Gait authentication method and system for shared device based on rotation-invariant features

Abstract

The application belongs to the technical field of implicit identity authentication. A shared device gait identity authentication method and system based on rotation invariant features are proposed. The three-axis acceleration original time sequence of a user walking is collected through the built-in three-axis acceleration sensor of a mobile device. The sequence is preprocessed to remove the gravity component and generate an acceleration window sequence by sliding division. A six-dimensional feature vector with three-dimensional rotation invariance is extracted by vector norm and inner product operation. The features are input into a neural network classifier to obtain an identity discrimination result. Hierarchical verification is performed according to the multi-window result and confidence, and authentication is completed. The application eliminates device wearing orientation interference from the input level, improves cross-session recognition stability, adapts to high-frequency handover scenarios of shared mobile devices, and ensures authentication security and efficiency.

Description

Technical Field

[0001] This invention relates to the field of implicit identity authentication technology, specifically to a shared device gait identity authentication method and system based on rotation-invariant features. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In collaborative scenarios such as hospital ward rounds, field operations, warehousing and logistics, and public services, the shared use of mobile terminals is becoming increasingly common, and frequent handover of devices between different staff members has become the norm. Traditional explicit authentication methods require users to actively enter passwords, verify fingerprints or faces, which are cumbersome and easily affected by the environment, seriously impacting collaboration efficiency. Implicit gait authentication technology based on inertial sensors has emerged to address this need. It relies on the triaxial accelerometer built into mobile devices to collect users' walking motion characteristics without physical contact, completing identity verification without additional interaction. Adaptable to various restricted collaborative scenarios, it has become an important research direction for identity authentication of shared devices.

[0004] Current mainstream gait authentication technologies directly use raw triaxial acceleration signals as feature inputs. However, the wearing orientation of users when carrying devices varies randomly, causing an overall rotational shift in the triaxial acceleration signals acquired by the sensors. Existing technologies lack effective means to eliminate orientation interference at the input representation level, making it impossible to guarantee the stability of features under arbitrary three-dimensional rotations. This makes it difficult to adapt to the complete authentication process of triaxial signal acquisition, preprocessing, rotation-invariant feature extraction, neural network classification, and hierarchical verification. Consequently, the authentication reliability across postures and sessions is insufficient, failing to meet the stable authentication requirements in high-frequency handover scenarios of shared devices. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a shared device gait authentication method and system based on rotation-invariant features. Through a complete process of three-axis acceleration signal acquisition, gravity removal and sliding window preprocessing, six-dimensional rotation-invariant feature extraction, neural network classification, and hierarchical verification, it effectively solves the core problems of signal offset and poor recognition stability caused by changes in device wearing orientation in existing gait authentication. The entire process is lightweight and adaptable to local deployment on mobile devices, requiring no uploading of raw data, while also protecting privacy. It is fully adaptable to high-frequency handover scenarios of shared devices, significantly improving the practicality and stability of gait authentication.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a shared device gait identity authentication method based on rotation-invariant features.

[0007] A shared device gait authentication method based on rotation-invariant features, comprising: The device uses a built-in triaxial accelerometer to collect raw time-series data of the user's triaxial acceleration during walking. Preprocessing is performed on the original triaxial acceleration time series to remove the gravity component and slide segmentation according to preset window parameters to generate an acceleration window series. Based on the acceleration window sequence, a six-dimensional rotation-invariant eigenvector is extracted, which is generated by the operation of vector norm and vector inner product. The six-dimensional rotation-invariant eigenvector is invariant to any three-dimensional spatial rotation. The six-dimensional rotation-invariant feature vector is input into a preset neural network classifier. After feature processing and classification calculation, the user identity determination result is output. Based on the identity recognition results and corresponding confidence levels of multiple consecutive windows, a hierarchical verification strategy is executed to complete the gait identity authentication of the mobile device.

[0008] In one implementation of the first aspect of the present invention, extracting the six-dimensional rotation-invariant feature vector includes: Calculate the magnitude of the triaxial acceleration vector to obtain the acceleration amplitude; Calculate the magnitude of the difference between the three-axis acceleration vectors at adjacent time points to obtain the first-order difference magnitude; Calculate the magnitude of the difference between the first-order difference vectors at adjacent time points to obtain the second-order difference magnitude; Calculate the normalized inner product of the triaxial acceleration vectors at adjacent time points; Calculate the variance of the acceleration amplitude within the preset sliding window to obtain the sliding variance; Calculate the mean of the squares of the magnitudes of the three-axis acceleration vectors within the preset sliding window to obtain the sliding composite power; The acceleration amplitude, first-order difference amplitude, second-order difference amplitude, normalized inner product, sliding variance, and sliding composite power constitute a six-dimensional rotationally invariant eigenvector.

[0009] In one implementation of the first aspect of the present invention, for the triaxial acceleration data within each acceleration window, the mean value of the corresponding axis is subtracted to eliminate the influence of gravity component on the signal; the acceleration data after removing gravity component is truncated according to a preset window length, and then the truncated data is slidably segmented according to a preset sliding step size to generate multiple continuous acceleration window sequences.

[0010] In one implementation of the first aspect of the present invention, the preset neural network classifier sequentially includes a multi-scale one-dimensional convolution module, a bidirectional recurrent unit module, an attention pooling module, and a cosine classification module, which are used to perform temporal feature extraction, dependency modeling, feature aggregation, and identity classification on a six-dimensional rotation-invariant feature vector.

[0011] As a further limitation of the first aspect of the present invention, the multi-scale one-dimensional convolution module is configured with two parallel one-dimensional convolution branches. Each branch includes multi-layer convolution, batch normalization and activation function processing. The output results of the two branches are spliced ​​and fused to output multi-scale temporal features. The bidirectional recurrent unit module receives multi-scale temporal features, models the long-term temporal dependencies of the feature sequences, and outputs temporal features; the attention pooling module assigns attention weights to each time step of the temporal features, and generates a fixed-dimensional gait embedding vector after weighted summation. The cosine classification module performs dimensional projection and normalization on the gait embedding vector, calculates its cosine similarity with the prototype vector of each user category, and outputs the final identity classification result in combination with the temperature coefficient.

[0012] In one implementation of the first aspect of the present invention, when the identity determination results of multiple consecutive windows are consistent and the confidence level is higher than a preset high threshold, the gait identity authentication is determined to be successful. When the confidence level is between a preset low threshold and a preset high threshold, read-only access is allowed for the mobile device and a lightweight explicit confirmation operation is triggered. When the confidence level is lower than a preset low threshold, restrict the mobile device's sensitive operation permissions until explicit identity authentication is completed.

[0013] Secondly, the present invention provides a shared device gait identity authentication system based on rotation-invariant features.

[0014] A shared device gait authentication system based on rotation-invariant features includes: The data acquisition unit is configured to: acquire the raw time series of triaxial acceleration during the user's walking process via the triaxial accelerometer built into the mobile device; The signal preprocessing unit is configured to perform preprocessing operations on the raw triaxial acceleration time series, remove the gravity component and slide segment according to preset window parameters to generate an acceleration window series. The feature extraction unit is configured to extract a six-dimensional rotation-invariant feature vector generated by the operation of vector norm and vector inner product based on the acceleration window sequence. The six-dimensional rotation-invariant feature vector is invariant to any three-dimensional spatial rotation. The identity classification unit is configured to: input a six-dimensional rotation-invariant feature vector into a preset neural network classifier, and output the user identity determination result after feature processing and classification calculation; The verification execution unit is configured to execute a hierarchical verification strategy based on the identity judgment results and corresponding confidence levels of multiple consecutive windows to complete the gait identity authentication of the mobile device.

[0015] Thirdly, the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and executed by the shared device gait authentication method based on rotation-invariant features according to the first aspect of the present invention.

[0016] Fourthly, the present invention provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the shared device gait authentication method based on rotation-invariant features according to the first aspect of the present invention.

[0017] Fifthly, the present invention provides a computer device, comprising: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the shared device gait authentication method based on rotation-invariant features according to the first aspect of the present invention.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This invention effectively solves the core problems of existing gait authentication, such as signal shift and poor recognition stability caused by changes in device wearing orientation, through a complete process including triaxial acceleration signal acquisition, gravity removal and sliding window preprocessing, six-dimensional rotation-invariant feature extraction, neural network classification, and hierarchical verification. The extracted six-dimensional features are based solely on vector norm and inner product operations, remaining invariant to rotations in any three-dimensional space, completely eliminating orientation interference at the input representation level. This significantly improves recognition reliability across conversations and in scenarios with changing orientations, without relying on complex model learning or data augmentation. The preprocessing operation accurately removes gravity components and standardizes the window sequence, ensuring the purity of the basic signal for feature extraction. Neural network classification can efficiently model temporal features, and the hierarchical verification strategy balances authentication security and ease of use. The entire process is lightweight and adaptable to local deployment on mobile devices, eliminating the need to upload raw data, thus protecting privacy and adapting to high-frequency handover scenarios with shared devices, greatly improving the practicality and stability of gait authentication.

[0019] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1A schematic diagram of the overall framework of a shared device gait identity authentication method based on rotation-invariant features provided as an exemplary embodiment of the present invention; Figure 2 A schematic diagram comparing the original signal and RI features before and after the device is rotated 45°, as provided in an exemplary embodiment of the present invention; Figure 3 A schematic diagram illustrating a dual-condition accuracy comparison, accuracy decline, and RI feature ablation is provided as an exemplary embodiment of the present invention, wherein... Figure 3 (a) in the figure is a diagram comparing the accuracy of two conditions. Figure 3 (b) shows the accuracy decrease caused by introducing a ±45° orientation change. Figure 3 (c) RI characteristic ablation schematic diagram; Figure 4 Stability analysis under various random seeds is provided as an exemplary embodiment of the present invention, wherein, Figure 4 (a) in the diagram is a schematic representation of the mean ± standard deviation for each method. Figure 4 (b) in the figure is a schematic diagram of the accuracy of various sub-subjects under the condition of changing orientation; Figure 5 A t-SNE visualization comparison of the original feature space and the RI feature space is provided as an exemplary embodiment of the present invention, wherein, Figure 5 (a) in the text represents the original feature (variable orientation). Figure 5 (b) in the figure represents the RI feature (orientation unchanged); Figure 6 A schematic diagram of a shared device gait authentication system based on rotation-invariant features, provided as an exemplary embodiment of the present invention; Figure 7 A schematic diagram of a computer device provided for an exemplary embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] With the widespread deployment of tablets, smartphones, and handheld devices in collaborative work scenarios such as hospital wards, field surveys, warehousing and logistics, and public service windows, frequent device handovers among multiple staff members have become the norm. Existing explicit authentication mechanisms (such as passwords, fingerprints, and facial recognition) require users to perform a clear authentication action after each handover. This repetitive and disruptive operation significantly reduces collaboration efficiency, especially in scenarios where staff wear gloves or are in restricted environments, such as during ward rounds, wetlands, or construction sites. Multiple studies have shown that in practice, the widespread adoption of workarounds such as sharing passwords, locking screens, and leaving devices unlocked for extended periods to maintain workflow actually severely undermines security benefits.

[0025] To alleviate the aforementioned contradictions, the industry has proposed implicit identity authentication technology based on behavioral characteristics. This technology continuously and seamlessly re-verifies the user's identity using behavioral signals (such as touch, button presses, gait, and grip posture) during device use. Among these, gait recognition based on smartphone accelerometers has received widespread attention due to its advantages such as high sensor availability, low acquisition cost, and strong feature discrimination. However, in shared device scenarios, gait-based implicit identity authentication has long faced a core obstacle: the sensitivity of device wearing orientation. The angle, depth, and orientation at which the same user places the device in their pocket will change in different sessions, causing the original three-axis acceleration signals to vary. The overall rotation occurs in the sensor coordinate system, and the recognition model's performance degrades significantly under cross-session test conditions.

[0026] To address this issue, existing technologies can be broadly categorized into two types. The first type is coordinate alignment methods, which rotate the sensor coordinate system to a standard coordinate system using gravity direction or principal component analysis before extracting features. However, this method fails when the user's walking direction changes or the device is in an unconventional position. The second type is data augmentation methods, which apply random rotations to samples during training, forcing the model to implicitly learn orientation invariance. However, these methods significantly increase training costs and cannot theoretically guarantee generalization ability to unseen orientations. Furthermore, some studies have attempted to design more complex network structures (such as domain adaptive networks and adversarial training networks) to improve cross-session robustness, but these methods have high computational overhead, are not suitable for resource-constrained mobile deployments, and do not eliminate orientation interference from the input representation.

[0027] To address the shortcomings of existing solutions, this invention proposes a shared device gait authentication method based on rotation-invariant features. It collects three-axis acceleration signals from the mobile device, performs gravity removal and sliding window segmentation, and extracts six-dimensional features (acceleration amplitude, first-order difference amplitude, second-order difference amplitude, normalized inner product of adjacent samples, sliding variance, and sliding composite power) consisting solely of vector norms and inner products. These six-dimensional features remain mathematically invariant to any three-dimensional rotation. The resulting features are then input into a neural network composed of multi-scale one-dimensional convolution, bidirectional GRU, attention pooling, and a cosine classifier to obtain the user's identity. This invention eliminates wearing orientation interference at the input representation level, maintaining a cross-session accuracy of 95.85% under ±45° orientation variations, outperforming the best existing methods by approximately 49 percentage points, and is suitable for collaborative scenarios with frequent device handover.

[0028] Specifically, the shared device gait authentication method of the present invention, such as Figure 1 As shown, the process includes the following: S101: The original time series of walking acceleration is collected by the built-in triaxial accelerometer of the mobile device at a sampling frequency f, and the collected signal is de-gravity component removed and segmented by a sliding window; S102: Extract six-dimensional rotation-invariant features for each window. The six-dimensional features are composed only of vector norms and vector inner products, thus satisfying the invariance under the three-dimensional special orthogonal group SO(3); S103: Input the six-dimensional feature sequence into a neural network classifier consisting of multi-scale one-dimensional convolution, bidirectional gated recurrent units, attention pooling and cosine classifier to obtain the identity discrimination result; S104: Execute a hierarchical verification strategy based on the continuous multi-window discrimination results and confidence level to complete implicit identity re-authentication.

[0029] In step S101 of this implementation, specifically, it includes: In this embodiment, the mobile device can be an Android smartphone, which collects triaxial acceleration signals output by its built-in MEMS triaxial accelerometer at a sampling frequency of f≈20Hz. ,in, This represents discrete time intervals. In practical deployments, the sampling frequency can also be 25Hz, 50Hz, 100Hz, etc. This invention does not limit the specific sampling frequency, only requiring that the window contains at least 2 to 3 complete gait cycles.

[0030] The acquired raw time series were processed according to a window length T=64 (corresponding to approximately 3.2 seconds, about 3 gait cycles) and a sliding step size during the training phase. Sliding step size during the reasoning phase The acceleration window sequence is obtained by segmentation. Subtract the mean value within each window from the values ​​on each of the three axes to obtain the pure acceleration after removing the gravitational component. This step also eliminates the different gravitational projection biases introduced by different device wearing orientations.

[0031] In step S102 of this implementation, specifically, it includes: For each time t within the window, extract the following six scalar features to form a six-dimensional feature vector sequence. : (1) Acceleration amplitude: (1); (2) First-order difference amplitude: (2); (3) Second-order difference amplitude: (3); (4) Normalized inner product of adjacent samples: (4); in, To prevent the denominator from being zero.

[0032] (5) Sliding window amplitude variance (take) ): (5); in, For this The mean of each sample.

[0033] (6) Sliding window composite power (take) ): (6); All six features mentioned above are determined solely by the vector norm. dot product with vector Composition. For any rotation matrix R∈SO(3) in three-dimensional space, since , Therefore, features (1)-(6) are present in The whole was replaced with The values ​​remain completely unchanged afterward, thus the six-dimensional features extracted in this invention are mathematically completely invariant to changes in device wearing orientation. For example... Figure 2 As shown, when the original triaxial acceleration signal is rotated by 45°, Figure 2 (a) and Figure 2 The original signal waveform shown in (b) has changed significantly, while Figure 2 (c) and Figure 2The six-dimensional invariant characteristic waveforms of the present invention shown in (d) are completely consistent (correlation coefficient is 1.0000), which intuitively verifies the above mathematical properties.

[0034] In step S103 of this implementation, specifically, it includes: The following neural network classifier is used to classify the six-dimensional invariant feature sequence F, which consists of four modules.

[0035] (1) Multi-scale one-dimensional convolution module. Two parallel one-dimensional convolutional branches with kernel sizes of 3 and 7 are set up to capture fine-grained gait phase features and coarser-grained gait periodic features, respectively. Each branch consists of two convolutional layers, followed by a batch normalization layer and a GELU activation function, with a total output channel number B=64. The outputs of the two branches are concatenated along the channel dimension and then fused by a 1×1 convolution to obtain an intermediate feature map of length T and number of channels B.

[0036] (2) Bidirectional gated recurrent unit module. A bidirectional GRU with 2 layers and 96 hidden dimensions in each direction is applied to the time dimension of the intermediate feature map, and the output dimension is 192, which is used to model the long-term temporal dependencies of gait sequences.

[0037] (3) Attention pooling module. For each time step of the GRU output, a scalar attention score is obtained using a shared fully connected layer. After being normalized by Softmax, the score is used as a weight to sum the GRU output along the time dimension, resulting in a gait embedding vector of fixed dimension.

[0038] (4) Projection layer and cosine classifier. The gait embedding vector is projected to a 192-dimensional space through layer normalization and multilayer perceptron. The 192-dimensional vector and the prototype vector of each category are L2 normalized respectively. The cosine similarity between the normalized vectors is calculated and then multiplied by the learnable temperature coefficient s to obtain the classification logits of the corresponding N types of users.

[0039] In step S104 of this implementation, specifically, it includes: Applying Softmax to the classification logits of each window yields the identity distribution of the current window. Take the category with the highest probability. and its confidence level When continuous (Typical example) All windows have the same discrimination result. And its average confidence level (Typically 0.9) indicates successful identity authentication; when (Typical example) When read-only access is allowed and a lightweight explicit confirmation (such as a click) is requested; when At this time, sensitive operations are restricted while routine functions are retained until explicit authentication is successful. The six-dimensional invariant features and classification process are all completed locally on the mobile device.

[0040] The training data for this invention comes from walking samples in the publicly available WISDM v1.1 dataset. Nineteen users, each with more than three independent walking segments, were selected. For each user, one segment was used as the test set, one segment as the validation set, and the rest as the training set, ensuring that training and testing always originate from different sessions (cross-session protocol). During the training phase, to simulate real-world wearing orientation changes, three Euler angles (α, β, γ) were independently sampled for each training segment. All three angles conform to the interval […]. The uniform distribution on [45°, +45°] is used to construct a rotation matrix R∈SO(3) according to the ZYX rotation order, and the same rotation is applied to the entire acceleration signal corresponding to this segment. It is worth noting that since the six-dimensional features of this invention are strictly invariant to R∈SO(3), the above rotation augmentation will not change the network input in theory, but it helps to construct a control experiment.

[0041] Training hyperparameters: The optimizer used was AdamW with an initial learning rate of 1×10^{-3}, weight decay of 0.01, and a cosine annealing learning rate strategy. The training epochs were 150. Data augmentation included additive Gaussian dithering (σ=0.03), random amplitude scaling (±10%), Mixup (α=0.2), and label smoothing (0.1). The loss function was normalized cross-entropy loss with a temperature coefficient. Training was repeated with five different random seeds (42, 123, 456, 789, 2024), and the mean accuracy and standard deviation are reported.

[0042] This embodiment compares the proposed method with four traditional machine learning methods (SVM-RBF, Random Forest, KNN-5, LDA) and four deep learning methods (LSTM, 1D-CNN, TCN, Transformer) on the WISDM v1.1 dataset. All comparison methods used the original three-axis acceleration signal (and its amplitude channels, a total of four channels) as input, while the proposed method uses six-dimensional rotation-invariant features as input. Two evaluation conditions were used: (Condition 1) fixed orientation, i.e., no rotation was applied; (Condition 2) variable orientation, i.e., each test segment was independently sampled with random rotations within a range of ±45°. The results are shown in Table 1 (all are the mean ± standard deviation of five random seeds, unit: %). Table 1: Comparison of recognition accuracy between the method of the present invention and existing typical methods under two orientation conditions.

[0043]

[0044] As shown in Table 1, under fixed orientation conditions, the accuracy of various deep learning methods ranges from 75% to 89%, while the method of this invention achieves 95.78%. However, under ±45° variable orientation conditions, the accuracy of all methods using the original triaxial acceleration signal as input drops sharply to 24%-47% (a decrease of 36-52 percentage points), while the method of this invention maintains 95.85%, with a change of only +0.08 percentage points, leading the best existing method by approximately 49 percentage points. Figure 3 As shown, Figure 3 (a) in the figure visually illustrates the accuracy distribution of each method under the two conditions. Figure 3 (b) shows the accuracy decrease introduced by the orientation change. Figure 3 (c) shows the ablation results after replacing the original triaxial signals with the six-dimensional rotation-invariant features of this invention in the input of each deep learning backbone network.

[0045] Furthermore, such as Figure 4 As shown, the accuracy of the method of the present invention is consistently between 94% and 98% under 5 random seeds, while the method using the original triaxial signal as input exhibits extremely high variance across different seeds; for example... Figure 5 As shown, the rotation-invariant features extracted by this invention clearly separate each user cluster in the t-SNE dimensionality reduction space, and the contour coefficients are significantly higher than those of features with the original triaxial signal statistics as input. This explains the fundamental reason for the high accuracy of the method of this invention under cross-session conditions from the perspective of feature geometry.

[0046] To further demonstrate that the key innovation of this invention lies in the six-dimensional rotation-invariant feature itself, rather than the specific network structure, this embodiment maintains the standard 1D-CNN, TCN, and Transformer as the backbone networks, only replacing the original three-axis acceleration signals with the six-dimensional rotation-invariant feature of this invention as the input. The network is then retrained and evaluated under varying orientation conditions. The results show that the accuracy of 1D-CNN improved from 47.07% to 95.24% (+48.2pp); TCN improved from 45.03% to 94.93% (+49.9pp); and Transformer improved from 24.43% to 84.38% (+60.0pp). The accuracy of the standard 1D-CNN after adopting the feature of this invention (95.24%) is only 0.66pp different from that of the complete network structure of this invention (Walk ID, 95.90%). Therefore, the six-dimensional rotation-invariant feature proposed in this invention is the fundamental reason for the high accuracy under varying orientation conditions and can serve as a preprocessing upgrade scheme for any existing deep gait recognition system without requiring any modification to its network structure.

[0047] In collaborative scenarios involving frequent equipment handover, such as hospital ward rounds, field surveys, and shared terminals, this invention can be combined with equipment motion state detection to achieve transparent re-authentication. Specifically, when a device switches from a stationary state to a walking state (detected by acceleration amplitude and variance exceeding a preset threshold), the method of this invention is automatically initiated to perform implicit identity re-authentication for the current holder. Since a single identification window only takes approximately 3.2 seconds, the new holder can usually complete authentication before reaching the destination, without interrupting the collaborative workflow. Combined with a hierarchical verification strategy, a balance between security and efficiency can be achieved through read-only access plus lightweight explicit confirmation when confidence is insufficient. Because the six-dimensional feature system of this invention consists of irreversible scalar statistics, it is impossible to reconstruct the user's original physical motion trajectory based on it. Therefore, the structure satisfies the principle of data minimization, and all feature extraction and classification calculations can be completed locally on the device, thereby avoiding the uploading of original acceleration data.

[0048] In collaborative scenarios involving frequent equipment handover, such as hospital ward rounds, field surveys, and shared terminals, this invention can be combined with equipment motion state detection to achieve transparent re-authentication. Specifically, when a device switches from a stationary state to a walking state (detected by acceleration amplitude and variance exceeding a preset threshold), the method of this invention is automatically initiated to perform implicit identity re-authentication for the current holder. Since a single identification window only takes approximately 3.2 seconds, the new holder can usually complete authentication before reaching the destination, without interrupting the collaborative workflow. Combined with a hierarchical verification strategy, a balance between security and efficiency can be achieved through read-only access plus lightweight explicit confirmation when confidence is insufficient. Because the six-dimensional feature system of this invention consists of irreversible scalar statistics, it is impossible to reconstruct the user's original physical motion trajectory based on it. Therefore, the structure satisfies the principle of data minimization, and all feature extraction and classification calculations can be completed locally on the device, thereby avoiding the uploading of original acceleration data.

[0049] Figure 6 A shared device gait authentication system based on rotation-invariant features is shown, comprising: The data acquisition unit 601 is configured to: acquire the raw time series of triaxial acceleration during the user's walking process through the triaxial accelerometer built into the mobile device; The signal preprocessing unit 602 is configured to perform preprocessing operations on the original triaxial acceleration time series, remove the gravity component and perform sliding segmentation according to preset window parameters to generate an acceleration window series. The feature extraction unit 603 is configured to: extract a six-dimensional rotation-invariant feature vector generated by the operation of vector norm and vector inner product based on the acceleration window sequence; the six-dimensional rotation-invariant feature vector is invariant to any three-dimensional spatial rotation. The identity classification unit 604 is configured to: input a six-dimensional rotation-invariant feature vector into a preset neural network classifier, and output the user identity determination result after feature processing and classification calculation; The verification execution unit 605 is configured to execute a hierarchical verification strategy based on the identity judgment results and corresponding confidence levels of multiple consecutive windows to complete the gait identity authentication of the mobile device.

[0050] It is understood that the aforementioned units can be individually or entirely merged into one or more other units, or some of the units can be further divided into multiple functionally smaller units. This achieves the same operation without affecting the technical effects of the embodiments of the present invention. The aforementioned units are based on logical functional division. In practical applications, the function of one unit can be implemented by multiple units, or the function of multiple units can be implemented by one unit. In other embodiments of the present invention, the system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0051] According to another embodiment of the present invention, the system of this embodiment can be constructed by running a computer program (including program code) capable of performing the steps involved in the corresponding method of the present invention on a general-purpose computing device, such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, loaded into the aforementioned computing device through the computer-readable recording medium, and run therein.

[0052] Figure 7 A computer device is shown, which includes a processor 701, a communication interface 702, and a computer-readable storage medium 703. The processor 701, communication interface 702, and computer-readable storage medium 703 can be connected via a bus or other means.

[0053] The communication interface 702 is used to receive and send data. The computer-readable storage medium 703 can be stored in the memory of the electronic device. The computer-readable storage medium 703 is used to store computer programs, which include program instructions. The processor 701 is used to execute the program instructions stored in the computer-readable storage medium 703.

[0054] The processor 701 is the computing and control core of an electronic device. It is suitable for implementing one or more instructions, specifically for loading and executing one or more instructions to achieve the corresponding method flow or corresponding function.

[0055] Processor 701 is configured to perform the following procedure: The device uses a built-in triaxial accelerometer to collect raw time-series data of the user's triaxial acceleration during walking. Preprocessing is performed on the original triaxial acceleration time series to remove the gravity component and slide segmentation according to preset window parameters to generate an acceleration window series. Based on the acceleration window sequence, a six-dimensional rotation-invariant eigenvector is extracted, which is generated by the operation of vector norm and vector inner product. The six-dimensional rotation-invariant eigenvector is invariant to any three-dimensional spatial rotation. The six-dimensional rotation-invariant feature vector is input into a preset neural network classifier. After feature processing and classification calculation, the user identity determination result is output. Based on the identity recognition results and corresponding confidence levels of multiple consecutive windows, a hierarchical verification strategy is executed to complete the gait identity authentication of the mobile device.

[0056] This invention also provides a computer-readable storage medium, which is a memory device in an electronic device for storing programs and data. It is understood that the computer-readable storage medium here may include both built-in storage media in the electronic device and extended storage media supported by the electronic device. The computer-readable storage medium provides storage space for storing the processing system of the electronic device.

[0057] Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory; alternatively, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0058] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to perform the following process: The device uses a built-in triaxial accelerometer to collect raw time-series data of the user's triaxial acceleration during walking. Preprocessing is performed on the original triaxial acceleration time series to remove the gravity component and slide segmentation according to preset window parameters to generate an acceleration window series. Based on the acceleration window sequence, a six-dimensional rotation-invariant eigenvector is extracted, which is generated by the operation of vector norm and vector inner product. The six-dimensional rotation-invariant eigenvector is invariant to any three-dimensional spatial rotation. The six-dimensional rotation-invariant feature vector is input into a preset neural network classifier. After feature processing and classification calculation, the user identity determination result is output. Based on the identity recognition results and corresponding confidence levels of multiple consecutive windows, a hierarchical verification strategy is executed to complete the gait identity authentication of the mobile device.

[0059] The present invention also provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the following process: The device uses a built-in triaxial accelerometer to collect raw time-series data of the user's triaxial acceleration during walking. Preprocessing is performed on the original triaxial acceleration time series to remove the gravity component and slide segmentation according to preset window parameters to generate an acceleration window series. Based on the acceleration window sequence, a six-dimensional rotation-invariant eigenvector is extracted, which is generated by the operation of vector norm and vector inner product. The six-dimensional rotation-invariant eigenvector is invariant to any three-dimensional spatial rotation. The six-dimensional rotation-invariant feature vector is input into a preset neural network classifier. After feature processing and classification calculation, the user identity determination result is output. Based on the identity recognition results and corresponding confidence levels of multiple consecutive windows, a hierarchical verification strategy is executed to complete the gait identity authentication of the mobile device.

[0060] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can implement the described functions using different methods for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0061] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic cable, digital cable) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A shared device gait authentication method based on rotation-invariant features, characterized in that, include: The device uses a built-in triaxial accelerometer to collect raw time-series data of the user's triaxial acceleration during walking. Preprocessing is performed on the original triaxial acceleration time series to remove the gravity component and slide segmentation according to preset window parameters to generate an acceleration window series. Based on the acceleration window sequence, a six-dimensional rotation-invariant eigenvector is extracted, which is generated by the operation of vector norm and vector inner product. The six-dimensional rotation-invariant eigenvector is invariant to any three-dimensional spatial rotation. The six-dimensional rotation-invariant feature vector is input into a preset neural network classifier. After feature processing and classification calculation, the user identity determination result is output. Based on the identity recognition results and corresponding confidence levels of multiple consecutive windows, a hierarchical verification strategy is executed to complete the gait identity authentication of the mobile device.

2. The shared device gait authentication method based on rotation-invariant features as described in claim 1, characterized in that, Extracting six-dimensional rotation-invariant feature vectors, including: Calculate the magnitude of the triaxial acceleration vector to obtain the acceleration amplitude; Calculate the magnitude of the difference between the three-axis acceleration vectors at adjacent time points to obtain the first-order difference magnitude; Calculate the magnitude of the difference between the first-order difference vectors at adjacent time points to obtain the second-order difference magnitude; Calculate the normalized inner product of the triaxial acceleration vectors at adjacent time points; Calculate the variance of the acceleration amplitude within the preset sliding window to obtain the sliding variance; Calculate the mean of the squares of the magnitudes of the three-axis acceleration vectors within the preset sliding window to obtain the sliding composite power; The acceleration amplitude, first-order difference amplitude, second-order difference amplitude, normalized inner product, sliding variance, and sliding composite power constitute a six-dimensional rotationally invariant eigenvector.

3. The shared device gait authentication method based on rotation-invariant features as described in claim 1, characterized in that, For the triaxial acceleration data within each acceleration window, the mean value of the corresponding axis is subtracted to eliminate the influence of gravity components on the signal; the acceleration data after removing gravity components is truncated according to the preset window length, and then the truncated data is slid segmented according to the preset sliding step size to generate multiple continuous acceleration window sequences.

4. The shared device gait authentication method based on rotation-invariant features as described in claim 1, characterized in that, The preset neural network classifier sequentially includes a multi-scale one-dimensional convolution module, a bidirectional recurrent unit module, an attention pooling module, and a cosine classification module, which are used to perform temporal feature extraction, dependency modeling, feature aggregation, and identity classification on six-dimensional rotation-invariant feature vectors.

5. The shared device gait authentication method based on rotation-invariant features as described in claim 4, characterized in that, The multi-scale one-dimensional convolution module is set up with two parallel one-dimensional convolution branches. Each branch contains multi-layer convolution, batch normalization and activation function processing. The output results of the two branches are spliced ​​and fused to output multi-scale temporal features. The bidirectional cyclic unit module receives multi-scale temporal features, models the long-term temporal dependencies of the feature sequences, and outputs temporal features; The attention pooling module assigns attention weights to each time step of the temporal features, and generates a fixed-dimensional gait embedding vector after weighted summation. The cosine classification module performs dimensional projection and normalization on the gait embedding vector, calculates its cosine similarity with the prototype vector of each user category, and outputs the final identity classification result in combination with the temperature coefficient.

6. The shared device gait authentication method based on rotation-invariant features as described in claim 1, characterized in that, When the identity determination results of multiple consecutive windows are consistent and the confidence level is higher than the preset high threshold, the gait identity authentication is deemed to be successful. When the confidence level is between a preset low threshold and a preset high threshold, read-only access is allowed for the mobile device and a lightweight explicit confirmation operation is triggered. When the confidence level is lower than a preset low threshold, restrict the mobile device's sensitive operation permissions until explicit identity authentication is completed.

7. A shared device gait authentication system based on rotation-invariant features, characterized in that, include: The data acquisition unit is configured to: acquire the raw time series of triaxial acceleration during the user's walking process via the triaxial accelerometer built into the mobile device; The signal preprocessing unit is configured to perform preprocessing operations on the raw triaxial acceleration time series, remove the gravity component and slide segment according to preset window parameters to generate an acceleration window series. The feature extraction unit is configured to extract a six-dimensional rotation-invariant feature vector generated by the operation of vector norm and vector inner product based on the acceleration window sequence. The six-dimensional rotation-invariant feature vector is invariant to any three-dimensional spatial rotation. The identity classification unit is configured to: input a six-dimensional rotation-invariant feature vector into a preset neural network classifier, and output the user identity determination result after feature processing and classification calculation; The verification execution unit is configured to execute a hierarchical verification strategy based on the identity judgment results and corresponding confidence levels of multiple consecutive windows to complete the gait identity authentication of the mobile device.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1 to 6, the shared device gait authentication method based on rotation-invariant features.

9. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the shared device gait authentication method based on rotation-invariant features as described in any one of claims 1 to 6.

10. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program that, when executed by the processor, implements the shared device gait authentication method based on rotation-invariant features as described in any one of claims 1 to 6.

Images