Joint estimation method for non-line-of-sight target position and wall concave-convex parameters in urban scene

Abstract

The invention provides a joint estimation method for a non-line-of-sight target position and wall concave-convex parameters in an urban scene, and aims to solve the problem that an existing non-line-of-sight target positioning method is large in positioning error in uneven building environments such as balconies and windows. In the method, a typical L-shaped non-line-of-sight scene is taken as an example, firstly, wall concave-convex characteristic parameters at an electromagnetic wave propagation reflection point are converted into wall irregular parameters, and a multi-path echo model containing the wall irregular parameters is constructed based on the sparsity of the scene. Secondly, constructing a target function by minimizing the received echo signal and reconstructing an error between multipath signals by using compressed sensing; and then, an ADMM algorithm and genetic algorithm optimization are adopted to carry out layered optimization on the target function. And finally, through verification of simulation and actual measurement experiments, the method provided by the invention has relatively high target positioning precision in a typical L-shaped non-line-of-sight scene containing an irregular wall.

Description

Technical Field

[0001] The present invention belongs to the technical field of shielded target detection, and in particular relates to a method for jointly estimating a non-line-of-sight target position and a wall concave-convex parameter in an urban scene. Background Art

[0002] With the acceleration of urbanization, pre-sensing target information in building blind spots has important research significance in the fields of smart cities, intelligent driving, and urban street fighting. Compared with traditional line-of-sight (LOS) target positioning, the propagation paths of electromagnetic wave diffraction, reflection, etc. in non-line-of-sight (NLOS) target positioning are highly dependent on the characteristics of the building environment, making the echo signal more complex.

[0003] At present, non-line-of-sight target positioning methods mainly include multi-path separation positioning methods based on geometric relationships, imaging methods based on image domain, and multi-sensor fusion using neural networks. Furthermore, researchers began to consider realizing non-line-of-sight target positioning when the building layout is unknown, such as using laser radar and millimeter-wave radar multi-source fusion technology to obtain wall information and target position, and using shape reconstruction group sparse constraint algorithm to reconstruct wall angle and target position. However, the existing non-line-of-sight target positioning methods all assume that the building surface is an ideal flat structure, but the actual building environment containing concave and convex structures such as balconies and windows will affect the propagation path of electromagnetic waves, thereby affecting the positioning accuracy. Summary of the invention

[0004] In order to solve the problem of the influence of the concave-convex structure of the building surface on the electromagnetic wave propagation path and positioning accuracy, the present invention proposes a method for jointly estimating the non-line-of-sight target position and the concave-convex parameters of the wall in urban scenes, which can not only obtain the concave-convex features of the wall at the reflection point, but also achieve high-precision target positioning. The method includes the following steps:

[0005] Step 1: Establish a multipath signal model in a non-line-of-sight scenario and parameterize the concave-convex features of the wall at the reflection point;

[0006] Step 2: Reconstruct the echo signal using compressed sensing;

[0007] Step 3, establishing an objective function for jointly estimating the concave-convex parameters of the target position and the wall at the reflection point;

[0008] Step 4: Use ADMM algorithm to optimize the target position σ;

[0009] Step 5: Use genetic algorithm to optimize the wall irregularity parameter ΔL at the reflection point.

[0010] Furthermore, after the concave-convex structure of the wall at the reflection point is parameterized in step 1, the mirror target position corresponding to each reflection path can be expressed as

[0011]

[0012] Among them, Q 1 is the target mirror coordinate of a reflection, Q 2 is the target mirror coordinate of the secondary reflection, Q 3 is the mirror coordinate of three reflections; L 2 is the distance between wall 2 and the coordinate origin, L 3 is the distance between wall 3 and the coordinate origin; ΔL 1 ,ΔL 2 ,ΔL 3 ,ΔL 4 Indicates the concave-convex parameters of the wall at the reflection point; specifically, ΔL 1 Indicates the horizontal coordinate of the electromagnetic wave reflected to a point on the wall 3 and the position L of the wall 3 3 The difference between .

[0013] Furthermore, the one-way propagation delay of the reflection path in step 1 can be expressed as

[0014]

[0015] Where c is the speed of light, Q p represents the target mirror coordinates corresponding to the pth path, T represents the transmitting antenna position, and R represents the receiving antenna position; Figure 1 As shown, C represents the corner of the building 1.

[0016] Furthermore, the echo signal in step 2 can be simplified to

[0017]

[0018] in, A dictionary representing the propagation mapping for each subpath; Represents the grid coefficient of each grid point in each path subgraph; get σ (p,q) After that, the location of the target can be obtained by fusing each imaging sub-image.

[0019] Furthermore, in step 3, by minimizing the received echo signal y t (f k ) and the echo signal y reconstructed based on scene sparsity e (f k ) to construct the objective function:

[0020]

[0021] The positioning accuracy of the target is highly dependent on the accuracy of the wall error estimation. There is a mutually coupled relationship between these two variables. Therefore, the internal and external layered optimization method is used to solve it. First, ΔL is optimized in the outer layer. For each candidate ΔL, the corresponding In addition, considering the sparsity of the scene, L1 regularization is introduced to the optimization problem to perform sparsity constraints.

[0022]

[0023] Among them, λ is the regularization parameter used to control The sparsity of is the sparsity constraint.

[0024] Furthermore, in step 4, by introducing auxiliary variables z and dual variables u, the original optimization problem is decomposed into sub-problems that are easier to solve:

[0025] (1) By introducing the auxiliary variable z, the original problem can be decomposed into the following two sub-problems:

[0026]

[0027] σ=z

[0028] Among them, one sub-problem is responsible for minimizing the L2 norm, and the other sub-problem is responsible for minimizing the L1 norm;

[0029] (2) Introducing the Lagrangian multiplier u, transforming the constrained optimization problem into an unconstrained problem, and constructing the corresponding augmented Lagrangian function

[0030]

[0031] Among them, ρ is the penalty parameter used to adjust the degree of violation of the original constraint; ADMM optimizes σ and z alternately and updates the dual variable u in each iteration.

[0032] Furthermore, each iteration task in step 4 performs the following three sub-questions:

[0033] (1) Update σ: In this step, the L2 norm part is optimized

[0034]

[0035] For this quadratic minimization problem, we construct a linear equation system to solve

[0036]

[0037] (2) Update z: Optimize the L1 regularization part and control sparsity through soft threshold operation

[0038]

[0039] (3) Update u: Update the dual variable u to gradually approach the original constraint and ensure the consistency of σ and z

[0040] u k+1 =u k +(σ k+1 -z k+1 ).

[0041] Furthermore, in step 5, the optimization of the wall error parameter ΔL is solved by using a genetic algorithm, including:

[0042] (1) Initialize the population to P = {p 1 ,p 2 ,...,p n}, randomly generate a set of initial solutions, and then evaluate the quality of each individual by calculating the fitness function. The invented fitness function is the objective function;

[0043] (2) In each generation, the individual with the highest fitness is selected as the optimal solution for the current iteration; based on the fitness value, the individuals are sorted and the first n / 2 individuals with higher fitness are selected as parents, that is,

[0044] P parents ={p i ∈P|i≤n / 2};

[0045] (3) The parental gene ΔL is transformed by a random factor α f , ΔL m Linear combination, generates new offspring through the following crossover operation

[0046] c 1 =αΔL f +(1-α)ΔL m

[0047] c 2 =αΔL m +(1-α)ΔL f

[0048] Among them, α∈[0,1] is a random number coefficient, indicating that the parental genes are combined in random proportions;

[0049] (4) Apply a random perturbation to each gene to achieve the p i Gaussian mutation increases population diversity and prevents the population from falling into local optimality:

[0050] p i '=p i +δ

[0051] Among them, δ is a disturbance; specifically, δ = γ(ΔL max -ΔL min ), γ is a random number in the range of [-0.5, 0.5], which is used to control the amplitude of the disturbance; ΔL max and ΔL min are the upper and lower limits of the individual’s ΔL, respectively, to ensure that the mutated individual is still within the search space.

[0052] Furthermore, when the genetic algorithm described in step 5 is iterating, the iteration rule is: when the fitness change is less than the set threshold, it is considered that there is no significant improvement in this generation, and the iteration is stopped; otherwise, it needs to continue to iterate to the maximum number of iterations.

[0053] Beneficial effects:

[0054] 1. The present invention provides a method for jointly estimating the non-line-of-sight target position and the concave-convex parameters of the wall in an urban scene, which fully considers the influence of the concave-convex structure of the wall surface such as doors and windows in the actual building environment on the electromagnetic wave propagation process, and parameterizes the concave-convex characteristics of the wall at the reflection point of the electromagnetic wave propagation into the wall irregularity parameters, which effectively improves the universality and positioning accuracy of the positioning algorithm;

[0055] 2. The present invention provides a method for jointly estimating the non-line-of-sight target position and the concave-convex parameters of the wall in an urban scene. In this method, a multipath echo model including irregular parameters of the wall is constructed based on the sparsity of the scene, and the error between the echo signal received by the target and the reconstructed multipath echo signal is hierarchically optimized by using the ADMM algorithm and the genetic algorithm to directly obtain the concave-convex parameters of the wall at the non-line-of-sight target position and the reflection point;

[0056] 3. Compared with the traditional ideal plane positioning method, the method proposed in the present invention can solve the positioning offset problem in the actual environment; simulation and actual measurement experiments show that the method proposed in the present invention has higher target positioning accuracy in a typical L-shaped non-line-of-sight scene containing irregular walls. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a schematic diagram of multipath propagation in a typical L-shaped scenario with concave and convex structures on the wall surface;

[0058] Figure 2 are the simulated and measured multipath echo signals, where (a) is the FDTD simulated echo signal and (b) is the measured experimental echo data;

[0059] Figure 3 It is a simulation scene;

[0060] Figure 4 It is the non-line-of-sight target imaging result of the simulation experiment;

[0061] Figure 5 are the subpath imaging results of the simulation experiment, where (a) is the imaging result of Path0-0, (b) is the imaging result of Path0-1, (c) is the imaging result of Path1-1, (d) is the imaging result of Path0-2, and (e) is the imaging result of Path1-2;

[0062] Figure 6 is a diagram of the measured experimental scene, where (a) is the measured scene 1, (b) is the measured scene 2, and (c) is the experimental equipment;

[0063] Figure 7 This is the imaging result of the traditional non-line-of-sight positioning method (assuming the wall is a smooth plane) in the measured scene 2;

[0064] Figure 8 is the experimental result of measured scene 1, where (a) is the multipath echo signal, (b) is the comparison between the multipath echo signal and the ideal echo signal, and (c) is the target imaging result;

[0065] Fig. 9 These are the experimental results of the measured scene 2, where (a) is the multipath echo signal, (b) is the comparison between the multipath echo signal and the ideal echo signal, and (c) is the target imaging result. DETAILED DESCRIPTION

[0066] The present invention proposes a method for jointly estimating the position of a non-line-of-sight target and the concave-convex parameters of a wall in an urban scene, which is suitable for positioning obscured targets in complex urban environments, and is particularly suitable for high-precision target positioning when there are concave-convex structures on the wall. In order to better describe the present application scheme, the following will combine the drawings in the embodiments of the present application to clearly and completely explain the technical scheme in the embodiments of the present application. The positioning method comprises the following steps:

[0067] Step 1: Establish a multipath signal model in a non-line-of-sight scenario and parameterize the concave-convex features of the wall at the reflection point

[0068] Figure 1 The study scene and the multipath propagation process of the signal between the radar and the target are shown. The brown part represents the concave and convex structure of the building surface in the actual scene, Path-i represents the one-way propagation path of the electromagnetic wave from the transmitting antenna to the target (i = 0, 1, 2, 3...), and i represents the number of reflections of the electromagnetic wave between the walls. In particular, Path-0 represents the path of the electromagnetic wave diffracted through the corner C of the building 1 to reach the target.

[0069] (1) Assume that the radar transmission signal is a single-transmit and single-receive stepped frequency continuous wave (SFCW) signal with a frequency range of (f 0 ,f K-1), the frequency point number is K. Then the echo expression of the received signal is

[0070] y(f k )=y w (f k )+y t (f k )+ξ(f k )k=1,2...K

[0071] Among them, y w (f k ) is the scattered clutter from other stationary objects such as walls, ξ(f k ) is the environmental noise. t (f k ) is the target reflected echo, which is composed of multipath propagation signals such as diffraction and reflection, and its expression is

[0072]

[0073] Where p represents the type of propagation path of the transmitted signal, and q represents the type of propagation path corresponding to the received signal. p Indicates the delay corresponding to the signal propagation path. p,q Represents the propagation attenuation coefficient corresponding to the p transmission path and the q reception path.

[0074] Figure 2 (a) is the FDTD simulation echo signal in this scenario. Figure 2 (b) is the echo data of the actual test. It can be seen that the high-order reflected signals are almost undetectable in the echo due to severe energy attenuation. Therefore, the invention mainly considers the propagation path echo signal with no more than three reflections. The one-way propagation of the transmitting antenna and the receiving antenna both meet the above path, so the two-way propagation path between the radar and the target is a combination of the above four paths, as shown in the following formula

[0075]

[0076] Among them, l pq It represents a two-way propagation path where electromagnetic waves are transmitted along the p path and received along the q path.

[0077] (2) The propagation delay of the diffraction path can be calculated by the uniform geometric theory of diffraction (UTD). The propagation delay of the reflection path requires the determination of the mirror image position of the target, and the reflection Householder transformation is used to determine the mirror image position of the target under different propagation paths. Furthermore, the concave-convex structure of the wall mainly affects the position of the reflection point, and then affects the propagation delay of the path. Therefore, we only need to pay attention to the concave-convex structure of the wall at the reflection point, and define it as the wall error parameter, such as Figure 1 The parameterization shown is ΔL 1 ,ΔL2 ,ΔL 3 ,ΔL 4 Specifically, ΔL 1 Indicates the horizontal coordinate of the electromagnetic wave reflected to a point on the wall 3 and the position L of the wall 3 3 Therefore, the mirror target position corresponding to each reflection path can be expressed as

[0078]

[0079] Among them, Q 1 is the target mirror coordinate of a reflection, Q 2 is the target mirror coordinate of the secondary reflection, Q 3 is the mirror coordinate of the three reflections. 2 is the distance between wall 2 and the coordinate origin, L 3 is the distance between wall 3 and the coordinate origin. The one-way propagation delay of the reflection path can be expressed as

[0080]

[0081]

[0082] Where c is the speed of light, Q p represents the target mirror coordinates corresponding to the pth path, T represents the position of the transmitting antenna, and R represents the position of the receiving antenna.

[0083] Step 2: Reconstruct the echo signal using compressed sensing

[0084] Based on the sparse characteristics of the scene, the non-line-of-sight imaging scene is discretized into S uniformly distributed grids. In this scene, each grid is assigned a grid coefficient to characterize whether there is a target in the area. Specifically, when the target is located at a specific grid position, the corresponding grid coefficient will be assigned a non-zero value. Based on this framework, the multipath echo model can be written as

[0085] y e (f k )=H (0,0) σ (0,0) + Η (0,1) σ (0,1) +…+Η (p,q) σ (p,q) +∈p,q∈{0,1,2,3}

[0086] Among them, σ (p,q) The corresponding imaging sub-image is formed by the electromagnetic wave through the pth transmission path and the qth receiving path. Specifically, σ (p,q) =(σ 1 (p,q) ,σ 2 (p,q),…,σ S (p,q) ). S (p,q) The grid coefficient corresponding to each grid point in each path sub-graph. According to the principle that the same target position shares the same position in the imaging results of different paths, the imaging sub-graphs corresponding to different paths are fused, that is, σ = σ (0,0) +σ (0,1) +...+σ (p,q) Indicates the location of the target. is a dictionary of propagation mappings for each imaging subimage, is the frequency domain signal of the electromagnetic wave reaching the sth grid point through the pth transmission path and returning to the receiving antenna through the qth receiving path. Its expansion is

[0087]

[0088] in, For each grid point s(x s ,y s ), taking a two-way single reflection as an example, the delay of each grid point can be expressed as

[0089]

[0090] Among them, d 1 (S,T) is the distance that the electromagnetic wave travels from the transmitting antenna T to the grid point s along the first reflection path. Similarly, d 1 (S, R) is the path of the electromagnetic wave returning from the grid point s to the receiving antenna R along the first reflection path, and its expansion is

[0091]

[0092] Frequency domain multipath echo signal based on scene sparsity e (f k ) contains the unknown parameter σ (p,q) and ΔL, the multipath echo signal can be further written as

[0093]

[0094] It can be further simplified to

[0095]

[0096] in, σ (p,q) =(σ 1 (0,0) ,σ 2 (0,1) ,...,σ S (p,q)). We get σ (p,q) After that, the location of the target can be obtained by fusing each imaging sub-image.

[0097] Step 3: Establish the objective function for joint estimation of the target position and the concave-convex parameters of the wall at the reflection point

[0098] The target positioning problem under the concave-convex structure of the wall is transformed into a joint estimation problem, that is, to estimate the grid coefficient of the scene and the wall irregularity parameters at the reflection point at the same time. Therefore, we minimize the received echo signal y t (f k ) and the echo signal y reconstructed based on scene sparsity e (f k ) to construct the following objective function

[0099]

[0100] The positioning accuracy of the target is highly dependent on the accuracy of the wall error estimation. There is a mutual coupling relationship between these two variables. Therefore, the internal and external layered optimization method is used to solve it. First, ΔL is optimized in the outer layer. For each candidate ΔL, the corresponding In addition, considering the sparsity of the scene, L1 regularization is introduced to the optimization problem to perform sparsity constraints.

[0101]

[0102] Among them, λ is the regularization parameter used to control The sparsity of is the sparsity constraint.

[0103] Step 4: Use ADMM algorithm to optimize the target position σ

[0104] The optimization of the intrinsic parameter sigma is a convex optimization problem, for which we use the alternating direction method of multipliers (ADMM). By introducing auxiliary variables z and dual variables u, the original optimization problem is decomposed into sub-problems that are easier to solve.

[0105] (1) By introducing the auxiliary variable z, the original problem can be decomposed into the following two sub-problems:

[0106]

[0107] σ=z

[0108] Among them, one subproblem is responsible for minimizing the L2 norm, and the other subproblem is responsible for minimizing the L1 norm;

[0109] (2) Introducing the Lagrangian multiplier u, transforming the constrained optimization problem into an unconstrained problem, and constructing the corresponding augmented Lagrangian function

[0110]

[0111] Where ρ is a penalty parameter used to adjust the degree of violation of the original constraint. ADMM optimizes σ and z alternately and updates the dual variable u in each iteration.

[0112] Specifically, each iteration task carries out the following three sub-problems:

[0113] (1) Update σ: In this step, the L2 norm part is optimized

[0114]

[0115] For this quadratic minimization problem, we construct a linear equation system to solve

[0116]

[0117] (2) Update z: Optimize the L1 regularization part and control sparsity through soft threshold operation

[0118]

[0119] (3) Update u: Update the dual variable u to gradually approach the original constraint and ensure the consistency of σ and z

[0120] u k+1 =u k +(σ k+1 -z k+1 ).

[0121] Step 5: Use genetic algorithm to optimize the wall irregularity parameter ΔL at the reflection point

[0122] Since different wall parameter errors will affect the image sparsity after path subgraph fusion, we use image sparsity as the optimization index. The objective function is as follows:

[0123]

[0124] Among them, 1 θ (x) is the sparsity control function, which can be expressed as

[0125]

[0126] The genetic algorithm is used to solve the optimization of the wall error parameter ΔL, including:

[0127] (1) Initialize the population to P = {p1 ,p 2 ,...,p n}, randomly generate a set of initial solutions, and then evaluate the quality of each individual by calculating the fitness function. The invented fitness function is the objective function;

[0128] (2) In each generation, the individual with the highest fitness is selected as the optimal solution for the current iteration. Based on the fitness value, the individuals are sorted and the first n / 2 individuals with higher fitness are selected as parents, that is,

[0129] P parents ={p i ∈P|i≤n / 2};

[0130] (3) The parental gene ΔL is transformed by a random factor α f , ΔL m Linear combination, generates new offspring through the following crossover operation

[0131] c 1 =αΔL f +(1-α)ΔL m

[0132] c 2 =αΔL m +(1-α)ΔL f

[0133] Among them, α∈[0,1] is a random number coefficient, indicating that the parental genes are combined in random proportions;

[0134] (4) Apply a random perturbation to each gene to achieve the p i Gaussian mutation increases population diversity and prevents the population from falling into local optimality:

[0135] p i '=p i +δ

[0136] Among them, δ is a disturbance. Specifically, δ = γ(ΔL max -ΔL min ), γ is a random number in the range of [-0.5, 0.5], which is used to control the amplitude of the disturbance. max and ΔL min are the upper and lower limits of the individual’s ΔL, respectively, to ensure that the mutated individual is still within the search space.

[0137] When performing iterations, the iteration rule is: when the fitness change is less than the set threshold, it is considered that there is no significant improvement in this generation, and the iteration is stopped; otherwise, it needs to continue to iterate to the maximum number of iterations.

[0138] The simulation and actual measurement verification are carried out below.

[0139] Simulation scenario such as Figure 3 As shown in the figure, an L-shaped non-line-of-sight scene model with concave and convex structures on the wall surface is constructed, and the electromagnetic echo in this scene is simulated using gprMax software. The size of the experimental scene is 6m×6m, with the lower left corner as the coordinate origin. The scene is divided into grids of 0.1m in size. The coordinates of the corner C of the left building 1 are (1.5m, 1.5m), the distance between the wall 3 and the coordinate origin is x=5m, the thickness of the wall protrusion is 0.2m, and the conductivity is 0.05S / m. The target is a uniform cylinder with a radius of 0.3m, simulating a human target, with a dielectric constant of 50 and a conductivity of 1.05S / m. It is located at (3m, 3.5m) and is in the non-line-of-sight blind area of ​​the radar. The transmitting antenna is located at (0.4m, 1m), the receiving antenna is located at (0.42m, 0.96m), the array element spacing of the transmitting and receiving antennas is 0.05m, and the transmitting signal is a stepped frequency continuous wave SFCW. The signal parameters are shown in Table 1.

[0140] Table 1. Radar signal parameters of simulation experiment

[0141]

[0142] The gprMax electromagnetic simulation data is read in MATLAB, and the multipath echo signal of the target is obtained after performing empty scene cancellation processing. Figure 5 Shown is the scene reconstructed image, and the highlighted area indicates the target positioning result. Figure 6 The imaging sub-image corresponding to each propagation path is shown, where the red circle indicates the true position of the target. It can be seen that the imaging results of all paths pass through the target location, laying the foundation for multi-path fusion and ensuring the accuracy of target location reconstruction. The simulation results of the target position and the wall parameters at the reflection point are shown in Table 2, and the error between the estimated value and the true value is also shown, where the error of the target position is the Euclidean distance between the two, and the error of the wall irregularity parameter is the absolute value of the difference between the two.

[0143] Table 2 Simulation results

[0144]

[0145] At the same time, in order to verify the feasibility of the proposed joint estimation algorithm in a practical environment, Figure 6 (a) and Figure 6 The experimental equipment is as follows: Figure 6 As shown in (c), a vector network analyzer is used as a signal transceiver equipped with a horn antenna. The experimental scene is a non-line-of-sight environment with two L-shaped streets. Figure 6 (a) shows a flat wall. Figure 6 (b) is a complex wall with recessed doors and windows. The specific experimental scene parameters are shown in Table 3, and the radar signal parameters are shown in Table 4.

[0146] Table 3 Parameters of the measured experimental scene

[0147]

[0148]

[0149] Table 4 Measured experimental radar signal parameters

[0150]

[0151] Based on the multipath echo data collected by field measurement, Figure 7 The figure shows the positioning results of the non-line-of-sight positioning algorithm using the traditional flat surface of the wall in experimental scene 2 with concave and convex structures on the wall. The red circle represents the real position of the target (2.65m, 2.35m), and the highlighted area is the target positioning result (3.3m, 2.7m). There is a significant deviation between the two, with an error value of about 0.74 meters. Obviously, the traditional positioning algorithm cannot solve the problem of non-line-of-sight target positioning under the existence of wall structures.

[0152] Subsequently, the positioning algorithm proposed in the present invention is used for improvement. First, we convert the collected 300 frames of frequency domain echo data into the time domain, and extract the multipath echo information of the target through the MTI technology. Figure 8 (a) and Fig. 9 (a) shows the distance-time diagram of the target echo, where multiple significant multipath echo signals can be clearly observed. Figure 8 (b) and Fig. 9 As shown in (b), we selected one frame of data to compare with the ideal multipath propagation signal. The blue curve represents the theoretical echo signal, and the orange curve represents the experimental echo signal. The propagation distance corresponding to the peak value of each path is basically consistent with the theoretical value, which conforms to the law of multipath propagation and verifies the accuracy of the measured data. Figure 8 (c) and Fig. 9 (c) is the target reconstructed image of the measured experiment, where the highlighted area indicates the target positioning result. Table 5 shows the estimation results and errors of the target position and wall irregularity parameters.

[0153] Table 5 Experimental results

[0154]

[0155] The present invention proposes a method for jointly estimating the position of a non-line-of-sight target and the concave-convex parameters of a wall in an urban scene. In this method, based on the sparsity framework of the scene, we propose an algorithm for jointly estimating the concave-convex structural parameters of the wall propagated by electromagnetic waves and the target position. Then, the ADMM algorithm and the genetic algorithm are used for hierarchical optimization to obtain high-precision target positions and concave-convex parameters of the wall. Finally, through simulation and actual experimental verification, the method proposed in the present invention has a high target positioning accuracy in a typical L-shaped non-line-of-sight scene containing non-concave-convex wall features.

[0156] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may certainly make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for jointly estimating non-line-of-sight target position and wall concave-convex parameters in urban scenes, characterized in that: The following steps are involved: Step 1: Establish a multipath signal model in a non-line-of-sight scenario and parameterize the concave-convex features of the wall at the reflection point; Step 2: Reconstruct the echo signal using compressed sensing; Step 3, establishing an objective function for jointly estimating the concave-convex parameters of the target position and the wall at the reflection point; Step 4: Use ADMM algorithm to optimize the target position σ; Step 5: Use genetic algorithm to optimize the wall irregularity parameter ΔL at the reflection point.

2. The method according to claim 1, characterized in that After parameterizing the concave-convex structure of the wall at the reflection point in step 1, the mirror target position corresponding to each reflection path can be expressed as Among them, Q1 is the target mirror coordinate of the first reflection, Q2 is the target mirror coordinate of the second reflection, and Q3 is the mirror coordinate of the third reflection; L2 is the distance between wall 2 and the coordinate origin, and L3 is the distance between wall 3 and the coordinate origin; ΔL1, ΔL2, ΔL3, and ΔL4 represent the concave and convex parameters of the wall at the reflection point; specifically, ΔL1 represents the difference between the horizontal coordinate of the electromagnetic wave reflected to a point on wall 3 and the position L3 of wall 3.

3. The method according to claim 1, characterized in that The one-way propagation delay of the reflection path in step 1 can be expressed as Where c is the speed of light, Q p represents the target mirror coordinates corresponding to the pth path, T represents the position of the transmitting antenna, and R represents the position of the receiving antenna; as shown in Figure 1, C represents the corner of building 1.

4. The method according to claim 1, characterized in that The echo signal in step 2 can be simplified to in, A dictionary representing the propagation mapping for each subpath; Represents the grid coefficient of each grid point in each path subgraph; get σ (p,q) After that, the location of the target can be obtained by fusing each imaging sub-image.

5. The method according to claim 1, characterized in that In step 3, the received echo signal y is minimized t (f k ) and the echo signal y reconstructed based on scene sparsity e (f k ) to construct the objective function: The positioning accuracy of the target is highly dependent on the accuracy of the wall error estimation. There is a mutually coupled relationship between these two variables. Therefore, the internal and external layered optimization method is used to solve it. First, ΔL is optimized in the outer layer. For each candidate ΔL, the corresponding In addition, considering the sparsity of the scene, L1 regularization is introduced to the optimization problem to perform sparsity constraints. Among them, λ is the regularization parameter used to control The sparsity of is the sparsity constraint.

6. The method according to claim 1, characterized in that In step 4, by introducing auxiliary variables z and dual variables u, the original optimization problem is decomposed into sub-problems that are easier to solve: (1) By introducing the auxiliary variable z, the original problem can be decomposed into the following two sub-problems: Among them, one subproblem is responsible for minimizing the L2 norm, and the other subproblem is responsible for minimizing the L1 norm; (2) Introducing the Lagrangian multiplier u, transforming the constrained optimization problem into an unconstrained problem, and constructing the corresponding augmented Lagrangian function Among them, ρ is the penalty parameter used to adjust the degree of violation of the original constraint; ADMM optimizes σ and z alternately and updates the dual variable u in each iteration.

7. The method according to claim 1, characterized in that Each iteration task in step 4 performs the following three sub-questions: (1) Update σ: In this step, the L2 norm part is optimized For this quadratic minimization problem, we construct a linear system of equations to solve (2) Update z: Optimize the L1 regularization part and control sparsity through soft threshold operation (3) Update u: Update the dual variable u to gradually approach the original constraint and ensure the consistency of σ and z at k+1 =u k +(σ k+1 -With k+1 )。 8. The method according to claim 1, characterized in that In step 5, the optimization of the wall error parameter ΔL is solved by using a genetic algorithm, including: (1) Initialize the population to P = {p1, p2, ..., p n }, randomly generate a set of initial solutions, and then evaluate the quality of each individual by calculating the fitness function. The invented fitness function is the objective function; (2) In each generation, the individual with the highest fitness is selected as the optimal solution for the current iteration; based on the fitness value, the individuals are sorted and the first n / 2 individuals with higher fitness are selected as parents, that is, P parents ={p i ∈P∣i≤n / 2}; (3) The parental gene ΔL is changed by a random factor α f , ΔL m Linear combination, generates new offspring through the following crossover operation c1=αΔL f +(1-a)ΔL m c2=αΔL m +(1-a)ΔL f Among them, α∈[0,1] is a random number coefficient, indicating that the parental genes are combined in random proportions; (4) Apply a random perturbation to each gene to achieve the p i Gaussian mutation increases population diversity and prevents the population from falling into local optimality: p' i =p i +d Among them, δ is a disturbance; specifically, δ = γ(ΔL max -ΔL min ), γ is a random number in the range of [-0.5, 0.5], which is used to control the amplitude of the disturbance; ΔL max and ΔL min are the upper and lower limits of the individual’s ΔL, respectively, to ensure that the mutated individual is still within the search space.

9. The method according to claim 1, characterized in that When the genetic algorithm described in step 5 is iterating, the iteration rule is: when the fitness change is less than the set threshold, it is considered that there is no significant improvement in this generation, and the iteration is stopped; otherwise, it needs to continue to iterate until the maximum number of iterations.

Images