or noise is added, attackers may still use machine learning technology to infer user behavior, and the risk of privacy leakage still exists. Many methods (such as differential privacy and adversarial sample generation) require a lot of computing resources and are not suitable for real-time HAR applications. In addition, existing technologies mainly focus on how to hide personal information rather than preventing the data itself from being abused. Attackers can still train their own HAR classifiers and extract useful information from the data.

[0003] As can be seen from the above, how to avoid information loss, protect the learnability of human behavior data, prevent unauthorized users from abusing data, and ensure that the data still has the analyzability of the original data in legal sharing and display scenarios are problems to be solved in this field. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method, apparatus, device, and medium for publishing and sharing human motion data, which can avoid information loss, protect the learnability of human motion data, prevent unauthorized users from abusing the data, and ensure that the data remains analyzable in accordance with the original data in legal sharing and display scenarios. The specific solution is as follows:

[0005] In a first aspect, the present application discloses a method for publishing and sharing human motion data, comprising:

[0006] Acquiring multivariate time series data of human motion, performing standardization processing, data segmentation, and feature extraction on the multivariate time series data to obtain processed data including motion categories and corresponding labels;

[0007] Constructing and training a preset autoencoder for each action category, and reconstructing the processed data using the trained preset autoencoder to obtain reconstructed data; the trained preset autoencoder includes an encoder and a decoder;

[0008] The reconstructed data is published and shared.

[0009] Optionally, the acquiring of multivariate time series data of human motion includes:

[0010] Multivariate time series data of human body movements are obtained from preset sensors; the preset sensors include sensors in portable communication devices, accelerometers and gyroscopes in inertial sensors; the multivariate time series data include three-axis acceleration and three-axis angular velocity.

[0011] Optionally, the performing standardization processing, data segmentation, and feature extraction on the multivariate time series data includes:

[0012] The Z-Score normalization technology is used to normalize the multivariate time series data to obtain the standardized data;

[0013] The sliding window method is used to divide the data segments in the standardized data to obtain the divided data, and the features of the divided data are extracted.

[0014] Optionally, build a pre-built autoencoder for each action category, including:

[0015] Construct an encoder based on one-dimensional convolutional layers and fully connected layers;

[0016] Build a decoder based on the deconvolution layer;

[0017] A preset autoencoder including the encoder and the decoder is constructed for each action category.

[0018] Optionally, the process of training a preset autoencoder is:

[0019] Using the mean square error of the preset autoencoder as a loss function;

[0020] The loss function is minimized and the preset autoencoder is iteratively trained using the Adam optimizer.

[0021] Optionally, reconstructing the processed data using the trained preset autoencoder includes:

[0022] The trained preset autoencoder is used to reconstruct the processed data, and the properties of the model fingerprint are used to introduce the preset model fingerprint perturbation during the data reconstruction process.

[0023] In a second aspect, the present application discloses a device for publishing and sharing human motion data, comprising:

[0024] A data acquisition and processing module is used to acquire multivariate time series data of human body movements, perform standardization processing, data segmentation, and feature extraction on the multivariate time series data to obtain processed data including movement categories and corresponding labels;

[0025] A data reconstruction module is used to construct and train a preset autoencoder for each action category, and use the trained preset autoencoder to reconstruct the processed data to obtain reconstructed data; the trained preset autoencoder includes an encoder and a decoder;

[0026] The reconstructed data publishing and sharing module is used to publish and share the reconstructed data.

[0027] Optionally, the data acquisition and processing module includes:

[0028] A data acquisition module is used to obtain multivariate time series data of human body movements from preset sensors; the preset sensors include sensors in portable communication devices, accelerometers and gyroscopes in inertial sensors; the multivariate time series data includes three-axis acceleration and three-axis angular velocity.

[0029] In a third aspect, the present application discloses an electronic device, comprising:

[0030] Memory, used to store computer programs;

[0031] The processor is used to execute the computer program to implement the aforementioned method for publishing and sharing human motion data.

[0032] In a fourth aspect, the present application discloses a computer storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the steps of the aforementioned method for publishing and sharing human motion data are implemented.

[0033] It can be seen that the present application provides a method for publishing and sharing human motion data, including obtaining multivariate time series data of human motion, standardizing, segmenting and extracting features of the multivariate time series data to obtain processed data including motion categories and corresponding labels; constructing and training a preset autoencoder for each motion category, and reconstructing the processed data using the trained preset autoencoder to obtain reconstructed data; the trained preset autoencoder includes an encoder and a decoder; and publishing and sharing the reconstructed data. This application performs standardization, data segmentation, and feature extraction on multivariate time series data to obtain processed data including action categories and corresponding labels, and then constructs and trains a preset autoencoder for each action category, uses the trained preset autoencoder to reconstruct the processed data, and uses the preset autoencoder to reconstruct the processed data so that the data of each category can only be restored by the corresponding autoencoder, and the data still retains the characteristics of the original category, preventing unauthorized users from abusing the data. In the process of data reconstruction, there is no need for additional calculation of complex perturbations, the computational cost is low, and it can be applicable to a variety of real-time human motion data tasks, thereby expanding the scope of application, avoiding information loss, protecting the learnability of human behavior motion data, and ensuring that the data is still analyzable in accordance with the original data in legal sharing and display scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0035] Figure 1 This is a flow chart of a method for publishing and sharing human motion data disclosed in this application;

[0036] Figure 2 A visualization diagram of part of the motion data collected by a sensor worn by a subject disclosed in this application;

[0037] Figure 3 A flow chart for training a class-specific autoencoder disclosed in this application;

[0038] Figure 4 This is a system framework diagram for publishing and sharing human motion data disclosed in this application;

[0039] Figure 5 A graph showing the classification accuracy of a random forest classifier trained on an original dataset and a reconstructed dataset disclosed in this application;

[0040] Figure 6 This is a schematic diagram of the structure of a device for publishing and sharing human motion data disclosed in this application;

[0041] Figure 7 This is a structural diagram of an electronic device provided in this application. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0043] HAR is an important research direction in machine learning and sensing technology, and is widely used in medical health monitoring, smart home, security monitoring and other fields. At present, HAR mainly relies on wearable sensors or computer vision technology to collect data and classify and identify it through machine learning or deep learning models. However, with the widespread application of HAR data, user privacy protection has become a key challenge. Existing related technologies mainly include the following categories: (1) Anonymization; (2) Differential Privacy; (3) Adversarial Learning; (4) Obfuscation. Existing methods (such as anonymization and differential privacy) usually reduce HAR recognition accuracy and make it difficult to balance data availability and security. Even if the data is anonymized or noise is added, attackers may still use machine learning technology to infer user behavior, and the risk of privacy leakage still exists. Many methods (such as differential privacy and adversarial sample generation) require a lot of computing resources and are not suitable for real-time HAR applications. In addition, existing technologies mainly focus on how to hide personal information rather than preventing the data itself from being abused. Attackers can still train their own HAR classifiers and extract useful information from the data. As can be seen from the above, how to avoid information loss, protect the learnability of human behavior data, prevent unauthorized users from abusing data, and ensure that the data still has the analyzability of the original data in legal sharing and display scenarios are problems to be solved in this field.

[0044] See also Figure 1 As shown, the embodiment of the present invention discloses a method for publishing and sharing human motion data, which may specifically include:

[0045] Step S11: acquiring multivariate time series data of human body movements, performing standardization processing, data segmentation, and feature extraction on the multivariate time series data to obtain processed data including movement categories and corresponding labels.

[0046] In this embodiment, multivariate time series data of human motion is obtained from preset sensors; the preset sensors include sensors in portable communication devices, accelerometers and gyroscopes in inertial sensors, the multivariate time series data is standardized using Z-Score normalization processing technology to obtain standardized data, the data segments in the standardized data are divided using a sliding window method to obtain divided data, and feature extraction is performed on the divided data; the multivariate time series data includes three-axis acceleration and three-axis angular velocity.

[0047] Taking the PAMAP2 dataset as an example, the sensors collect multivariate time series data of human motion, such as three-axis acceleration and three-axis angular velocity. The visualization of some motion data collected by the subjects wearing sensors is shown in the following figure: Figure 2 As shown in the figure, the sensors come from the accelerometer and gyroscope in the inertial sensor of a smartphone or IMU (Inertial Measurement Unit). The raw data is normalized using Z-Score to balance the dimensions, and the sliding window method is used to divide the data segments, and the action categories are encoded into integer labels (such as: walking = 0, sitting = 1, etc.).

[0048] Step S12: constructing and training a preset autoencoder for each action category, and reconstructing the processed data using the trained preset autoencoder to obtain reconstructed data; the trained preset autoencoder includes an encoder and a decoder.

[0049] In this embodiment, the process of constructing and training a preset autoencoder is as follows: constructing an encoder based on a one-dimensional convolutional layer and a fully connected layer; constructing a decoder based on a deconvolutional layer; constructing a preset autoencoder including the encoder and the decoder for each action category, using the mean square error of the preset autoencoder as a loss function, minimizing the loss function, and iteratively training the preset autoencoder using the Adam optimizer.

[0050] In this embodiment, after constructing and training the preset autoencoder, the trained preset autoencoder is used to reconstruct the processed data, and in the process of data reconstruction, the properties of the Model Fingerprint are used to introduce the preset model fingerprint perturbation to obtain reconstructed data.

[0051] The core of this application is a class-specific autoencoder training method for protecting HAR data and preventing unauthorized users from abusing the data. By training a separate autoencoder for each action category, while ensuring the learnability of the data, the model fingerprint perturbation is introduced, making it difficult for unauthorized classifiers to generalize. An autoencoder is trained for each action category, and data reconstruction is performed as follows: Figure 3 As shown:

[0052] Given a human action recognition dataset:

[0053] ;

[0054] in, Input data for HAR sensors (such as acceleration, gyroscope data), is the corresponding action category, and C is the total number of categories.

[0055] The goal of this invention is to train a separate autoencoder for each category c and use it to reconstruct the data to add model fingerprint perturbations, causing the data to lose generalization ability during the test phase. For each action category c, a class-specific autoencoder is defined, which consists of an encoder and a decoder:

[0056] (1) Encoder: The encoder is responsible for converting the original data Mapping to low-dimensional latent variable space :

[0057] ;

[0058] in, is a shaper of category c, is the parameter of the encoder;

[0059] The encoder usually consists of three layers of Conv1D (one-dimensional convolutional layers) to extract temporal features and reduce the dimension to the latent variable space through FC (Fully Connected Layers);

[0060] (2) Decoder: The decoder is used to extract latent variables Restore input data:

[0061] ;

[0062] in, is the decoder for category c, is the parameter of the decoder;

[0063] The decoder usually uses 3 layers of Deconv1D (deconvolution) to restore the original data format.

[0064] In addition, the training process of the preset autoencoder is:

[0065] (1) Loss function

[0066] The goal of the autoencoder is to minimize the reconstruction error, using MSE (Mean-Square Error) as the loss function:

[0067] ;

[0068] (2) Training algorithm:

[0069] # Input: training dataset D = {(x_i, y_i)}, number of training rounds Tmax, learning rate eta

[0070] # Output: trained class-specific autoencoder AE_c

[0071] initialize autoencoders AE_c for each class c

[0072] for c in range(1, C+1):

[0073] extract class-specific data D_c = {x_i | y_i = c}

[0074] for epoch in range(1, Tmax+1):

[0075] for x_i in D_c:

[0076] Encoding and Decoding

[0077] z_i = AE_c.encoder(x_i)

[0078] x_hat_i = AE_c.decoder(z_i)

[0079] # Calculate loss

[0080] loss = MSE(x_i, x_hat_i)

[0081] # Back propagation update parameters

[0082] AE_c.update_parameters(loss, learning_rate=eta)

[0083] (3) Training steps:

[0084] (a) Initialize C independent autoencoders ;

[0085] (b) Traverse all categories c;

[0086] Extract the dataset of category c ,train , use Adam optimizer (learning rate 0.001) to train for 200 rounds or stop when the loss converges;

[0087] (c). After training, each class c corresponds to an autoencoder .

[0088] Step S13: publishing and sharing the reconstructed data.

[0089] In this embodiment, the trained autoencoder described above is used to reconstruct the data to obtain a reconstructed output, which is published and shared as protected data instead of the original data, thereby achieving learnability protection.

[0090] To overcome the contradiction between privacy protection and data availability in existing technologies, the present invention proposes a method for publishing and sharing human motion data to protect the learnability of HAR data, prevent unauthorized users from abusing the data, and ensure that the data remains analyzable in accordance with the original data in legal sharing and display scenarios. Existing technologies mainly use methods such as anonymization, differential privacy, adversarial samples, and data obfuscation to protect the privacy of HAR data. The present invention reconstructs data through a class-specific autoencoder and introduces model fingerprint perturbations, making the data usable during training but losing its generalization ability during testing. This effectively prevents unauthorized users from using the data for learning while not affecting the normal use of legitimate users.

[0091] The system modules of this application are as follows Figure 4 Shown, including:

[0092] Data acquisition module: obtains raw human motion data, usually from accelerometers, gyroscopes or smart device sensors;

[0093] Data preprocessing module: standardizes, segments, and extracts features from raw data to create a format suitable for machine learning training;

[0094] Class-specific autoencoder module: trains an independent autoencoder for each class to learn class-specific features and perform data reconstruction;

[0095] Classifier training module: uses the reconstructed data to train the classifier to ensure high classification accuracy on the training set;

[0096] Testing and evaluation module: Verify that the generalization ability of unauthorized classifiers on the test set is reduced to ensure data privacy protection.

[0097] The present invention introduces a classifier training module and a test and evaluation module in the simulation link to verify whether the data processed by the present invention can effectively prevent unauthorized classifier generalization learning:

[0098] Classifier training module:

[0099] (1) Objective: To train a classifier using the data processed by the present invention, ensuring that it achieves high classification accuracy on the training set, but the accuracy decreases on the test data;

[0100] (2) Training process: Initialize the classifier (such as the LSTM-CNN combined model); use the data processed by the present invention for training, and the optimization goal is to minimize the cross entropy loss:

[0101] ;

[0102] in, The class probability predicted by the model;

[0103] (3) Optimization algorithm: Adam is used, the learning rate is set to 0.001, and training is performed for 100 rounds or until the loss converges;

[0104] Testing and Assessment Modules:

[0105] (1) Objective: To test the data processed by the present invention on the classifier to verify its generalization ability is reduced and ensure data privacy protection. The main indicator for measuring the effectiveness of the learning data protection mechanism is to compare the classification accuracy on the training dataset and the test dataset to ensure that the protected data achieves the same high performance during training but cannot generalize on the test set;

[0106] (2) Evaluation process: Use A to train the classifier; use the test data to test the classification performance of the classifier.

[0107] In order to evaluate the performance of the present invention, experiments were conducted on three widely used HAR datasets: UCI HAR dataset, WISDM dataset, and PAMAP2 dataset. All experiments were run 5 times on machines with the following specifications: Intel Xeon Gold 5120@CPU, NVIDIA GeForce 3090@GPU, 128 GB@RAM. The PyTorch framework was used to train the autoencoder and classifier. For each dataset, a classifier was trained on the original data and the reconstructed data, and the classification accuracy was evaluated on the training set and the test set. The classification accuracy results of the random forest classifier trained on the original dataset and the reconstructed dataset are shown in Figure 2. Figure 5 As shown in Figure 2. It can be observed that in the training set, the classifier achieves the same high accuracy for both the original data and the reconstructed data, with little difference in performance. In the test set, the performance of the classifier on the reconstructed data is significantly lower than that on the original data, indicating that the reconstructed data is more difficult to generalize. Figure 5It can be seen that although the reconstructed data achieves a higher consistency than the original data, its generalization ability on the test set is obviously impaired, which shows that the present invention can effectively prevent unauthorized classifiers from learning meaningful patterns in protected data.

[0108] The key points of this application are: class-specific autoencoder reconstruction, using independent class-specific autoencoders to reconstruct HAR data, so that each category of data can only be restored by the corresponding autoencoder, and the data still retains the characteristics of the original category, but because the reconstruction process introduces model-specific perturbations, unauthorized classifiers find it difficult to learn useful features and lose generalization, thereby preventing data abuse; with the help of model fingerprint characteristics, during the reconstruction process, the model is guided to learn "shortcut learning (shortcut features)", so that it can learn efficiently during training, but loses generalization ability during testing. Unauthorized models perform well on training data, but the accuracy on test data drops significantly, fundamentally eliminating data abuse; low computational overhead, suitable for real-time applications. Compared with adversarial samples or differential privacy methods, data reconstruction only requires training independent class-specific autoencoders, without the need for additional calculation of complex perturbations, and has low computational cost. It can be applied to real-time HAR tasks such as smart homes, medical monitoring, and mobile devices, thereby expanding the scope of application.

[0109] In this embodiment, multivariate time series data of human motion is obtained, and the multivariate time series data is subjected to standardization, data segmentation, and feature extraction to obtain processed data including motion categories and corresponding labels; a preset autoencoder is constructed and trained for each motion category, and the processed data is reconstructed using the trained preset autoencoder to obtain reconstructed data; the trained preset autoencoder includes an encoder and a decoder; and the reconstructed data is published and shared. This application performs standardization, data segmentation, and feature extraction on multivariate time series data to obtain processed data including action categories and corresponding labels, and then constructs and trains a preset autoencoder for each action category, uses the trained preset autoencoder to reconstruct the processed data, and uses the preset autoencoder to reconstruct the processed data so that the data of each category can only be restored by the corresponding autoencoder, and the data still retains the characteristics of the original category, preventing unauthorized users from abusing the data. In the process of data reconstruction, there is no need for additional calculation of complex perturbations, the computational cost is low, and it can be applicable to a variety of real-time human motion data tasks, thereby expanding the scope of application, avoiding information loss, protecting the learnability of human behavior motion data, and ensuring that the data is still analyzable in accordance with the original data in legal sharing and display scenarios.

[0110] See also Figure 6As shown, the embodiment of the present invention discloses a device for publishing and sharing human motion data, which may specifically include:

[0111] The data acquisition and processing module 11 is used to acquire multivariate time series data of human body movements, perform standardization processing, data segmentation, and feature extraction on the multivariate time series data to obtain processed data including movement categories and corresponding labels;

[0112] A data reconstruction module 12 is configured to construct and train a preset autoencoder for each action category, and to reconstruct the processed data using the trained preset autoencoder to obtain reconstructed data; the trained preset autoencoder includes an encoder and a decoder;

[0113] The reconstructed data publishing and sharing module 13 is used to publish and share the reconstructed data.

[0114] In this embodiment, multivariate time series data of human motion is obtained, and the multivariate time series data is subjected to standardization, data segmentation, and feature extraction to obtain processed data including motion categories and corresponding labels; a preset autoencoder is constructed and trained for each motion category, and the processed data is reconstructed using the trained preset autoencoder to obtain reconstructed data; the trained preset autoencoder includes an encoder and a decoder; and the reconstructed data is published and shared. This application performs standardization, data segmentation, and feature extraction on multivariate time series data to obtain processed data including action categories and corresponding labels, and then constructs and trains a preset autoencoder for each action category, uses the trained preset autoencoder to reconstruct the processed data, and uses the preset autoencoder to reconstruct the processed data so that the data of each category can only be restored by the corresponding autoencoder, and the data still retains the characteristics of the original category, preventing unauthorized users from abusing the data. In the process of data reconstruction, there is no need for additional calculation of complex perturbations, the computational cost is low, and it can be applicable to a variety of real-time human motion data tasks, thereby expanding the scope of application, avoiding information loss, protecting the learnability of human behavior motion data, and ensuring that the data is still analyzable in accordance with the original data in legal sharing and display scenarios.

[0115] In some specific embodiments, the data acquisition and processing module 11 may specifically include:

[0116] A data acquisition module is used to obtain multivariate time series data of human body movements from preset sensors; the preset sensors include sensors in portable communication devices, accelerometers and gyroscopes in inertial sensors; the multivariate time series data includes three-axis acceleration and three-axis angular velocity.

[0117] In some specific embodiments, the data acquisition and processing module 11 may specifically include:

[0118] The standardization processing module is used to standardize the multivariate time series data using the Z-Score normalization processing technology to obtain the standardized data;

[0119] The partitioning module is used to partition the data segments in the standardized data using a sliding window method to obtain the partitioned data and perform feature extraction on the partitioned data.

[0120] In some specific embodiments, the data reconstruction module 12 may specifically include:

[0121] Encoder building module, used to build an encoder based on one-dimensional convolutional layers and fully connected layers;

[0122] Decoder building module, used to build a decoder based on the deconvolution layer;

[0123] The preset autoencoder construction module is used to construct a preset autoencoder including the encoder and the decoder for each action category.

[0124] In some specific embodiments, the data reconstruction module 12 may specifically include:

[0125] A loss function determination module, configured to use the mean square error of the preset autoencoder as a loss function;

[0126] The iterative training module is used to minimize the loss function and iteratively train the preset autoencoder using the Adam optimizer.

[0127] In some specific embodiments, the data reconstruction module 12 may specifically include:

[0128] The perturbation introduction module is used to reconstruct the processed data using the trained preset autoencoder, and introduce the preset model fingerprint perturbation by using the properties of the Model Fingerprint during the data reconstruction process.

[0129] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps of the method for publishing and sharing human motion data performed by an electronic device as disclosed in any of the aforementioned embodiments.

[0130] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0131] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon include an operating system 221, a computer program 222 and data 223, etc. The storage method can be temporary storage or permanent storage.

[0132] Among them, the operating system 221 is used to manage and control the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to calculate and process the data 223 in the memory 22. It can be Windows, Unix, Linux, etc. In addition to including computer programs that can be used to complete the human body motion data publishing and sharing method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include computer programs that can be used to complete other specific tasks. In addition to including data transmitted from external devices received by the human body motion data publishing and sharing device, the data 223 can also include data collected by its own input and output interface 25.

[0133] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0134] Furthermore, an embodiment of the present application also discloses a computer-readable storage medium, in which a computer program is stored. When the computer program is loaded and executed by a processor, the steps of the method for publishing and sharing human motion data disclosed in any of the aforementioned embodiments are implemented.

[0135] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0136] The above is a detailed introduction to the method, device, equipment and storage medium for publishing and sharing human motion data provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for publishing and sharing human motion data, characterized in that: include: Acquiring multivariate time series data of human motion, performing standardization processing, data segmentation, and feature extraction on the multivariate time series data to obtain processed data including motion categories and corresponding labels; Constructing and training a preset autoencoder for each action category, and reconstructing the processed data using the trained preset autoencoder to obtain reconstructed data; The trained preset autoencoder includes an encoder and a decoder; The reconstructed data is published and shared.

2. The method for publishing and sharing human motion data according to claim 1, characterized in that: The method of obtaining multivariate time series data of human body movements includes: Multivariate time series data of human body movements are obtained from preset sensors; the preset sensors include sensors in portable communication devices, accelerometers and gyroscopes in inertial sensors; the multivariate time series data include three-axis acceleration and three-axis angular velocity.

3. The method for publishing and sharing human motion data according to claim 1, characterized in that: The standardization processing, data segmentation and feature extraction of the multivariate time series data include: The Z-Score normalization technology is used to normalize the multivariate time series data to obtain the standardized data; The sliding window method is used to divide the data segments in the standardized data to obtain the divided data, and the features of the divided data are extracted.

4. The method for publishing and sharing human motion data according to claim 1, characterized in that: Build preset autoencoders for each action category, including: Construct an encoder based on one-dimensional convolutional layers and fully connected layers; Build a decoder based on the deconvolution layer; A preset autoencoder including the encoder and the decoder is constructed for each action category.

5. The method for publishing and sharing human motion data according to claim 1, characterized in that: The process of training the preset autoencoder is: Using the mean square error of the preset autoencoder as a loss function; The loss function is minimized and the preset autoencoder is iteratively trained using the Adam optimizer.

6. The method for publishing and sharing human motion data according to any one of claims 1 to 5, characterized in that: The reconstructing the processed data by using the trained preset autoencoder includes: The trained preset autoencoder is used to reconstruct the processed data, and the properties of the model fingerprint are used to introduce the preset model fingerprint perturbation during the data reconstruction process.

7. A device for publishing and sharing human motion data, characterized in that: include: A data acquisition and processing module is used to acquire multivariate time series data of human body movements, perform standardization processing, data segmentation, and feature extraction on the multivariate time series data to obtain processed data including movement categories and corresponding labels; A data reconstruction module is used to construct and train a preset autoencoder for each action category, and reconstruct the processed data using the trained preset autoencoder to obtain reconstructed data; The trained preset autoencoder includes an encoder and a decoder; The reconstructed data publishing and sharing module is used to publish and share the reconstructed data.

8. The device for publishing and sharing human motion data according to claim 7, characterized in that: The data acquisition and processing module includes: A data acquisition module is used to obtain multivariate time series data of human body movements from preset sensors; the preset sensors include sensors in portable communication devices, accelerometers and gyroscopes in inertial sensors; the multivariate time series data includes three-axis acceleration and three-axis angular velocity.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method for publishing and sharing human motion data as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that Used to store a computer program; wherein, when the computer program is executed by a processor, the method for publishing and sharing human motion data as described in any one of claims 1 to 6 is implemented.

Images