Identity authentication method and system based on sensor gait data

Abstract

The invention relates to the technical field of identity authentication, and provides an identity authentication method and system based on sensor gait data, and the method comprises the steps: extracting gait biological features through a gait mixed feature extraction model based on the sensor gait data of a to-be-authenticated person; similarity comparison is carried out on the gait biological characteristics of the person to be authenticated in a characteristic comparison library, and identity authentication is carried out; according to the gait mixed feature extraction model, a full convolutional network feature extractor is adopted to extract spatial local features in sensor gait data, and a bidirectional gating loop network feature extractor is adopted to capture a time sequence dependency relationship from bidirectional time sequence changes of the sensor gait data. A depth separable convolutional network feature extractor is adopted to extract spatial features and channel features of sensor gait data in sequence, and a gating network is adopted to dynamically adjust the input weight and the output weight of each feature extractor according to the features of the sensor gait data. And the identity authentication efficiency and precision are ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of identity authentication, and particularly relates to an identity authentication method and system based on sensor gait data. Background Art

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Identity authentication technology plays a crucial role in modern society. With the in-depth development of digitalization, networking, and intelligence, the demands and challenges of identity authentication are increasing day by day. The core task of identity authentication technology is to ensure that the user is the claimed identity when accessing the system, prevent unauthorized access, and thus guarantee the security of data, systems, and services.

[0004] Since human gait is formed by the coordination between the human skeletal, nervous, and muscular systems, gait features are a unique and difficult-to-imitate biometric feature. In recent years, gait-based identity authentication methods have received extensive attention. Compared with video gait, sensor gait recognition methods are not easily affected by external interference and have a lower computational complexity, which is considered a very promising solution in real-time applications.

[0005] Due to the powerful feature representation and extraction capabilities of deep neural networks, deep learning methods have been widely applied to the field of sensor gait recognition, but there are certain disadvantages:

[0006] The method based on sequence classification uses sensor gait to train an identity recognition classifier and performs identity recognition through this classifier. The advantage of this type of method is high identity recognition accuracy, and the disadvantage is that the model cannot register new users and has poor scalability;

[0007] The method based on feature matching can flexibly add new users and new gait feature templates, but the disadvantage is lower recognition accuracy. Summary of the Invention

[0008] In order to solve the technical problems existing in the above background art, the present invention provides an identity authentication method and system based on sensor gait data, which uses different feature extractors to respectively extract the spatial local features, temporal dependence relationships, spatial features, and channel features of sensor gait data, and dynamically adjusts the input weights and output weights of each feature extractor according to the features of sensor gait data through a gating network, realizing the expansion of the richness of extracted features while consuming lower computing resources, and at the same time ensuring the efficiency and accuracy of identity authentication.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] A first aspect of the present invention provides an identity authentication method based on sensor gait data, comprising:

[0011] Obtain sensor gait data of the person to be authenticated;

[0012] Based on the sensor gait data, gait biometrics are extracted through the gait hybrid feature extraction model;

[0013] For the gait biometric features of the person to be authenticated, a similarity comparison is performed in the feature comparison library to perform identity authentication;

[0014] Among them, the gait hybrid feature extraction model uses a fully convolutional network feature extractor to extract spatial local features in sensor gait data, a bidirectional gated recurrent network feature extractor to capture temporal dependencies from the bidirectional temporal changes of sensor gait data, and a deep separable convolutional network feature extractor to sequentially extract spatial features and channel features of sensor gait data. A gated network is used to dynamically adjust the input weights and output weights of each feature extractor according to the characteristics of sensor gait data.

[0015] Furthermore, the gait hybrid feature extraction model uses triplet loss and classification loss for joint training during the training process, and adjusts the weights of the two losses through learnable weight parameters. The learnable weight parameters separately configure an optimizer and dynamically learn and adjust according to the gradient size of the two parts of the loss.

[0016] Furthermore, the sensor gait data is input into a gait mixed feature extraction model after data preprocessing, and the data preprocessing step includes: performing Butterworth filtering on the accelerometer signal in the sensor gait data and then splicing it into the sensor gait data; and using a sliding window method to segment the spliced ​​sensor gait data.

[0017] Furthermore, the similarity comparison step includes: performing a similarity comparison between the gait biometric features of the person to be authenticated and all gait biometric features in a constructed feature comparison library; comparing the maximum similarity in the similarity comparison result with a set threshold value, if the maximum similarity is greater than the threshold value, the authentication is successful; otherwise, it is determined to be an illegal intrusion.

[0018] Furthermore, the sensor gait data is Among them, acc i is the accelerometer signal, The dimensions are 1×L i vector; gry i is the gyroscope signal, The dimensions are 1×L i Vector; L iDetermined by the acquisition duration and the sampling frequency; x, y, and z respectively represent the x-axis signal, y-axis signal, and z-axis signal of the sensor.

[0019] Furthermore, the gait hybrid feature extraction model further includes a spatio-temporal attention mechanism, a convolutional layer, and a fully connected network connected in sequence; after the output features of each feature extractor are concatenated in the channel dimension, they are processed by the spatio-temporal attention mechanism; the concatenated features after passing through the spatio-temporal attention mechanism are fused by the convolutional layer; the features after fusion are converted into gait biometric features by the fully connected network.

[0020] Furthermore, it further includes: obtaining the sensor gait data of the registered person, processing the sensor gait data of the registered person by using the feature extraction model, extracting the gait biometric features, and storing the gait biometric features of the registered person in the feature comparison library.

[0021] The second aspect of the present invention provides an identity authentication system based on sensor gait data, which includes:

[0022] A data acquisition module, which is configured to: acquire the sensor gait data of the person to be authenticated;

[0023] A feature extraction module, which is configured to: based on the sensor gait data, extract the gait biometric features through the gait hybrid feature extraction model;

[0024] A feature comparison module, which is configured to: perform similarity comparison on the gait biometric features of the person to be authenticated in the feature comparison library for identity authentication;

[0025] Among them, the gait hybrid feature extraction model uses a fully convolutional network feature extractor to extract the spatial local features in the sensor gait data, uses a bidirectional gated recurrent network feature extractor to capture the temporal dependence relationship from the bidirectional temporal changes of the sensor gait data, uses a depthwise separable convolutional network feature extractor to sequentially extract the spatial features and channel features of the sensor gait data, and uses a gated network to dynamically adjust the input weights and output weights of each feature extractor according to the features of the sensor gait data.

[0026] The third aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps in an identity authentication method based on sensor gait data as described above.

[0027] The fourth aspect of the present invention provides a computer device, including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, and when the processor executes the program, it implements the steps in an identity authentication method based on sensor gait data as described above.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] The present invention uses different feature extractors to respectively extract the spatial local features, temporal dependence relationships, spatial features, and channel features of sensor gait data, and dynamically adjusts the input weights and output weights of each feature extractor according to the features of the sensor gait data through a gating network, achieving the expansion of the richness of the extracted features while using relatively low computational resource consumption, and at the same time ensuring the computational efficiency and authentication accuracy.

[0030] The present invention jointly trains with a triplet loss structure and a classification loss, and adjusts the proportion of the two in the total loss through learnable weight parameters, preventing the possible loss imbalance problem in manually adjusting the loss weights. The joint training of the two takes into account both the accuracy of the gait mixed feature extraction model in the identity recognition task and the generalizability of flexibly dealing with the addition of unfamiliar samples in the identity authentication task. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0032] Figure 1 It is a flowchart of registration and identity authentication provided in Embodiment 1 of the present invention;

[0033] Figure 2 It is a schematic diagram of the data preprocessing process provided in Embodiment 1 of the present invention;

[0034] Figure 3 It is a schematic diagram of the structure of a hybrid feature extraction network (MFEN) provided in Embodiment 1 of the present invention;

[0035] Figure 4 It is a schematic diagram of the structures of three feature extractors provided in Embodiment 1 of the present invention;

[0036] Figure 5 It is a schematic diagram of the loss structure provided in Embodiment 1 of the present invention;

[0037] Figure 6 It is a flowchart of identity authentication provided in Embodiment 1 of the present invention;

[0038] Figure 7 It is a schematic diagram of the structure of a computer device in Embodiment 4 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0040] It should be noted that the following detailed description is illustrative and aims to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0041] Example 1

[0042] This example provides an identity authentication method based on sensor gait data.

[0043] An identity authentication method based on sensor gait data provided in this example, taking the gait data of 156 users in a walking scenario as an example, includes the following steps:

[0044] Step (1), collect the inertial sensor gait data of N (N = 156 in this example) users, and construct a labeled gait sequence set S for training a gait hybrid feature extraction model, specifically as follows:

[0045] Considering the built-in accelerometer and gyroscope of the mobile phone as input data sources, the gait sequence of the i-th user i = 1, 2,..., N, acc i is the accelerometer signal, gry i is the gyroscope signal, represents the transpose operation; thus, construct a gait sequence set S = {S1, S2,..., S N}.

[0046] Among them, the accelerometer signal of the i-th user where is a vector with a dimension of 1 × L i , acc i is a matrix with a dimension of L i ×3, L i is determined by the acquisition duration d i (d i = 15s in this example) and the sampling frequency f s (f s = 50Hz in this example), L i = d i ×f s (L i = 750 in this example), and x, y, and z respectively represent the x-axis signal, y-axis signal, and z-axis signal of the sensor.

[0047] The gyroscope signal of the i-th user where is a vector with a dimension of 1 × L i , gry iis a matrix of dimension L i ×3.

[0048] Thus, S i is a matrix of dimension 6×L i .

[0049] For the gait sequence S of the i-th user i add the user identity label Tag i,user (in this embodiment, Tag i,user ∈{0,1,2,…,155}), and the gait sequence label T of the i-th user i = Tag i,user . Thus, the label set T corresponding to the gait sequence set S of N gait sequences is formed as T = {T1, T2, …, T N}}.

[0050] Step (2): Perform data preprocessing on the original data set S. Specifically: The original gait sequence S i is regarded as being composed of 6 subsequences s i of length L i,h . where s i,h represents the h-th subsequence in the gait sequence of the i-th user, and is a vector of dimension 1×L i , and h = 1, 2, …, 6; Perform data preprocessing on the gait sequence S i , take the data of the accelerometer part among them for Butterworth filtering to remove the low-frequency components, and splice them into the original data to obtain Perform data preprocessing on each subsequence s i,h using a sliding window for data segmentation operation.

[0051] The following is the specific process of data preprocessing, and the process is as Figure 2 shown:

[0052] Step 201: For the walking data, each subject uploaded 8 - 12 pieces of data, the acquisition time of each piece of data was 15s, the sampling frequency was 50Hz, and the length of each piece of data was 750 time points;

[0053] Step 202: Perform Butterworth filtering on the data of the accelerometer part in each piece of original data to filter out the gravity component, and splice it with the original data to form a data set of 9×L i ;

[0054] Step 203: Use the sliding window method to segment 1716 pieces of original data (there were 156 participants in the data acquisition process, and each participant uploaded 11 pieces of walking data). The sliding window size was 128 and the step size was 58, obtaining 15444 pieces of segmented gait data.

[0055] Step (3): Construct and train a gait hybrid feature extraction network to process gait sequence data and extract the gait features contained in the gait sequence, facilitating subsequent feature template registration and comparison.

[0056] As Figure 3 shown, the constructed gait hybrid feature extraction network (MFEN) includes the following parts:

[0057] Construct a fully convolutional network (FCN), a bidirectional gated recurrent unit network (Stack BiGRU Network), and a depthwise separable convolutional network (DSCNN) as three feature extractors;

[0058] Construct a gate network (Gate Network) to assign the input weights of the original data and the output weights of different feature extractors to the three feature extractors, and together with the feature extractors, form the feature extraction layer in the feature extraction network;

[0059] Construct a spatio-temporal attention mechanism (CBAM). After concatenating the output features of various feature extractors in the channel dimension, input them into the spatio-temporal attention mechanism to focus on the more important parts of the features;

[0060] Construct a convolutional layer (CNN, with a convolutional kernel size of 3 and the number of convolutional kernels being 1) to fuse the concatenated features after passing through the spatio-temporal attention mechanism, and together with the spatio-temporal attention mechanism, form the feature fusion layer in the feature extraction network;

[0061] Construct a fully connected network (FC) as the feature output layer to convert the fused attention features into the finally output gait features.

[0062] It should be noted that both the input weight and the output weight are the output values of the gate network. The dimension of the input weight is 1×3, and the sum of the weight values is 1. For example: the input weight is [i1, i2, i3], and the original data input matrix is inp, then i1 + i2 + i3 = 1, and the inputs of the three feature extractors are inp * i1, inp * i2, inp * i3 respectively. Similarly, the dimension of the output weight is 1×3, and the sum of the weight values is 1. For example: the output weight is [o1, o2, o3], and the output feature vectors of the three feature extractors are out1, out2, out3 respectively, then o1 + o2 + o3 = 1, and before the output vectors of the three feature extractors are concatenated, they need to be weighted and transformed into out1 * o1, out2 * o2, out3 * o3.

[0063] In this embodiment, the construction processes of the three feature extractors are as Figure 4 shown:

[0064] The structure of the fully convolutional network (FCN) includes two sets of convolutional networks composed of one-dimensional convolutional layers (CNN), batch normalization layers (BN), activation layers (Relu), and channel attention mechanisms (SE). Then, a convolutional network composed of one-dimensional convolutional layers, batch normalization layers, and activation layers is added. Finally, the feature map passes through a global average pooling layer in the spatial dimension (GlobalAverage Pooling) to obtain the feature extraction result of the fully convolutional network. Among them, the kernel sizes of the three one-dimensional convolutional layers are 7, 5, and 3 respectively, and the numbers of kernels are 64, 128, and 128 respectively.

[0065] The structure of the stacked bidirectional gated recurrent unit network (Stack BiGRU Network) includes a convolutional network composed of a convolutional layer (CNN, kernel size is 3, number of kernels is 64), a batch normalization layer (BN), an activation layer (Relu), and a channel attention mechanism (SE). Then, two bidirectional gated recurrent units (BiGru) are added. The output of the first bidirectional gated recurrent unit passes through a spatial attention module (SAM) and is skip-connected and added to the output of the second bidirectional gated recurrent unit to obtain the feature extraction result of the stacked bidirectional gated recurrent unit network.

[0066] The structure of the depthwise separable convolutional network (DSCNN) includes two sets of depthwise separable convolutions composed of depthwise convolutions and 1×1 convolutions (Pointwise-Conv). Batch normalization (BN) and activation layers (Relu) are added after the depthwise convolutional layer. Finally, the feature map passes through a global average pooling layer in the spatial dimension (Global AveragePooling) to obtain the feature extraction result of the depthwise separable convolutional network.

[0067] The gait feature extraction network is trained using the training set to obtain the trained gait hybrid feature extraction network.

[0068] Furthermore, the training process of the gait hybrid feature extraction network includes:

[0069] (a) Mine data triples in a dataset containing known identity samples. The triples include anchor data, positive data with the same identity label as the anchor data, and negative data with a different identity label from the anchor data. They are respectively trained through a gait hybrid feature extraction network with shared weights to output corresponding gait features. For example, the training set contains gait sensor data of 20 subjects, and the dimension of each data is 9 * 128 after data preprocessing. If the subject identity labels are represented by [p1, p2,..., p20], and anchor is a randomly selected gait sensor data, assuming the true identity label of the anchor data is p1, then the identity label of the positive data should be p1, and the identity label of the negative data cannot be p1;

[0070] (b) Use the Triplet loss function to perform supervised training on the mined data pairs. Its formula is: L T = max(d(a, p) - d(a, n) + margin, 0), where d(a, p) represents the Euclidean distance between the gait features of the anchor data and the positive data, and conversely, d(a, n) represents the Euclidean distance between the gait features of the anchor data and the negative data, and margin represents the margin;

[0071] (c) Extract the gait features of the positive data samples in each group of triple data groups and input them into the classification layer. Use the Softmax loss for multi-classification training to assist the hybrid feature extractor in improving the feature extraction ability.

[0072] Among them, the classification layer is constructed by a fully connected layer and a softmax function. Taking the data of 20 subjects in the training set and the feature dimension extracted by the hybrid feature extraction network being 1 * 128 as an example, the input and output dimensions of the classification layer are 128 and 20 respectively; among them, the meaning of the output vector with a dimension of 20 of the classification layer is the probability that a certain positive sample belongs to the identities of 20 subjects, and the Softmax loss focuses on its probability on the correct label. The greater the probability on the correct label, the smaller the loss.

[0073] In summary, the data sample mining process and the overall loss structure of the model are as Figure 5 shown.

[0074] Furthermore, in the joint training process of Softmax loss and Triplet loss, add a learnable dynamic weight parameter ω, and the total training loss is Loss = L T + ω·L S, where ω can be dynamically learned and adjusted according to the gradient magnitudes of the two-part loss function during the training process. In this embodiment, an optimizer is separately configured for the dynamic weight parameter ω, and a learning rate of 0.0003 is used to prevent the problem of excessive loss gradient oscillation caused by excessive fluctuations.

[0075] Step (4), register and authenticate the personnel identity. The specific steps are as Figure 1 shown, including:

[0076] Step (401), registration phase. Save and use the gait feature extraction network trained in step (3) to extract the gait identity features of legal personnel for template registration. Specifically: obtain the sensor gait data of the registered personnel, and after data preprocessing, form an input matrix; use the trained feature extraction model to process the input of the registered personnel and extract the gait biometric features; store the gait biometric features of the registered personnel in the database to construct a feature comparison library.

[0077] Step (402), authentication phase. Extract the gait identity features of the personnel to be authenticated and compare them with the features in the template library. If the similarity reaches a certain threshold, it is considered that the authentication is passed; otherwise, the authentication fails. Specifically: obtain the sensor gait data of the personnel to be authenticated, and after data preprocessing, form an input matrix; use the trained feature extraction model to process the input of the personnel to be authenticated and extract the gait biometric features; compare the gait biometric features of this person with all the features in the constructed feature comparison library; calculate the maximum similarity s in the similarity comparison results; use the maximum similarity s and the set threshold value to authenticate the identity of this person. If the maximum similarity s is greater than the threshold value, the authentication is successful; otherwise, it is determined as an illegal intrusion, as Figure 6 shown.

[0078] An identity authentication method based on sensor gait data provided by this embodiment solves the problems of low identity recognition accuracy and poor generalization ability on an untrained stranger dataset through a hybrid feature extraction network (MFEN).

[0079] This embodiment provides an identity authentication method based on sensor gait data, which uses a hybrid feature extraction network to extract gait identity features. In the hybrid feature extraction network, the full convolutional network feature extractor focuses on the spatial local features in the gait signal, the bidirectional gated recurrent unit network feature extractor can capture the temporal dependency from the bidirectional temporal changes of the signal, and the deep separable convolutional network feature extractor can extract spatial features and channel features in sequence. Moreover, due to the small number of convolution kernels, it has a small number of parameters and a faster computing speed. The function of the gated network is to dynamically adjust and configure the input and output weights of each feature extractor according to the characteristics of different input data. In some cases, the gated network can only activate a part of the feature extractors for feature extraction. The effect is to expand the richness of the extracted features of the gait hybrid feature extraction model while using lower computing resource consumption, while ensuring computing efficiency and authentication accuracy.

[0080] This embodiment provides an identity authentication method based on sensor gait data, which adopts joint training of triplet loss structure and classification loss, and adjusts the proportion of the two in the total loss through a learnable weight parameter ω, thereby preventing the loss imbalance problem that may occur in manual adjustment of loss weights. The joint training of the two takes into account the accuracy of the model in identity recognition tasks and the generalizability of the model to flexibly cope with the addition of unfamiliar samples in identity authentication tasks.

[0081] Embodiment 2

[0082] This embodiment provides an identity authentication system based on sensor gait data, which specifically includes:

[0083] A registration module, which is configured to: obtain sensor gait data of the registered person, process the sensor gait data of the registered person using a feature extraction model, extract gait biometric features, and store the gait biometric features of the registered person in a feature comparison library;

[0084] A data acquisition module, which is configured to: acquire sensor gait data of the person to be authenticated;

[0085] A feature extraction module is configured to: extract gait biometric features based on sensor gait data of the person to be authenticated by using a gait hybrid feature extraction model;

[0086] A feature comparison module is configured to: perform similarity comparison on the gait biometric features of the person to be authenticated in a feature comparison library to perform identity authentication;

[0087] Among them, the gait mixed feature extraction model uses a fully convolutional network feature extractor to extract spatial local features from sensor gait data, uses a bidirectional gated recurrent network feature extractor to capture temporal dependencies from the bidirectional temporal changes of sensor gait data, uses a depthwise separable convolutional network feature extractor to sequentially extract spatial features and channel features of sensor gait data, and uses a gated network to dynamically adjust the input weights and output weights of each feature extractor according to the features of sensor gait data.

[0088] Furthermore, during the training process of the gait mixed feature extraction model, triplet loss and classification loss are used for joint training, and the weights of the two losses are adjusted by learnable weight parameters. The learnable weight parameters are configured with a separate optimizer and are dynamically learned and adjusted according to the gradient magnitudes of the two parts of the losses.

[0089] Furthermore, the sensor gait data is input into the gait mixed feature extraction model after data preprocessing. The steps of the data preprocessing include: performing Butterworth filtering on the accelerometer signal in the sensor gait data and then splicing it to the sensor gait data; using the sliding window method to segment the spliced sensor gait data.

[0090] Furthermore, the steps of the similarity comparison include: comparing the gait biometric features of the person to be authenticated with all the gait biometric features in the pre-constructed feature comparison library; comparing the maximum similarity in the similarity comparison results with the set threshold value. If the maximum similarity is greater than the threshold value, the authentication is successful; otherwise, it is determined as an illegal intrusion.

[0091] Furthermore, the sensor gait data is where acc i is the accelerometer signal, both are vectors with a dimension of 1×L i gry i is the gyroscope signal, both are vectors with a dimension of 1×L i L i is determined by the acquisition duration and the sampling frequency; x, y, and z respectively represent the x-axis signal, y-axis signal, and z-axis signal of the sensor.

[0092] Furthermore, the gait mixed feature extraction model further includes a spatio-temporal attention mechanism, a convolutional layer, and a fully connected network connected in sequence; the output features of each feature extractor are spliced in the channel dimension and then processed by the spatio-temporal attention mechanism; the spliced features after passing through the spatio-temporal attention mechanism are fused through the convolutional layer; the features after fusion are converted into gait biometric features through the fully connected network.

[0093] It should be noted here that each module in this embodiment corresponds to each step in Embodiment 1 one by one, and the specific implementation process is the same, so it will not be repeated here.

[0094] Embodiment 3

[0095] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in an identity authentication method based on sensor gait data as described in Embodiment 1 above.

[0096] Embodiment 4

[0097] This embodiment provides a computer device, as Figure 7 shown, including a display device, an input device, a computer-readable storage medium (volatile memory and non-volatile storage medium), a processor, a communication interface (i.e., a network interface), and a computer program stored on the computer-readable storage medium and executable on the processor. Among them, the processor, the communication interface, and the computer-readable storage medium can be connected through a bus or other means. Among them, the communication interface is used to receive and send data, and when the processor executes the program, it implements the steps in an identity authentication method based on sensor gait data as described in Embodiment 1 above.

[0098] Among them, any reference to a memory, storage, database, or other medium provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or an external cache. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0099] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce means for implementing the functions specified in the flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0100] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in the flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0101] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0102] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An identity authentication method based on sensor gait data, characterized in that: include: Obtain sensor gait data of the person to be authenticated; Based on the sensor gait data, gait biometrics are extracted through the gait hybrid feature extraction model; For the gait biometric features of the person to be authenticated, a similarity comparison is performed in the feature comparison library to perform identity authentication; Among them, the gait hybrid feature extraction model uses a fully convolutional network feature extractor to extract spatial local features in sensor gait data, a bidirectional gated recurrent network feature extractor to capture temporal dependencies from the bidirectional temporal changes of sensor gait data, and a deep separable convolutional network feature extractor to sequentially extract spatial features and channel features of sensor gait data. A gated network is used to dynamically adjust the input weights and output weights of each feature extractor according to the characteristics of sensor gait data.

2. The identity authentication method based on sensor gait data as claimed in claim 1, characterized in that: The gait hybrid feature extraction model uses triplet loss and classification loss for joint training during the training process, and adjusts the weights of the two losses through learnable weight parameters. The learnable weight parameters separately configure an optimizer and dynamically learn and adjust according to the gradient size of the two parts of the loss.

3. The identity authentication method based on sensor gait data as claimed in claim 1, characterized in that: The sensor gait data is input into the gait mixed feature extraction model after data preprocessing, and the data preprocessing step includes: performing Butterworth filtering on the accelerometer signal in the sensor gait data and splicing it to the sensor gait data; and using a sliding window method to segment the spliced ​​sensor gait data.

4. The identity authentication method based on sensor gait data as claimed in claim 1, characterized in that: The similarity comparison step includes: performing a similarity comparison between the gait biometric features of the person to be authenticated and all the gait biometric features in the constructed feature comparison library; comparing the maximum similarity in the similarity comparison result with the set threshold value. If the maximum similarity is greater than the threshold value, the authentication is successful; otherwise, it is determined to be an illegal intrusion.

5. The identity authentication method based on sensor gait data as claimed in claim 1, characterized in that: The sensor gait data is Among them, acc i is the accelerometer signal, The dimensions are 1×L i vector; gry i is the gyroscope signal, The dimensions are 1×L i Vector; L i Determined by the acquisition duration and sampling frequency; x, y, z represent the x-axis signal, y-axis signal, and z-axis signal of the sensor, respectively.

6. The identity authentication method based on sensor gait data as claimed in claim 1, characterized in that: The gait hybrid feature extraction model also includes a spatiotemporal attention mechanism, a convolutional layer and a fully connected network connected in sequence; the output features of each feature extractor are spliced ​​in the channel dimension and processed through the spatiotemporal attention mechanism; the spliced ​​features after the spatiotemporal attention mechanism are fused through the convolutional layer; the fused features are converted into gait biological features through the fully connected network.

7. The identity authentication method based on sensor gait data as claimed in claim 1, characterized in that: Also includes: The sensor gait data of the registered person is obtained, the sensor gait data of the registered person is processed using a feature extraction model, the gait biometric features are extracted, and the gait biometric features of the registered person are stored in a feature comparison library.

8. An identity authentication system based on sensor gait data, characterized in that: include: A data acquisition module, which is configured to: acquire sensor gait data of the person to be authenticated; A feature extraction module is configured to: extract gait biometric features based on sensor gait data through a gait hybrid feature extraction model; A feature comparison module is configured to: perform similarity comparison on the gait biometric features of the person to be authenticated in a feature comparison library to perform identity authentication; Among them, the gait hybrid feature extraction model uses a fully convolutional network feature extractor to extract spatial local features in sensor gait data, a bidirectional gated recurrent network feature extractor to capture temporal dependencies from the bidirectional temporal changes of sensor gait data, and a deep separable convolutional network feature extractor to sequentially extract spatial features and channel features of sensor gait data. A gated network is used to dynamically adjust the input weights and output weights of each feature extractor according to the characteristics of sensor gait data.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the identity authentication method based on sensor gait data as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a computer-readable storage medium, a processor, and a computer program stored in the computer-readable storage medium and executable on the processor, characterized in that: When the processor executes the program, the steps in the identity authentication method based on sensor gait data as described in any one of claims 1-7 are implemented.

Images