Obstacle super-resolution perception method and device based on constant false alarm detector

Abstract

The present invention discloses an obstacle super-resolution perception method based on a constant false alarm detector, comprising the following steps: step 1, obtaining a spectrum signal of a multidimensional line spectrum signal in a radar signal; step 2, selecting a unit with the largest square amplitude modulus value as a first detection unit, and inputting it into a constant false alarm detector for detection; step 3, removing the first detection unit from the spectrum signal, selecting a unit with the largest square amplitude modulus value in the removed spectrum signal as a second detection unit, and inputting it into the constant false alarm detector to determine whether the second detection unit is a valid target, and to review the first detection unit determined as an invalid target; step 4, repeating the above steps to output a final valid target set; step 5, obtaining the spatial position information of the obstacle based on the final valid target set. The present invention also provides an obstacle false alarm device. The method provided by the present invention can effectively improve the feedback speed and accuracy of false alarm results.

Description

Technical Field

[0001] The present invention belongs to the technical field of line spectrum estimation, and in particular relates to an obstacle super-resolution perception method and device based on a constant false alarm detector. Background Art

[0002] Line spectrum estimation problems are widely used, such as the environmental perception problem of linear frequency modulated continuous wave radar in autonomous driving and the multipath channel estimation problem in orthogonal frequency division multiplexing (OFDM) communication.

[0003] There are many mature solutions to this problem at this stage, such as conventional constant false alarm rate (CFAR) detection based on fast Fourier transform (FFT) and NOMP-CFAR algorithm.

[0004] Among them, conventional CFAR detection based on FFT converts the line spectrum signal in the time domain into a frequency domain signal after FFT, and then detects each unit in the frequency domain through the CFAR method. Conventional CFAR detection estimates the level of interference from data samples based on methods such as cell average (CA) and order statistic (OS), and the threshold factor is determined by the estimation method and the set average false alarm rate. The estimation result is multiplied by the threshold factor to obtain an adaptive threshold, so as to detect each frequency domain unit.

[0005] However, the algorithm has limitations: First, FFT is a grid-based algorithm, and grid mismatch may occur. In addition, the resolution of FFT is limited by the Rayleigh criterion. Second, when stronger and weaker targets are adjacent, the FFT algorithm may have a phenomenon where the strong target may cover the weak target, making it difficult for the algorithm to detect the weak target and reducing the detection probability. The above describes some inherent problems of the FFT algorithm, and the FFT-based CFAR detection method may cause further problems, such as the CA-CFAR method may cause missed detection due to mutual masking between targets or false alarms due to clutter edges.

[0006] The NOMP-CFAR algorithm includes an initialization step and a detection step, or a model order estimation (MOE) step. In the initialization step, the algorithm uses the Newton orthogonal matching algorithm to pre-estimate the target so that the number of candidate targets reaches the set maximum target number upper limit. In the detection step, a constant false alarm detector is used to evaluate whether the target exists. Unlike the traditional CFAR detector that outputs a binary decision result, the detector in NOMP-CFAR outputs a soft threshold, which represents the amplitude (in decibels) of the target intensity crossing the set threshold. According to the soft threshold, only one frequency component is activated or deactivated in each iteration.

[0007] However, NOMP-CFAR includes forward iterations and backward iterations involving multiple FFT operations, which has high computational complexity and requires a long running time.

[0008] Patent document CN111830465A discloses a two-dimensional Newton orthogonal matching pursuit compressed beamforming method. The method includes: step 1, constructing a mathematical model and a sound source estimation expression, step 2, roughly estimating the sound source position coordinates from the discrete grid nodes based on the discretized grid model, step 3, optimizing the roughly estimated single sound source position coordinates by Newton method, and step 4, further optimizing the identified s sound source position coordinates and sound source intensity estimation on the basis of completing the single optimization of step 3.

[0009] The global loop optimization feedback process introduced by this method enables the identified close sound sources to be updated and have a chance to be accurately separated. However, the iterative step stopping criterion adopted by this method is that the residual reduction is less than the set threshold (10 -4 ), there are large false alarms (that is, too sensitive) in actual use. In addition, no optimization is done in the calculation process, which makes the result converge slowly in actual use. Summary of the invention

[0010] The purpose of the present invention is to provide an obstacle super-resolution perception method based on a constant false alarm detector, which can effectively improve the feedback speed of false alarm detection results while also ensuring the accuracy of the final result.

[0011] An obstacle super-resolution perception method based on a constant false alarm detector comprises the following steps:

[0012] Step 1: Obtain a multidimensional line spectrum signal in the radar signal, and obtain a frequency spectrum signal of the multidimensional line spectrum signal through an FFT algorithm.

[0013] Step 2: Calculate the square of the amplitude modulus value in each unit in the spectrum signal, select the unit with the largest square of the amplitude modulus value as the first detection unit, and input it into the pre-built constant false alarm detector for detection to determine whether there is a valid target in the first detection unit:

[0014] When the first detection unit is a valid target, the corresponding target complex amplitude estimation value and angular frequency estimation value are calculated based on the coordinates of the unit where the valid target is located, the calculated angular frequency estimation value is corrected using the Newton orthogonal matching tracking method, and the target complex amplitude estimation value and the corrected angular frequency estimation value are stored in the valid target set;

[0015] When the first detection unit is an invalid target, the number of invalid targets is counted and added by 1. At the same time, the corresponding target complex amplitude estimation value and angular frequency estimation value are calculated based on the coordinates of the unit where the invalid target is located. The calculated angular frequency estimation value is corrected using the Newton orthogonal matching tracking method, and the target complex amplitude estimation value and the corrected angular frequency estimation value are stored in the invalid target set.

[0016] Step 3: remove the first detection unit from the spectrum signal, select the unit with the largest square of amplitude modulus value in the spectrum signal after removal as the second detection unit, and input it into the constant false alarm detector to determine whether the second detection unit is a valid target, and at the same time review the first detection unit determined as an invalid target:

[0017] When the second detection unit is a valid target, the corresponding target complex amplitude estimation value and the corrected angular frequency estimation value are calculated and stored in the valid target set, and the target complex amplitude estimation value and the corrected angular frequency estimation value corresponding to the first detection unit in the non-valid target set are stored in the valid target set, and the number of non-valid targets is cleared;

[0018] When the second detection unit is an invalid target, the number of invalid targets is counted plus 1, and the corresponding target complex amplitude estimation value and the corrected angular frequency estimation value are calculated and stored in the invalid target set.

[0019] Step 4: Repeat steps 2 to 3 until the judgment termination condition of the constant false alarm detector is met to output a final valid target set.

[0020] Step 5: Based on the final valid target set obtained in step 4, determine the spatial position information of the obstacle in the feedback direction of the radar signal.

[0021] The present invention introduces Newton's method on the basis of NOMP-CFAR algorithm to make fine correction on the frequency, and only needs to perform one FFT operation, while also avoiding the problem of backward iteration.

[0022] Specifically, in step 1, the radar signal includes vehicle-mounted radar signals, navigation radar and flight radar.

[0023] Specifically, the constant false alarm detector needs to be initialized with parameters before use, and the parameters include false alarm probability, maximum number of iterations, maximum number of consecutive ineffective targets, oversampling factor, number of corrections, number of protection units in the detection process, and number of training units.

[0024] Specifically, the determination of the constant false alarm detector includes the following process:

[0025] The unit with the largest square amplitude modulus value is taken as the detection unit, and a corresponding data set is constructed with the preset number of protection units and the number of training units.

[0026] The interference level estimation value of the data set is calculated, and its specific expression is as follows:

[0027]

[0028] Where N r represents the number of training units, Representing a collection The unit number in , i and j represent the coordinates of the detection unit, Represents the signal spectrum.

[0029] The threshold of the effective target is determined by the interference level estimation value, which is specifically expressed as follows:

[0030]

[0031] In the formula, α represents the multiplication factor.

[0032] The square of the amplitude modulus value of the detection unit is compared with the threshold value to output a judgment result.

[0033] Specifically, in step 2, the target complex amplitude estimation value and the angular frequency estimation value are calculated as follows:

[0034]

[0035] Where i and j represent the coordinates of the detection unit, represents the target complex amplitude estimate, N represents the number of samples in the fast time domain, M represents the number of samples in the slow time domain, γ x represents the oversampling factor in the fast time domain, γ y represents the oversampling factor in the slow time domain, and Represents the angular frequency estimate.

[0036] Specifically, in step 2, the Newton orthogonal matching tracking method is used to correct the calculated angular frequency, and its expression is as follows:

[0037]

[0038] Where Y r (k-1) represents the residual of the line spectrum signal after removing the influence of (k-1) estimated targets. represents the estimated value of the kth target frequency, Indicates that S is at the iteration point The gradient at Indicates that S is at the iteration point The Hessian matrix at , (·) T represents the transpose of a matrix or vector, (·) -1 represents the inverse of the matrix, ‖·‖ F Represents the F-norm of the matrix.

[0039] Preferably, symmetry processing is used to simplify the multidimensional line spectrum signal in the Newton orthogonal matching tracking method, and its expression is as follows:

[0040]

[0041] In the formula, represents a multidimensional line spectrum atom, a N (·) represents the atomic vector in the fast time domain, represents the atomic vector transpose in the slow time domain, Represents the estimated value of the kth target angular frequency.

[0042] At the same time, the phase correction is also performed on the complex amplitude of the target, which can be specifically expressed as:

[0043]

[0044] This reduces the computational complexity of the entire optimization and correction process.

[0045] Specifically, in step 3, the elimination includes eliminating the influence of the first detection unit on the time domain line spectrum signal and the influence of the first detection unit on the frequency domain line spectrum signal.

[0046] The specific expression is as follows:

[0047]

[0048] Where Y r (k) represents the time domain line spectrum signal, represents the frequency domain line spectrum signal, A(·) represents the line spectrum atomic matrix, represents the frequency domain form of the line spectrum atomic matrix, represents the target complex amplitude estimate of the kth target, Represents the corrected angular frequency estimate of the kth target.

[0049] Specifically, in step 4, the determination termination condition includes:

[0050] a. The number of repetitions has reached the maximum number of iterations preset by the constant false alarm detector.

[0051] b. The number of marked ineffective targets has reached the maximum number of consecutive ineffective targets preset by the constant false alarm detector.

[0052] Preferably, step 4 also includes using the least squares method to correct all target complex amplitude estimates in the final valid target set.

[0053] The present invention also provides an obstacle false alarm device, comprising a computer memory, a computer processor, and a computer program stored in the computer memory and executable on the computer processor, wherein the computer program adopts the above-mentioned obstacle super-resolution perception method based on a constant false alarm detector.

[0054] The computer processor executes the computer program to implement the following steps: acquiring a radar signal received by a vehicle-mounted radar, analyzing and processing the radar signal using the obstacle false alarm method, and outputting spatial position information of obstacles around the vehicle.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] (1) Compared with the FFT-based CFAR method, the present invention uses the NOMP method on a grid-based basis to make corrections, thereby breaking through the limitation of Rayleigh resolution in terms of accuracy performance and approaching the Cramer-Rao bound.

[0057] (2) Secondly, for the problem of strong targets covering weak targets in FFT, the algorithm of the present invention eliminates the targets one by one. When the strong targets are eliminated by the algorithm, the probability of successful detection of weak targets is greatly improved. Thirdly, for the problem of mutual covering between targets in CA-CFAR, the present invention stores the ineffective targets first. If effective targets are detected in the next round (or several rounds) of detection, the stored ineffective targets are regarded as effective targets and corrected and least squares calculated. Such algorithm design better solves the problem of mutual covering between targets.

[0058] (3) Compared with the NOMP-CFAR algorithm, the present invention cyclically deletes the time-consuming target estimation initialization step in the original NOMP-CFAR, and changes the original algorithm from removing invalid targets from the result set to adding valid targets. This change greatly reduces the complexity of the algorithm and reduces the repeated calls of functions. In addition, in order to reduce the occurrence of missed detection events, the present algorithm introduces a non-valid target number marker, thereby avoiding the occurrence of missed detection events caused by strong interference and improving the detection probability of the algorithm.

[0059] (4) In addition to removing invalid targets and adding valid targets, the present invention also updates the residuals in the time domain and frequency domain at the same time, and downsamples the residuals in the frequency domain to the input required for CFAR detection, thereby reducing the number of times the FFT algorithm is called in the algorithm, and also reducing the complexity of the algorithm in the present invention.

[0060] (5) The algorithm of the present invention adopts symmetric multi-dimensional line spectrum atoms. Such algorithm design can appropriately increase the iterative step size of each step of the NOMP algorithm, thereby accelerating the convergence speed of the NOMP algorithm and reducing the iterative steps, thereby reducing the algorithm execution time. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A flowchart of an obstacle super-resolution perception method based on a constant false alarm detector provided in this embodiment;

[0062] Figure 2 A determination flow chart of a constant false alarm detector provided in this embodiment;

[0063] Figure 3 This is a schematic diagram comparing the obstacle false alarm result provided in this embodiment with the actual situation. DETAILED DESCRIPTION

[0064] like Figure 1 As shown in FIG. 1 , a method for false alarm of obstacles is shown, and the specific process is as follows:

[0065] Step 1: Obtain a multidimensional line spectrum signal in the radar signal, and obtain a frequency spectrum signal of the multidimensional line spectrum signal through an FFT algorithm.

[0066] More specifically, the multidimensional line spectrum signal is a two-dimensional line spectrum signal, and its expression is as follows:

[0067]

[0068] In the formula, represents a two-dimensional line spectrum signal, N represents the number of samples in the fast time domain, and M represents the number of samples in the slow time domain. represents the complex amplitude of each target, K represents the total number of targets, ω k =[ω x,k,ω y,k ] T ∈[0,2π) 2 represents the angular frequency vector of the Kth target, ω x,k and ω y,k Represent the angular frequencies in the fast and slow time domains respectively, or the matrix ω=[ω 1 ,ω 2 ,ω 3 …ω K ] T to include the angular frequency vectors of all K targets, represents noise, σ 2 is the noise variance, A(·) represents the two-dimensional line spectrum atom, which is expressed as follows:

[0069]

[0070] a N (ω)=[1,e jω ,…e j(N-1)ω ] T

[0071] a M (ω)=[1,e jω ,…e j(M-1)ω ] T

[0072] Then the spectrum of the line spectrum signal is obtained by FFT calculation The specific expression is as follows:

[0073]

[0074] In the formula, is a two-dimensional Fourier transform, which can be implemented by using the fft2 function in matlab or the fft library in numpy. The transformation in is a unitary transformation, while the matlab fft2 function is not, so it needs to be properly adjusted in actual operation. At the same time, initialize the variables required by the algorithm, such as: set the target frequency and complex amplitude estimation result sets to empty sets, set the target number estimation value to And the non-valid target number flag f is initialized to 0.

[0075] Step 2: Calculate the square of the amplitude modulus value in each unit in the spectrum signal, select the unit with the largest square of the amplitude modulus value as the first detection unit, and input it into a pre-built constant false alarm detector for detection to determine whether there is a valid target in the first detection unit.

[0076] When the first detection unit is a valid target, the corresponding target complex amplitude estimation value and angular frequency estimation value are calculated based on the coordinates of the unit where the valid target is located, the calculated angular frequency estimation value is corrected using the Newton orthogonal matching tracking method, and the target complex amplitude estimation value and the corrected angular frequency estimation value are stored in the valid target set.

[0077] When the first detection unit is an invalid target, the number of invalid targets is counted and added by 1. At the same time, the corresponding target complex amplitude estimation value and angular frequency estimation value are calculated based on the coordinates of the unit where the invalid target is located. The calculated angular frequency estimation value is corrected using the Newton orthogonal matching tracking method, and the target complex amplitude estimation value and the corrected angular frequency estimation value are stored in the invalid target set.

[0078] The constant false alarm detector needs to be initialized before use, including the false alarm probability Enter the maximum allowed number of iterations K max , the maximum allowed number of consecutive ineffective targets f max , oversampling factor (γ x ,γ y ), single target correction times R s , the number of cycle corrections R c , the number of protection units and training units of the CFAR detector and other parameter settings required by the algorithm

[0079] More specifically, if Figure 2 The following is the judgment process of the constant false alarm detector:

[0080] The unit with the largest square amplitude modulus value is taken as the detection unit, and a corresponding data set is constructed with the preset number of protection units and the number of training units.

[0081] The interference level estimation value of the data set is calculated, and its specific expression is as follows:

[0082]

[0083] Where N r represents the number of training units, Representing a collection The unit number in is i and j, which are the detection unit positions.

[0084] The threshold of the effective target is determined by the interference level estimation value, which is specifically expressed as follows:

[0085]

[0086] In the formula, α represents the multiplication factor. α is only determined by N, M and N r Determine, its specific relationship can be expressed as:

[0087]

[0088] The anomalous integral can be calculated by the integral function of Matlab.

[0089] Compare the amplitude modulus of the cell pointed to by the maximum value with the threshold and make a decision. The decision can be expressed as

[0090]

[0091] That is, output the judgment result. If If true, the output is 1; otherwise, the output is 0.

[0092] The Newton orthogonal matching tracking method is used to correct the calculated angular frequency, and its expression is as follows:

[0093]

[0094] Where Y r (k-1) represents the residual of the line spectrum signal after removing the influence of (k-1) estimated targets. represents the estimated value of the kth target frequency, Indicates that S is at the iteration point The gradient at Indicates that S is at the iteration point The Hessian matrix at , (·) T represents the transpose of a matrix or vector, (·) -1 represents the inverse of the matrix, ‖·‖ F Represents the F-norm of the matrix.

[0095] At the same time, the multi-dimensional line spectrum signal in the Newton orthogonal matching tracking method is simplified by symmetry processing, and its expression is as follows:

[0096]

[0097] In the formula, represents a multidimensional line spectrum atom, a N (·) represents the atomic vector in the fast time domain, represents the atomic vector transpose in the slow time domain, It represents the estimated value of the kth target angular frequency, and at the same time performs phase correction on the complex amplitude of the target, which can be specifically expressed as:

[0098]

[0099] Step 4: Repeat steps 2 to 3 until the judgment termination condition of the constant false alarm detector is met to output a final valid target set.

[0100] More specifically, the least squares method is used to correct all target complex amplitude estimates in the final effective target set, and its expression is as follows:

[0101]

[0102] At the same time, vectors and matrices can be used in parallel during the correction process to reduce the complexity of the algorithm.

[0103] Step 5: Based on the final valid target set obtained in step 4, determine the spatial position information of the obstacle in the feedback direction of the radar signal.

[0104] The termination conditions include:

[0105] a. The number of repetitions has reached the maximum allowed number of iterations K for the constant false alarm detector input. max .

[0106] b. The number of marked ineffective targets has reached the number of continuous ineffective targets of the constant false alarm detector f. max .

[0107] This embodiment also provides an obstacle false alarm device, including a computer memory, a computer processor, and a computer program stored in the computer memory and executable on the computer processor, wherein the computer program adopts the obstacle false alarm method described in the above embodiment.

[0108] The computer processor executes the computer program to implement the following steps: acquiring a radar signal received by a vehicle-mounted radar, analyzing and processing the radar signal using the obstacle false alarm method, and outputting spatial position information of obstacles around the vehicle.

[0109] In order to better reflect the technical effects of the present invention, this embodiment also provides some specific applications.

[0110] The AWR-1642 radar chip observes stationary targets and moving targets in the scene, and after the radar line spectrum signal received by the radar chip, the radar line spectrum signal is processed by the obstacle false alarm method provided in the above embodiment.

[0111] Setting of constant term: number of fast time domain samples N x =128, the number of slow time domain samples is M y =64, so the total number of line spectrum signal samples is N = 8192, and the average false alarm rate is set The number of training units N r =72, enter the maximum allowed number of iterations K max =32, the maximum number of consecutive ineffective targets allowed f max=3, oversampling factor γ = 4, single target correction times R s =1, the number of cycle corrections R c =3.

[0112] like Figure 3 As shown in the figure, the dotted line represents the true radial distance and azimuth value of the obstacle, while the blue solid line represents the result estimated by the obstacle false alarm method and the corresponding target intensity. In addition, the "×" sign represents the threshold value corresponding to each estimation result.

[0113] More specifically, for the AWR-1642 radar chip, combined with the matching mmwave studio software, line spectrum data can be acquired. The AWR-1642 chip has four receiving antennas, each of which can obtain a two-dimensional line spectrum signal. After stacking the signals obtained by each antenna, a three-dimensional line spectrum can be obtained, including the radial distance r, radial velocity v and azimuth angle θ information of each target within the measurable range.

[0114] Estimate the angular frequency vector ω from the line spectrum k =[ω x,k ,ω y,k ,ω z,k ] T The corresponding target k ,v k ,θ k ] T The relationship is

[0115]

[0116] Where c is the propagation speed of electromagnetic waves, T s is the fast time sampling period, T r is a chirp period, f c is the carrier frequency and d is the distance between antennas.

[0117] For each antenna, the two-dimensional line spectrum is analyzed by the obstacle false alarm method provided in this example, using a short time (about 0.1s) to finally obtain the number of iterations. Radial distance Radial velocity and target amplitude Accurate estimate.

[0118] Note that for adjacent antennas, the same target complex amplitude phase difference is Therefore, after matching the estimated values ​​of the same target at different antennas, their complex amplitude values ​​are written as vectors in sequence:

[0119]

[0120] Then apply the result to FFT to get the corresponding Then, according to the process provided in the above embodiment, the corresponding and The corresponding number of targets, as well as the distance, speed and azimuth of each target can be obtained.

[0121] This will enable more accurate obstacle false alarms, facilitate subsequent services such as tracking, cross-domain fusion and other functions, and effectively assist drivers in safe driving or achieve autonomous driving.

Claims

1. An obstacle super-resolution perception method based on a constant false alarm detector, characterized in that: The following steps are involved: Step 1: obtaining a multidimensional line spectrum signal in the radar signal, and obtaining a frequency spectrum signal of the multidimensional line spectrum signal through an FFT algorithm; Step 2: Calculate the square of the amplitude modulus value in each unit in the spectrum signal, select the unit with the largest square of the amplitude modulus value as the first detection unit, and input it into the pre-built constant false alarm detector for detection to determine whether there is a valid target in the first detection unit: When the first detection unit is a valid target, the corresponding target complex amplitude estimation value and angular frequency estimation value are calculated based on the coordinates of the unit where the valid target is located, the calculated angular frequency estimation value is corrected using the Newton orthogonal matching tracking method, and the target complex amplitude estimation value and the corrected angular frequency estimation value are stored in the valid target set; When the first detection unit is an invalid target, the number of invalid targets is counted and added by 1, and the corresponding target complex amplitude estimation value and angular frequency estimation value are calculated based on the coordinates of the unit where the invalid target is located, and the calculated angular frequency estimation value is corrected using the Newton orthogonal matching tracking method, and the target complex amplitude estimation value and the corrected angular frequency estimation value are stored in the invalid target set; Step 3: remove the first detection unit from the spectrum signal, select the unit with the largest square of amplitude modulus value in the spectrum signal after removal as the second detection unit, and input it into the constant false alarm detector to determine whether the second detection unit is a valid target, and at the same time review the first detection unit determined as an invalid target: When the second detection unit is a valid target, the corresponding target complex amplitude estimation value and the corrected angular frequency estimation value are calculated and stored in the valid target set, and the target complex amplitude estimation value and the corrected angular frequency estimation value corresponding to the first detection unit in the non-valid target set are stored in the valid target set, and the number of non-valid targets is cleared; When the second detection unit is an invalid target, the number of invalid targets is counted plus 1, and the corresponding target complex amplitude estimation value and the corrected angular frequency estimation value are calculated and stored in the invalid target set; Step 4, repeating steps 2 to 3 until the judgment termination condition of the constant false alarm detector is met to output a final valid target set; Step 5: Based on the final valid target set obtained in step 4, determine the spatial position information of the obstacle in the feedback direction of the radar signal.

2. The obstacle super-resolution perception method based on constant false alarm detector according to claim 1 is characterized in that: In step 1, the radar signal includes vehicle radar signal, navigation radar and flight radar.

3. The obstacle super-resolution perception method based on constant false alarm detector according to claim 1 is characterized in that: In step 2, the constant false alarm detector needs to be initialized before use, and the parameters include false alarm probability, maximum number of iterations, maximum number of consecutive ineffective targets, oversampling factor, number of corrections, number of protection units in the detection process, and number of training units.

4. The obstacle super-resolution perception method based on a constant false alarm detector according to claim 1 or 3, characterized in that: The determination of the constant false alarm detector includes the following process: The unit with the largest square amplitude modulus value is taken as the detection unit, and a corresponding data set is constructed with the preset number of protection units and the number of training units; The interference level estimation value of the data set is calculated, and its specific expression is as follows: Where N r represents the number of training units, Representing a collection The unit number in , i and j represent the coordinates of the detection unit, is the spectrum of the input signal; The threshold of the effective target is determined by the interference level estimation value, which is specifically expressed as follows: In the formula, α represents the multiplication factor; The square of the amplitude modulus value of the detection unit is compared with the threshold value to output a judgment result.

5. The obstacle super-resolution perception method based on constant false alarm detector according to claim 1 is characterized in that: In step 2, the target complex amplitude estimate and angular frequency estimate are calculated as follows: Where i and j represent the coordinates of the detection unit, represents the target complex amplitude estimate, n represents the number of samples in the fast time domain, M represents the number of samples in the slow time domain, γ x represents the oversampling factor in the fast time domain, γ y represents the oversampling factor in the slow time domain, and Represents the angular frequency estimate.

6. The obstacle super-resolution perception method based on constant false alarm detector according to claim 1, characterized in that: In step 2, the Newton orthogonal matching tracking method is used to correct the calculated angular frequency, and its expression is as follows: Where Y r (k-1) represents the residual of the line spectrum signal after removing the influence of (k-1) estimated targets. represents the estimated value of the kth target frequency, Indicates that S is at the iteration point The gradient at Indicates that S is at the iteration point The Hessian matrix at , (·) T represents the transpose of a matrix or vector, (·) -1 represents the inverse of the matrix, ‖·‖ F represents the F-norm of the matrix, represents the target complex amplitude estimate of the kth target, Represents a multidimensional line spectrum atom.

7. The obstacle super-resolution perception method based on a constant false alarm detector according to claim 1 or 6, characterized in that: Symmetry processing is used to simplify the multidimensional line spectrum signal in the Newton orthogonal matching tracking method, and its expression is as follows: In the formula, represents a multidimensional line spectrum atom, a N (·) represents the atomic vector in the fast time domain, represents the atomic vector transpose in the slow time domain, represents the estimated value of the kth target angular frequency, N represents the number of samples in the fast time domain, and M represents the number of samples in the slow time domain.

8. The obstacle super-resolution perception method based on constant false alarm detector according to claim 1, characterized in that: In step 4, the determination termination condition includes: a. The number of repetitions has reached the maximum number of iterations preset by the constant false alarm detector; b. The number of marked ineffective targets has reached the maximum number of consecutive ineffective targets preset by the constant false alarm detector.

9. The obstacle super-resolution perception method based on constant false alarm detector according to claim 1, characterized in that: Step 4 also includes using the least square method to correct all target complex amplitude estimates in the final valid target set.

10. An obstacle false alarm device, comprising a computer memory, a computer processor, and a computer program stored in the computer memory and executable on the computer processor, characterized in that: The computer program adopts the obstacle super-resolution perception method based on a constant false alarm detector as described in any one of claims 1 to 9; The computer processor executes the computer program to implement the following steps: acquiring a radar signal received by a vehicle-mounted radar, analyzing and processing the radar signal using the obstacle super-resolution perception method, and outputting spatial position information of obstacles around the vehicle.

Images