Vehicle key accurate tracking method based on short-distance wireless signal

Abstract

The invention provides a decimeter-level vehicle key positioning and tracking method based on a short-distance wireless signal, belongs to the technical field of intelligent network connection vehicle positioning, can realize real-time accurate tracking of the position of a vehicle key, and meets the actual application requirements of vehicle unlocking and the like. In order to improve the tracking precision, an anchor node selection method is designed, LoS anchor nodes can be effectively identified, nodes with significant ranging errors can be eliminated, and thus the reliability and accuracy of the system are improved. In addition, an AEPF algorithm is provided, the particle calculation complexity is remarkably reduced in combination with a localized extended EKF, meanwhile, the real-time performance is optimized, and technical support is provided for efficient positioning in a complex scene. The method has a wide application prospect, and can be widely applied to safety management, accurate control and personalized service of intelligent network connection vehicles. For example, the system can be used for achieving the functions of keyless entry and starting of vehicles, precise parking position positioning, rapid tracking of lost vehicle keys in complex scenes and the like, provides key technical support for development of intelligent traffic and the Internet of Vehicles technology, and has remarkable market value and popularization potential.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communication and positioning technology and intelligent networked new energy vehicles, and is a method for accurately tracking a car key based on short-range wireless signals. Background Art

[0002] With the rapid development of intelligent networked new energy vehicles, vehicles are no longer isolated embedded systems, but an important part of the intelligent transportation network. The application of wireless communication technology makes vehicle operation more convenient, and users can remotely control and manage vehicles through smart devices. In this context, the car key tracking system has gradually become a hot research topic. By accurately tracking the location of the car key, security can be enhanced, that is, the vehicle will only be unlocked when the car key is within a specific range. At the same time, knowing the exact location of the car key can improve user convenience, such as automatically unlocking the door when the user approaches the vehicle, or activating specific settings of the vehicle based on the distance of the car key.

[0003] Currently, there are many positioning technologies based on short-range wireless signals, such as positioning technologies based on WiFi, Bluetooth, Ultra-Wideband (UWB) and Radio Frequency Identification (RFID). However, when these positioning technologies are applied to car key positioning and tracking systems, they all have certain limitations. For example, WiFi has a large coverage range and is suitable for medium and long-range positioning, especially in scenarios where WiFi networks are already available, and it is easy to deploy and use. However, WiFi has low positioning accuracy and high power consumption, and is not suitable for devices that are powered by batteries for a long time. Low-power Bluetooth has low power consumption and wide device compatibility, so it is suitable for short-range positioning and can be widely used at low cost. However, its positioning accuracy is low, and it usually relies on signal strength (Received Signal Strength Indicator, RSSI) to estimate the distance, resulting in large errors. UWB positioning technology can provide centimeter-level ranging accuracy, which is particularly suitable for high-precision applications, and has strong anti-interference ability, and can maintain stable positioning in complex environments. However, its main disadvantage is its high cost. RFID is a low-cost, low-power technology suitable for close-range object tracking and identity authentication. Passive RFID tags do not require battery support and can be used for a long time, but their positioning range is very short, usually only a few meters. In summary, these short-range wireless positioning technologies are limited in the application of car key positioning and tracking.

[0004] The present invention proposes a decimeter-level car key tracking system based on a single-carrier frequency hopping signal, which can realize real-time positioning and tracking of car keys. For the frequency hopping system, a new anchor node selection method is proposed, which can reliably identify the line-of-sight (LoS) anchor nodes, while excluding the anchor nodes with large ranging errors caused by non-line-of-sight (NLoS) conditions. By analyzing the changing trend of the channel impulse response (CIR) amplitude, the method can effectively distinguish between LoS and NLoS anchor nodes. In addition, an adaptive extended Kalman particle filter (AEKPF) algorithm is proposed, which combines the efficiency of the extended Kalman filter (EKF) and the flexibility of the particle filter (PF). EKF can provide local state estimation and optimize particle distribution, thereby reducing computational complexity and improving real-time performance. This hybrid method can maintain tracking accuracy even when pedestrian motion is unpredictable. Summary of the invention

[0005] The present invention aims to provide a car key tracking method based on short-range wireless signals, which can realize accurate tracking of the car key position.

[0006] The invention discloses a method for accurately tracking a car key based on a short-range wireless signal, comprising the following steps:

[0007] Step 1: Set the anchor node as Figure 1 Layout in the following way;

[0008] Step 2: Remove NLoS anchor nodes and anchor nodes with large ranging errors to maximize tracking accuracy. The specific steps are as follows:

[0009] Step 2 (1), preprocessing the CIR amplitude, including moving average smoothing and normalization operations, to remove noise and standardize the data;

[0010] Step 2 (2), using a binary hypothesis test formula, distinguishing LoS and NLoS anchor nodes based on the peak number feature;

[0011] Step 2 (3), further eliminating anchor nodes with large ranging errors through the triangle inequality to ensure that valid anchor nodes that meet the distance constraints are selected;

[0012] Step 3: Use iterative least squares method to obtain the initial position of the tracking algorithm;

[0013] Step 4: In order to further improve the tracking accuracy and enhance the real-time performance of the algorithm, the target position is tracked using the AEKPF algorithm. The specific steps of the AEKPF algorithm are as follows:

[0014] Step 4 (1) Initialize the particle state. Initialize the particle set according to the known initial position point;

[0015] Step 4 (2), adaptively adjust the noise matrix. Adaptively adjust the observation noise matrix according to the anchor nodes selected in step 2;

[0016] Step 4 (3), state prediction. Use EKF to predict the state and estimate the position of the target at the current moment based on the state transition equation of the system;

[0017] Step 4 (4), particle weight update. Use the observed data to calculate the likelihood of each particle and update the particle weight to ensure that the particle state reflects the true position;

[0018] Step 4 (5), resampling. Resample the particles to avoid the degradation problem of low-weight particles, and regenerate the particle set based on high-weight particles;

[0019] Step 4 (6), position estimation: The estimated position of the target is calculated based on the states of all particles by weighted averaging, thereby achieving accurate tracking of the vehicle key.

[0020] Beneficial Effects

[0021] The present invention has the following advantages:

[0022] 1. Efficient and reliable: The anchor node selection method can reliably identify the LoS anchor node, and can exclude points with large ranging errors, thereby improving the positioning accuracy of the system;

[0023] 2. Strong real-time performance: The AEKPF algorithm combines the high efficiency of EKF with the flexibility of PF, which not only reduces the computational complexity but also enhances the real-time performance;

[0024] 3. No need to model pedestrian trajectories: Accurate tracking can be achieved regardless of the direction or trajectory of pedestrians approaching the vehicle;

[0025] 4. Low cost and high precision: By utilizing short-range wireless frequency-hopping signals, accurate tracking can be achieved on a low-cost hardware platform, providing feasibility for large-scale application of car key positioning systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Place the anchor nodes for this system;

[0027] Figure 2 Comparison of CIR amplitude spectra before and after preprocessing;

[0028] Figure 3 Comparison of CIR amplitude spectrum fluctuations in LoS and NLoS. Specific implementation plan

[0029] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings. The method for accurately tracking a car key based on a short-range wireless signal of the present invention specifically comprises the following steps:

[0030] Step 1: Follow Figure 1 The layout shown here arranges the anchor nodes to ensure that no matter from which direction the pedestrian carrying the car key approaches the vehicle, at least one anchor node maintains a LoS connection with the car key. However, this layout may also cause some anchor nodes to be in NLoS connection with the car key. Directly using the ranging data of all anchor nodes for tracking may introduce significant errors due to NLoS conditions and multipath effects. Therefore, it is very necessary to select appropriate anchor nodes before tracking the car key.

[0031] Step 2: Remove NLoS anchor nodes and anchor nodes with large ranging errors to maximize tracking accuracy. The specific steps are as follows:

[0032] Step 2 (1), in the frequency hopping system, the center frequency changes over time, introducing different channel conditions, which in turn affects the path delay and gain. Under LoS conditions, due to the small attenuation of the direct path, the CIR amplitude remains relatively consistent between adjacent frequencies, and the impact of the reflected path is relatively weak. Under NLoS conditions, the multipath effect causes the CIR amplitude to fluctuate significantly. This difference can be used to distinguish between LoS anchor nodes and NLoS anchor nodes, specifically manifested as: there are more detectable peaks in the CIR amplitude curve under NLoS conditions. However, before peak detection, a series of preprocessing of the CIR is required to remove interference and improve detection accuracy. In order to reduce the fluctuations caused by noise, the CIR amplitude is processed using moving average smoothing. The CIR amplitude after moving average smoothing can be expressed as:

[0033]

[0034] Where w is the moving average smoothing coefficient, |h i (t)| represents the amplitude signal of the CIR at the i-th frequency point, and S represents the total number of frequency points. i (t)|The expression normalized by the maximum amplitude is:

[0035]

[0036] After preprocessing, the amplitude spectrum of CIR becomes smoother, such as Figure 2 shown.

[0037] Step 2 (2) uses the binary hypothesis test formula to distinguish LoS and NLoS anchor nodes based on the number of peak features. After preprocessing, under LoS conditions, the CIR amplitude becomes smoother, while under NLoS conditions, the CIR amplitude shows large fluctuations, resulting in more peaks, such as Figure 3 To maximize the accuracy of peak detection, only peaks with prominence greater than 0.05 are detected.

[0038] Based on the above analysis, the difference in the number of detected peaks can be used to identify LoS anchor nodes. Specifically, the LoS anchor node identification problem is formulated as a binary hypothesis test, where the LoS condition is H0 and the NLoS condition is H1. As mentioned earlier, the number of peaks under the LoS condition is significantly less than the number of peaks under the NLoS condition. Therefore, we define the hypothesis test as follows:

[0039]

[0040] Where P represents the number of peaks detected, P th Represents the recognition threshold.

[0041] Step 2 (3), after processing the CIR, the LoS anchor node can be identified. However, even under LoS conditions, due to the influence of multipath effects, the distance of the direct path may still not be accurately measured. Therefore, it is necessary to exclude anchor nodes with large ranging errors based on the principle of triangle inequality. Only anchor nodes that meet at least two of the following triangle inequalities are selected:

[0042] d i +a ij ≥d j (4)

[0043] Among them, d i and d j Represent the distance measurement results of the i-th and j-th anchor nodes respectively, a ij The distance between the i-th and j-th anchor nodes.

[0044] Step 3: Use iterative least squares to obtain the initial position of the tracking algorithm. The core of the tracking model is to predict the current state of the target based on its previous state. However, before the target starts to move, there is no prior state to refer to, so an initial point needs to be determined first. i ,y i ) should be:

[0045]

[0046] First, determine an initial point (x0, y0), which does not need to be very accurate and is mainly used for algorithm iteration. The estimated distance to the i-th anchor node is:

[0047]

[0048] Then, the error between the true distance and the estimated distance is linearized using a first-order Taylor expansion:

[0049]

[0050] Among them, the coefficient matrix H is:

[0051]

[0052] Finally, after multiple iterations, the most accurate initial estimated position is obtained:

[0053]

[0054] Here, (x0,y0) is the initial position of the tracking algorithm.

[0055] Step 4: Through the selection of the above anchor nodes, even if there is only one accurate ranging result, it can usually be used for tracking. However, there are still many challenges in tracking the trajectory of car keys. The motion trajectory of car keys is usually difficult to model. In other words, the trajectory of pedestrians is difficult to predict. For example, pedestrians approaching a vehicle may suddenly change direction or even stay still due to emergencies. In addition, their movement speed may vary from slow walking to fast running. Therefore, for pedestrians with unpredictable direction and speed, EKF is difficult to achieve the optimal tracking effect. Using PF tracking requires setting a larger particle range and number, which significantly increases the complexity and calculation time of the algorithm, which is not conducive to meeting the high real-time requirements of the car key tracking system. To this end, it is proposed to combine EKF in the local area to optimize the generation and distribution of particles, thereby reducing the computational complexity and improving the real-time performance. At the same time, the observation noise matrix will be adaptively adjusted according to the selection of anchor nodes. This algorithm is called the AEPF algorithm.

[0056] In reality, although it is impossible to accurately model pedestrian trajectories, their motion can be simplified to uniform motion because the walking speed is usually relatively stable:

[0057] x t =[x t v x,t y t v y,t ] T (10)

[0058] Among them, x t and tis the 2D coordinate of the target, v x,t and v y,t is the corresponding velocity component. The state transfer equation is as follows:

[0059] x t =Fx t-1 +w t (11)

[0060] Among them, w t is process noise, which follows Gaussian distribution The state transfer matrix F is:

[0061]

[0062] Where Δt is the sampling period. Assume that the positions of the four anchor nodes are (x1, y1), (x2, y2), (x3, y3) and (x4, y4). The observation vector z t for:

[0063] z t =[d1 d2 d3 d4] T (13)

[0064] Among them, each observation distance is:

[0065]

[0066] Here, the observation noise follows a Gaussian distribution

[0067] The specific steps of the AEKPF algorithm are as follows:

[0068] Step 4 (1) Initialize the particle state. Initialize the particle set based on the known initial position point. Assume that the first positioning point is x0 = [x0 0 y0 0] T , set the initial state of all particles to this anchor point:

[0069]

[0070] Step 4 (2), adaptively adjust the noise matrix. Adaptively adjust the observation noise matrix according to the anchor node selected in step 2. First, initialize the observation noise matrix:

[0071]

[0072] Then, define a 4×4 coefficient diagonal matrix S, whose diagonal elements are Represents the situation where the anchor node is selected. If the i-th anchor node is selected, is set to a smaller value m; otherwise is set to a larger value n (n>>m). In order to adaptively adjust the observation noise matrix, it is only necessary to multiply the observation noise matrix with the coefficient matrix:

[0073] R new =R t-1 S (17)

[0074] Step 4 (3), state prediction. Use EKF to predict the state and estimate the position of the target at the current moment based on the state transition equation of the system. First, calculate the posterior distribution x through the standard EKF process. t|t and the covariance matrix P t|t , so as to provide the particles with a density distribution that conforms to Gauss's recommendation. The state of each particle is updated as follows:

[0075]

[0076] Step 4 (4), particle weight update. Use the observed data to calculate the likelihood of each particle and update the particle weight to ensure that the particle state reflects the true position. The weight of each particle is updated according to the observed likelihood value:

[0077]

[0078] in, is based on the state of the i-th particle The predicted observed value, represents the weight of the i-th particle at time t, R new is the observed noise matrix after adjustment.

[0079] Step 4 (5), resampling. Resample the particles to avoid the degradation problem of low-weight particles, and regenerate the particle set based on high-weight particles. After resampling, the weights of all particles are reset to equal values:

[0080]

[0081] Step 4 (6): Position estimation. The estimated position of the target is calculated based on the states of all particles by weighted averaging, thereby achieving accurate tracking of the vehicle key:

[0082]

[0083] The above description is only a specific implementation mode of the present invention. Any feature disclosed in this specification, unless otherwise stated, can be replaced by other alternative features that are equivalent or have similar purposes; all the disclosed features, or all the steps in the methods or processes, except for mutually exclusive features and / or steps, can be combined in any way.

Claims

1. A decimeter-level car key tracking method based on short-range wireless signals, characterized in that: a) Reasonably arrange anchor nodes on the vehicle; b) Filter NLoS anchor nodes and anchor nodes with large ranging errors; c) using iterative least squares method to obtain the initial position of the tracking algorithm; d) Track the target position through the AEPF algorithm.

2. According to claim 1, the anchor nodes are reasonably arranged on the vehicle, characterized in that It can ensure that no matter from which direction the pedestrian carrying the car key approaches the vehicle, at least one anchor node maintains a LoS connection with the car key. Arranging the anchor nodes according to the layout shown in Figure 1 can ensure that no matter from which direction the pedestrian carrying the car key approaches the vehicle, at least one anchor node maintains a LoS connection with the car key. However, this layout may also cause some anchor nodes to be in NLoS connection with the car key. Directly using the ranging data of all anchor nodes for tracking may introduce significant errors due to NLoS conditions and multipath effects. Therefore, it is very necessary to select appropriate anchor nodes before tracking the car key.

3. According to claim 2, the NLoS anchor nodes and the anchor nodes with large ranging errors are selected, characterized in that The erroneous ranging information of these anchor nodes can be eliminated to improve the tracking accuracy. The specific steps include: ① In a frequency hopping system, the center frequency changes over time, introducing different channel conditions, which in turn affects the path delay and gain. Under LoS conditions, due to the small attenuation of the direct path, the CIR amplitude remains relatively consistent between adjacent frequencies, and the impact of the reflected path is relatively weak. Under NLoS conditions, however, the multipath effect causes significant fluctuations in the CIR amplitude. This difference can be used to distinguish between LoS anchor nodes and NLoS anchor nodes, specifically manifested as more detectable peaks in the CIR amplitude curve under NLoS conditions. However, before peak detection, a series of preprocessing of the CIR is required to remove interference and improve detection accuracy. In order to reduce the fluctuations caused by noise, the CIR amplitude is processed using moving average smoothing. The CIR amplitude after moving average smoothing can be expressed as: Where w is the moving average smoothing coefficient, |h i (t)| represents the amplitude of the CIR of the ith frequency point, and S represents the total number of frequency points. i (t)|The expression normalized by the maximum amplitude is: After preprocessing, the amplitude spectrum of CIR becomes smoother, as shown in Figure 2. ② The binary hypothesis test formula is used to distinguish LoS and NLoS anchor nodes based on the number of peak features. After preprocessing, the CIR amplitude becomes smoother under LoS conditions, while under NLoS conditions, the CIR amplitude shows large fluctuations, resulting in more peaks, as shown in Figure 3. In order to maximize the accuracy of peak detection, only peaks with a prominence greater than 0.05 are detected. Based on the above analysis, the difference in the number of detected peaks can be used to identify LoS anchor nodes. Specifically, the LoS anchor node identification problem is formulated as a binary hypothesis test, where the LoS condition is H0 and the NLoS condition is H1. As mentioned earlier, the number of peaks under the LoS condition is significantly less than the number of peaks under the NLoS condition. Therefore, we define the hypothesis test as follows: Where P represents the number of peaks detected, P th Represents the recognition threshold. ③ After processing the CIR, the LoS anchor nodes can be identified. However, even under LoS conditions, the distance of the direct path may not be accurately measured due to the multipath effect. Therefore, it is necessary to exclude anchor nodes with large ranging errors based on the triangle inequality principle. Only anchor nodes that satisfy at least two of the following triangle inequalities are selected: d i +a ij ≥d j (4) Among them, d i and d j Represent the distance measurement results of the i-th and j-th anchor nodes respectively, a ij The distance between the i-th and j-th anchor nodes.

4. According to claim 3, the initial position of the tracking algorithm is obtained using an iterative least squares method, characterized in that The initial position of the tracking algorithm is given. The core of the tracking model is to predict the current state of the target based on its previous state. However, before the target starts to move, there is no previous state to refer to, so an initial point needs to be determined first. i ,y i ) should be: First, determine an initial point (x0, y0), which does not need to be very accurate and is mainly used for algorithm iteration. The estimated distance to the i-th anchor node is: Then, the error between the true distance and the estimated distance is linearized using a first-order Taylor expansion: Among them, the coefficient matrix H is: Finally, after multiple iterations, the most accurate initial estimated position is obtained: Here, (x′0, y′0) is the initial position of the tracking algorithm.

5. According to claim 4, the target position is tracked by the AEPF algorithm, characterized in that It can improve the tracking accuracy and enhance the real-time performance of the algorithm. Through the selection of the above anchor nodes, even if there is only one accurate ranging result, it can usually be used for tracking. However, there are still many challenges in tracking the trajectory of car keys. The motion trajectory of car keys is usually difficult to model. In other words, the trajectory of pedestrians is difficult to predict. For example, pedestrians approaching vehicles may suddenly change direction or even stay still due to emergencies. In addition, their moving speed may vary from slow walking to fast running. Therefore, for pedestrians with unpredictable direction and speed, EKF is difficult to achieve the optimal tracking effect. Using PF tracking requires setting a larger particle range and number, which significantly increases the complexity and calculation time of the algorithm, which is not conducive to meeting the high real-time requirements of the car key tracking system. To this end, it is proposed to combine EKF in the local area to optimize the generation and distribution of particles, thereby reducing the computational complexity and improving the real-time performance. At the same time, the observation noise matrix will be adaptively adjusted according to the selection of anchor nodes. This algorithm is called the AEPF algorithm. In reality, although it is impossible to accurately model pedestrian trajectories, their motion can be simplified to uniform motion because the walking speed is usually relatively stable: x t =[x t v x,t y t v y,t ] T (10) Among them, x t and t is the 2D coordinate of the target, v x,t and v y,t is the corresponding velocity component. The state transfer equation is as follows: x t =Fx t-1 +w t (11) Among them, w t is process noise, which follows Gaussian distribution The state transfer matrix F is: Here, Δt is the sampling period. Assume that the positions of the four anchor nodes are (x1, y1), (x2, y2), (x3, y3) and (x4, y4). The observation vector z t for: z t [d1 d2 d3 d4 T (13) Among them, each observation distance is: Here, the observation noise follows a Gaussian distribution The specific steps of the AEFP algorithm are as follows: ① Initialize the particle state. Initialize the particle set based on the known initial position point. Assume that the first positioning point is x0 = [x′0 0 y′0 0] T , set the initial state of all particles to this anchor point: ② Adaptively adjust the noise matrix. Adaptively adjust the observation noise matrix according to the anchor node selected in step 2. First, initialize the observation noise matrix: Then, define a 4×4 coefficient diagonal matrix S, whose diagonal elements are Represents the situation where the anchor node is selected. If the i-th anchor node is selected, is set to a smaller value m; otherwise is set to a larger value n (n>>m). In order to adaptively adjust the observation noise matrix, it is only necessary to multiply the observation noise matrix with the coefficient matrix: R new =R t-1 S (17) ③ State prediction. Use EKF to predict the state and estimate the position of the target at the current moment based on the state transition equation of the system. First, calculate the posterior distribution x through the standard EKF process. t|t and the covariance matrix P t|t , so as to provide the particles with a density distribution that conforms to Gauss's recommendation. The state of each particle is updated as follows: ④ Particle weight update. Use the observed data to calculate the likelihood of each particle and update the particle weight to ensure that the particle state reflects the true position. The weight of each particle is updated according to the observed likelihood value: in, is based on the state of the i-th particle The predicted observed value, represents the weight of the i-th particle at time t, R new is the observed noise matrix after adjustment. ⑤ Resampling. Resample the particles to avoid the degradation problem of low-weight particles, and regenerate the particle set based on high-weight particles. After resampling, the weights of all particles are reset to equal values: ⑥ Position estimation. The estimated position of the target is calculated based on the state of all particles by weighted average, thereby achieving accurate tracking of the vehicle key: .

Images