4D radar angle correction method, electronic equipment and storage medium

Abstract

The embodiment of the invention relates to the technical field of radars, and discloses a 4D radar angle correction method, electronic equipment and a storage medium. The neural network is trained by taking a radar measurement angle as a training sample and a turntable rotation angle corresponding to the radar measurement angle as a training label; wherein the radar and the rotary table rotate synchronously, and the neural network corrects an azimuth angle and a pitch angle in a radar measurement angle in a coupling mode to obtain a radar correction angle; the target radar measurement angle is corrected based on the neural network obtained after training is finished, and a target radar correction angle corresponding to the target radar measurement angle is obtained, so that the azimuth angle and the pitch angle of the radar are corrected at the same time, and the azimuth angle precision and the pitch angle precision of the radar are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of radar technology, and in particular to a 4D radar angle correction method, electronic device, and storage medium. Background Art

[0002] With the wide application of various radars in fields such as autonomous driving, the performance requirements for various radars are becoming higher and higher. The requirements of various manufacturers for radars have also been upgraded from the original 2D (2-Diemensional Radar) radars that can measure distance and speed to 4D (4-Diemensional Radar) radars that can measure distance, speed, azimuth, and pitch angles. The point cloud density of 4D radars is much higher than that of 2D radars, so as the point cloud density increases, the accuracy requirements for 4D radars are also getting higher and higher.

[0003] When 4D radar is used in the automotive field, if the radar's angle measurement error is large, the measured target vehicle may deviate from the lane, and the measured road sign height may be much higher than the actual road sign height, resulting in a vehicle accident. In the field of drones, the angle deviation of the radar may cause the drone to crash.

[0004] The angle correction of traditional radars mainly focuses on correcting the signal before solving the angle and compensating for the inconsistency between signals. However, since the inconsistency of signals is different at different angles, it is impossible to perform different compensations on signals at different angles. And considering that the azimuth and elevation angles of the 4D radar are coupled with each other, only compensating the signal in the azimuth direction or only compensating the signal in the elevation direction cannot effectively improve the azimuth and elevation angle accuracy at the same time. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a 4D radar angle correction method, electronic device and storage medium, which can simultaneously improve the radar azimuth angle accuracy and pitch angle accuracy.

[0006] In order to solve the above technical problems, the embodiment of the present application provides a 4D radar angle correction method, including:

[0007] The radar measurement angle is used as a training sample, and the turntable rotation angle corresponding to the radar measurement angle is used as a training label to train the neural network;

[0008] The radar rotates synchronously with the turntable, and the neural network corrects the azimuth and elevation angles in the radar measurement angle in a coupled manner to obtain a radar correction angle;

[0009] The target radar measurement angle is corrected based on the neural network obtained after the training is completed to obtain a target radar correction angle corresponding to the target radar measurement angle.

[0010] Another aspect of the present application provides a 4D radar angle correction device, including:

[0011] A training module, used to train a neural network using radar measurement angles as training samples and turntable rotation angles corresponding to the radar measurement angles as training labels;

[0012] The radar rotates synchronously with the turntable, and the neural network corrects the azimuth and elevation angles in the radar measurement angle in a coupled manner to obtain a radar correction angle;

[0013] The correction module is used to correct the target radar measurement angle based on the neural network obtained after the training is completed, and obtain the target radar correction angle corresponding to the target radar measurement angle.

[0014] Another aspect of the present application provides an electronic device, including:

[0015] at least one processor; and,

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the 4D radar angle correction method as described above.

[0018] On the other hand, an embodiment of the present application provides a computer-readable storage medium storing a computer program, which implements the 4D radar angle correction method as described above when executed by a processor.

[0019] Compared with the related art, the embodiment of the present application trains the neural network by using the radar measurement angle as the training sample and the turntable rotation angle corresponding to the radar measurement angle as the training label; wherein the radar and the turntable rotate synchronously, and the neural network corrects the azimuth and elevation angles in the radar measurement angle in a coupled manner to obtain the radar correction angle; based on the neural network obtained after the training, the target radar measurement angle is corrected to obtain the target radar correction angle corresponding to the target radar measurement angle, thereby realizing the simultaneous correction of the radar azimuth angle and the pitch angle, and improving the radar azimuth angle accuracy and the pitch angle accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0021] Figure 1 It is a schematic diagram of the angle accuracy of each orientation angle when the pitch angle is 0° provided by the present application;

[0022] Figure 2 It is a schematic diagram of the angle accuracy of each pitch angle when the azimuth angle is 0° according to the application;

[0023] Figure 3 is a schematic diagram of the angular accuracy of the radar in the azimuth plane according to an embodiment of the present application;

[0024] Figure 4 is a statistical histogram of the angular accuracy of the radar azimuth plane provided according to an embodiment of the present application;

[0025] Figure 5 is a schematic diagram of the angular accuracy of the radar in the pitch plane according to an embodiment of the present application;

[0026] Figure 6 is a statistical histogram of the angle accuracy of the radar elevation plane provided according to an embodiment of the present application;

[0027] Figure 7 is a flow chart of a 4D radar angle correction method provided according to an embodiment of the present application;

[0028] Figure 8 is a schematic diagram of a neural network structure provided according to an embodiment of the present application;

[0029] Fig. 9 It is a schematic diagram of the angle accuracy of each orientation angle when the pitch angle is 0° after the neural network correction provided by the present application;

[0030] Fig.10 It is a schematic diagram of the angle accuracy of each pitch angle when the azimuth angle is 0° after neural network correction provided by the present application;

[0031] Fig.11 is a schematic diagram of the angular accuracy of the radar in the azimuth plane after correction by a neural network according to an embodiment of the present application;

[0032] Fig.12 is a statistical histogram of the angle accuracy of the radar azimuth plane after neural network correction provided by an embodiment of the present application;

[0033] Fig.13 is a schematic diagram of the angle accuracy of the radar in the pitch plane after correction by a neural network according to an embodiment of the present application;

[0034] Fig.14 is a statistical histogram of the angle accuracy of the radar pitch plane after neural network correction provided by an embodiment of the present application;

[0035] Fig.15 is a schematic diagram of a 4D radar angle correction device provided according to an embodiment of the present application;

[0036] Fig.16 It is a schematic diagram of the structure of an electronic device provided according to the present application. DETAILED DESCRIPTION

[0037] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, each embodiment of the present application will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in each embodiment of the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is for the convenience of description, and the specific implementation of the present application should not be construed as any limitation, and the various embodiments can be combined and referenced with each other under the premise of no contradiction.

[0038] 4D radar, also known as imaging radar, adds height analysis of the target on the basis of the original distance, speed and direction data, integrating the fourth dimension into the traditional millimeter-wave radar to better understand and map the environment, making the measured data more accurate.

[0039] Traditional 4D radar uses compensation signals to correct azimuth and elevation angles. When the elevation angle (Ele) is 0°, the angular accuracy of each azimuth angle (Azi) can be as follows: Figure 1 As shown; when the azimuth angle (Azi) is 0°, the angular accuracy of each elevation angle (Ele) can be as follows Figure 2 shown.

[0040] For the convenience of explanation, in this embodiment, the symbol "θ" represents the azimuth angle and "φ" represents the elevation angle. Various angle symbols are added to the symbols to represent the relevant statistics of the two angles. For example, the angle accuracy is expressed as It is defined as the difference between the turntable rotation angle [θ, φ] and the radar measurement angle [θ', φ'], as shown in formula (1):

[0041]

[0042] When the radar is fixed on a turntable and all azimuth angles are collected at each pitch angle, the angular accuracy of the radar in the azimuth plane of the entire plane can be measured, such as Figure 3 As shown. Figure 3 By performing histogram statistics on the angle accuracy in the radar azimuth plane, we can obtain the angle accuracy statistical histogram of the radar azimuth plane as follows: Figure 4 shown.

[0043] When the radar is fixed on a turntable and each elevation angle is collected at each azimuth angle, the angular accuracy of the radar's elevation plane on the entire plane can be measured, such as Figure 5 As shown. Figure 5 By performing histogram statistics on the angle accuracy in the radar, we can get the angle accuracy statistical histogram of the radar pitch plane as shown in Figure 6 shown.

[0044] The above radar angle accuracies are all measured using traditional correction methods. In an embodiment of the present application, in order to further solve the nonlinear problem in the traditional angle correction method, a method using a neural network will be provided to simultaneously re-correct the azimuth angle and pitch angle measured by the radar, so as to simultaneously improve the radar azimuth angle accuracy and pitch angle accuracy.

[0045] An embodiment of the present application relates to a 4D radar angle correction method, such as Figure 7 As shown, the method comprises the following steps.

[0046] Step 101: Using the radar measurement angle as a training sample and the turntable rotation angle corresponding to the radar measurement angle as a training label, the neural network is trained;

[0047] The radar rotates synchronously with the turntable, and the neural network corrects the azimuth and elevation angles in the radar measurement angle in a coupled manner to obtain the radar correction angle.

[0048] The present embodiment does not limit the network structure of the neural network. For example, the present embodiment may adopt Figure 8 The neural network structure shown in Figure 1 contains an input layer, a hidden layer, and an output layer.

[0049] Among them, the training samples used to train the neural network are the sampled radar measurement angles. In theory, the radar measurement angle and the radar's own rotation angle should be consistent, and the radar rotates synchronously with the turntable, so in theory the radar measurement angle and the turntable rotation angle should also be consistent.

[0050] Figure 8 In the input layer of the neural network, the input is the radar measurement angle [θ', φ'], and the dimension is [k*2], that is, k groups of radar measurement angles are available [θ′ k ,φ′ k ] indicates that the number of hidden layers is unlimited, and the number of neurons in each hidden layer is unlimited. In this embodiment, the structure mainly provides a hidden layer with two layers. Among them, the number of neurons in the first layer of the hidden layer is n, and the number of neurons in the second layer of the hidden layer is m. The activation function of each hidden layer can be customized. In this embodiment, the sigmoid activation function is used, as shown in the following formula:

[0051]

[0052] like Figure 8 As shown in , the first hidden layer can be expressed as:

[0053]

[0054] The second hidden layer can be expressed as:

[0055]

[0056] The output layer is represented as:

[0057]

[0058] The number of neurons in the first hidden layer is n, n = 1, 2, 3, ..., N; the number of neurons in the second hidden layer is m, m = 1, 2, 3, ..., M; θ′ k and φ′ k are the azimuth angle and elevation angle in the radar measurement angle, with a total of k pairs of values, k = 1, 2, 3, ..., K; ω, ω' and ω" are all weight parameters of the neural network (where the subscript of the weight parameter represents the position of the weight parameter in the matrix), b, b' and b" are all bias parameters of the neural network (where the subscript of the bias parameter represents the position of the bias parameter in the matrix); θ" k and φ″ k They are the azimuth angle and pitch angle of the radar correction angle calculated by the neural network.

[0059] When training the above neural network, the radar measurement angle is used as a training sample, and the turntable rotation angle corresponding to each radar measurement angle is used as a training label. Since the radar and the turntable rotate synchronously, the turntable rotation angle can be considered as the real angle of the radar, that is, the real radar measurement angle. The radar measurement angle is input to the neural network to be trained so that it learns the label value of the turntable rotation angle, thereby completing the training process of the neural network. The neural network obtained after the training is completed can correct the azimuth and pitch angle in the radar measurement angle in a coupled manner, thereby obtaining the corresponding radar correction angle. It should be noted here that the various angles involved in this embodiment all contain two components, namely the azimuth angle and the pitch angle. For example, the turntable rotation angle contains the turntable rotation azimuth angle and the turntable rotation pitch angle, and the radar correction angle contains the radar correction azimuth angle and the radar correction pitch angle.

[0060] In order to allow the azimuth angle and the elevation angle in the radar measurement angle to be corrected at the same time and overcome the nonlinear problem, the present embodiment sets the neural network to correct the azimuth angle and the elevation angle in the radar measurement angle in a coupled manner to obtain the radar correction angle. That is, the data of the azimuth angle and the elevation angle are processed simultaneously in each neuron to solve the nonlinear problem in the correction process.

[0061] Since different radar solutions have differences in hardware, algorithms, etc., the neuron parameters are all different. Considering that the sigmoid activation function contains exponential operations, it is impossible to directly solve the weight parameters and bias parameters of the neural network. Therefore, this embodiment can use the back propagation method to solve the weight parameters and bias parameters of the neural network.

[0062] Specifically, the radar measurement angle is used as a training sample, and the turntable rotation angle corresponding to the radar measurement angle is used as a training label to train the neural network, including: inputting the training sample into the neural network to obtain the radar correction angle θ″ k and φ″ k The corresponding turntable rotation angle θ k and φ k The neural network is iteratively trained by obtaining the error through the following loss function:

[0063]

[0064] In formula (6), ω, b generally refer to all weighted parameters and bias parameters in the neural network to be trained. The training sample is the radar measurement angle [θ′ k ,φ′ k ], the training samples are processed by the neural network and output as the radar correction angle [θ″ k ,φ″ k ], the corresponding turntable rotation angle [θ k ,φ k During the training process, the radar is corrected to an angle [θ″ k ,φ″ k ] and the turntable rotation angle [θ k ,φ k ] is the smallest, that is, the error loss obtained by calculating formula (6) is minimized.

[0065] After the training, the radar correction angle [θ″ k ,φ″ k ], when the pitch angle is 0°, the angle accuracy of each direction is as follows Fig. 9 As shown, when the azimuth angle is 0°, the angular accuracy of each elevation direction is as follows: Fig.10 shown.

[0066] Correspondingly, when the radar is fixed on the turntable, all azimuth angles are collected at each pitch angle, and then the radar measurement angles [θ′ k ,φ′ k ] is input into the trained neural network to obtain the radar correction angle [θ″] in each direction k ,φ″ k ], then the angle accuracy of the radar azimuth plane after neural network correction can be obtained as Fig.11 As shown. Fig.11 By performing histogram statistics on the angle accuracy in the azimuth plane of the radar after neural network correction, the angle accuracy statistical histogram of the radar azimuth plane can be obtained as follows: Fig.12 shown.

[0067] When the radar is fixed on the turntable, each pitch angle is collected at each azimuth angle, and then the radar measurement angle [θ′ k ,φ′ k ] is input into the trained neural network to obtain the radar correction angle [θ″] in each direction k ,φ″ k ], then the angle accuracy of the radar pitch plane after neural network correction can be obtained as Fig.13 As shown. Fig.13 By performing histogram statistics on the angle accuracy in the radar, we can get the angle accuracy statistical histogram of the radar pitch plane after neural network correction as shown in Fig.14 shown.

[0068] By comparison Figure 4 and 12 , and contrast Figure 6 and Fig.14 It can be clearly found that for the number of angles with an angle accuracy between [-0.2°, 0.2°], the proportion of angles after neural network correction is significantly higher than the proportion of angles without neural network correction.

[0069] Step 102: Correct the target radar measurement angle based on the neural network obtained after the training, and obtain a target radar correction angle corresponding to the target radar measurement angle.

[0070] Specifically, after the neural network is obtained after the training is completed, the neural network can be directly used to calculate the target radar measurement angle that needs to be corrected to obtain the target radar correction angle, that is: the target radar measurement angle is input into the neural network obtained after the training is completed to obtain the target radar correction angle corresponding to the target radar measurement angle; the neural network obtained after the training is completed can also be used as a priori to correct the target radar measurement angle, so as to obtain a target radar correction angle with higher accuracy.

[0071] This embodiment does not limit the specific correction and application process of using the neural network obtained after the training as a priori.

[0072] In some application scenarios, considering that the actual direct use of neural networks for correction calculations involves exponential and division operations, this may be a heavy load on the radar's processor, or even unbearable. Therefore, this embodiment provides a method of using neural networks as a priori to reduce the amount of calculation by constructing a two-dimensional lookup table.

[0073] Correspondingly, in some embodiments, correcting the target radar measurement angle based on the neural network obtained after the training is completed to obtain the target radar correction angle corresponding to the target radar measurement angle may include steps one and two.

[0074] Step 1: Perform two-dimensional sampling on the neural network obtained after the training to form a two-dimensional lookup table including a two-dimensional sampling point table and a two-dimensional correction point table; the sampling points in the two-dimensional sampling point table represent the radar measurement angle sampling points, and the correction points in the two-dimensional correction point table are the radar correction angles calculated and output by the neural network obtained after the radar measurement angle sampling points are trained.

[0075] Specifically, after the training is completed, the neural network is obtained, the weight parameters and bias parameters of the neural network are obtained, and the neural network is sampled in two dimensions, that is, the angles of all directional planes and elevation planes are sampled and combined to obtain the angles of all directions of the full plane (each direction includes an azimuth angle and an elevation angle), which are used as radar measurement angle sampling points to form a two-dimensional sampling point table. The sampling points in the two-dimensional sampling point table can be arranged in a two-dimensional plane. The radar measurement angle corresponding to each sampling point in the two-dimensional sampling point table is input into the above-mentioned neural network, the corresponding radar correction angle is output, and a two-dimensional correction point table is constructed with each radar correction angle as a correction point. The correction points in the two-dimensional correction point table can also be arranged in a two-dimensional plane.

[0076] For example: The sampling points of the neural network for the azimuth angle can be constructed as η m , m = 1, 2, 3, ..., M; the neural network sampling point for constructing the pitch angle is ε n , n = 1, 2, 3, ..., N; the neural network sampling points of each azimuth angle are arranged and combined with the neural network sampling points of each pitch angle to generate the two-dimensional sampling points [Γ k , Υ k ], where Γ k =Γ m,n ,k=1,2,3,...,K;Γ k The neural network sampling points corresponding to the azimuth angle; k =Y m,n,k=1,2,3,...,K; Y k The neural network sampling points corresponding to the pitch angle; the two-dimensional sampling points [Γ k , Υ k ] is input into the neural network obtained after training to obtain the corresponding two-dimensional correction point [A k , E k ]:

[0077] (A k ,E k ) = f NN (Γ k ,Υ k )

[0078] A k =A m,n

[0079] E k =E m,n ,

[0080] Among them, f NN () represents the neural network algorithm, A k The azimuth angle value of the neural network output after training the two-dimensional sampling points at different azimuths and elevation angles, E k It represents the pitch angle value output by the neural network after training two-dimensional sampling points at different azimuths and pitch angles.

[0081] Among them, after permuting and combining the neural network sampling points of each azimuth angle and the neural network sampling points of each elevation angle, k combinations (k=m*n) are obtained, that is, k sampling points, and each k corresponds to a group of (m, n). Both k and (m, n) can be understood as the position of the sampling point or the correction point in the corresponding table. The sampling points and correction points corresponding to the same position have a corresponding search relationship.

[0082] A k and E k The corresponding two-dimensional lookup table can be represented by two tables AT and ET respectively. Table AT is used to store Γ k With A k The corresponding search relationship, that is, AT = {Γ k |A k}={Γ m,n |A m,n}; Table ET is used to store Υ k With E k The corresponding search relationship, that is, ET = {Υ k |E k}={Υ m,n |E m,n}.

[0083] Step 2: According to the position of the target radar measurement angle in the two-dimensional sampling point table, the target radar correction angle corresponding to the target radar measurement angle is calculated based on the correction point corresponding to the position in the two-dimensional correction point table.

[0084] Specifically, based on the corresponding same position in the two-dimensional sampling point table and the two-dimensional correction point table, that is, there is a corresponding search relationship between the sampling point and the correction point of the same k or (m, n). Therefore, the position of the target radar measurement angle to be corrected in the two-dimensional sampling point table can be locked first, and then the correction point at the position (or the nearby position) can be extracted from the two-dimensional correction point table. The target radar correction angle corresponding to the target radar measurement angle can be calculated based on these correction points.

[0085] In some embodiments, the target radar correction angle corresponding to the target radar measurement angle may be calculated according to the following formula:

[0086]

[0087]

[0088] in, They are the azimuth angle and elevation angle in the target radar correction angle respectively; α' and β' are the azimuth angle and elevation angle in the target radar measurement angle respectively.

[0089] For example, assuming that the target radar measurement angle to be corrected is [α', β'], the radar real angle (turntable rotation angle) is [α, β], the target radar correction angle after correction by the two-dimensional lookup table is

[0090] The specific correction process is: perform a two-dimensional search on the target radar measurement angle [α', β'] through the lookup tables AT and ET to find a set of Γ m ≤α'≤Γ m+1 , and satisfies Υ n ≤β'≤Υ n+1 , determine the angles of the four sampling points: Γ m,n , Γ m+1,n , Υ m,n , Υ m,n+1 , and by looking up the AT and ET tables, we can get the corresponding angles of the four correction points: A m,n , A m+1,n 、E m,n 、E m,n+1 Finally, the output angle after correction by the lookup table is obtained through formulas (7) and (8):

[0091] After the radar calibration based on the neural network obtained after the training, the target radar angle measurement error is defined as the difference (Ε) between the target radar correction angle obtained after the neural network correction and the true angle (turntable rotation angle). A , E E )(angle accuracy):

[0092]

[0093]

[0094] Among them, A Indicates the angular accuracy of the direction angle, E E Indicates the angular accuracy of the pitch angle.

[0095] pass Figure 11 to Figure 14 It can be seen that the neural network obtained after training in this embodiment has greatly improved the accuracy of the azimuth and pitch angles in the radar measurement angles. 80% of the measurement error azimuth angle falls between [-0.2°, +0.2°], and 90% of the measurement error pitch angle falls between [-0.5°, +0.5°], which is a significant improvement over the traditional calibration method.

[0096] Compared with the prior art, this embodiment trains the neural network by using the radar measurement angle as a training sample and the turntable rotation angle corresponding to the radar measurement angle as a training label; wherein the radar and the turntable rotate synchronously, and the neural network corrects the azimuth and elevation angles in the radar measurement angle in a coupled manner to obtain the radar correction angle; based on the neural network obtained after the training, the target radar measurement angle is corrected to obtain the target radar correction angle corresponding to the target radar measurement angle, thereby realizing the simultaneous correction of the radar azimuth angle and the pitch angle, and improving the radar azimuth angle accuracy and the pitch angle accuracy.

[0097] Another embodiment of the present invention relates to a 4D radar angle correction device, which is used to perform the 4D radar angle correction method as described above, such as Fig.15 As shown, the device comprises:

[0098] A training module 11 is used to train a neural network using radar measurement angles as training samples and turntable rotation angles corresponding to the radar measurement angles as training labels;

[0099] The radar rotates synchronously with the turntable, and the neural network corrects the azimuth and elevation angles in the radar measurement angle in a coupled manner to obtain a radar correction angle;

[0100] The correction module 12 is used to correct the target radar measurement angle based on the neural network obtained after the training is completed, and obtain the target radar correction angle corresponding to the target radar measurement angle.

[0101] In some embodiments, the neural network comprises: an input layer, a hidden layer and an output layer; the hidden layer comprises two layers;

[0102] The activation function of each hidden layer is expressed as:

[0103]

[0104] The first layer of the hidden layer is expressed as:

[0105]

[0106] The second layer of the hidden layer is expressed as:

[0107]

[0108] The output layer is expressed as:

[0109]

[0110] The number of neurons in the first layer of the hidden layer is n, where n=1, 2, 3, ..., N; the number of neurons in the second layer of the hidden layer is m, where m=1, 2, 3, ..., M; θ′ k and φ′ k are the azimuth angle and the elevation angle in the radar measurement angle, with a total of k pairs of values, k = 1, 2, 3, ..., K; ω, ω' and ω" are all weight parameters of the neural network, b, b' and b" are all bias parameters of the neural network; θ k ” and φ k ” are respectively the azimuth angle and the pitch angle of the radar correction angle calculated by the neural network.

[0111] In some embodiments, the training module 11 is specifically configured to input the training sample into the neural network to obtain the radar correction angle θ″ k and φ″ k The corresponding turntable rotation angle θ k and φ k The neural network is iteratively trained by obtaining the error through the following loss function:

[0112]

[0113] In some embodiments, the correction module 12 is specifically used to perform two-dimensional sampling on the neural network obtained after the training is completed to form a two-dimensional lookup table including a two-dimensional sampling point table and a two-dimensional correction point table; the sampling points in the two-dimensional sampling point table represent radar measurement angle sampling points, and the correction points in the two-dimensional correction point table are radar correction angles calculated and output by the neural network obtained after the radar measurement angle sampling points are trained;

[0114] According to the position of the target radar measurement angle in the two-dimensional sampling point table, the target radar correction angle corresponding to the target radar measurement angle is calculated based on the correction point corresponding to the position in the two-dimensional correction point table.

[0115] In some embodiments, the correction module 12 performs two-dimensional sampling on the neural network obtained after training to form a two-dimensional lookup table including a two-dimensional sampling point table and a two-dimensional correction point table, including:

[0116] The sampling points of the neural network for constructing the azimuth angle are η m , m = 1, 2, 3, ..., M; the neural network sampling point for constructing the pitch angle is ε n , n = 1, 2, 3, ..., N; the neural network sampling points of each azimuth angle are arranged and combined with the neural network sampling points of each pitch angle to generate the two-dimensional sampling points [Γ k , Υ k ], where Γ k =Γ m,n ,k=1,2,3,...,K;Γ k The neural network sampling points corresponding to the azimuth angle; k =Y m,n ,k=1,2,3,...,K; Y k Neural network sampling points corresponding to pitch angles;

[0117] The two-dimensional sampling point [Γ k , Υ k ] is input into the neural network obtained after the training to obtain the corresponding two-dimensional correction point [A k , E k ]:

[0118] (A k ,E k ) = f NN (Γ k ,Υ k )

[0119] A k =A m,n

[0120] E k=E m,n ,

[0121] Among them, f NN () represents the neural network algorithm, A k The azimuth angle value of the neural network output after training the two-dimensional sampling points at different azimuths and elevation angles, E k It represents the pitch angle value output by the neural network after training two-dimensional sampling points at different azimuths and pitch angles.

[0122] In some embodiments, the correction module 12 calculates the target radar correction angle corresponding to the target radar measurement angle based on the correction point corresponding to the position in the two-dimensional correction point table according to the position of the target radar measurement angle in the two-dimensional sampling point table, including:

[0123] The target radar correction angle corresponding to the target radar measurement angle is calculated according to the following formula:

[0124]

[0125]

[0126] in, They are respectively the azimuth angle and the elevation angle in the target radar correction angle; α' and β' are respectively the azimuth angle and the elevation angle in the target radar measurement angle.

[0127] In some embodiments, the correction module 12 is further specifically configured to input the target radar measurement angle into the neural network obtained after training, to obtain a target radar correction angle corresponding to the target radar measurement angle.

[0128] Compared with the prior art, this embodiment uses a training module to train a neural network using radar measurement angles as training samples and a turntable rotation angle corresponding to the radar measurement angles as training labels; wherein the radar and the turntable rotate synchronously, and the neural network corrects the azimuth and elevation angles in the radar measurement angles in a coupled manner to obtain a radar correction angle; and the correction module corrects the target radar measurement angle based on the neural network obtained after the training to obtain a target radar correction angle corresponding to the target radar measurement angle, thereby achieving simultaneous correction of the radar azimuth angle and the pitch angle, and improving the radar azimuth angle accuracy and the pitch angle accuracy.

[0129] Another embodiment of the present invention relates to an electronic device, such as Fig.16As shown, it includes at least one processor 202; and a memory 201 that is communicatively connected to the at least one processor 202; wherein the memory 201 stores instructions that can be executed by the at least one processor 202, and the instructions are executed by the at least one processor 202 so that the at least one processor 202 can execute any of the above method embodiments.

[0130] Among them, the memory 201 and the processor 202 are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects one or more processors 202 and various circuits of the memory 201 together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, and therefore, are not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor 202 is transmitted on a wireless medium through an antenna, and further, the antenna also receives data and transmits the data to the processor 202.

[0131] The processor 202 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management and other control functions. The memory 201 can be used to store data used by the processor 202 when performing operations.

[0132] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program, which implements any of the above method embodiments when executed by a processor.

[0133] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0134] Those skilled in the art will appreciate that the above-mentioned embodiments are specific examples for implementing the present invention, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A 4D radar angle correction method, characterized in that: include: The radar measurement angle is used as a training sample, and the turntable rotation angle corresponding to the radar measurement angle is used as a training label to train the neural network; The radar rotates synchronously with the turntable, and the neural network corrects the azimuth and elevation angles in the radar measurement angle in a coupled manner to obtain a radar correction angle; The target radar measurement angle is corrected based on the neural network obtained after the training is completed to obtain a target radar correction angle corresponding to the target radar measurement angle.

2. The method according to claim 1, characterized in that The neural network comprises: an input layer, a hidden layer and an output layer; the hidden layer comprises two layers; The activation function of each hidden layer is expressed as: The first layer of the hidden layer is expressed as: The second layer of the hidden layer is expressed as: The output layer is expressed as: The number of neurons in the first layer of the hidden layer is n, where n=1, 2, 3, ..., N; the number of neurons in the second layer of the hidden layer is m, where m=1, 2, 3, ..., M; θ′ k and φ′ k are the azimuth angle and the elevation angle in the radar measurement angle, with a total of k pairs of values, k = 1, 2, 3, ..., K; ω, ω' and ω" are all weight parameters of the neural network, b, b' and b" are all bias parameters of the neural network; θ" k and φ″ k They are respectively the azimuth angle and the pitch angle in the radar correction angle calculated by the neural network.

3. The method according to claim 2, characterized in that The method of using the radar measurement angle as a training sample and the turntable rotation angle corresponding to the radar measurement angle as a training label to train the neural network includes: The radar correction angle θ″ obtained after the training sample is input into the neural network k and φ″ k The corresponding turntable rotation angle θ k and φ k The neural network is iteratively trained by obtaining the error through the following loss function:

4. The method according to claim 1, characterized in that: The correcting the target radar measurement angle based on the neural network obtained after the training is completed to obtain the target radar correction angle corresponding to the target radar measurement angle includes: Performing two-dimensional sampling on the neural network obtained after the training, forming a two-dimensional lookup table including a two-dimensional sampling point table and a two-dimensional correction point table; the sampling points in the two-dimensional sampling point table represent radar measurement angle sampling points, and the correction points in the two-dimensional correction point table are radar correction angles calculated and output by the neural network obtained after the radar measurement angle sampling points are trained; According to the position of the target radar measurement angle in the two-dimensional sampling point table, the target radar correction angle corresponding to the target radar measurement angle is calculated based on the correction point corresponding to the position in the two-dimensional correction point table.

5. The method according to claim 4, characterized in that The two-dimensional sampling of the neural network obtained after the training is performed to form a two-dimensional lookup table including a two-dimensional sampling point table and a two-dimensional correction point table, including: The sampling points of the neural network for constructing the azimuth angle are η m , m = 1, 2, 3, ..., M; the neural network sampling point for constructing the pitch angle is ε n , n = 1, 2, 3, ..., N; the neural network sampling points of each azimuth angle are arranged and combined with the neural network sampling points of each pitch angle to generate the two-dimensional sampling points [Γ k , γ k ], where Γ k =Γ m,n ,k=1,2,3,...,K;Γ k The neural network sampling points corresponding to the azimuth angle; γ k =γ m,n ,k=1,2,3,...,K; γ k Neural network sampling points corresponding to pitch angles; The two-dimensional sampling point [Γ k , γ k ] is input into the neural network obtained after the training to obtain the corresponding two-dimensional correction point [A k , E k ]: (A k ,E k )=f NN (C k ,c k ) A k =A m,n AND k =And m,n , Among them, f NN () represents the neural network algorithm, A k The azimuth angle value of the neural network output after training the two-dimensional sampling points at different azimuths and elevation angles, E k It represents the pitch angle value output by the neural network after training two-dimensional sampling points at different azimuths and pitch angles.

6. The method according to claim 5, characterized in that The step of calculating the target radar correction angle corresponding to the target radar measurement angle based on the position of the target radar measurement angle in the two-dimensional sampling point table and the correction point corresponding to the position in the two-dimensional correction point table includes: The target radar correction angle corresponding to the target radar measurement angle is calculated according to the following formula: in, They are respectively the azimuth angle and the elevation angle in the target radar correction angle; α' and β' are respectively the azimuth angle and the elevation angle in the target radar measurement angle.

7. The method according to claim 1, characterized in that The correcting the target radar measurement angle based on the neural network obtained after the training is completed to obtain the target radar correction angle corresponding to the target radar measurement angle includes: The target radar measurement angle is input into the neural network obtained after the training is completed to obtain the target radar correction angle corresponding to the target radar measurement angle.

8. A 4D radar angle correction device, characterized in that: include: A training module, used to train a neural network using radar measurement angles as training samples and turntable rotation angles corresponding to the radar measurement angles as training labels; The radar rotates synchronously with the turntable, and the neural network corrects the azimuth and elevation angles in the radar measurement angle in a coupled manner to obtain a radar correction angle; The correction module is used to correct the target radar measurement angle based on the neural network obtained after the training is completed, and obtain the target radar correction angle corresponding to the target radar measurement angle.

9. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the 4D radar angle correction method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the 4D radar angle correction method according to any one of claims 1 to 7 is implemented.

Images