Multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter

Abstract

The invention proposes a multi-target trajectory tracking and fusion method based on a multi-observation trajectory probability hypothesis density filter, and the method comprises the steps: transmitting a scanning beam to a detection region through a 5G base station, receiving an echo signal of the scanning beam through a plurality of radar receivers, and extracting the information of the echo signal; each radar receiver carries out calculation to obtain parameters of a target and clutter and estimation values of Gaussian distribution of the target and the clutter, a multi-observation trajectory probability hypothesis density filter is used for calculating the parameters of the target and the clutter and the estimation values of Gaussian distribution of the target and the clutter, and an estimation result of the multi-observation trajectory probability hypothesis density filter on a multi-target trajectory is obtained; the track fusion center fuses the estimation results of the multi-target tracks obtained by all the radar receivers to obtain a fusion result; and each radar receiver updates the trajectory state of the radar receiver according to the fusion result, and multi-target trajectory tracking and fusion based on the multi-observation trajectory probability hypothesis density filter are completed.

Description

Technical Field

[0001] The present invention relates to a multi-target trajectory tracking and fusion method, in particular to a multi-target trajectory tracking and fusion method based on a multi-observation trajectory probability hypothesis density filter. Background Art

[0002] The information provided in this section is only background information related to the present disclosure, and it is not necessarily prior art.

[0003] In a passive radar system, in addition to digital audio / video broadcast signals, MIMO-OFDM (Multi-input Multi-output Orthogonal Frequency Division Multiplexing) signals transmitted by 5G base stations are also important sensing signals. The integration of MIMO technology and OFDM signals has advantages such as high arrival angle estimation resolution and robustness against frequency-selective fading. Currently, there are many methods for studying MIMO-OFDM signals to improve sensing performance, such as compressive sensing algorithms based on MIMO-OFDM signals, optimizing power or carrier resource balance for communication and sensing performance. Recently, some researchers have studied the existing 5G base station cell synchronization signals to detect vehicles. When the signal cyclic prefix length is long enough, MIMO-OFDM signals can even be used for imaging.

[0004] In addition to detection, MIMO-OFDM signals can also be used for multi-target tracking. When targets exist in complex multi-reflection scenarios, multi-target tracking algorithms can effectively eliminate clutter and accurately estimate the states and quantities of targets. Traditional multi-target algorithms mostly assume the Multiple Hypothesis Tracker (MHT) and the Joint Probabilistic Data Association (JPDA). However, these two algorithms require prior knowledge of the number of targets and have high algorithmic complexity in multi-clutter scenarios. Another type is the multi-target tracking algorithm based on random finite sets. The three basic algorithms include the Probability Hypothesis Density Filter (PHDF), the Cardinalized PHD Filter (CPHDF), and the Multi-Bernoulli Filter (MBF). Other algorithms based on random finite sets are derived and improved from these three algorithms. Such algorithms have low algorithmic complexity and can automatically estimate the number of targets, but accurate modeling of targets and clutter is required. Therefore, studying the working principle of 5G broadcast signals, analyzing the characteristics of target parameters and clutter parameters generated under this principle, and designing a multi-target tracking filter are of great significance for realizing multi-target tracking in 5G scenarios.

[0005] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] Objective of the Invention: The technical problem to be solved by the present invention is to provide a multi-target trajectory tracking and fusion method based on a multi-observation trajectory probability hypothesis density filter in view of the deficiencies of the prior art.

[0007] To solve the above technical problem, the present invention discloses a multi-target trajectory tracking and fusion method based on a multi-observation trajectory probability hypothesis density filter, including the following steps:

[0008] Step 1: Use a 5G base station to emit scanning beams to a detection area, use multiple radar receivers to receive the echo signals of the scanning beams, and extract the information on the echo signals.

[0009] Step 2: Each radar receiver calculates the information to obtain the parameters of targets and clutter and the estimated values of their Gaussian distributions.

[0010] Step 3: Each radar receiver uses the MD-TPHD filter to calculate the parameters of the target and clutter and their estimated values of Gaussian distribution, and obtains its estimation result of the multi-target trajectory.

[0011] Step 4: The trajectory fusion center fuses the estimation results of the multi-target trajectories obtained by all radar receivers to obtain a fusion result.

[0012] Step 5: Each radar receiver updates the trajectory state of the radar receiver according to the fusion result, and completes the multi-target trajectory tracking and fusion based on the multi-observation trajectory probability hypothesis density filter.

[0013] Beneficial effects:

[0014] The present invention provides a solution that uses the 5G base station broadcast signal as the sensing signal, solves the multi-target trajectory tracking in a multi-clutter environment and fuses the tracking trajectories of multiple receivers by designing a filter, which can effectively eliminate clutter and relatively accurately estimate the state and quantity of the target trajectory, has a fast response to the appearance and disappearance of the target trajectory. Considering the limited sensor field of view of the radar receiver, the tracking of the target trajectory in the entire detection area is realized by fusing the tracking trajectories of multiple radar receivers and the missed detection is reduced. Description of the drawings

[0015] The following further specifically describes the present invention in conjunction with the drawings and specific embodiments, and the above and / or other advantages of the present invention will become clearer.

[0016] Figure 1 is the system model diagram of the method of the present invention.

[0017] Figure 2 is the work flow chart of the present invention.

[0018] Figure 3a is the comparison diagram of the trajectory quantity estimation between the present invention and other methods.

[0019] Figure 3b is the comparison diagram of the root mean square error of the trajectories between the present invention and other methods.

[0020] Figure 4a is the comparison diagram of the target quantity estimation between the single receiver and the multi-receiver fusion of the present invention.

[0021] Figure 4b is the comparison diagram of the root mean square error of the trajectory sets between the single receiver and the multi-receiver fusion of the present invention. Specific embodiments

[0022] When using 5G base station broadcast signal sensing, multiple observation values may be generated by the detected targets and there is a large amount of clutter in the environment. The present invention designs an MD-TPHD (Multi-detection Trajectory Probability Hypothesis Density) filter to eliminate clutter and estimate the states and quantities of multi-target trajectories. Considering the limited field of view of the radar receiver sensors, the entire sensing area is detected by fusing the trajectories from all radar receivers at the trajectory fusion center, and the fusion result is fed back to the radar receivers to improve the tracking performance.

[0023] The present invention includes a 5G base station, multiple radar receivers, several targets, and a trajectory fusion center. Among them, the 5G base station covers the entire detection area in a beam scanning manner, and there is a common coverage area between adjacent beams. Multiple targets move within the entire detection area, and the targets appear and disappear at different times. The radar receivers receive the echo signals from the targets, which contain the clutter signals in the environment. First, a Bayesian learning method is designed to obtain the estimated values and their distributions of the targets and clutter, and then it is processed by the designed MD-TPHD filter (Multi-observation Trajectory Probability Hypothesis Density filter) to achieve clutter elimination and multi-target trajectory tracking. Finally, the tracking results of different radar receivers are fused and the results are fed back. The steps are as follows:

[0024] Step 1: The 5G base station scans a total of B beams in a scanning cycle. Each beam modulates and transmits an OFDM signal. The rth radar receiver receives the echo signals from the B beams, and extracts the information on each subcarrier of the echo signals through matched filtering.

[0025] Among them, the baseband signal of the bth beam transmitted by the base station is:

[0026]

[0027] Among them, Q is the number of OFDM symbols, N s is the number of subcarriers, γ b is the beamforming vector, s b,q (n) is the symbol modulated on the nth subcarrier of the qth symbol of the bth beam (such as QAM modulation), Δf is the bandwidth of each subcarrier, T s =T + T cp is the length of each symbol, where T = 1 / Δf, T cp is the length of the cyclic prefix, ξ(t - (q - 1)T s ) is a rectangular pulse with a duration of T s When (q - 1)T s < t < qT s ξ(t - (q - 1)T s) = 1, otherwise ξ(t - (q - 1)T s ) = 1. The baseband signal of the b-th beam is transmitted after upconversion, and is where f c is the carrier frequency, and Re(·) is to take the real part of the signal.

[0028] The r-th radar receiver receives the target echo and clutter of the b-th beam. After downconversion, the signal is:

[0029]

[0030] O r,b is the sum of the number of targets and the number of clutter with Poisson distribution within the coverage of the b-th beam, is the antenna array gain, M T 、M R are the number of transmit antennas and receive antennas respectively, A r,b,o is the attenuation factor, f r,b,o 、β r,b,o and τ r,b,o are the Doppler frequency shift, angle of arrival and time delay corresponding to the target or clutter respectively, α b is the transmission angle,

[0031] v r,b (t) is Gaussian white noise with a mean of 0 and a variance of n 0 . After matched filtering, the signal of the n-th subcarrier on the q-th symbol of the b-th beam received by the r-th radar receiver is:

[0033]

[0034] where, is an M R ×1 vector, p n is the power on the n-th subcarrier, is Gaussian white noise with a mean of 0 and a variance of p n n 0 / T.

[0035] Step 2: Use the Complex Multi-task Bayesian Compressed Sensing (CMT-BCS) method to roughly estimate the target and clutter parameters for the signal after matched filtering, and use the estimation results as the prior information of the Bayesian Learning (BL) method to further estimate the parameters of the target and clutter and their Gaussian distributions

[0036] where, let It can be reformulated as follows:

[0037]

[0038] Among them, the o-th column vector of the matrix Φ r,b is:

[0039]

[0040] The o-th element of the vector is

[0041] First, use the CMT-BCS method to perform a rough estimation on the target and clutter. Set the dictionary matrix Φ r,b , where the arrival angle β r,b,o and time delay τ r,b,o of each column of the dictionary are fixed and known, and their values are the angles and time delays where the target and clutter may appear (the possible area is determined by two-dimensional fast Fourier transform (2D-FFT)). Use the CMT-BCS method to estimate Its mean value is Average the calculation results of Q symbols to obtain which is Set a threshold for each element of and form a new vector with the elements exceeding the threshold The arrival angle and time delay of each column in the dictionary corresponding to these elements are which can be expressed by and Φ r,b as:

[0042]

[0043] Substitute each element of into Φ r,b and expand the real and imaginary parts of the signal to obtain:

[0044]

[0045] Among them, takes the real part of the complex signal, takes the imaginary part of the complex signal, is a vector of with all elements being 1. Perform a first-order Taylor expansion on in to obtain:

[0046]

[0047] Let ω r,b contain all the true parameters and clutter parameters of the targets, and assume that its prior distribution follows a Gaussian distribution with a mean of and a variance of . By performing a first-order Taylor expansion on all columns of , can be written in the linear form of ω r,b as follows:

[0048]

[0049] where Assume that follows a Gaussian distribution with a mean of 0 and a variance of σ 1 I (I is the identity matrix). According to the prior information of ω r,b and the linear expression form of ω r,b , the posterior Gaussian distribution of ω r,b can be calculated using Bayes' criterion as follows:

[0051]

[0052] where

[0053]

[0054] Given and , σ 1 and Λ r,b can be obtained by maximum likelihood estimation as follows:

[0055]

[0056] where is a preset large constant. By alternately calculating and δ i , σ 1 until convergence, the estimated value of ω r,b for Q symbols is obtained The final estimated result is obtained by averaging the results of Q symbols:

[0057] According to the relationship between and , we get:

[0058]

[0059] in,

[0060] make z r,b Yes r,b The estimated value of is Gaussian white noise, and the likelihood function of the estimated value can be obtained:

[0061]

[0062] in,

[0063] Step 3: Input the B beam echo estimation results of the rth radar receiver as a single observation value into the designed MD-TPHD filter to obtain the multi-target trajectory tracking results of the rth radar receiver.

[0064] Among them, the results of B beam estimation are used as the primary input of the MD-TPHD filter, and the results of B beam estimation are:

[0065] Z k = {Z k,1 ,...,Z k,m ,...,Z k,M(k)}

[0066] Among them, Z k,m =z r,b,o (z r,b,o Yes r,b The oth element in and Define the state of the oth target scanned by the bth beam at time k as x b,o (k) = [d x,k,b,o ,v x,k,b,o ,d y,k,b,o ,v y,k,b,o ,Ω k,b,o ] T , where d x,k,b,o and d y,k,b,o is the distance of the target in the x-axis and y-axis directions, v x,k,b,o and v y,k,b,o is the speed of the target along the x-axis and y-axis, Ω k,b,o is the target turning speed. b,o (k) has a functional relationship with angle and time delay, set as [β r,b,o ,τ r,b,o ] T =g(x b,o (k)), so each observation Z k,m About x b,o The likelihood function of (k) can be approximately written as follows:

[0067]

[0068] where R m is the o-th 2×2 sub-block on the main diagonal.

[0069] The MD-TPHD filter is derived based on the multi-target Bayesian criterion, which consists of two steps: prediction and update, as follows:

[0070] f(X k |Z (k-1) ) = ∫f(X k |X k-1 )f(X k-1 |Z (k-1) )δX k-1

[0071]

[0072] where f(X k |Z (k-1) ) is the multi-target probability density function in the prediction step, f(X k |Z (k) ) is the multi-target probability density function in the update step, X k is the set of multi-target trajectory states at time k, and Z (k) is the set of all observations from time 1 to k (including targets and clutter).

[0073] The prediction and update steps of the MD-TPHD filter are the first moments of f(X k |Z (k-1) ) and f(X k |Z (k) ), which are D k|k-1 (X k ) and D k (X k ), respectively, as follows:

[0074] D k|k-1 (X k ) = ∫f({X k} ∪ X k |Z (k--) )δX k

[0075] D k (X k ) = ∫f({X k} ∪ X k |Z (k) )δX k

[0076] The Gaussian implementation of the MD-TPHD filter is as follows:

[0077] Prediction step:

[0078]

[0079] Among them, is the first moment of the probability density of the multi-target trajectory surviving from time k - 1 to time k in the prediction, is the first moment of the probability density of the newly emerged trajectory of the multi-target trajectory at time k.

[0080]

[0081] Among them, X k =(t, k, x) represents the set of variables of the trajectory, t is the starting time of the trajectory, k is the current time, x is the trajectory state, x(k) represents the state of the trajectory at the k-th moment, p S is the probability that the trajectory survives from time k - 1 to time k, J k-1 is the number of Gaussian components in the Gaussian mixture distribution at time k - 1, is the weight of the j-th surviving trajectory, t j is the starting time of the j-th surviving trajectory, and are the mean and covariance of the state of the j-th surviving trajectory respectively; is the number of Gaussian components in the first moment of the newly born trajectory at time k, is the weight of the j-th newly born trajectory, the starting time and the current time of the trajectory at time k are both k, and are the mean and covariance of the state of the j-th newly born trajectory.

[0082] Update step:

[0083]

[0084] Among them, is the detection probability of the trajectory, Z k is the set of observation values, P∠Z k means that the set P divides the set Z k into non-overlapping non-empty cells, J k represents the number of updated Gaussian components, is the weight of the j-th Gaussian component, t j is the starting time of the trajectory of the j-th Gaussian component, and P j kThe trajectory state mean and covariance of the j-th Gaussian component. Finally, the Gaussian components with weights greater than 0.5 are extracted as the target trajectories of the final estimate. Among them,

[0085]

[0086] where |W| represents the number of elements in the set W. It is defined that a target generates at most L observations (L ≤ B), and the l-th observation of this target is defined as Z k,θ(l) , which corresponds to the θ(l)-th observed value in Z k . If the l-th observation is not detected, then θ(l) = 0. Then the set of all detected observations of this target is W = ∪ θ(l)＞0 {Z k,θ(l)}. Assume that the probability that the l-th observation is observed is

[0087] Assume that the number of clutters in the environment follows a Poisson distribution with a mean of λ c , and its distribution in space is c(Z k ). After Bayesian learning estimation, the number of clutters may increase. |C f | represents the number of newly added clutters, and P(|C f | = j) represents the probability that the number of clutters increases by j. c[g] = ∫g(Z)c(Z)dZ is a functional of c(Z).

[0088] Step 4: All radar receivers send the multi-target trajectories they are tracking to the trajectory fusion center. The trajectory fusion center fuses the multi-target trajectories of each radar receiver in turn. By comparing the Mahalanobis distance between the trajectories it retains and each trajectory of the radar receiver, the trajectory AA fusion algorithm is used to fuse the trajectories of the radar receiver.

[0089] Among them, after all radar receivers complete the trajectory tracking at time k, they send the first-order moment of the posterior probability of the updated multi-target trajectories to the trajectory fusion center. To distinguish the first-order moments of the posterior probabilities of the multi-target trajectories of different radar receivers and the trajectory fusion center, it is defined that is the first-order moment from the r-th radar receiver, is the first-order moment of the fusion center. and can be simplified into the following form:

[0090]

[0091] When fusing the of the r-th radar receiver and the of the fusion center, the trajectory AA fusion algorithm is used. The fusion algorithm is as follows:

[0092]

[0093] where f c,r and f r are fusion weights, satisfying f c,r , f r > 0, f c,r + f r = 1.

[0094] The expanded form is:

[0095]

[0096] and The values of c the j r -th trajectory of the fusion center and the j

[0097]

[0098] -th trajectory of the r-th radar receiver are obtained by calculating the Mahalanobis distance from the radar trajectory and the fusion center trajectory. For the j -th trajectory of the radar receiver and the j r -th trajectory of the fusion center, the Mahalanobis distance between them is: c trajectory distribution is the Mahalanobis distance between the j -th trajectory of the radar receiver and the j c -th trajectory of the fusion center, r trajectory distribution is the Mahalanobis distance between the j -th trajectory of the fusion center and the j r -th trajectory of the radar receiver. When any one of these two Mahalanobis distances is less than a preset threshold, it is considered that these two trajectories are generated by the same target. When the Mahalanobis distance between the j c -th trajectory of the radar receiver and any trajectory of the fusion center is greater than the preset threshold, it is considered that this trajectory is a newly generated trajectory. When the Mahalanobis distance between the j

[0099] -th trajectory of the fusion center and any trajectory of the radar receiver is greater than the threshold, if this radar receiver is not the last radar receiver, it is considered that this trajectory may be observed by the remaining radar receivers. If the radar receiver is the last one, it is considered that this trajectory has not been observed by the radar receiver and this trajectory disappears.

[0099] Step 5: After fusing the tracks of all radar receivers, the track fusion center feeds back the results to each radar receiver. Each radar receiver also calculates the Mahalanobis distance between each of its own tracks and the tracks transmitted by the track fusion center, and uses the track AA fusion algorithm to fuse the tracks of the track fusion center.

[0100] Among them, when the fusion center finishes fusing the tracks of all radar receivers, it feeds back the fusion results to each radar receiver, and each radar receiver fuses the tracks from the fusion center using the same method as in Step 4.

[0101] Embodiment:

[0102] As Figure 1 shown, the system designed by the present invention for realizing perception based on 5G broadcast signals includes a 5G base station, three radar receivers, and a track fusion center. The 5G base station detects targets by periodically scanning four beams. The radar receivers receive the echoes and clutter reflected by the targets, and each radar receiver establishes a communication connection with the track fusion center. As Figure 2 shown, the working process of the radar receiver and the track fusion center designed by the present invention is as follows: First, R radar receivers estimate the target and clutter parameters of the echoes of each beam (combining the CMT-BCS and BL methods). Then, the estimation results of the B-beam echoes of each radar receiver are integrated and filtered through the designed MD-TPHD filter. Then, the processing results of the R radar receivers are transmitted to the track fusion center for fusion. Finally, the track fusion center outputs the fusion results and feeds them back to each radar receiver.

[0103] The following gives the system signal model. The signal of the b-th beam transmitted by the base station is:

[0104]

[0105] where Q is the number of OFDM symbols, N s is the number of subcarriers, γ b is the beamforming vector, s b,q (n) is the symbol modulated on the n-th subcarrier of the q-th symbol of the b-th beam (e.g., QAM modulation), Δf is the bandwidth of each subcarrier, T s =T + T cp is the length of each symbol, where T = 1 / Δf, T cp is the length of the cyclic prefix, ξ(t - (q - 1)T s ) is a rectangular pulse with a duration of T s , and when (q - 1)T s < t < qT s ξ(t - (q - 1)T s) = 1, otherwise ξ(t - (q - 1)T s ) = 1, f c is the carrier frequency, and Re(·) is to take the real part of the signal. Each radar receiver receives the target echoes and clutter from B beams. After down-conversion and matched filtering, the signal of the nth subcarrier on the qth symbol of the bth beam received by the rth radar receiver is

[0106] Taking the result estimated by the CMT - BCS algorithm as the prior, using the BL algorithm to estimate the target and clutter parameters of the echoes of B beams from including the angle of arrival and the distance, as well as the distribution of each parameter. All the parameters estimated by the B beams form the set Z and use it as the input of the MD - TPHD filter. k

[0107] The MD - TPHD filter is divided into prediction and update steps. The prediction step is as follows:

[0108]

[0109] Among them, is the first - order moment of the probability density of the multi - target trajectory surviving from the (k - 1)th moment to the kth moment in prediction, is the first - order moment of the probability density of the newly - emerged trajectory at the kth moment in the multi - target trajectory.

[0110]

[0111] Among them, p S is the probability that the trajectory survives from the (k - 1)th moment to the kth moment, J k-1 is the number of Gaussian components in the mixture Gaussian distribution at the (k - 1)th moment, is the weight of the jth surviving trajectory, t j is the starting time of the jth surviving trajectory, and are the mean and covariance of the state of the jth surviving trajectory respectively; is the number of Gaussian components in the first - order moment of the newly - born trajectory at the kth moment, is the weight of the jth newly - born trajectory. The starting time and the current time of the trajectory at the kth moment are both k, and are the mean and covariance of the state of the jth newly - born trajectory.

[0112] Using Z k to update the result of the prediction step as follows:

[0113]

[0114] Among them,​ is the detection probability of the trajectory, Z k is the set of observations, P∠Z k indicates that set P divides set Z k into disjoint non - empty cells, J k represents the number of updated Gaussian components, is the weight of the j - th Gaussian component, t j is the starting time of the trajectory of the j - th Gaussian component, and P j k The mean and covariance of the trajectory state of the j - th Gaussian component. Extract the Gaussian components with weights greater than 0.5 as the target trajectories of the final estimate. Round the sum of these weights to estimate the number of target trajectories surviving at the current moment.

[0115] The final estimation results of each radar receiver are transmitted to the trajectory fusion center. The trajectory fusion center calculates the Mahalanobis distance between its own retained trajectories and the trajectories transmitted by the radar receivers, and uses the trajectory AA algorithm for fusion as follows: Define as the first - order moment from the r - th radar receiver, as the first - order moment of the fusion center. and can be simplified into the following form:

[0116]

[0117] When fusing the of the r - th radar receiver and the of the fusion center, use the trajectory AA fusion algorithm as follows

[0118]

[0119] and The values of are obtained by calculating the Mahalanobis distance between the radar trajectory and the fusion center trajectory. For the j - th c trajectory of the fusion center and the j - th r trajectory of the r - th radar receiver, the Mahalanobis distance between them is:

[0120]

[0121] where is the Mahalanobis distance between the j - th r trajectory of the radar receiver and the j - th c trajectory distribution of the fusion center, and is the Mahalanobis distance between the j - th c trajectory of the fusion center and the j - th r trajectory distribution of the radar receiver Mahalanobis distance. When any one of these two Mahalanobis distances is less than a preset threshold, it is considered that these two trajectories are generated by the same target. When the j r th trajectory of the radar receiver and the Mahalanobis distance between any trajectory of the fusion center are greater than the preset threshold, it is considered that this trajectory is a newly generated trajectory. When the j c th trajectory of the fusion center and the Mahalanobis distance between any trajectory of the radar receiver are greater than the threshold, if this radar receiver is not the last radar receiver, it is considered that this trajectory may be observed by the remaining radar receivers. If the radar receiver is the last one, it is considered that this trajectory has not been observed by the radar receiver and this trajectory disappears. Finally, the fusion center outputs the fusion result and feeds it back to the radar receiver, and the radar receiver fuses the trajectory of the fusion center in the same way.

[0122] Figure 2 is the system working flow chart. The input of this process is the After the matched filtering process, first, each radar receiver independently estimates the target state and clutter of the B-beam echoes. Then, each radar receiver inputs the estimation results of the B beams into its own MD-TPHD filter, and each filter transmits the tracking results to the fusion center. Finally, the trajectory fusion center outputs the fusion result and feeds it back to each radar receiver.

[0123] Finally, the proposed method is verified by simulation. Let the total transmission power p T = 1.92W, N s = 256, and the power of each subcarrier is evenly distributed, p n = p T / N s , Δf = 120kHz, Q = 4, f c = 26GHz, M T = 192, M R = 64, T cp = 3.4us, the target radar cross section RCS = 0.1m 2 , n 0 = 0.25×10 -22 W / Hz. The 5G base station is located at (0m, 0m), and the three radar receivers are located at (50m, -250m), (50m, 0m), and (50m, 250m) respectively. The detection range of the base station is [-75°, 75°]×[0, 1000]m, and the field of view of the bth radar receiver sensor is [(b - 1)·35° - 75°, (b - 1)·35° - 30°]×[100, 1000]m, λ c= 4. Assume that the clutter distribution c(·) in space is uniformly distributed, and the calculation method is 1 / (angle range / 180 * distance range). In the non-beam overlap region λ c c(·) = 5.659×10 -3 , in the beam overlap region λ c c(·) = 11.32×10 -3 , P(|C f | = 0) = 0.5, P(|C f | = 1) = 0.3, P(|C f | = 2) = 0.2. To improve the trajectory detection probability, the survival probability p S and the detection probability p D are generally above 0.9. p S = 0.99, p D = 0.98. The newborn model of the target is based on the regions where the target may appear. In the simulation, it is assumed that the target appears from four regions of the same size, that is The target is randomly generated in the above four regions and can be designed according to needs. Among them, the variance The larger it is, the wider the coverage of the four regions. The preset fusion threshold is 100, and the value range is 50 - 300. A too large or too small threshold may lead to missed detections or false alarms.

[0125] Such as Figure 3a and Figure 3b shown, compare the performance of multi-target trajectory tracking using the MD-TPHD filter and tracking using the TPHD filter for the 5G broadcast signal tracking proposed in the present invention. It is mainly compared from two aspects: the trajectory number estimation and the root mean square error of the trajectory. The number of targets is achieved by summing the weights of all Gaussian components, that is This formula represents the number of trajectories at time k, and round(·) represents rounding. The root mean square error of the trajectory is where N mc is the number of Monte Carlo experiments, is the error between the estimated trajectory set and the actual trajectory set. Among them, Figure 3a is the number estimation result, Figure 3b is the root mean square error of the trajectory, and the abscissa is the tracking time. From Figure 3a it can be seen that the MD-TPHD filter has better estimation accuracy in terms of quantity than the TPHD filter, is closer to the true number of targets, and has a faster appearance and response speed to the target. From Figure 3b it can be seen that the root mean square error of the trajectory tracked by the MD-TPHD filter is smaller than that of the TPHD filter.

[0126] Such as Figure 4a andFigure 4b As shown, the present invention compares the trajectory tracking results of a single radar receiver with those of a trajectory fusion center, mainly from two aspects: target number estimation and the root mean square error of the trajectory set. Figure 4a This is the result comparison of number estimation. The abscissa is the tracking time, and the ordinate is the estimated target number. It can be seen that the number estimation after fusion is better than the tracking results of a single radar receiver. Figure 4b This compares the root mean square error of the trajectory set estimated by a single radar receiver with that of the trajectory set at the fusion center. The abscissa is the tracking time, and the ordinate is the root mean square error of the trajectory set. It can be seen that the root mean square error after fusion is smaller than that of a single radar receiver.

[0127] In specific implementation, the present application provides a computer storage medium and a corresponding data processing unit. Among them, the computer storage medium can store a computer program, and when the computer program is executed by the data processing unit, it can run the inventive content of a multi-target trajectory tracking and fusion method based on a multi-observation trajectory probability hypothesis density filter and some or all of the steps in each embodiment. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0128] Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present invention can be implemented by means of a computer program and its corresponding general hardware platform. Based on such an understanding, the technical solutions in the embodiments of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer program, that is, a software product. This computer program software product can be stored in a storage medium and includes several instructions to enable a device (which can be a personal computer, a server, a single-chip microcomputer, an MCU, or a network device, etc.) including a data processing unit to execute the methods described in each embodiment or some parts of the embodiments of the present invention.

[0129] The present invention provides an idea and method of a multi-target trajectory tracking and fusion method based on a multi-observation trajectory probability hypothesis density filter. There are many methods and ways to specifically implement this technical solution. The above is only the preferred implementation mode of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. Each component not clearly defined in this embodiment can be implemented by the prior art.

Claims

1. A multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter, characterized in that: The following steps are involved: Step 1: Use a 5G base station to send a scanning beam to the detection area, use multiple radar receivers to receive the echo signal of the scanning beam, and extract information on the echo signal; Step 2, each radar receiver calculates the information to obtain estimated values ​​of target and clutter parameters and their Gaussian distribution; Step 3, each radar receiver uses the MD-TPHD filter to calculate the estimated values ​​of the parameters of the target and clutter and their Gaussian distribution to obtain its estimation results for the multi-target trajectories; Step 4: The trajectory fusion center fuses the estimated results of the multi-target trajectories obtained by all radar receivers to obtain a fusion result; Step 5: Each radar receiver updates the trajectory state of the radar receiver according to the fusion result, and completes the multi-target trajectory tracking and fusion based on the multi-observation trajectory probability hypothesis density filter.

2. According to claim 1, a multi-target trajectory tracking and fusion method based on a multi-observation trajectory probability hypothesis density filter is characterized in that: The step of extracting information from the echo signal in step 1 includes: Step 1-1, assuming that the 5G base station sends a total of B beams in one scanning cycle, where the baseband signal s of the bth beam is b (t) is: Where Q is the number of symbols in the OFDM signal of the beam, N s is the number of subcarriers, γ b is the beamforming vector, s b,q (n) is the symbol modulated on the nth subcarrier on the qth symbol of the bth beam, Δf is the bandwidth of each subcarrier, T s =T+T cp is the length of each symbol, where T = 1 / Δf is the effective length of each symbol, T cp is the length of the cyclic prefix, ξ(t-(q-1)T s ) is the duration of T s Rectangular pulses; The baseband signal s of the bth beam b (t) After up-conversion, the transmitted signal for: Among them, f c is the carrier frequency, Re(·) is the real part of the signal; Step 1-2: Assume that there are R radar receivers in total, where the rth radar receiver receives the echo of the bth beam, and after down-conversion, the signal y r,b (t) is: Among them, O r,b is the sum of the number of targets within the coverage area of ​​the bth beam and the number of Poisson-distributed clutter, is the antenna array gain, M T and M R are the number of transmitting antennas and receiving antennas, respectively, r,b,o is the attenuation factor, f r,b,o , β r,b,o and τ r,b,o are the Doppler shift, arrival angle and delay of the target, α b is the launch angle, v r,b (t) is a Gaussian white noise with a mean of 0 and a variance of n0. r (β r,b,o ) is the angle of arrival vector, a t (α b ) is the launch angle vector; Step 1-3, performing matched filtering to obtain information on the echo signal, that is, the signal of the nth subcarrier on the qth symbol of the bth beam received by the rth radar receiver for: in, Indicates b,q The conjugate of (n), It is M R ×1 vector, p n =|s b,q (n)| 2 is the power on the nth subcarrier, It is subject to the mean of 0 and the variance of p n Gaussian white noise of n0 / T.

3. The multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter according to claim 2 is characterized in that: The parameters of the target and clutter and their Gaussian distribution obtained in step 2, that is, the information on the echo signal obtained in step 1, is estimated using a complex multi-task Bayesian compressed sensing method to estimate the parameters of the target and clutter, and the estimation result is used as the prior information of the Bayesian learning method, and the parameters of the target and clutter and their Gaussian distribution are further estimated using the Bayesian learning method. The specific method includes: Step 2-1, using the signal of the nth subcarrier on the qth symbol of the bth beam received by the rth radar receiver Calculate N on the qth symbol of the bth beam received by the rth radar receiver s Subcarrier stacking signal as follows: in,[·] T represents the transpose of a vector, N s is the number of subcarriers on the qth symbol of the bth beam received by the rth radar receiver, Re-expressed as follows: in, and They are dictionary matrix, amplitude vector and noise vector respectively. The dictionary matrix Φ r,b The oth column vector of vector The oth element of Step 2-2, use the complex multi-task Bayesian compressed sensing method to estimate the target and clutter parameters in the first step, as follows: Set up the dictionary matrix Φ r,b , the arrival angle β of each column in the dictionary matrix r,b,o and time delay τ r,b,o The values ​​of the angles and delays where the target and clutter may appear are fixed and estimated using the complex multi-task Bayesian compressed sensing method. Its mean is The average result of the calculation of Q symbols is is the mean Set a threshold for each element of , and set the number of elements exceeding the threshold to Make it into a new vector The arrival angle and delay of each column in the dictionary matrix corresponding to these elements are Let the true parameters and clutter parameters of all targets be vectors, expressed as follows: Assume that its prior distribution follows the mean: The variance is Gaussian distribution, after first-order Taylor expansion, the subcarrier stacking signal Written as the parameter to be estimated ω r,b The linear form of is as follows: in, is to get the real part of the complex signal, is the imaginary part of the complex signal, H r,b and d r,b is about the mean of the prior distribution function, and the parameter to be estimated ω r,b No matter, is the noise term, assuming It obeys a Gaussian distribution with a mean of 0 and a variance of σ1I, where I is the unit matrix; Step 2-3, based on the estimation result of the first step ω r,b Prior information and About Omega r,b Linear expression, using the Bayesian criterion to calculate the estimated parameter ω r,b The posterior Gaussian distribution of is as follows: in, In a given and After that, σ1 and Λ r,b It is obtained by maximum likelihood estimation; after obtaining Q symbols about ω r,b Estimated value of The results of Q symbols are averaged to obtain the final estimated result μ r,b : According to the estimation result of the qth symbol and The relationship is: in, make z r,b is the first step estimation result ω r,b The estimated values ​​of , that is, the estimated values ​​of the parameters of the target and clutter and their Gaussian distribution, is Gaussian white noise, and the likelihood function of the estimated value is obtained: in, 4. The multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter according to claim 3 is characterized in that: The step 3 of obtaining the estimation results of the multi-target trajectories includes: The result of estimating the echo of B beams received by the rth radar receiver is used as the primary input of the MD-TPHD filter. The estimated result Z of the echo of B beams is k for: WITH k ={Z k,1 ,...,WITH k,m ,...,WITH k,M(k) } Among them, Z k,m =z r,b,o , z r,b,o are the parameters of the target and clutter and their estimated Gaussian distribution z r,b The oth element in Represents the estimated result Z k The mth element in is the oth element estimated in the bth beam, Represents the total number of elements of B beam estimates; The state of the oth target scanned by the bth beam at time k is defined as: x b,o (k)=[d x,k,b,o ,v x,k,b,o ,d y,k,b,o ,v y,k,b,o ,Ω k,b,o ] T Among them, d x,k,b,o and d y,k,b,o is the distance of the target in the x-axis and y-axis directions, v x,k,b,o and v y,k,b,o is the speed of the target along the x-axis and y-axis, Ω k,b,o is the target turning speed; The state x of the oth target is known b,o (k) has a functional relationship g() with angle and time delay, which is set as: [b r,b,o ,t r,b,o ] T =g(x b,o (k)) Then each observation Z k,m About x b,o The likelihood function of (k) is written as follows: Among them, R m yes The oth 2×2 sub-block on the main diagonal; The final estimation result of the multi-target trajectory after passing through the MD-TPHD filter is: Among them, X k =(t,k,x) represents the variable set of the trajectory, t is the starting time of the trajectory, k is the current time, x is the state of the trajectory, and x(k) represents the state of the trajectory at the kth time. is the trajectory detection probability, is the Gaussian distribution of the trajectory, and its parameters represent The starting time of the trajectory, the current time of the trajectory, the mean of the trajectory, the variance of the trajectory), is the weight of the Gaussian distribution as the true target trajectory.

5. The multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter according to claim 4 is characterized in that: The MD-TPHD filter described in step 3, including: The MD-TPHD filter is derived according to the multi-objective Bayesian criterion and consists of two steps: prediction and update. The Gaussian implementation of the prediction step is as follows: in, It is the first-order moment of the probability density of multi-target trajectories predicted from time k-1 to time k. is the first-order moment of the probability density of the newly emerged multi-target trajectory at time k, and the calculation methods are as follows: Among them, p S is the probability that the trajectory survives from time k-1 to time k, J k-1 is the number of Gaussian components in the mixed Gaussian distribution at time k-1, is the weight of the jth survival trajectory, t j is the starting time of the jth survival trajectory, and are the mean and covariance of the jth survival trajectory state respectively; is the number of Gaussian components in the first-order moment of the new trajectory at time k, is the weight of the jth new trajectory, the starting time and current time of the k-th trajectory are both k, and is the mean and covariance of the jth newborn trajectory state; The Gaussian implementation of the update step is as follows: in, is the detection probability of the trajectory, Z k is the set of observations, P∠Z k Denotes that set P will be set Z k Split into disjoint non-empty cells, J k represents the number of updated Gaussian components, is the weight of the jth Gaussian component, t j is the starting time of the trajectory of the jth Gaussian component, and The trajectory state mean and covariance of the jth Gaussian component; Extract the Gaussian component whose weight is greater than the preset weight value as the estimation result of the multi-target trajectory 6. The multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter according to claim 5 is characterized in that: Extract the Gaussian component whose weight is greater than the preset weight value as the estimation result of the multi-target trajectory The specific process is as follows: in, is the weight, and is the trajectory mean and variance corresponding to the weight, is a Gaussian distribution related to W, Z W is a vector formed by stacking the elements in the set W, and its mean is The variance is S W Gaussian distribution, τ W is with q j (W) related function, κ W is a function related to clutter, w P It is about τ W and κ W function.

7. The multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter according to claim 6 is characterized in that: The estimation results of the multi-target trajectories obtained by all radar receivers described in step 4 are fused, including: Step 4-1, after all radar receivers obtain the estimated results of the multi-target trajectories at time k, the estimated results are The estimated results of the multi-target trajectories are sent to the trajectory fusion center in sequence for fusion and update; Step 4-2, Definition is the first-order moment from the rth radar receiver, i.e., the estimated result of the multi-target trajectory, is the first-order moment of the fusion center, and the trajectory AA fusion method is used for fusion.

8. The multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter according to claim 7 is characterized in that: The fusion is performed using the trajectory AA fusion method described in step 4-2, including: in, is the fusion result, f c,r and f r is the fusion weight, satisfying f c,r ,f r >0,f c,r +f r =1.

9. The multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter according to claim 8 is characterized in that: The trajectory AA fusion method described in step 4-2 is expanded as follows: in, and The value of is obtained by calculating the Mahalanobis distance between the estimated result of the multi-target trajectory from the radar receiver and the estimated result of the fused multi-target trajectory temporarily stored in the trajectory fusion center. c The jth trajectory of the rth radar receiver r The Mahalanobis distance between them is: in, is the radar receiver j r Trajectory and trajectory fusion center j c Trajectory distribution The Mahalanobis distance, It is the trajectory fusion center j c Track and radar receiver r Track distribution When any of the two Mahalanobis distances is less than the preset threshold, it is considered that the two trajectories are generated by the same target. When the radar receiver's j r If the Mahalanobis distance between a trajectory and any trajectory of the trajectory fusion center is greater than the preset threshold, the trajectory is considered to be a new trajectory. When the jth c When the Mahalanobis distance between the trajectory and any trajectory of the radar receiver is greater than the threshold, and the radar receiver is not the last radar receiver, it is considered that the trajectory may be observed by the remaining radar receivers. If the radar receiver is the last one, it is considered that the track is not observed by the radar receiver and the track disappears.

10. The multi-target trajectory tracking and fusion method based on multi-observation trajectory probability hypothesis density filter according to claim 9 is characterized in that: The updating of the track state of the radar receiver described in step 5 includes: Each radar receiver uses the trajectory AA fusion method described in step 4-2 according to the fusion result sent by the trajectory fusion center to fuse the fusion result from the trajectory fusion center and its own trajectory, and uses the fusion result to update its own trajectory.

Images