RFID trajectory tracking system based on least square speed self-adaption

Abstract

The invention provides an RFID trajectory tracking system based on least square speed adaptation, and the method comprises the steps: firstly building a half-wavelength observation distance expansion model, and effectively solving the problem of error divergence caused by a conventional phase unwrapping scheme; then, generating an initial position candidate set by using a triangular centroid weighting algorithm, performing instantaneous speed estimation by using a least square algorithm, and introducing a dynamic threshold-weight constraint mechanism to obtain a high-quality initial state set; thirdly, speed estimation based on least square is fused in the EKF, speed data are updated through the least square method before each round of antenna scheduling is started, target tracking trajectories estimated based on different initial positions are obtained, and the problem that in the target tracking process, the speed suddenly changes or the tracking error is large during steering is solved; and finally, dynamically deleting the tracking trajectory with the zero weight, and selecting the trajectory with the maximum weight value as the final tracking trajectory of the target, thereby effectively reducing the calculation complexity of the algorithm. The invention provides an RFID trajectory tracking system which is convenient and simple to deploy, can realize trajectory tracking of a target tag, and is suitable for most indoor positioning scenes.

Description

Technical Field

[0001] The invention belongs to the field of RFID indoor positioning, and in particular relates to an RFID trajectory tracking system based on least squares velocity adaptation. Background Art

[0002] Radio Frequency Identification (RFID) is an automatic identification technology that uses radio frequency (RF) for contactless, two-way data communication. It relies on data communication between an RFID reader and a passive tag to identify the target. Passive RFID technology offers advantages such as low cost, fast read speed and wide range, no need for additional equipment, and high positioning accuracy. It has been widely used in various fields, including supply chain management, logistics management, access control, production line automation, and warehouse management. RFID-based positioning technology has also become a research hotspot and is one of the information technologies with great development potential.

[0003] The sampling frequency of the Impinj R420 is approximately 30Hz. If a passive tag moves at a high speed within the coverage area of ​​the device's antenna array, some phase measurement data may be lost. This data loss phenomenon will directly affect the continuity after phase expansion, which in turn has an adverse effect on the accuracy of distance and position calculations. In addition, when the motion trajectory of the passive tag is more complex, the superposition of multipath effects will be more significant. Currently, in tracking systems that use antenna arrays, additional hardware configuration is usually required to increase the sampling frequency. This solution not only increases the complexity of the system, but also brings additional economic costs. Therefore, developing a tracking system that can achieve high-precision passive tag positioning without increasing the hardware burden remains a technical problem that needs to be solved urgently. Summary of the Invention

[0004] Based on this, the present invention proposes an RFID trajectory tracking system based on least squares (LS) speed adaptation to address the problem of system tracking instability due to phase unwrapping failure, decreased tracking accuracy when performing complex trajectory tracking tasks or when the passive tag's moving speed changes. The system can provide the real-time position of the positioning target.

[0005] The present invention first provides an RFID trajectory tracking system based on least squares velocity adaptation:

[0006] An RFID trajectory tracking system based on least squares speed adaptation includes an RFID reader, a passive tag, four directional antennas, several network cables, several radio frequency cables, and a data processing terminal.

[0007] The RFID reader and directional antenna are used to collect the phase information required for positioning. The directional antennas are distributed in the XOY plane in a rectangular distribution. The data processing terminal is responsible for data processing and positioning solution.

[0008] Optionally, the RFID reader is Impinj R420;

[0009] Optionally, the passive tag model is the commercial tag Alien-9746;

[0010] Optionally, the commercial directional antenna is VIKITEK VA094;

[0011] Optionally, the data processing terminal is configured with an i5-13500H processor, 16G running memory, and a storage device required for hardware control and positioning;

[0012] To solve the technical problem, the present invention provides an RFID trajectory tracking method based on least squares velocity adaptation, which specifically includes the following steps:

[0013] Step S1, the antenna array is composed of four directional antennas distributed in the XOY plane. The directional antennas are arranged clockwise along a rectangle with a side length of 80 cm, namely directional antenna 1, directional antenna 2, directional antenna 3 and directional antenna 4. The navigation coordinate system is established with directional antenna 1 as the origin of the coordinate system. The passive tag is fixed on the top of the writing tool, and the experimenter writes at the midpoint of any matrix edge. The RFID reader and directional antenna transmit signals through the passive tag. The passive tag loads its own modulated signal onto the signal, which is then transmitted to the directional antenna, and the collected phase information is transmitted to the data processing terminal in real time through the RFID reader.

[0014] Step S2: Using RFID ranging and a half-wavelength observation distance expansion model, a specific phase value may correspond to multiple distance values, and these distances differ by integer multiples of a half-wavelength. The maximum half-wavelength expansion number determined by the measurement range requires expanding the measured phase to a corresponding set of distances. Each observed distance will result in a different positioning result, thus completing the distance expansion corresponding to each phase.

[0015] Step S3: Estimate the initial state of the passive tag using the triangle weighted centroid positioning algorithm and the LS algorithm, including the X-axis and Y-axis coordinates and velocity. Then, perform a preliminary screening of candidate locations by establishing a spatial constraint model, and perform a secondary filtering of the location set using the non-negative condition of the weight value, thereby obtaining a small number of high-quality initial state sets.

[0016] Step S4: The calculated initial position and the speed fitted by LS are used as the initial state input of the Kalman filter, and the speed is updated after each round of antenna scheduling, until the passive tag position coordinates corresponding to each timestamp are solved, thereby completing the tracking of all candidate trajectories;

[0017] Step S5: perform trajectory pruning during the tracking process, that is, gradually reduce the number of trajectories according to the calculated real-time weights, and select the optimal trajectory as the result output according to the weight values ​​of the remaining candidate trajectories.

[0018] The beneficial effects of the present invention are as follows: the RFID reader and directional antenna of the present invention are both commercial devices, with a simple system structure and low cost, suitable for most typical indoor environments, and can achieve trajectory tracking with simple deployment. Compared with general tracking systems, it has a wider range of applications in actual scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a system structure diagram of the present invention.

[0020] Figure 2 This is a diagram of the half-wavelength observation distance expansion model of the present invention.

[0021] Figure 3 It is the initial position positioning model diagram of the present invention.

[0022] Figure 4 This is a speed estimation model diagram of the present invention.

[0023] Figure 5 Flowchart of the speed state update algorithm. DETAILED DESCRIPTION

[0024] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, which are only one of the embodiments of the present invention and are not all embodiments.

[0025] The present invention provides an RFID trajectory tracking system based on least squares speed adaptation, the system structure is as follows Figure 1 In this embodiment, the RFID reader in the figure uses the Impinj R420, the passive tag model in the tag array is Alien-9746, four commercial VIKITEKVA094 directional antennas are used, and the data processing terminal is configured with an i5-13500H processor, 16G RAM, and storage devices required for hardware control and positioning.

[0026] When using the present invention, the passive tag is fixed on the upper end of the writing tool, and the experimenter performs a writing operation at the coordinate position of (0cm, 40cm). The terminal is used to control the RFID reader to collect the passive tag phase and implement the positioning algorithm, such as Figure 1 shown.

[0027] The detailed steps for implementing the positioning algorithm are as follows:

[0028] Step S1: In order to track the passive tag, we first need to expand the observation distance corresponding to each phase. Assuming d is the distance between the directional antenna and the passive tag, the total backscatter communication distance is 2d. The relationship between phase and distance can be expressed as

[0029]

[0030] where the wavelength is λ, is the phase value received by the RFID reader, Phase offset introduced by the system hardware, including the phase offset introduced by the RFID reader transmitter circuit Phase offset introduced by the reader receiving circuit and phase shift introduced by passive tag circuits

[0031] The half-wavelength phase unwrapping model is used to extend the phase distance. Figure 2 As shown, due to the periodicity of the signal, a specific phase value There may be multiple distance values ​​corresponding to the phase difference, and the distances differ by an integer multiple of half a wavelength. Therefore, according to the theoretical relationship between phase and distance, the phase value is measured. Corresponding observation distance Z t Can be expanded to:

[0032]

[0033] Where n is the maximum half-wavelength extension number determined by the measurement range. It is worth noting that at time t, n independent observation distances are solved, each of which will lead to a different positioning result. Only one observation distance needs to be selected as the positioning result at time t;

[0034] Step S2: Use the triangle centroid weighted algorithm and the least squares algorithm to complete the initial state estimation of the passive tag. When using three directional antennas to realize the initial position positioning of the passive tag using the triangle weighted centroid algorithm, the coordinates of directional antenna 1, directional antenna 2, and directional antenna 3 are the center of the circle, and the distance d1, d2, and d3 from each directional antenna to the passive tag are the radius of the circle to form three intersection points o1, o2, and o3. These intersection points will form a triangular area. By calculating the centroid o of the triangle, the position of the passive tag can be determined. The coordinates (x0, y0) of the passive tag to be tested are:

[0035]

[0036] in, is the intersection o i However, in the actual environment, due to the influence of uncertainty noise, the three circles may not form a common intersection area. Therefore, the present invention first determines whether the two circles intersect. If they do not intersect, it means that no valid intersection point can be obtained under the current distance parameter. The system will automatically discard the current distance data and wait for the next more accurate distance parameter input. If they intersect, the intersection point of the two circles is calculated. When there are 4 directional antennas, at most 1 intersection point is generated. From each group, three points are selected to form a triangle. Then, by comparing all possible triangles, the one with the shortest perimeter is selected because its centroid is closer to the true location. Finally, the centroid of this triangle is calculated, and the result is the optimal coordinate of the target location.

[0037] For each directional antenna, the algorithm initially calculates multiple possible distances based on the current phase measurement. To obtain all possible positions, the algorithm must consider all possible observation distances. However, as the number of phase readings increases, the number of candidate initial positions becomes excessive, which leads to significant computational overhead and latency. Therefore, a method combining spatial constraints with weighted filtering is used to optimize the selection of candidate positions. The key is to initially screen candidate positions by establishing a spatial constraint model and then perform a secondary filtering of the position set using the non-negativity condition of the weight values.

[0038] like Figure 4 As shown, the radial displacement Δd of the passive tag can be expressed as:

[0039]

[0040] Assumptions is the ith directional antenna at time stamp t n The phase measurement value at , the additional phase shift is naturally removed due to the phase difference, so the radial velocity of the passive tag of directional antenna i Can be calculated as:

[0041]

[0042] The relationship between radial velocity and instantaneous velocity is:

[0043]

[0044] where α is the radial velocity With instantaneous speed The angle between . Decomposing the velocity is Substituting into formula (5) we can get:

[0045]

[0046] Assuming that the RFID reader is equipped with N directional antennas, N linear functions can be constructed and expressed in matrix form as follows:

[0047]

[0048] Where Λ is the system matrix, with a size of N×2, where each row corresponds to an observation of a directional antenna, v is the instantaneous velocity vector to be solved in this paper, with a size of 2×1, and b is the observation vector, with a size of N×1, where each element is the square of the modulus of the radial velocity observed by the corresponding directional antenna. Therefore, the instantaneous velocity vector v is solved as:

[0049] v=(Λ T Λ) -1 Λ T b (9)

[0050] The v calculated by the formula is the estimated instantaneous speed (v x (t n ),v y (t n ), for any sampling point t in the continuous time series n , the same calculation process is sampled and iterated to estimate the instantaneous speed of the passive tag at that moment;

[0051] Step S3: First, the obtained initial state value is used as the input of the EKF() algorithm to achieve accurate tracking of the system state. Before each directional antenna scheduling process is started, the velocity information received at directional antenna 0 is updated using the least squares method to generate multiple candidate trajectory sets based on different initial position estimates.

[0052] This patent models the passive tag mobility state as:

[0053] X t =[P x,t ,P y,t ,V x,t,V y,t ] T (10)

[0054] Among them, P x,t and P y,t is the position coordinate of the passive tag at time t, V x,t and V y,t is the corresponding speed value, Σ t is the state covariance matrix, so the state transfer equation of the passive tag at time t is:

[0055]

[0056] Where Δt is the time difference between adjacent moments, Α is the system matrix describing the mobility of the passive tag, s t is the system Gaussian noise, that is Q t is the process noise covariance matrix.

[0057] Perform two-dimensional tracking on the passive tag and construct the observation model at time t as follows:

[0058]

[0059] Among them, A x,i and A y,i is the position of directional antenna i at the current moment, which is a known condition. R t is the observation noise covariance matrix, which is linearized using the extended Kalman filter and converted into a discrete-time linear representation. Specifically, the observation equation after first-order Taylor series expansion near the current estimation point becomes:

[0060] z t =H t X t +u t (13)

[0061]

[0062] Among them H t is the observation Jacobian matrix, which involves computing the partial derivatives of each element of the observation function with respect to each state variable.

[0063] Therefore, for the candidate trajectory l, the EKF first uses the state transfer matrix to predict the state and state covariance at time t+1 as follows:

[0064]

[0065] The Kalman gain is calculated as:

[0066]

[0067] Then the observation equation and the predicted Combined update

[0068]

[0069] In order to solve the problem that the position estimated by the traditional EKF algorithm gradually deviates from the true position as the observation value increases, a velocity update mechanism is introduced. This mechanism is activated after each round of directional antenna scheduling starts. After the start of a new round of directional antenna scheduling cycle, the estimated state of the candidate trajectory l at the next moment can be obtained. for:

[0070]

[0071] The velocity value in the state is determined by EKF. After inputting the new observation value, the instantaneous velocity at time t+1 can be estimated by the least squares fitting method. The velocity vector solved by LS is for:

[0072]

[0073] The estimated velocity value is then incorporated into the state vector At this time, the state vector It can be expressed as:

[0074]

[0075] The observation model is updated as follows:

[0076]

[0077] During this round of directional antenna scheduling cycle, the next state estimation is continued based on EKF through the updated state vector and observation model, such as Figure 5 Shown is a flow chart of the speed status update method.

[0078] Step S4: By performing EKF iteration on all candidate initial positions, the state estimation of each trajectory can be continuously adjusted and optimized according to the new measurement data. After correctly estimating all possible states, the weight of each trajectory is calculated, and the trajectory with the largest weight is taken as the final tracking result.

[0079] Assume that the RFID tracking system has N directional antennas and the initial state of the candidate trajectory l is for:

[0080]

[0081] This patent normalizes the initial weight of trajectory l Expressed as:

[0082]

[0083] in, Indicates the phase of the ith directional antenna at a given initial moment Under the condition of , the initial position of trajectory l is The probability of is a Gaussian probability density function with mean μ and variance R, represents the distance between directional antenna i and the initial position The probability density of is the observation distance based on the phase model when the trajectory l determines the initial position, is the observation noise covariance matrix.

[0084] The weight at time t has been obtained during the tracking process, and the weight at time t+1 can be expressed as:

[0085]

[0086]

[0087] Among them, Q t+1 is the process noise covariance matrix at time t+1, and the observation likelihood Indicates a known state A specific observation distance is observed under the condition of The probability of state transition Indicates that the state at the previous moment is known Transfer to the next state probability.

[0088] Based on the above analysis, the weight of each state is the accumulation of the historical weights of the trajectory. When there is a significant deviation between the system's estimated position and the actual position, the weight value of the current state will approach zero, causing the weight values ​​of all subsequent states to be zero, and the trajectory will therefore lose the opportunity to be selected. Based on this, this patent designs an iterative optimization strategy that uses the dynamic value of the weight to gradually reduce the number of trajectories processed by the system. Finally, the system selects the optimal trajectory as the result output based on the weight values ​​of the remaining candidate trajectories, thereby achieving real-time positioning and tracking of the target.

[0089] The above-described implementation examples are merely preferred implementation examples of the present invention and are not intended to limit the present invention. The scope of protection of the present invention is not limited to the above-described embodiments. For technicians in the same field, all equivalent replacements, modifications, and modifications made based on the disclosure of the present invention should be included in the scope of protection of the present invention.

Claims

1. An RFID trajectory tracking system based on least squares velocity adaptation, characterized in that: The system includes an RFID reader / writer, a passive tag, four directional antennas, several network cables and radio frequency cables, and a data processing terminal. The RFID reader / writer and directional antennas are used to collect phase information required for positioning. The directional antennas are distributed in an XOY plane in a rectangular shape. The data processing terminal is responsible for data processing and positioning calculation. The system also includes a method for constructing an RFID antenna array: Step S1: An RFID antenna array is formed by four directional antennas distributed in the XOY plane. The directional antennas are arranged clockwise along a rectangle with a side length of 80 cm, namely directional antenna 1, directional antenna 2, directional antenna 3, and directional antenna 4. A navigation coordinate system is established with directional antenna 1 as the origin of the coordinate system. The passive tag was fixed on the top of the writing tool, and the experimenter wrote at the midpoint of any matrix edge; In step S2, the RFID reader and directional antenna transmit signals through the passive tag, which adds its own modulated signal to the signal and transmits it to the directional antenna. The RFID reader transmits the collected phase information to the data processing terminal in real time.

2. The RFID trajectory tracking system based on least squares velocity adaptation according to claim 1, characterized in that: A trajectory tracking method for avoiding phase unwrapping is provided, comprising the following steps: Step S1: Distance extension based on the RFID phase ranging model; assuming d is the distance between the directional antenna and the passive tag, the total backscatter communication distance is 2d, and the relationship between phase and distance can be expressed as: where the wavelength is λ, is the phase value received by the RFID reader, Phase offset introduced by the system hardware, including the phase offset introduced by the RFID reader transmitter circuit Phase offset introduced by the reader receiving circuit and phase shift introduced by passive tag circuits Due to the periodicity of the signal, a specific phase value There may be multiple distance values ​​corresponding to these distances, and the difference between these distances is an integer multiple of half a wavelength. Therefore, the measured phase value Corresponding observation distance Z t Can be expanded to: Where n is the maximum half-wavelength extension number determined by the measurement range. At time t, n independent observation distances are solved. Each observation distance will lead to a different positioning result. Only one observation distance needs to be selected as the positioning result at time t. At this point, the distance extension corresponding to the RFID phase is completed. Step S2, initial state estimation of the passive tag; first, the initial position set is estimated. When three directional antennas are used to realize the initial position positioning of the passive tag using the triangle weighted centroid algorithm, the coordinates of directional antenna 1, directional antenna 2, and directional antenna 3 are the center of the circle, and the distance d1, d2, and d3 from each directional antenna to the passive tag are the radius of the circle to form three intersection points o1, o2, and o3. These intersection points will form a triangular area. By calculating the centroid o of the triangle, the position of the passive tag can be determined. The coordinates (x0, y0) of the passive tag to be tested are: in, is the intersection o i However, in the actual environment, due to the influence of uncertainty noise, the three circles may not form a common intersection area. Therefore, this patent first determines whether the two circles intersect. If not, the system will automatically discard the current distance data and wait for the next more accurate distance parameter input. If they intersect, the intersection point of the two circles is calculated. When positioning using four directional antennas, at most one intersection point is generated. Group, select three points from each group to form a triangle, then compare all possible triangles and select the one with the shortest perimeter. Finally, calculate the center of mass of the triangle according to formula (3), and the result is the optimal coordinate of the target position; According to the distance expansion model, each directional antenna has a different observation distance from the passive tag. Combining these four directional antennas in groups of three can construct several potential locations. A spatial constraint model is then used to initially screen the candidate locations, and the location set is then secondary filtered using the non-negative condition of the weight values. Thus, a small number of high-quality initial location sets have been estimated. The second is the estimation of the initial velocity. The radial displacement Δd during the movement of the passive tag can be expressed as: Assumptions is the ith directional antenna at time stamp t n The phase measurement value at , the additional phase shift is naturally removed due to the phase difference, so the radial velocity of the passive tag of directional antenna i Can be calculated as: The relationship between radial velocity and instantaneous velocity is: where α is the radial velocity With instantaneous speed The angle between the two, the speed is decomposed into Substituting into formula (5) we can get: Assuming that the RFID reader is equipped with N directional antennas, N linear functions can be constructed and expressed in matrix form as follows: Where Λ is the system matrix, with a size of N×2, where each row corresponds to an observation of a directional antenna, v is the instantaneous velocity vector to be solved in this paper, with a size of 2×1, and b is the observation vector, with a size of N×1, where each element is the square of the modulus of the radial velocity observed by the corresponding directional antenna. Therefore, the instantaneous velocity vector v is solved as: The v calculated by the formula is the estimated instantaneous speed (v x (t n ),v y (t n ), for any sampling point t in the continuous time series n , the same calculation process is sampled and iterated to estimate the instantaneous speed of the passive tag at that moment; At this point, the estimation of the initial state of the passive tag has been completed, including the initial position and initial velocity set; Step S3, candidate trajectory tracking; This patent models the movement state of the passive tag as: X t =[P x,t ,P y,t ,V x,t ,V y,t ] T (10) Among them, P x,t and P y,t is the position coordinate of the passive tag at time t, V x,t and V y,t is the corresponding speed value, Σ t is the state covariance matrix, so the state transfer equation of the passive tag at time t is: Where Δt is the time difference between adjacent moments, Α is the system matrix describing the mobility of the passive tag, s t is the system Gaussian noise, that is Q t is the process noise covariance matrix; Perform two-dimensional tracking on the passive tag and construct the observation model at time t as follows: Among them, A x,i and A y,i is the position of directional antenna i at the current moment, which is a known condition. R t is the observation noise covariance matrix, which is linearized using the extended Kalman filter and converted into a discrete-time linear representation. Specifically, the observation equation after first-order Taylor series expansion near the current estimation point becomes: z t =H t X t +u t (13) Among them, H t is the observation Jacobian matrix, which involves computing the partial derivatives of each element in the observation function with respect to each state variable; For the candidate trajectory l, the Kalman filter is first extended to use the state transfer matrix to predict the state and state covariance at time t+1: The Kalman gain is calculated as: Then the observation equation and the predicted Combine to update: In order to solve the problem that the position estimated by the traditional extended Kalman filter algorithm gradually deviates from the true position as the observation value increases, a velocity update mechanism is introduced. This mechanism is activated after each round of directional antenna scheduling starts. After the start of a new round of directional antenna scheduling cycle, the estimated state of the candidate trajectory l at the next moment can be obtained. for: The velocity value in the state is determined by the extended Kalman filter. After inputting the new observation value, the instantaneous velocity at time t+1 can be estimated by the least squares fitting method. The velocity vector solved by the least squares fitting method is for: The estimated velocity value is then incorporated into the state vector At this time, the state vector It can be expressed as: The observation model is updated as follows: During this round of directional antenna scheduling, the next step of state estimation is continued based on the extended Kalman filter through the updated state vector and observation model. At this point, the trajectory tracking of all candidate positions is completed; Step S4, target trajectory output: Assuming that the RFID tracking system has N directional antennas, the initial state of the candidate trajectory l for: This patent normalizes the initial weight of trajectory l Expressed as: in, Indicates the phase of the ith directional antenna at a given initial moment Under the condition of , the initial position of trajectory l is The probability of represents the distance between directional antenna i and the initial position The Gaussian probability density of is the observation distance based on the phase model when the trajectory l determines the initial position, is the observation noise covariance matrix; Assuming that the weight at time t has been obtained during the tracking process, the weight at time t+1 can be expressed as: Among them, Q t+1 is the process noise covariance matrix at time t+1, and the observation likelihood Indicates a known state A specific observation distance is observed under the condition of The probability of state transition Indicates that the state at the previous moment is known Transfer to the next state probability; According to the above analysis, the weight of each state is the accumulation of the historical weights of the trajectory. When there is a significant deviation between the estimated position and the true position, the weight value of the current state will approach zero, causing the weight values ​​of all subsequent states to be zero, and the trajectory will therefore lose the opportunity to be selected. Based on this, the dynamic value of the weight is used to gradually reduce the number of trajectories processed by the system, and then the trajectory with the best weight value is selected from the remaining candidate trajectories as the result output, thereby completing real-time tracking of the target.

Images