A Wi-Fi-based method and system for simultaneous indoor passive tracking and gait recognition

Abstract

This invention provides a Wi-Fi-based indoor method and system for simultaneous passive tracking and gait recognition. The method includes: generating a PLCR spectrum; generating a Fast PPVP spectrum; passive tracking and gait separation; and gait recognition. This invention solves the technical problems of relying on known or fixed trajectories for gait recognition, inaccurate gait features, and difficult deployment.

Description

Technical Field

[0001] This invention relates to the field of Wi-Fi passive tracking and gait recognition in indoor environments, specifically to a method and system for simultaneous Wi-Fi passive tracking and gait recognition indoors. Background Technology

[0002] As a key application of the Internet of Things (IoT), smart homes have attracted widespread attention. Smart homes can monitor and control home attributes such as lighting, temperature, and appliances. Such intelligent systems benefit from user location and identity information. For example, we can determine where to adjust the temperature based on the user's location. Furthermore, after identifying the user, it is feasible to provide customized environmental configurations for specific individuals.

[0003] Vision and sound are two typical methods for locating and identifying users. However, these technologies have the following weaknesses: vision-based methods are affected by ambient light and cannot work in dark environments or when there are obstructions. Moreover, deploying a large number of cameras to cover the entire home is impractical; sound-based methods have limited coverage, for example, the device may not be able to be woken up when we leave. In contrast, Wi-Fi has a wider coverage range and is not affected by ambient light. Therefore, Wi-Fi-based location and identification is a promising solution for the future of smart homes.

[0004] To facilitate smart home applications, existing technologies have proposed a series of methods to achieve deviceless tracking and gait recognition separately. However, they have failed to effectively integrate these two methods. First, current deviceless tracking systems typically use speed, distance, and angle to track targets. While the proposed methods are physically plausible, the resulting trajectories may conflict with human gait. For example, the direction of movement often changes within a stride. Therefore, gait features can help correct inconsistencies in the trajectories. Second, most gait recognition systems ignore the influence of the user's trajectory on gait features. For example, recent research shows that the observed gait features for a pair of transceivers depend on the user's position. That is, even if the user walks with the same gait along different trajectories, we still observe different features. Therefore, some existing systems rely on known or fixed trajectories for gait recognition.

[0005] In summary, simultaneous tracking and gait recognition are essential. Specifically, compared to gesture recognition applications, gait recognition typically implies a constantly changing user position. Since the same gait will exhibit different observed Wi-Fi characteristics at different locations, it is necessary to determine the current user's location to extract path-independent gait features. GaitSense was the first method to attempt to integrate tracking and gait recognition technologies. However, this method has the following limitations: GaitSense only uses location to extract path-independent gait features but does not use gait features to correct the user's trajectory. Therefore, positional errors accumulate rapidly, leading to inaccurate gait features; GaitSense's method for acquiring path-independent gait features requires dense Wi-Fi equipment and is very time-consuming to implement, making the system difficult to deploy in real-world environments.

[0006] The existing invention patent application document CN110738130A, entitled "A Path-Independent Gait Recognition Method Based on Wi-Fi," describes a path-independent gait recognition method based on commercial Wi-Fi. This method involves collecting Physical Layer Channel State Information (CSI) data at two receiving ends, removing CSI signal redundancy and noise; detecting the start of motion based on the Frequency Spectral Density (PSD) distribution; calculating the motion direction within a window; mapping a sub-spectral map to the subject's motion direction; stitching together the sub-spectral maps within all window slices to generate a path-independent complete spectral map; and extracting features from the path-independent sub-spectral maps for gait recognition. Based on Fresnel theory, the number of Fresnel zones crossed and the phase delay of the amplitude signal are measured from the CSI signal, and the subject's walking direction within each window slice is derived from these data. And the existing invention patent application document CN109711251A, entitled "A Direction-Independent Gait Recognition Method Based on Commercial Wi-Fi", includes the following steps: the subject repeatedly walks a specified distance within a predetermined area, and the subject's CSI signal is collected through a commercial Wi-Fi device; using the PCA-reduced data, motion detection is performed based on the different states of the PSD distribution of the CSI signal in walking and stationary states to deduce the direction of the subject's walking; STFT is performed on the first 10 principal component data after dimensionality reduction to transform the time-domain signal into the time-frequency domain, and 10 spectra are obtained with time as the horizontal axis and frequency as the vertical axis. The spectra are then denoised using a two-dimensional Gaussian low-pass filter; based on the deduced direction, the spectra are mapped to the direction reflecting the actual human movement, and error correction and fusion are performed to obtain the spectra; manual features are extracted from the spectra, and automatic features are obtained from the Gabor filter; the two types of features are input into a support vector machine with radial basis functions to achieve gait recognition.

[0007] However, the aforementioned existing solutions only achieve gait recognition and fail to reduce recognition errors. The two comparative papers use STFT to extract spectral images, but the limited number of frequency points in the spectrograms restricts frequency resolution. Furthermore, the two papers use different methods to process the spectral images: one combines velocity feature extraction with an SVM classifier, while the other directly feeds the spectral image into a CNN and LSTM network. The former's recognition accuracy needs improvement (80% for 6 people), while the latter does not further disclose the specific details of temporal and spatial features. The aforementioned existing solutions struggle to guarantee the interpretability of features in DNN classification, resulting in poor recognition performance and a low average recognition accuracy.

[0008] In summary, existing technologies suffer from technical problems such as reliance on known or fixed trajectories for gait recognition, inaccurate gait features, and difficulties in deployment. Summary of the Invention

[0009] The technical problem to be solved by this invention is: how to solve the technical problems of relying on known or fixed trajectories for gait recognition, inaccurate gait features, and difficult deployment in the prior art.

[0010] This invention solves the above-mentioned technical problems by employing the following technical solution: A Wi-Fi-based indoor simultaneous passive tracking and gait recognition method includes:

[0011] S1. Define transmitter coordinates and receiver coordinates, obtain and calculate the signal reflection path length based on the Wi-Fi signal to construct a Fresnel zone model, obtain and determine the signal reflection path change rate PLCR based on the correlation between Doppler frequency shift (DFS) and signal reflection path change rate (PLCR), use the Fresnel zone model to convert the signal reflection path change rate PLCR into human motion speed, integrate the human motion speed to obtain the user's real-time position and displacement change data;

[0012] S2. Establish a mapping rule between the signal reflection path change rate (PLCR) and the actual velocity in the polar coordinate system. Based on no less than two PLCR spectra, generate a Fast PPVP spectra in the polar coordinate system.

[0013] S3. Based on the Fast PPVP spectrum, obtain the velocity change data and angle change data of the maximum energy value position under the preset time series, and use them for passive tracking and gait separation. Extract the motion direction and velocity change in polar coordinates, obtain and passively track according to the user's initial position to obtain the velocity curve; integrate the velocity curve to track the target, obtain and process the displacement change to obtain the gait change feature dataset.

[0014] S4. Analyze the gait feature dataset and perform gait recognition based on clustering and classification operations.

[0015] This invention utilizes Wi-Fi signals to simultaneously perform passive tracking and gait recognition indoors. By establishing a theoretical model, it can directly convert path-dependent spectra into path-independent spectra at a low cost, tracking targets and extracting gait features for identification. This invention employs an ensemble method to obtain stable trajectories without destroying gait features, ensuring that tracking and gait recognition can mutually correct each other. The tracking process applies a smoother velocity curve to obtain a stable trajectory, while gait recognition retains detailed features to distinguish different users.

[0016] In a more specific technical solution, S1 includes:

[0017] S11. Define the transmitter coordinates and i receiver coordinates using the following logic:

[0018]

[0019] S12. Obtain and calculate the signal reflection path length based on the CSI information of the Wi-Fi signal, construct no less than 2 Fresnel ellipses, and obtain the Fresnel region model;

[0020] S13. Using the following logic, express the relationship between Doppler frequency shift (DFS) and signal reflection path change rate (PLCR):

[0021]

[0022] In the formula, f represents the communication frequency and c represents the speed of light;

[0023] S14. Differentiate the correlation relationship to obtain the signal reflection path change rate PLCR;

[0024] S15. In the Fresnel zone model, based on the speed of human movement... The projection onto the normal vector of the concentric ellipse yields the signal reflection path change rate PLCR, defined as the velocity projection vector of the i-th receiver. Establish signal reflection path change rate PLCR and human motion speed The relationship between them;

[0025] S16. Based on the signal reflection path change rate PLCR and the human movement speed The relationship between these factors is used to determine the speed of human movement.

[0026] In a more specific technical solution, S12 uses the following logic to express CSI:

[0027]

[0028] In the formula, H s (f,t) and H d(f,t) represent the static and dynamic reflection components of CSI, respectively. L represents the amplitude and phase of a dynamic signal. d (t) represents the path length of the dynamic signal reflection, and λ represents the wavelength of the wireless signal.

[0029] In a more specific technical solution, in S15, the velocity projection vector... The following formula is obtained after processing:

[0030]

[0031] In the formula, r i It refers to the one-dimensional PLCR sequence calculated by the i-th receiver;

[0032] Given human body location The velocity projection vector is processed using the following logic.

[0033]

[0034] Establish the signal reflection path change rate (PLCR) and human movement speed through a Wi-Fi transceiver link. The relationship between them.

[0035] In a more specific technical solution, in S16, at least two velocity projection components are integrated to obtain a one-dimensional PLCR time series, a velocity model is established, and the human motion velocity is solved. The following logical expression is used. The optimal solution:

[0036]

[0037] In the formula, It is a matrix composed of all projection vectors, r = (r1, r2, ..., r...). I ) T It is a matrix composed of all PLCRs.

[0038] To address the issue that raw CSI information does not contain features directly related to velocity, this invention extracts effective physical features from the raw data. This invention utilizes a PLCR spectrum generation method based on conjugate multiplication and Hilbert transform, and employs FFT zero-padding to fill CSI data to obtain a more accurate velocity curve.

[0039] In a more specific technical solution, S2 includes:

[0040] S21. Perform a segmentation operation. For each time period t, obtain the value and index of the PLCR spectrum.

[0041] S22. Perform a mapping operation for each user location in the PPVP spectrum. Calculate the corresponding signal reflection path change rate (PLCR);

[0042] S23. Perform a fusion operation. Using the signal reflection path change rate (PLCR) of the receiver as an indicator, find the corresponding energy value in the PLCR spectrum, multiply it by the energy of the corresponding transceiver pair, and determine the energy value of all user locations in the PPVP spectrum.

[0043] S24. Perform a sorting operation. Repeat S21 to S23 for different time periods t to obtain continuous PPVP spectrum.

[0044] S25. Perform tracking operation and derive user speed characteristics and user gait characteristics based on the peak values ​​of the continuous PPVP spectrum for tracking target.

[0045] Compared to the time-consuming methods of traditional velocity feature acquisition, this invention develops a fast physical polar coordinate velocity profile to extract path-independent velocity features. Compared to techniques such as BVP, Fast PPVP can significantly save execution time at the same spectral resolution.

[0046] In a more specific technical solution, S22 uses a polar coordinate system to represent the walking gait of the test object;

[0047] Let the target location be: Among them, v ρ and It refers to the walking speed and direction at time t;

[0048] In polar coordinates, the following logic can be used to express the velocity of human movement.

[0049]

[0050] In the formula, the position of any point P on the plane can be determined by the length ρ of line segment OP and the angle from Ox to OP. Let P be the coordinates of point P in the polar coordinate system. ρ is called the polar radius along the radial direction. The angle along the horizontal direction is called the polar angle.

[0051] In the polar coordinate system of this invention, polar angles and polar radii represent the position of a point using angles and lengths. In polar coordinates, the angular changes of the limbs relative to the torso can be further calculated. When dealing with angle calculations, the polar coordinate system is more convenient and efficient.

[0052] In a more specific technical solution, in S22, the following logic is used to obtain the polar coordinate system mapping expression for the signal reflection path change rate PLCR:

[0053]

[0054] In the formula, This is the normal vector calculated for the location by the Wi-Fi transceiver based on the target and the location.

[0055] Based on the polar coordinate system mapping expression of the signal reflection path change rate PLCR, the human motion speed in the polar coordinate system is derived.

[0056] Based on the PLCR polar coordinate system mapping expression for the signal reflection path change rate, the PLCR spectrum is converted into the PPVP spectrum.

[0057] In a more specific technical solution, S4 also includes:

[0058] S41. Extract gait features from the less feature dataset, perform clustering and classification operations on the gait features, and output the k-class centroids and clustering results as classification labels;

[0059] S42. Using principal component analysis (PCA) and Gaussian noise, expand the gait feature dataset to obtain an augmented dataset;

[0060] S43. Train a deep neural network (DNN) based on the classification labels.

[0061] In more specific technical solutions, Wi-Fi-based indoor simultaneous passive tracking and gait recognition systems include:

[0062] The PLCR conversion module is used to define the transmitter coordinates and receiver coordinates, acquire and calculate the signal reflection path length based on the Wi-Fi signal to construct a Fresnel zone model, acquire and determine the signal reflection path change rate PLCR based on the correlation between Doppler frequency shift (DFS) and signal reflection path change rate PLCR, and use the Fresnel zone model to convert the signal reflection path change rate PLCR into human motion speed. By integrating the human motion speed, the real-time position and displacement change data of the user are obtained.

[0063] The Fast PPVP generation module is used to establish the mapping rule between the signal reflection path change rate (PLCR) and the actual velocity in the polar coordinate system. Based on no less than two PLCR spectra, it generates a Fast PPVP spectrum in the polar coordinate system. The Fast PPVP generation module is connected to the PLCR conversion module.

[0064] The passive tracking and gait separation module is used to obtain velocity and angle change data of the maximum energy value location under a preset time series based on the Fast PPVP spectrum. It performs passive tracking and gait separation, extracts the motion direction and velocity change in polar coordinates, obtains and performs passive tracking based on the user's initial position to obtain a velocity curve, integrates the velocity curve to track the target, obtains and processes the displacement change to obtain a gait change feature dataset, and the gait change feature acquisition module is connected to the Fast PPVP generation module.

[0065] The gait recognition module is used to analyze gait feature datasets and perform gait recognition based on clustering and classification operations. The gait recognition module is connected to the passive tracking and gait separation module.

[0066] The present invention has the following advantages over the prior art:

[0067] This invention utilizes Wi-Fi signals to simultaneously perform passive tracking and gait recognition indoors. By establishing a theoretical model, it can directly convert path-dependent spectra into path-independent spectra at a low cost, tracking targets and extracting gait features for identification. This invention employs an ensemble method to obtain stable trajectories without destroying gait features, ensuring that tracking and gait recognition can mutually correct each other. The tracking process applies a smoother velocity curve to obtain a stable trajectory, while gait recognition retains detailed features to distinguish different users.

[0068] To address the issue that raw CSI information does not contain features directly related to velocity, this invention extracts effective physical features from the raw data. This invention utilizes a PLCR spectrum generation method based on conjugate multiplication and Hilbert transform, and employs FFT zero-padding to fill CSI data to obtain a more accurate velocity curve.

[0069] Compared to the time-consuming methods of traditional velocity feature acquisition, this invention develops a fast physical polar coordinate velocity profile to extract path-independent velocity features. Compared to techniques such as BVP, Fast PPVP can significantly save execution time at the same spectral resolution.

[0070] In the polar coordinate system of this invention, polar angles and polar radii represent the position of a point using angles and lengths. In polar coordinates, the angular changes of the limbs relative to the torso can be further calculated. When dealing with angle calculations, the polar coordinate system is more convenient and efficient.

[0071] This invention primarily combines passive tracking and gait recognition technologies. Compared to simple gait recognition, this invention can reduce recognition errors by tracking and identifying the current position and extracting gait features independent of the path. The two comparative documents in the background art, publication numbers CN110738130A and CN109711251A, use STFT to extract the spectrum, while this invention applies FFT zero-padding technology to extract the spectrum, increasing the number of frequency points in the spectrum and further improving frequency resolution. The two comparative documents also differ in their methods for processing the spectrum: one solves for velocity features using an SVM classifier, while the other directly puts the spectrum into a CNN and LSTM network. The former's recognition accuracy needs improvement (6 people, 80%), while the latter does not further disclose the specific details of temporal and spatial features. Compared to the aforementioned existing solutions, this invention, based on the PPVP spectrum, obtains the maximum frequency components in two spatial dimensions within a unit period and analyzes the periodic features of the corresponding angle and velocity, performing DNN classification on the features. This approach ensures the interpretability of DNN classification features while achieving good recognition results, with an average recognition accuracy of 95.3% across 10 users.

[0072] This invention solves the technical problems of existing technologies, such as reliance on known or fixed trajectories for gait recognition, inaccurate gait features, and difficulties in deployment. Attached Figure Description

[0073] Figure 1 These are the basic steps of the Wi-Fi-based indoor simultaneous passive tracking and gait recognition method of Embodiment 1 of the present invention.

[0074] Schematic diagram;

[0075] Figure 2 This is the Fast PPVP spectrum over continuous time in Embodiment 1 of the present invention;

[0076] Figure 3 This is a velocity curve corresponding to the maximum energy value in the polar coordinate system of Embodiment 1 of the present invention;

[0077] Figure 4 This is a schematic diagram illustrating the specific steps of converting the PLCR spectrum into the PPVP spectrum in Embodiment 1 of the present invention;

[0078] Figure 5 This is a positioning effect diagram of Embodiment 1 of the present invention;

[0079] Figure 6 This is a 10-person identity recognition confusion matrix diagram of Embodiment 2 of the present invention. Detailed Implementation

[0080] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0081] Example 1

[0082] like Figure 1 As shown, the Wi-Fi-based indoor simultaneous passive tracking and gait recognition method provided by this invention includes the following basic steps:

[0083] S1, PLCR spectrum generation stage;

[0084] In this embodiment, CSI information of human movement is collected through multi-link Wi-Fi signals in the system. However, the raw CSI information does not contain features directly related to speed, so effective physical features are extracted from the raw data first. To this end, a PLCR spectrum generation method based on conjugate multiplication and Hilbert transform is proposed, and FFT zero-padding is used to fill the CSI data to obtain a more accurate speed curve.

[0085] In this embodiment, Wi-Fi passive tracking technology is a technique that uses Wi-Fi signals to track the location of users. It extracts features of signal propagation pattern changes caused by human movement from channel state information, such as signal strength, time of arrival, and the Doppler effect, thereby analyzing the position and movement trajectory of the human body in an indoor environment. Passive tracking differs from active tracking in that it does not require the user to carry intrusive devices such as mobile phones or sensors; it completes the tracking work solely by analyzing channel state information reflected from the human body.

[0086] In this embodiment, Channel State Information (CSI) describes the propagation process of a wireless signal between the transmitter and receiver. This channel state information reflects the combined effects of factors such as distance, power attenuation, and scattering on the signal, and includes basic information such as the channel's amplitude, phase, and frequency.

[0087] In this embodiment, the Fresnel zone is a series of concentric ellipsoidal regions with equal focal length formed when an electromagnetic wave signal propagates between the transmitter and receiver. The transmitter and receiver are located at the two foci of the ellipse. When a human body moves across multiple Fresnel zones, the propagation path lengths of the reflected and direct signals differ, resulting in different superposition effects between the transmitter and receiver. As an object passes through multiple Fresnel zones, the signal strength periodically increases and decreases; the location is determined by calculating the signal period. The Fresnel zone is the theoretical basis upon which this tracking and gait recognition system relies.

[0088] In this embodiment, Doppler Frequency Shift (DFS) is the change in wireless signal frequency caused by the movement of a human body relative to a Wi-Fi link in an indoor positioning and tracking scenario. This change can be used to detect the motion state, speed, and direction of an object.

[0089] In this embodiment, the coordinates of one transmitter and i receivers are respectively... and This is indicated so that subsequent formulas can be defined.

[0090] In this embodiment, the signal reflection path length is calculated based on the CSI of the Wi-Fi signal, and multiple Fresnel ellipses are constructed. The expression for CSI is as follows:

[0091]

[0092] Among them, H s (f,t) and H d (f,t) represent the static and dynamic reflection components of CSI, respectively. L represents the amplitude and phase of a dynamic signal. d (t) represents the path length of the dynamic signal reflection, and λ represents the wavelength of the wireless signal. The propagation path of the reflected signal is from the transmitter, reflected by the human body, and then received by the receiver. Therefore, the path length is the sum of the distances from the human body to the transmitter and receiver. The innermost ellipse is defined as the first Fresnel zone. The human body passing through each Fresnel zone layer will cause the signal reflection path to change by λ. When the human body moves by d along the reflection path position with the dynamic component, the phase of the dynamic component will shift by 2πd / λ.

[0093] In this embodiment, the human body traversing multiple Fresnel zones within the perception range causes changes in Doppler frequency shift and the rate of change of reflection path length. The change in reflection path length is the fundamental cause of Doppler frequency shift, and the two can be calculated and converted into each other. We use f... dLet r(t) and r(t) represent DFS and PLCR, and the relationship between them is as follows:

[0094]

[0095] Where f represents the communication frequency and c represents the speed of light. The above formula represents the rate of change of the signal reflection path, or PLCR, which can be obtained by differentiating the signal reflection path.

[0096] In this embodiment, since the PLCR only reflects the Doppler frequency shift, it is not the true velocity of the moving target. The target's velocity and position jointly determine the PLCR of the target's reflection path. Using the Fresnel zone model, the PLCR can be converted into the true velocity of human motion. In Widar, human motion velocity... The projection onto the normal vector of the concentric ellipse is PLCR. For the projection vector defined as the i-th receiver... We can conclude that:

[0097]

[0098] Where, r i This refers to the one-dimensional PLCR sequence calculated from the i-th receiver. Given the human's location... to indicate The formula is as follows:

[0099]

[0100] In this embodiment, PLCRr and the actual speed are established through all Wi-Fi transceiver links. The relationship between these components. By integrating multiple velocity projection components, i.e., the time series of a one-dimensional PLCR, we can establish a velocity model to solve for the actual human motion velocity. The optimal solution is expressed as:

[0101]

[0102] in, It is a matrix composed of all projection vectors, and r = (r1, r2, ..., r I ) T It is a matrix composed of all PLCRs. By integrating the velocity, we can obtain the real-time user position. Therefore, the PLCR spectrum can be obtained through CSI, and the velocity can be further integrated to obtain the displacement change.

[0103] S2, Fast PPVP spectrum generation stage;

[0104] In this embodiment, the PLCR spectrum depends on the positional relationship between the user and the transceiver. Different spectra can be observed even for the same user when the trajectories differ. Therefore, the PLCR spectrum cannot be used as a gait feature to distinguish users. A mapping rule between PLCR and actual velocity is established in polar coordinates, and a Fast PPVP spectrum in physical polar coordinates is generated based on multiple PLCR spectra.

[0105] In this embodiment, the Path Length Change Rate (PLCR) describes the rate of change of the signal path after reflection from the human body, and it is the cause of the Doppler frequency shift. In connection with Fresnel zone theory, PLCR reflects the change in the signal path caused by the human target relative to the link, and is the inner product of the human velocity vector and the ellipse normal vector.

[0106] In this embodiment, commonly used coordinate systems include rectangular coordinates, polar coordinates, natural coordinates, and spherical coordinates. In principle, any reference system can be chosen for the same problem, but different reference systems have a significant impact on the simplicity or complexity of the model, and the convenience or redundancy of calculations. In a rectangular coordinate system, the horizontal and vertical coordinates represent the position of a point using horizontal and vertical component vectors; while in a polar coordinate system, the polar angle and polar radius represent the position of a point using angle and length. Therefore, when dealing with problems involving angle calculations, using the polar coordinate system is more convenient and faster.

[0107] like Figure 2 and Figure 3 As shown, in this embodiment, a fixed point O is taken on the plane. A ray Ox emanating from point O, a unit of length, and the positive direction for calculating angles (usually counterclockwise) are collectively called a polar coordinate system. Point O is called the pole, and Ox is called the polar axis. The position of any point P on the plane can be determined by the length ρ of line segment OP and the angle from Ox to OP. Let P be the coordinates of point P in the polar coordinate system. ρ is called the polar radius along the radial direction. The angle along the horizontal direction is called the polar angle.

[0108] In the gait recognition of this embodiment, a polar coordinate system can be used to represent the walking gait of the test subject. Assume the target location is... v ρ and This refers to the walking speed and direction at time t. Target speed. In polar coordinates, this can also be expressed as the following equation:

[0109]

[0110] Based on the formula, the formula for PLCR can also be expressed as:

[0111]

[0112] in, This is the normal vector calculated from the position of the target and the Wi-Fi transceiver. The PLCR spectrum represents the energy of the PLCR at different body parts, in m / s. The target velocity in polar coordinates can be derived from the PLCR using the formula.

[0113] like Figure 4 As shown, in this embodiment, based on the aforementioned mapping formula, a conversion method for converting the PLCR spectrum to the PPVP spectrum is designed, including the following specific steps:

[0114] S21, Fragmentation operation;

[0115] In this embodiment, for each time period t, we can obtain the value and index of the PLCR spectrum. In this embodiment, the value of the PLCR spectrum describes the energy of different PLCR components, and the index is the PLCR in m / s.

[0116] S22, Mapping operation;

[0117] In this embodiment, for each position in the PPVP spectrum The corresponding PLCR is calculated using formula (5-2). The user's location is provided by the positioning process, and the transceiver's location can be known in advance.

[0118] Therefore, i transceiver pairs can be obtained through mapping. 1,2,...,i .

[0119] S23, Fusion Operation;

[0120] In this embodiment, r i The corresponding energy value can be found in the PLCR spectrum as an indicator. By multiplying by the energy relative to different transceiver pairs, the energy value at the current position in the PPVP spectrum is further determined. Finally, the energy values ​​at all positions in the PPVP spectrum are determined.

[0121] S24, Sorting operation;

[0122] In this embodiment, the above steps are repeated for different time periods t to obtain continuous PPVP spectra. As shown in the figure, the portion of the PPVP spectrum with the largest energy value represents the motion speed and direction with the highest probability.

[0123] S25, Tracking Operation;

[0124] In this embodiment, the peak value of each spectrum can derive partial characteristics of the user's speed and gait, which can be used to track the target.

[0125] In this embodiment, the PLCR spectrum is converted into a PPVP spectrum, with the higher energy portion of the spectrum representing the velocity and direction of motion. Finally, the peak values ​​of the velocity spectrum at each time t are concatenated, filtered, and a velocity curve in polar coordinates is generated.

[0126] S3, Passive tracking and gait separation phase;

[0127] In this embodiment, the velocity magnitude and angular changes at the location of the maximum energy value in the time series can be obtained through the Fast PPVP spectrum, and the stable motion direction and detailed velocity changes in polar coordinates can be extracted. Given the user's initial position, passive tracking can be performed, and the velocity curve can be obtained.

[0128] In this embodiment, the target can be tracked by integrating the velocity curve, and gait changes can be investigated based on displacement variations. In this embodiment, gait refers to the posture and behavioral characteristics of a human body when walking, such as walking speed, stride length, and leg swing amplitude. Specific measurement indicators include spatiotemporal characteristics, kinematic characteristics, and dynamic characteristics.

[0129] In this embodiment, the spatiotemporal features include gait cycle, walking speed, stride length, and other temporal and spatial characteristics. The gait cycle is the duration *t* of each step, and the stride length is the integral of the velocity curve for each step. Based on these two spatiotemporal features, the average speed of the target is calculated. Simultaneously, the velocity curve visually displays the maximum speed, minimum speed, and speed change trend within a step. Overall, six gait features can be extracted: gait cycle, stride length, speed difference within a step, maximum speed, minimum speed, and speed variance. Kinematic features are the angular changes of different limbs relative to the body, such as the swing amplitude of the legs during walking. In the PPVP spectrum, it can be observed that higher energy areas correspond to the torso, and lower energy areas correspond to the limbs. In polar coordinates, the angular changes of the limbs relative to the torso can be further calculated. This embodiment primarily analyzes the velocity curve information of different limbs, using the speed of different limbs to represent the angular changes.

[0130] S4, Gait recognition stage;

[0131] In this embodiment, the changes in gait characteristics within the gait cycle, such as velocity extrema, variance, and stride length, can be obtained based on the velocity curve. By analyzing the gait feature dataset, a gait recognition method based on clustering and classification is designed.

[0132] In this embodiment, after extracting gait features individually, the extracted gait features are first clustered and then classified. This eliminates interference from unpredictable factors and ensures that the input features are all gait-related. Therefore, the k-medoids algorithm is chosen to perform k-medoids clustering on the extracted gait features. By establishing a k-medoids clusterer, six classes of gait features are input and recognition results are generated. The clustering algorithm uses Euclidean distance, randomly selecting k cluster centers each time, repeating the iteration 20 times, and displaying the final iteration result. Applying k-medoids clustering to the input gait feature dataset, the k class center points and their clustering results are output as classification labels.

[0133] Then, Principal Components Analysis (PCA) and Gaussian noise are used to expand the dataset. A straightforward solution is to add random Gaussian noise n to the input vector K, where n ~ N(0, σ). 2 However, since different features in K lie in different ranges, it is unreasonable to use a fixed σ for all features. Therefore, this section utilizes PCA to generate transformation matrices, transforming a series of K into a low-dimensional vector K. ′ And add random Gaussian noise n to K ′ In China, it is possible to augment 200 actual cases into 2000 cases.

[0134] like Figure 5 As shown, this embodiment utilizes labels provided by a clustering method to train a deep neural network (DNN). The DNN structure used in the system is as follows: First, the features and labels of the training set are input into a feature input layer, and the z-score data normalization method is applied in this network layer. Next, a fully connected layer with an output size of 100 is connected. To accelerate the training speed of the neural network and reduce its sensitivity to network initialization, a batch normalization layer is used before the ReLU layer. Finally, a typical classification structure is used to implement classification, which consists of a fully connected layer, a softmax layer, and a classification layer. A fully connected layer with an output size of k is used to specify the number of classifications in the deep neural network. The final average recognition accuracy for 10 users is 95.3%.

[0135] Example 2

[0136] like Figure 6As shown, in this embodiment, one transmitter and two receivers were deployed in two open indoor environments of 4.5m×4.5m and 4m×5m, with the angle between the lines connecting the two receivers and the transmitter being 90 degrees. All transmitters and receivers were pre-installed with the Ubuntu 14.04.4 operating system and equipped with an Intel 5300 series wireless network card and CSI tools. Each transmitter was equipped with a single antenna, and each receiver had three antennas linearly mounted, with a spacing of 2.5cm between the antennas. All transmitters and receivers were remotely connected to the host via the SSH protocol. The AP's transmission and reception could be controlled via remote commands from the host, with the data packet rate set to 1000Hz. To verify performance, we used 10 volunteers in the experiment. Each volunteer had to follow four different tracks, each track repeated five times. We will first verify the system's tracking performance and then calculate the accuracy of gait recognition.

[0137] In summary, this invention performs passive tracking and gait recognition simultaneously indoors using Wi-Fi signals. By establishing a theoretical model, it can directly convert path-dependent spectra into path-independent spectra at a low cost, track targets, and extract gait features for identification. This invention uses an ensemble method to obtain stable trajectories without destroying gait features, ensuring that tracking and gait recognition can mutually correct each other. The tracking process applies a smoother velocity curve to obtain a stable trajectory, while gait recognition retains detailed features to distinguish different users.

[0138] To address the issue that raw CSI information does not contain features directly related to velocity, this invention extracts effective physical features from the raw data. This invention utilizes a PLCR spectrum generation method based on conjugate multiplication and Hilbert transform, and employs FFT zero-padding to fill CSI data to obtain a more accurate velocity curve.

[0139] Compared to the time-consuming methods of traditional velocity feature acquisition, this invention develops a fast physical polar coordinate velocity profile to extract path-independent velocity features. Compared to techniques such as BVP, Fast PPVP can significantly save execution time at the same spectral resolution.

[0140] In the polar coordinate system of this invention, polar angles and polar radii represent the position of a point using angles and lengths. In polar coordinates, the angular changes of the limbs relative to the torso can be further calculated. When dealing with angle calculations, the polar coordinate system is more convenient and efficient.

[0141] This invention solves the technical problems of existing technologies, such as reliance on known or fixed trajectories for gait recognition, inaccurate gait features, and difficulties in deployment.

[0142] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A Wi-Fi based indoor simultaneous passive tracking and gait recognition method, characterized in that, The method includes: S1. Define transmitter coordinates and receiver coordinates, obtain and calculate the signal reflection path length based on the Wi-Fi signal to construct a Fresnel zone model, obtain and determine the signal reflection path change rate PLCR based on the correlation between Doppler frequency shift (DFS) and signal reflection path change rate PLCR, use the Fresnel zone model to convert the signal reflection path change rate PLCR into human motion speed, integrate the human motion speed to obtain the user's real-time position and displacement change data; S2. Establish the mapping rule between the signal reflection path change rate (PLCR) and the actual velocity in the polar coordinate system, and generate a Fast PPVP spectrum in the polar coordinate system based on no less than two PLCR spectra. S3. Based on the Fast PPVP spectrum, obtain the velocity change data and angle change data of the maximum energy value position under the preset time series, and perform passive tracking and gait separation accordingly. Extract the motion direction and velocity change in polar coordinates, obtain and perform passive tracking based on the user's initial position to obtain the velocity curve; integrate the velocity curve to track the target, obtain and process the displacement change to obtain the gait change feature dataset; S4. Analyze the gait change feature dataset and perform gait recognition based on clustering and classification operations.

2. The Wi-Fi based indoor simultaneous passive tracking and gait recognition method of claim 1, wherein, S1 includes: S11. Defining the transmitter coordinates, using the logic: one of the receiver coordinates: S12. Obtain and calculate the signal reflection path length based on the CSI information of the Wi-Fi signal, construct no less than 2 Fresnel ellipses, and obtain the Fresnel region model; S13. The relationship between the Doppler frequency shift (DFS) and the signal reflection path change rate (PLCR) is expressed using the following logic: ， wherein denotes the communication frequency, denotes the speed of light; S14. Differentiate the correlation to obtain the signal reflection path change rate PLCR; S15. In the Fresnel zone model, based on the human body movement speed... The projection onto the normal vector of the concentric ellipse yields the signal reflection path change rate PLCR, defined as the first... The velocity projection component normal vector of the receiver Establish the signal reflection path change rate PLCR and the human movement speed. The relationship between them; S16. Based on the signal reflection path change rate PLCR and the human movement speed... The relationship between these factors is used to determine the speed of the human body movement.

3. The Wi-Fi-based indoor simultaneous passive tracking and gait recognition method according to claim 2, characterized in that, In S12, the CSI is expressed using the following logic: ， In the formula, and These represent the static and dynamic reflection components of CSI, respectively. This represents the amplitude and phase of a dynamic signal. Indicates the path length of the dynamic signal reflection. Indicates the wavelength of the wireless signal.

4. The Wi-Fi-based indoor simultaneous passive tracking and gait recognition method according to claim 2, characterized in that, In step S15, the normal vector of the velocity projection component... The following formula is obtained after processing: ， In the formula, It refers to the first One-dimensional PLCR sequence calculated by each receiver; Given human body location The velocity projection component normal vector is processed using the following logic. : The signal reflection path change rate (PLCR) and human movement speed are established via a Wi-Fi transceiver link. The relationship between them.

5. The Wi-Fi-based indoor simultaneous passive tracking and gait recognition method according to claim 2, characterized in that, In step S16, the normal vectors of at least two velocity projection components are integrated to obtain a one-dimensional PLCR time series, a velocity model is established, and the velocity of the human body movement is solved. The following logical expression is used. The optimal solution: ， In the formula, It is a matrix composed of the normal vectors of all velocity projection components. It is a matrix composed of all PLCRs.

6. The Wi-Fi-based indoor simultaneous passive tracking and gait recognition method according to claim 1, characterized in that, S2 includes: S21. Perform segmentation operation, for each time period Obtain the value and index of the PLCR spectrum; S22. Perform a mapping operation for each user location in the PPVP spectrum. Calculate the corresponding signal reflection path change rate PLCR; S23. Perform a fusion operation, using the signal reflection path change rate (PLCR) of the receiver as an indicator, find the corresponding energy value in the PLCR spectrum, multiply it by the energy of the corresponding transceiver pair, and determine the energy value of all user locations in the PPVP spectrum. S24. Perform a sorting operation on the different time periods. Repeat steps S21 to S23 to obtain a continuous PPVP spectrum; S25. Perform tracking operation, and derive user speed characteristics and user gait characteristics based on the peak values ​​of the continuous PPVP spectrum for tracking target.

7. The Wi-Fi-based indoor simultaneous passive tracking and gait recognition method according to claim 6, characterized in that, In step S22, the walking gait of the test object is represented using a polar coordinate system; Let the target location be: ,in, and It is time The walking speed and direction; In the polar coordinate system, the human body movement speed is expressed using the following logic. ： ， In the formula, any point on the plane The position can be determined by a line segment length and from arrive Angle To represent, in polar coordinates, point The coordinates are , The radial direction is called the polar radius. The angle along the horizontal direction is called the polar angle.

8. The Wi-Fi-based indoor simultaneous passive tracking and gait recognition method according to claim 6, characterized in that, In step S22, the following logic is used to obtain the polar coordinate system mapping expression for the signal reflection path change rate PLCR: ， In the formula, The normal vector is calculated for the location of the Wi-Fi transceiver based on the target and the sum of the targets. Based on the polar coordinate system mapping expression of the signal reflection path change rate PLCR, the human body movement speed in the polar coordinate system is derived. Based on the PLCR polar coordinate system mapping expression for the signal reflection path change rate, the PLCR spectrum is converted into the PPVP spectrum.

9. The Wi-Fi-based indoor simultaneous passive tracking and gait recognition method according to claim 1, characterized in that, S4 further includes: S41. Extract gait features from the gait change feature dataset, perform the clustering and classification operations on the gait features, and output the results. k The class centroids and clustering results are used as classification labels; S42. Using principal component analysis (PCA) and Gaussian noise, the gait change feature dataset is expanded to obtain an augmented dataset; S43. Train a deep neural network (DNN) based on the classification labels.

10. A Wi-Fi-based indoor simultaneous passive tracking and gait recognition system, characterized in that, The system includes: The PLCR conversion module is used to define transmitter coordinates and receiver coordinates, acquire and calculate the signal reflection path length based on the Wi-Fi signal to construct a Fresnel zone model, acquire and determine the signal reflection path change rate PLCR based on the correlation between Doppler frequency shift (DFS) and signal reflection path change rate PLCR, use the Fresnel zone model to convert the signal reflection path change rate PLCR into human motion speed, and integrate the human motion speed to obtain the user's real-time position and displacement change data. The Fast PPVP generation module is used to establish the mapping rule between the signal reflection path change rate (PLCR) and the actual velocity in the polar coordinate system. Based on no less than two PLCR spectra, it generates a Fast PPVP spectrum in the polar coordinate system. The Fast PPVP generation module is connected to the PLCR conversion module. The passive tracking and gait separation module is used to obtain velocity change data and angle change data of the maximum energy value position under a preset time series based on the Fast PPVP spectrum, and to perform passive tracking and gait separation accordingly. It extracts the motion direction and velocity change in polar coordinates, obtains and performs passive tracking based on the user's initial position to obtain a velocity curve; integrates the velocity curve to track the target, obtains and processes the displacement change to obtain a gait change feature dataset, and the gait change feature acquisition module is connected to the Fast PPVP generation module. The gait recognition module is used to analyze the gait change feature dataset and perform gait recognition based on clustering and classification operations. The gait recognition module is connected to the passive tracking and gait separation module.

Images