Method for carrying out environment-independent gait authentication by using RFID (Radio Frequency Identification Device) single antenna

Abstract

The invention belongs to the technical field of user authentication, and discloses a method for performing environment-independent gait authentication by using a radio frequency identification (RFID) single antenna, which is abbreviated as RFID. The method comprises the following steps: firstly, acquiring wireless signal data backscattered by an RFID tag array in an indoor environment by using an RFID single antenna to obtain phase and received signal strength information (Received Signal Strength Indication, abbreviated as RSSI); secondly, data preprocessing is conducted on the phase data and the RSSI data, and preprocessing comprises the steps of phase unwrapping, data smoothing and denoising, data linear interpolation and motion detection; then, static components of the recovered wireless signals are removed through a tangent vector model scheme, phases generated by dynamic component changes are obtained, and human body Doppler frequency shift features are further calculated through the phases; and forming a user gait contour matrix by using Doppler frequency shifts of a plurality of labels. And finally, training the gait contour matrix by using a deep learning model to realize identity authentication by using the gait information of the user.

Description

Technical Field

[0001] The present invention belongs to the technical field of user authentication, and more specifically, relates to a method for contactless and environment-independent user authentication by using a tangent vector model to remove static components from data collected by a single-antenna radio frequency identification (RFID) device to obtain dynamic components, extracting Doppler information representing user gait characteristics from the dynamic components. Background Art

[0002] In the era of the Internet of Things, human-computer interaction has become the key to integrating the physical world and the information world. User identity authentication, as the core technology to ensure the security of human-computer interaction, plays a vital role. Traditional user authentication methods usually rely on personal identification codes for authentication, but this method is vulnerable to dictionary attacks. In recent years, some works have used biometrics (e.g., face, fingerprint) for user identity authentication, but biometric-based authentication methods are vulnerable to replay attacks.

[0003] Gait is the posture of a person walking. It is unique and reliable, and is an effective feature for distinguishing individuals. Therefore, it is used in user authentication. Early gait-based authentication methods mainly rely on cameras or wearable devices. Although cameras are effective, they may cause privacy issues, and wearable devices require users to carry additional devices. In order to achieve non-invasive privacy protection, researchers have recently explored the use of radar technology to perceive human gait and perform gait-based user identity authentication. However, due to the relatively high deployment cost and reliance on special hardware, it does not meet practical needs. Radio Frequency Identification (RFID) has become a research hotspot in user gait authentication due to its low cost, flexible deployment, and no need for batteries.

[0004] In recent years, the need for RFID-based gait authentication systems has been proposed. For example, Qian Zhang et al. from Shanghai Jiao Tong University used the similarity of RFID phases for user authentication in a paper titled Unobtrusive and robust human identification using COTS RFID published in Computer Networks, and adopted a carefully designed WMD-DTW algorithm to achieve user identification. Shang Jiang et al. from the University of Chinese Academy of Sciences used a multivariate variational mode decomposition (MVMD) algorithm and an intrinsic mode screening criterion in a paper titled RF-Gait: Gait-Based Person Identification with COTS RFID published in Wireless Communications & Mobile Computing to extract feature vectors from the collected RF signals to represent the intrinsic gait pattern of the human body. Chao Feng et al. from Northwestern University developed a hybrid deep learning model to extract and integrate human behavioral and physiological features for identity recognition in a paper titled RF-Identity: Non-Intrusive Person Identification Based on Commodity RFID Devices published in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies. Yunzhong Chen et al. from Shanghai Jiao Tong University published a paper titled RFPass: Towards environment-independent gait-based user authentication leveraging RFID at the IEEE International Conference on Sensing, Communication, and Networking. The paper uses multiple antennas to collect data and adopts a multipath arrival selection algorithm to extract environment-independent features to construct user gait features.

[0005] However, when only using data from a single antenna to extract environment-independent gait features, the above-mentioned existing technical solutions may have errors and cannot guarantee the accuracy of gait feature extraction. Summary of the invention

[0006] In order to overcome the defects of the prior art and extract environment-independent gait features in the case of a single antenna, the present invention proposes a method for extracting environment-independent gait features using a single RFID antenna. The present invention first preprocesses the data collected by the single antenna, and then removes the static component through a tangent vector model. Then, by calculating the Doppler frequency shift representing the gait characteristics of the human body, a feature matrix is ​​formed using multiple tags. Finally, the matrix is ​​trained using a deep learning model to achieve user gait authentication.

[0007] In order to achieve the above object, the present invention is achieved through the following technical solutions:

[0008] The present invention is a method for extracting environment-independent gait features using a single RFID antenna, which specifically includes the following steps:

[0009] Step 1: Use a single RFID antenna to collect wireless signal data backscattered by an RFID tag array in an indoor environment to obtain phase and received signal strength information (Received Signal Strength Indication, abbreviated as RSSI);

[0010] Step 2: Preprocess the phase data and RSSI data. The preprocessing steps include: phase unwrapping, data smoothing and denoising, data linear interpolation, and motion detection.

[0011] Step 3: Project the preprocessed signal onto the IQ plane, which is used to describe the two-dimensional representation of complex numbers, where the I axis (horizontal axis) represents the real part of the complex data and the Q axis (vertical axis) represents the imaginary part of the complex data; calculate the tangent vector using the tangent vector method to obtain the tangent vector phase, and then calculate the human body Doppler frequency shift based on the tangent vector phase;

[0012] Step 4: Use the spatial characteristics of multiple tags to collect Doppler frequency shifts at different heights to form a gait profile matrix. Use a deep learning model to train the gait profile matrix to achieve identity authentication using user gait information.

[0013] In step 1, a single RFID antenna is used to collect wireless signal data backscattered by an RFID tag array in an indoor environment to obtain phase and RSSI, specifically:

[0014] The communication equipment of this method includes an RFID reader and 8 RFID tags. The RFID reader is connected to a LairdS9028PCL circularly polarized antenna. The RFID reader antenna receives the signal data backscattered by the RFID tag and obtains the phase and RSSI data. The phase and RSSI represent the link state changes of the wireless signal in the space propagation, which can reflect the highly sensitive changes in the surrounding environment.

[0015] A further improvement of the present invention is that in step 2, data preprocessing is performed on the phase and RSSI, and the preprocessing includes phase unwrapping, data smoothing and denoising, and data linear interpolation, specifically:

[0016] Step 2-1, phase unwrapping: The phase value received by the reader antenna is periodic, and the measured phase angle is limited to the range of 0 to 2π, so the observed data has a jump from 0 to 2π; through phase unwrapping, continuous phase information can be restored to reflect the actual phase change of the signal;

[0017] Step 2-2, data smoothing and denoising: Apply moving average filter to denoise RSSI data and unwrapped phase data. Moving average filter has a good suppression effect on high-frequency noise;

[0018] Step 2-3, data linear interpolation: During the transmission of RF signals, the RFID antenna first activates the RFID tag through electromagnetic coupling, and then the tag transmits the RF signal back to the antenna through backscattering; however, due to conflicts and packet loss between tags, the tags do not send RF signals at the same time, resulting in non-uniform sampling points in the received RF signal; in order to synchronize the sampling points of all tags, linear interpolation is applied to the received signal for resampling;

[0019] Step 2-4, motion detection: The RSSI and phase data are detected using a variance-based threshold monitoring method, and the phase and RSSI data of the human body during walking are obtained by segmentation.

[0020] The further improvement of the present invention is that: in step 3, the static component of the restored wireless signal is removed by using the tangent vector solution to obtain the phase generated by the dynamic component change, and the phase is further used to calculate the Doppler frequency shift characteristics representing the human gait; the Doppler frequency shifts of multiple tags are used to form a user gait feature matrix. Specifically:

[0021] Step 3-1: First, use the phase and RSSI data obtained during human walking to restore the wireless signal. The calculation formula is:

[0022]

[0023] Where θ represents the phase, t represents the time t, and R(t) represents the recovered signal;

[0024] Step 3-2: Introduce the tangent vector model and calculate the tangent vector V(t) of the signal. The calculation method is to subtract two adjacent data of the wireless signal. The calculation formula is:

[0025] V(t)=R(t+△t)-R(t)

[0026] Where △t represents the time interval;

[0027] Step 3-3: Calculate the phase of the tangent vector and untangle it. The angle between the vector and the I axis in the IQ plane is its phase value. Since the phase information is periodic, the phase of the tangent vector is further untangled.

[0028] Step 3-4: Use the phase to calculate the Doppler frequency shift characteristic f associated with human walking D , and its calculation formula is:

[0029]

[0030] Where: θ(t) is the phase of the signal changing with time t, is the time derivative of the phase, which represents the rate of change of the phase.

[0031] The further improvement of the present invention is: Step 4: using the spatial characteristics of multiple tags to collect Doppler frequency shifts at different heights, forming a gait profile matrix, using a deep learning model to train the gait profile matrix, and realizing identity authentication using user gait information. Specifically:

[0032] Step 4-1: Use the attention mechanism and deep residual network (ResNet) to train the gait profile matrix to extract the user's unique gait features.

[0033] Step 4-2: During the testing phase, the trained network is used to perform user authentication.

[0034] The beneficial effects of the present invention are as follows: the present invention uses data collected by a single RFID antenna, and after preprocessing, removes the static component through a tangent vector model. This process reduces the influence of the static component on the signal during human walking, making the measurement of the dynamic component changes during human walking more accurate. Then, the Doppler frequency shift characteristics related to human walking are extracted to form a feature matrix. Such a method meets the requirements of environmental independence and single antenna, and realizes user identity authentication. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flow chart of the method of the present invention.

[0036] Figure 2 It is a schematic diagram of the relationship between the signal and the tangent vector model of the present invention.

[0037] Figure 3 is the user gait authentication confusion matrix. DETAILED DESCRIPTION

[0038] The following will disclose the embodiments of the present invention with drawings. For the purpose of clear description, many practical details will be described together in the following description. However, it should be understood that these practical details should not be used to limit the present invention. That is to say, in some embodiments of the present invention, these practical details are not necessary.

[0039] like Figure 1 As shown, after the RFID reader transmits and receives the signal backscattered by the RFID tag, RSSI and phase data are collected. The present invention first performs data preprocessing on the original phase data and RSSI data, including phase unwrapping, data smoothing and denoising, data interpolation, and motion detection. Then the present invention uses the preprocessed phase and RSSI data to restore the signal, uses the data of adjacent sampling points to calculate the tangent vector, then calculates the phase of the tangent vector, and further uses the phase to obtain the Doppler shift. Finally, the present invention combines the Doppler shifts of multiple tags to form a Doppler shift matrix. A deep learning model is used to extract features to implement a user gait authentication scheme. The gait authentication method of the present invention specifically includes the following steps:

[0040] Step 1: Use a single RFID antenna to collect wireless signal data backscattered by the RFID tag array in an indoor environment to obtain the phase and RSSI. The specific steps are:

[0041] The communication equipment of this method includes an RFID reader and 8 RFID tags. The RFID reader is connected to a LairdS9028PCL circularly polarized antenna. The RFID reader antenna receives the signal data backscattered by the RFID tag and obtains the phase and RSSI data. The phase and RSSI represent the link state changes of the wireless signal in the space propagation, which can reflect the highly sensitive changes in the surrounding environment.

[0042] Step 2: Preprocess the phase data and RSSI data. The preprocessing steps include: phase unwrapping, data smoothing and denoising, data linear interpolation, and motion detection. The specific steps are:

[0043] Step 2-1: Phase unwrapping: The phase value received by the reader antenna is periodic, and the measured phase angle is limited to the range of 0 to 2π, so the observed data has a jump from 0 to 2π. Through phase unwrapping, continuous phase information can be restored to reflect the actual phase change of the signal.

[0044] Step 2-2: Data smoothing and denoising: Apply a moving average filter to denoise the RSSI data and the unwrapped phase data. The moving average filter has a good suppression effect on high-frequency noise.

[0045] Step 2-3: Data Linear Interpolation: During the transmission of RF signals, the RFID antenna first activates the RFID tag through electromagnetic coupling. Subsequently, the tag transmits the RF signal back to the antenna through backscattering. However, due to collisions and packet loss between tags, the tags do not send RF signals at the same time, resulting in non-uniform sampling points in the received RF signal. In order to synchronize the sampling points of all tags, linear interpolation is applied to the received signal for resampling.

[0046] Step 2-4: Motion detection: The RSSI and phase data are detected using a variance-based threshold monitoring method, and the phase and RSSI data of the human body during walking are obtained by segmentation.

[0047] Step 3: Project the preprocessed signal onto the IQ plane, which is used to describe the two-dimensional representation of complex numbers, where the I axis (horizontal axis) represents the real part of the complex data and the Q axis (vertical axis) represents the imaginary part of the complex data. Figure 2 As shown in (a), the restored signal R(t) can be divided into the static component H s (t) and the dynamic component H d (t). According to Figure 2 (b) shows the relationship between the tangent vector phase difference and the dynamic component phase difference. The tangent vector is calculated using the tangent vector method to obtain the tangent vector phase, and then the Doppler shift is calculated based on the tangent vector phase.

[0048] Step 3-1: First, use the segmented phase and RSSI data obtained during human walking to restore the wireless signal. The calculation formula is:

[0049]

[0050] Where θ represents the phase, t represents the time t, and R(t) represents the recovered signal.

[0051] Step 3-2: Introduce the tangent vector model and calculate the tangent vector V(t) of the signal by subtracting two adjacent data of the wireless signal. The calculation formula is:

[0052] V(t)=R(t+△t)-R(t)

[0053] Where △t represents the time interval.

[0054] Step 3-3: Calculate the phase of the tangent vector and unwrap it. The angle between the vector and the I-axis in the IQ plane is its phase value. Since the phase information is periodic, the phase of the tangent vector is further unwrapped.

[0055] Step 3-4: Use the phase to calculate the Doppler frequency shift characteristic f associated with human walking D The calculation formula is:

[0056]

[0057] Where: θ(t) is the phase of the signal changing with time t. is the time derivative of the phase, which represents the rate of change of the phase.

[0058] Step 4: Use the spatial characteristics of multiple tags to collect Doppler frequency shifts at different heights to form a gait profile matrix. Use a deep learning model to train the gait profile matrix to achieve identity authentication using user gait information.

[0059] Specifically:

[0060] Step 4-1: Use the attention mechanism and ResNet network to train the gait profile matrix and extract the user's unique gait features.

[0061] Step 4-2: During the testing phase, the trained network is used to perform user authentication.

[0062] The environment in which the data is collected in this invention is an indoor corridor environment. During the experiment, the reader antenna and the tag are facing each other, the distance is 1.8 meters, and the human body passes between the reader antenna and the tag. A total of 8 people, 50 walking data of each person, are collected, of which 80% are used for training and 20% are used for testing. The test set confusion matrix results are as follows: Figure 3 As shown, the average recognition accuracy can reach 98%.

[0063] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for environment-independent gait authentication using a single RFID antenna, characterized in that: The gait authentication method specifically comprises the following steps: Step 1: Use a single radio frequency identification (RFID) antenna to collect wireless signal data backscattered by an RFID tag array in an indoor environment to obtain phase and received signal strength information (RSSI). Step 2: Preprocess the phase data and RSSI data. The preprocessing steps include: phase unwrapping, data smoothing and denoising, data linear interpolation, and motion detection. Step 3: Project the preprocessed signal onto the IQ plane, which is used to describe the two-dimensional representation of complex numbers, where the I axis (horizontal axis) represents the real part of the complex data and the Q axis (vertical axis) represents the imaginary part of the complex data. The tangent vector is calculated using the tangent vector method to obtain the tangent vector phase, and then the human body Doppler shift is calculated based on the tangent vector phase. Step 4: Use the spatial characteristics of multiple tags to collect Doppler frequency shifts at different heights to form a gait profile matrix. Use a deep learning model to train the gait profile matrix to achieve identity authentication using user gait information.

2. The method for environment-independent gait authentication using a single RFID antenna according to claim 1, characterized in that: In step 1, the communication device includes an RFID reader and 8 RFID tags. The RFID reader is connected to a Laird S9028PCL circularly polarized antenna. The RFID reader antenna receives signal data backscattered by the RFID tag to obtain phase and RSSI data. The phase and RSSI represent the link state changes of the wireless signal in space propagation and can reflect the highly sensitive changes in the surrounding environment.

3. The method for environment-independent gait authentication using a single RFID antenna according to claim 1, characterized in that: The step 2 specifically includes the following steps: Step 2-1, phase unwrapping: The phase value received by the reader antenna is periodic, and the measured phase angle is limited to the range of 0 to 2π, so the observed data has a jump from 0 to 2π; Through phase unwrapping, continuous phase information can be restored to reflect the actual phase change of the signal; Step 2-2, data smoothing and denoising: Apply moving average filter to denoise RSSI data and unwrapped phase data. Moving average filter has a good suppression effect on high-frequency noise; Step 2-3, data linear interpolation: During the transmission of the RF signal, the RFID antenna first activates the RFID tag through electromagnetic coupling, and then the tag transmits the RF signal back to the antenna through backscattering; However, due to collisions and packet loss between tags, tags do not send RF signals at the same time, resulting in non-uniform sampling points in the received RF signals. In order to synchronize the sampling points of all tags, linear interpolation is applied to the received signals for resampling. Step 2-4, motion detection: The RSSI and phase data are detected using a variance-based threshold monitoring method, and the phase and RSSI data of the human body during walking are obtained by segmentation.

4. The method for environment-independent gait authentication using a single RFID antenna according to claim 1, characterized in that: The step 3 specifically includes the following steps: Step 3-1: First, use the phase and RSSI data obtained during human walking to restore the wireless signal. The calculation formula is: Where θ represents the phase, t represents the time t, and R(t) represents the recovered signal; Step 3-2: Introduce the tangent vector model and calculate the tangent vector y(t) of the signal. The calculation method is to subtract two adjacent data of the wireless signal. The calculation formula is: V(t)=R(t+Δt)-R(t) Where Δt represents the time interval; Step 3-3: Calculate the phase of the tangent vector and untangle it. The angle between the vector and the I axis in the IQ plane is its phase value. Since the phase information is periodic, the phase of the tangent vector is further untangled. Step 3-4: Use the phase to calculate the Doppler frequency shift characteristic f associated with human walking D , and its calculation formula is: Where: θ(t) is the phase of the signal changing with time t, is the time derivative of the phase, which represents the rate of change of the phase.

5. The method for environment-independent gait authentication using a single RFID antenna according to claim 1, characterized in that: The step 4 specifically includes the following steps: Step 4-1: Use the attention mechanism and deep residual network (ResNet) to train the gait profile matrix and extract the user's unique gait features; Step 4-2: During the testing phase, the trained network is used to perform user authentication.

Images