Non-line-of-sight unmanned aerial vehicle positioning method based on genetic algorithm

Abstract

The invention provides a non-line-of-sight unmanned aerial vehicle positioning method based on a genetic algorithm, and the method comprises the steps: obtaining the ghosting position of a target through BP imaging based on the multi-path echo data characteristics of a millimeter-wave radar, and enabling the principle to accord with a multi-path propagation model in a non-line-of-sight environment; then, the street width is adopted to represent building street layout information, an objective function containing the unknown street width is established based on the geometrical relationship between the positions of the target and the ghosting, the objective function is optimized through a genetic algorithm, and therefore the accurate position of the target and the street width are obtained; that is to say, the non-line-of-sight target positioning can be realized under the condition that the building layout is unknown, the method is an effective target positioning method, and the target positioning precision is higher; the method is especially suitable for a typical L-shaped city block model, and provides technical support for target positioning under a generalized non-line-of-sight model.

Description

Technical Field

[0001] The invention belongs to the technical field of shielded target detection, and in particular relates to a non-line-of-sight unmanned aerial vehicle positioning method based on a genetic algorithm. Background Art

[0002] In recent years, drone technology has been widely used in express logistics, disaster relief, smart cities and other fields due to its unique advantages. However, in complex urban environments, the increase in the number of drones inevitably brings potential airspace conflicts and security risks. Therefore, accurate acquisition of drone positions is the key to improving the safety of drone systems, promoting efficient collaboration between drones, and optimizing flight path planning.

[0003] Radar technology is a key means of target positioning, and there have been a lot of studies on positioning under line-of-sight (LOS) conditions. However, in complex urban environments, electromagnetic waves cannot reach the target location directly due to the obstruction of buildings, but need to reach the target through multi-path propagation such as diffraction and reflection on the surface of buildings. Currently, research on drone positioning in non-line-of-sight (NLOS) scenarios is relatively limited. Existing methods mainly rely on known building layout information to study drone signal extraction and positioning algorithms, but their applicability to unknown scenarios in practical applications, especially non-cooperative environments, is relatively limited. Summary of the invention

[0004] To solve the above problems, the present invention provides a non-line-of-sight UAV positioning method based on genetic algorithm, which can accurately obtain the target position of the UAV behind the non-line-of-sight corner under the condition of unknown building street layout.

[0005] A non-line-of-sight UAV positioning method based on a genetic algorithm, wherein a target to be measured is located in a non-line-of-sight blind area formed by a corner occlusion of an L-shaped street; the positioning method comprises the following steps:

[0006] S1: Obtain a set of multipath ghost positions of the target to be measured through the BP image containing the multipath ghost information of the UAV;

[0007] S2: Obtain the mapping relationship between the real position of the ghost of the target to be measured formed on each reflection path of the radar electromagnetic wave reflected multiple times by the L-shaped street and the real position of the target to be measured, wherein each reflection forms a real position of the ghost, and the real position of the ghost formed on the first P reflection paths with the largest reflection intensity is used to construct the objective function, wherein P is the number of paths in the multipath ghost position;

[0008] S3: The objective function is constructed based on the multipath ghost position and the real ghost position of the target to be measured as follows:

[0009]

[0010] Among them, (x e ,y e ) is the estimated position of the target to be measured, G est is a position set composed of the real positions of each ghost represented by the mapping relationship, G ref is the location set consisting of multipath ghost locations, ||·|| represents the bi-norm, D1 is the street width, and argmin represents the time when the objective function reaches the minimum value (x e ,y e ), the value of D1;

[0011] S4: Use genetic algorithm to optimize the objective function and obtain the position estimate of the target to be measured (x e ,y e ) and street width D1.

[0012] Furthermore, a method for obtaining a set of multipath ghost positions of the target to be measured through a BP image containing multipath ghost information of a drone is as follows:

[0013] Using Gaussian smoothing filter to process BP image I BP , and obtain the smoothed image I S ;

[0014] Using CA-CFAR threshold detection method to suppress smooth image I S The grating lobe in the image is obtained by c , where the grating lobe suppression image I c The pixel value I corresponding to each grid point c The calculation method of (q) is:

[0015]

[0016] Among them, I s (q) is the suppressed smoothed image I S The pixel value of the qth grid point on the left is α, which is the set detection threshold in the CA-CFAR threshold detection method;

[0017] Grating lobe suppression image I c Perform binarization processing to obtain a binary image;

[0018] The eight-connectedness algorithm is used to obtain the connected domains in the binary image, and the centroid of each connected domain is used as the multipath ghost position, wherein the number of connected domains is at most 3.

[0019] Furthermore, the method for obtaining the BP image containing the multipath ghost information of the UAV is as follows:

[0020] The millimeter-wave radar system with M transmission and N reception is used to collect multipath echo signals of the target to be measured in the non-line-of-sight blind area;

[0021] The background static clutter of multipath echo signals is suppressed by moving target detection technology MTI;

[0022] Perform FFT processing on the multipath echo signal after clutter suppression to obtain the distance-time diagram of M×N channels;

[0023] The imaging area of ​​the millimeter wave radar system is divided into X×Y grids, and the time delay τ corresponding to each grid q on each channel is calculated m,n (q):

[0024]

[0025] Among them, τ m (q) represents the electromagnetic wave transmitted from the mth transmitting antenna T m The delay along the straight line to the grid q, τ n (q) indicates that the electromagnetic wave propagates from the grid q along a straight line to the nth receiving antenna R n The delay, ||qT m || is the grid q to the mth transmitting antenna T m The distance, ||qR n || is the grid q to the nth receiving antenna R n distance, c is the speed of light, m ​​= 1, 2, .., M, n = 1, 2, .., N;

[0026] Match the multi-channel amplitudes for each grid point in the imaging area to obtain a BP image containing the multipath ghost information of the drone. The method for determining the amplitude corresponding to any grid point on any channel is:

[0027] Get the delay τ corresponding to the current grid point on the current channel m,n (q), obtain the delay τ on the distance-time graph m,n (q) The mth transmitting antenna T m With the nth receiving antenna R n The distance of the formed channel is taken as the amplitude of the current grid point on the current channel.

[0028] Furthermore, the method for obtaining the centroid of any connected domain is:

[0029]

[0030] Among them, G p (x p ,y p ) is the centroid coordinate of the pth connected domain, (x k,p ,yk,p ) is the coordinate of the kth grid in the pth connected domain, N p is the number of grids in the pth connected domain, p = 1, 2, …, P.

[0031] Furthermore, M=3, then the mapping relationship between the real position of each ghost image and the real position of the target to be measured is as follows:

[0032] G 11 (x 11 ,y 11 )=(x t -2D2-x u ,y u )

[0033] G 22 (x 22 ,y 22 )=(x u -2D1,y u )

[0034] G 33 (x 33 ,y 33 )=(2x t -2D2-2D1-x u ,y u )

[0035] Among them, G 11 (x 11 ,y 11 ) is the real position of the ghost image formed by the target on the reflection path of the first reflection of the radar electromagnetic wave, G 22 (x 22 ,y 22 ) is the real position of the ghost image formed by the target on the reflection path of the second reflection of the radar electromagnetic wave, G 33 (x 33 ,y 33 ) is the real position of the ghost image formed by the target on the reflection path of the third reflection of the radar electromagnetic wave, (x u ,y u ) is the actual position of the target to be measured, (x t ,y t ) is the position of the radar, D1 is the street width of the L-shaped street in the non-line-of-sight blind spot, and D2 is the distance from the radar to the wall of the L-shaped street in the non-line-of-sight blind spot.

[0036] Furthermore, the genetic algorithm is used to optimize the objective function to obtain the position estimate of the target to be measured (x e ,y e ) and the street width estimate D1, specifically:

[0037] S41: (x with different initial values e ,y e ) and D1 as different individuals to obtain the initial population, where the number of individuals is G;

[0038] S42: Determine whether each individual satisfies two set conditions x c <x e <x c +D1 and Where D1 is the width of the L-shaped street in the non-line-of-sight blind area, x c is the horizontal coordinate of the corner of the L-shaped street, is the average value of the ordinate of the real position of each ghost represented by the mapping relationship; the constraint violation counter conv corresponding to the individual that does not meet the two set conditions is recorded as 2, the constraint violation counter conv corresponding to the individual that meets one of the set conditions is recorded as 1, and the constraint violation counter conv corresponding to the individual that meets both set conditions is recorded as 0;

[0039] S43: Calculate the objective function value corresponding to each individual;

[0040] S44: Use simulated binary crossover and uniform mutation to evolve each individual of the previous generation population to obtain a new population after crossover and mutation;

[0041] S45: Calculate the objective function value corresponding to each individual in the new population, and determine the constraint violation counter conv corresponding to each individual in the new population according to the number of set conditions satisfied by the new population;

[0042] S46: Select the first G individuals with the smallest objective function value from the new population individuals whose constraint violation counter conv is 0 to form the next generation population;

[0043] S47: The next generation population obtained in step S46 is used as the previous generation population in step S44, and then steps S44 to S47 are repeated until the set number of iterations is reached, and then step S48 is entered;

[0044] S48: Select the best individual in all iterations as the final position estimate of the target to be measured (x e ,y e ) and the street width estimate D1, where the optimal individual satisfies: the constraint violation counter conv is 0 and the objective function value is the smallest.

[0045] Beneficial effects:

[0046] The present invention provides a non-line-of-sight UAV positioning method based on a genetic algorithm. Based on the characteristics of millimeter-wave radar multipath echo data, the ghost position of the target is obtained through BP imaging, and the principle conforms to the multipath propagation model in a non-line-of-sight environment; then, the street width is used to characterize the building street layout information, and based on the geometric relationship between the position of the target and the ghost, an objective function including the unknown street width is established, and the objective function is optimized through a genetic algorithm, so as to obtain the accurate position and street width of the target; that is, the present invention can realize non-line-of-sight target positioning under the condition of unknown building layout, is an effective target positioning method, and has higher target positioning accuracy; the present invention is particularly suitable for a typical L-shaped urban block model, and provides technical support for target positioning under a generalized non-line-of-sight model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a ghost and target position relationship diagram and a scene diagram simulated by the method of the present invention;

[0048] Figure 2 is the ghost imaging result of the simulation experiment;

[0049] Figure 3 This is the ghost position extraction result diagram of the simulation experiment;

[0050] Figure 4 It is the genetic algorithm optimization process curve of the simulation experiment;

[0051] Figure 5 It is the result of 50 Monte Carlo experiments in simulation;

[0052] Figure 6 It is a scene graph taken by the method of the present invention in actual measurement;

[0053] Figure 7 This is the ghost position extraction result diagram of the actual experiment;

[0054] Figure 8 It is the genetic algorithm optimization process curve of the measured experiment. DETAILED DESCRIPTION

[0055] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0056] A non-line-of-sight UAV positioning method based on a genetic algorithm, wherein a target to be measured is located in a non-line-of-sight blind area formed by a corner occlusion of an L-shaped street; the positioning method comprises the following steps:

[0057] S1: Obtaining a set of multipath ghost positions of the target to be measured through the BP image containing the multipath ghost information of the UAV, specifically including the following steps:

[0058] S11: Processing BP image I using Gaussian smoothing filter BP , and obtain the smoothed image I S ; At the same time, the method for obtaining the BP image containing the multipath ghost information of the UAV is:

[0059] S11a: A millimeter-wave radar system with M transmitters and N receivers is used to collect multipath echo signals of the target to be measured in the non-line-of-sight blind area;

[0060] S11b: Suppress the background static clutter of multipath echo signals through moving target detection technology MTI;

[0061] S11c: Perform FFT processing on the multipath echo signal after clutter suppression to obtain a distance-time diagram of M×N channels;

[0062] S11d: Divide the imaging area of ​​the millimeter wave radar system into X×Y grids, and calculate the time delay τ corresponding to each grid q on each channel m,n (q):

[0063]

[0064] Among them, τ m (q) represents the electromagnetic wave transmitted from the mth transmitting antenna T m The delay along the straight line to the grid q, τ n (q) indicates that the electromagnetic wave propagates from the grid q along a straight line to the nth receiving antenna R n The delay, ||qT m || is the grid q to the mth transmitting antenna T m The distance, ||qR n || is the grid q to the nth receiving antenna R n distance, c is the speed of light, m ​​= 1, 2, .., M, n = 1, 2, .., N;

[0065] S11e: Match the multi-channel amplitude for each grid point in the imaging area to obtain a BP image containing multipath ghost information of the drone, wherein the amplitude corresponding to any grid point on any channel is determined as follows:

[0066] Get the delay τ corresponding to the current grid point on the current channel m,n (q), obtain the delay τ on the distance-time graph m,n (q) The mth transmitting antenna T m With the nth receiving antenna R n The distance of the formed channel is taken as the amplitude of the current grid point on the current channel;

[0067] S12: Suppressing smooth image I using CA-CFAR threshold detection methodS The grating lobe in the image is obtained by c , where the grating lobe suppression image I c The pixel value I corresponding to each grid point c The calculation method of (q) is:

[0068]

[0069] Among them, I s (q) is the suppressed smoothed image I S The pixel value of the qth grid point on the left is α, which is the set detection threshold in the CA-CFAR threshold detection method;

[0070] S13: Grating lobe suppression image I c Perform binarization processing to obtain a binary image;

[0071] S14: using the eight-connectivity algorithm to obtain connected domains in the binary image, and taking the centroid of each connected domain as the multipath ghost position, wherein the number of connected domains is at most 3; the method for obtaining the centroid of any connected domain is:

[0072]

[0073] Among them, G p (x p ,y p ) is the centroid coordinate of the pth connected domain, (x k,p ,y k,p ) is the coordinate of the kth grid in the pth connected domain, N p is the number of grids in the pth connected domain, p = 1, 2, …, P.

[0074] S2: Obtain the mapping relationship between the real position of the ghost of the target to be measured formed on each reflection path of the radar electromagnetic wave reflected multiple times by the L-shaped street and the real position of the target to be measured, wherein each reflection forms a real position of the ghost, and the real position of the ghost formed on the first P reflection paths with the largest reflection intensity is used to construct the objective function, wherein P is the number of paths in the multipath ghost position;

[0075] Construct a typical non-line-of-sight urban environment with an "L"-shaped street layout, such as Figure 1 As shown, the street width is represented by D1, and the radar T(x t ,y t ) to wall 1 is represented by D2, and radar T(x t ,y t ) to wall 3 is represented by D3, and the position of corner C is (x c ,y c ), set the target real position U(xu ,y u ) is located at the radar T(x t ,y t )(location is known) in the non-line-of-sight region. Among them, D3 is a known quantity, and D1 and D2 are unknown quantities;

[0076] The transformation relationship for determining building layout parameters is:

[0077]

[0078] It should be noted that the radar electromagnetic wave will be reflected multiple times in the L-shaped street and have different reflection paths, but the energy of the radar electromagnetic wave is severely attenuated after three reflections and can be ignored. Therefore, the present invention only calculates the ghost position within three reflections.

[0079] Based on this, the mapping relationship between the real position of each ghost image and the real position of the target to be measured is as follows:

[0080] G 11 (x 11 ,y 11 )=(x t -2D2-x u ,y u )

[0081] G 22 (x 22 ,y 22 )=(x u -2D1,y u )

[0082] G 33 (x 33 ,y 33 )=(2x t -2D2-2D1-x u ,y u )

[0083] Among them, G 11 (x 11 ,y 11 ) is the real position of the ghost image formed by the target on the reflection path of the first reflection of the radar electromagnetic wave, G 22 (x 22 ,y 22 ) is the real position of the ghost image formed by the target on the reflection path of the second reflection of the radar electromagnetic wave, G 33 (x 33 ,y 33 ) is the real position of the ghost image formed by the target on the reflection path of the third reflection of the radar electromagnetic wave, (x u ,y u) is the actual position of the target to be measured, (x t ,y t ) is the position of the radar, D1 is the street width of the L-shaped street in the non-line-of-sight blind spot, and D2 is the distance from the radar to the wall of the L-shaped street in the non-line-of-sight blind spot.

[0084] S3: The objective function is constructed based on the multipath ghost position and the real ghost position of the target to be measured as follows:

[0085]

[0086] Among them, (x e ,y e ) is the estimated position of the target to be measured, G est is a position set composed of the real positions of each ghost represented by the mapping relationship, G ref is the location set consisting of multipath ghost locations, ||·|| represents the bi-norm, D1 is the street width, and argmin represents the time when the objective function reaches the minimum value (x e ,y e ), the value of D1;

[0087] S4: Use genetic algorithm to optimize the objective function and obtain the position estimate of the target to be measured (x e ,y e ) and street width D1, specifically comprising the following steps:

[0088] S41: (x with different initial values e ,y e ) and D1 as different individuals to obtain the initial population, where the number of individuals is G;

[0089] S42: Determine whether each individual satisfies two set conditions x c <x e <x c +D1 and Where D1 is the width of the L-shaped street in the non-line-of-sight blind area, x c is the horizontal coordinate of the corner of the L-shaped street, is the average value of the ordinate of the real position of each ghost represented by the mapping relationship; the constraint violation counter conv corresponding to the individual that does not meet the two set conditions is recorded as 2, the constraint violation counter conv corresponding to the individual that meets one of the set conditions is recorded as 1, and the constraint violation counter conv corresponding to the individual that meets both set conditions is recorded as 0;

[0090] S43: Calculate the objective function value corresponding to each individual;

[0091] S44: Use simulated binary crossover and uniform mutation to evolve each individual of the previous generation population to obtain a new population after crossover and mutation;

[0092] S45: Calculate the objective function value corresponding to each individual in the new population, and determine the constraint violation counter conv corresponding to each individual in the new population according to the number of set conditions satisfied by the new population;

[0093] S46: Select the first G individuals with the smallest objective function value from the new population individuals whose constraint violation counter conv is 0 to form the next generation population;

[0094] S47: The next generation population obtained in step S46 is used as the previous generation population in step S44, and then steps S44 to S47 are repeated until the set number of iterations is reached, and then step S48 is entered;

[0095] S48: Select the best individual in all iterations as the final position estimate of the target to be measured (x e ,y e ) and the street width estimate D1, where the optimal individual satisfies: the constraint violation counter conv is 0 and the objective function value is the smallest.

[0096] The following takes the number of iterations H=50 and the number of individuals G=50 as an example to describe in detail the process of optimizing the objective function by the genetic algorithm of the present invention.

[0097] First, the parameters of the genetic algorithm are initialized, and the following constraints are set using the positional relationship between the target location and the building block:

[0098] (1) The target is located within the street interval, that is, x c <x u <x c +D1;

[0099] (2) The difference between the y-axis coordinates of the target and the extracted ghost point is no more than 0.2, that is,

[0100] If an individual violates a constraint, the constraint violation counter conv is incremented by 1, and the individual is evolved by simulating binary crossover (SBX) and uniform mutation. The crossover probability P1 is set to 0.9, that is, when the random number generated between (0,1) is greater than P1, crossover evolution is performed. If two parent individuals p1 and p2 cross, two offspring c1 and c2 can be generated, which can be expressed as

[0101]

[0102] Where a is a constant with a value of 0.5. β is dynamically and randomly controlled by the parameters of the crossover distribution and is calculated by the following formula:

[0103]

[0104] Wherein, rand is a random number in the interval (0,1). T is a constant, and its value is 0.5. μ is a parameter that controls the crossover distribution. The larger its value, the smaller the difference between the offspring and the parent individuals. In the present invention, μ=10 is set. The probability of mutation P2 is set to 0.04. For each individual after the crossover, a random number in the interval (0,1) is generated. If the random number is greater than P2, the individual mutates. The present invention adopts a uniform mutation method, that is, by randomly extracting a mutation amount from a uniform distribution in the interval (0,1), and adding the mutation amount to the individual, a new individual after mutation is generated. After obtaining the new population after the crossover mutation, a competitive selection method is used for environmental selection. The individual with the highest fitness and the lowest conv value is selected as the winner, and is used as the individual of the next generation population.

[0105] Repeat the above crossover, mutation and selection operations. After 50 iterations, select the best individual as the parameter estimation result.

[0106] The simulation and actual measurement verification are carried out below.

[0107] Simulation scenario such as Figure 1 As shown in the figure, an L-shaped non-line-of-sight scene model is constructed, and the gprMax software is used to simulate the electromagnetic echo in this scene. The target is set in the non-line-of-sight blind area of ​​the radar. The specific simulation parameters are shown in Table 1.

[0108] Table 1 Simulation experiment parameter settings

[0109]

[0110] Figure 2 is a ghost image processed by the method proposed by the present invention, Figure 3 This is the ghost position extraction result after processing in step 4. Considering the serious attenuation problem of high-order reflection signals in actual measurement, only the ghosts formed by the first three paths are used. Figure 4 is the optimization process curve of the genetic algorithm, where the blue line is the average degree of constraint violation. As the iterations increase, the degree of constraint violation decreases rapidly and approaches 0, indicating that the optimization process is effectively looking for the optimal solution that meets the constraints; the red line is the objective function value, indicating that the genetic algorithm is constantly looking for the optimal solution. The target position estimation result is (2.55m, 2.01m), and the error compared with its true value (2.45m, 2.0m) is 0.11m. The street width estimation value is 3.03m, and the error compared with the true value is 0.03m. In order to further analyze the feasibility of the method proposed in this invention, the target position is randomly generated for Monte Carlo experiment verification. Figure 5The experimental error curve is unfolded, where the blue curve represents the street width error and the red curve represents the drone position error. After 50 experiments, the root mean square error (RMES) of the drone position is 0.086, and the RMES of the street width is 0.128m.

[0111] At the same time, in order to verify the practicability of the method proposed in the present invention, the IWR6843ISK millimeter wave radar is used to carry out actual measurement experiments. Figure 6 As shown, the rotorcraft UAV is placed in the non-line-of-sight blind spot of the radar, and the parameter settings of the actual measurement experiment are shown in Table 2.

[0112] Table 2 Parameter settings of the measured experiment

[0113]

[0114] Based on the multipath echo data collected by field measurement, Figure 7 is the extracted ghost position, Figure 8 For the optimization process curve, as the number of iterations increases, the average constraint violation and the optimal objective function value gradually converge and tend to 0, indicating that the optimal solution satisfies both the objective function and the constraint conditions. The target position estimation result is (1.08m, 1.33m), and the street width estimation value is 1.60m.

[0115] The present invention provides a non-line-of-sight UAV positioning method based on a genetic algorithm. In this method, the ghost position is calculated by using the traditional BP imaging method, which reflects the virtual target position corresponding to each propagation path in the imaging area. Then, the building layout information is parameterized, and the street spacing is used to characterize the building layout information in combination with the building geometry. An objective function is established based on the positional relationship between the ghost and the target, and a genetic algorithm is used for optimization to obtain an accurate UAV target position. Compared with the existing methods, the proposed method can effectively solve the problem of UAV target positioning when the building layout is unknown. Both simulation experiments and actual measurement experiments show that the present invention can accurately obtain the target position and is an effective non-line-of-sight UAV target positioning method.

[0116] 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 non-line-of-sight UAV positioning method based on genetic algorithm, characterized in that: The target to be measured is located in a non-line-of-sight blind area formed by the corner occlusion of an L-shaped street; the positioning method comprises the following steps: S1: Obtain a set of multipath ghost positions of the target to be measured through the BP image containing the multipath ghost information of the UAV; S2: Obtain the mapping relationship between the real position of the ghost of the target to be measured formed on each reflection path of the radar electromagnetic wave reflected multiple times by the L-shaped street and the real position of the target to be measured, wherein each reflection forms a real position of the ghost, and the real position of the ghost formed on the first P reflection paths with the largest reflection intensity is used to construct the objective function, wherein P is the number of paths in the multipath ghost position; S3: The objective function is constructed based on the multipath ghost position and the real ghost position of the target to be measured as follows: Among them, (x e ,y e ) is the estimated position of the target to be measured, G est is a position set composed of the real positions of each ghost represented by the mapping relationship, G ref is the location set consisting of multipath ghost locations, ||·|| represents the bi-norm, D1 is the street width, and argmin represents the time when the objective function reaches the minimum value (x e ,y e ), the value of D1; S4: Use genetic algorithm to optimize the objective function and obtain the position estimate of the target to be measured (x e ,y e ) and street width D1.

2. The non-line-of-sight UAV positioning method based on genetic algorithm as claimed in claim 1, characterized in that: The method for obtaining a set of multipath ghost positions of the target to be measured through the BP image containing the multipath ghost information of the UAV is: Using Gaussian smoothing filter to process BP image I BP , and obtain the smoothed image I S ; Using CA-CFAR threshold detection method to suppress smooth image I S The grating lobe in the image is obtained by c , where the grating lobe suppression image I c The pixel value I corresponding to each grid point c The calculation method of (q) is: Among them, I s (q) is the suppressed smoothed image I S The pixel value of the qth grid point on the left is α, which is the set detection threshold in the CA-CFAR threshold detection method; Grating lobe suppression image I c Perform binarization processing to obtain a binary image; The eight-connectedness algorithm is used to obtain the connected domains in the binary image, and the centroid of each connected domain is used as the multipath ghost position, wherein the number of connected domains is at most 3.

3. A non-line-of-sight UAV positioning method based on genetic algorithm as claimed in claim 1 or 2, characterized in that: The method for obtaining the BP image containing the multipath ghost information of the UAV is as follows: The millimeter-wave radar system with M transmission and N reception is used to collect multipath echo signals of the target to be measured in the non-line-of-sight blind area; The background static clutter of multipath echo signals is suppressed by moving target detection technology MTI; Perform FFT processing on the multipath echo signal after clutter suppression to obtain the distance-time diagram of M×N channels; The imaging area of ​​the millimeter wave radar system is divided into X×Y grids, and the time delay τ corresponding to each grid q on each channel is calculated m,n (q): Among them, τ m (q) represents the electromagnetic wave transmitted from the mth transmitting antenna T m The delay along the straight line to the grid q, τ n (q) indicates that the electromagnetic wave propagates from the grid q along a straight line to the nth receiving antenna R n The delay, ||qT m || is the grid q to the mth transmitting antenna T m The distance, ||qR n || is the grid q to the nth receiving antenna R n distance, c is the speed of light, m ​​= 1, 2, .., M, n = 1, 2, .., N; Match the multi-channel amplitudes for each grid point in the imaging area to obtain a BP image containing the multipath ghost information of the drone. The method for determining the amplitude corresponding to any grid point on any channel is: Get the delay τ corresponding to the current grid point on the current channel m,n (q), obtain the delay τ on the distance-time graph m,n (q) The mth transmitting antenna T m With the nth receiving antenna R n The distance of the formed channel is taken as the amplitude of the current grid point on the current channel.

4. The non-line-of-sight UAV positioning method based on genetic algorithm as claimed in claim 2, characterized in that: The method to obtain the centroid of any connected domain is: Among them, G p (x p ,y p ) is the centroid coordinate of the pth connected domain, (x k,p ,y k,p ) is the coordinate of the kth grid in the pth connected domain, N p is the number of grids in the pth connected domain, p = 1, 2, …, P.

5. The non-line-of-sight UAV positioning method based on genetic algorithm as claimed in claim 1, characterized in that: M = 3, then the mapping relationship between the real position of each ghost image and the real position of the target to be measured is as follows: G 11 (x 11 ,y 11 )=(x t -2D2-x u ,y u ) G 22 (x 22 ,and 22 )=(x u -2D1,and u ) G 33 (x 33 ,y 33 )=(2x t -2D2-2D1-x u ,y u ) Among them, G 11 (x 11 ,y 11 ) is the real position of the ghost image formed by the target on the reflection path of the first reflection of the radar electromagnetic wave, G 22 (x 22 ,y 22 ) is the real position of the ghost image formed by the target on the reflection path of the second reflection of the radar electromagnetic wave, G 33 (x 33 ,y 33 ) is the real position of the ghost image formed by the target on the reflection path of the third reflection of the radar electromagnetic wave, (x u ,y u ) is the actual position of the target to be measured, (x t ,y t ) is the position of the radar, D1 is the street width of the L-shaped street in the non-line-of-sight blind spot, and D2 is the distance from the radar to the wall of the L-shaped street in the non-line-of-sight blind spot.

6. The non-line-of-sight UAV positioning method based on genetic algorithm as claimed in claim 1, characterized in that: The genetic algorithm is used to optimize the objective function and obtain the position estimate of the target to be measured (x e ,y e ) and the street width estimate D1, specifically: S41: (x with different initial values e ,y e ) and D1 as different individuals to obtain the initial population, where the number of individuals is G; S42: Determine whether each individual satisfies two set conditions x c <x e <x c +D1 and Where D1 is the width of the L-shaped street in the non-line-of-sight blind area, x c is the horizontal coordinate of the corner of the L-shaped street, is the average value of the ordinate of the real position of each ghost represented by the mapping relationship; the constraint violation counter conv corresponding to the individual that does not meet the two set conditions is recorded as 2, the constraint violation counter conv corresponding to the individual that meets one of the set conditions is recorded as 1, and the constraint violation counter conv corresponding to the individual that meets both set conditions is recorded as 0; S43: Calculate the objective function value corresponding to each individual; S44: Use simulated binary crossover and uniform mutation to evolve each individual of the previous generation population to obtain a new population after crossover and mutation; S45: Calculate the objective function value corresponding to each individual in the new population, and determine the constraint violation counter conv corresponding to each individual in the new population according to the number of set conditions satisfied by the new population; S46: Select the first G individuals with the smallest objective function value from the new population individuals whose constraint violation counter conv is 0 to form the next generation population; S47: The next generation population obtained in step S46 is used as the previous generation population in step S44, and then steps S44 to S47 are repeated until the set number of iterations is reached, and then step S48 is entered; S48: Select the best individual in all iterations as the final position estimate of the target to be measured (x e ,y e ) and the street width estimate D1, where the optimal individual satisfies: the constraint violation counter conv is 0 and the objective function value is the smallest.

Images