A millimeter wave radar azimuth super-resolution method based on multi-frame reconstruction

Abstract

The present invention discloses a millimeter wave radar azimuth super-resolution method based on multi-frame reconstruction. First, a millimeter wave radar azimuth multi-frame super-resolution imaging model is established, and then displacement estimation is performed to obtain an initial high-resolution estimation frame. Then, a convex set constraint is constructed, and the azimuth residual between multiple frames of echo data is used to perform azimuth projection on the initial high-resolution estimation frame, and finally azimuth super-resolution imaging is realized. The method of the present invention solves the problem of low azimuth resolution of the traditional millimeter wave real beam imaging method, and can make full use of the complementary information between multiple frames of echo data to obtain a higher azimuth resolution than the existing super-resolution method, thereby improving the azimuth imaging resolution of the millimeter wave radar.

Description

Technical Field

[0001] The invention belongs to the technical field of radar imaging, and in particular relates to a millimeter wave radar azimuth super-resolution method based on multi-frame reconstruction. Background Art

[0002] Millimeter-wave radar has the advantages of miniaturization, low cost, and strong anti-interference ability. It can work all day and all night and has been widely used in traffic monitoring and assisted driving in recent years. However, due to the limitation of antenna aperture size, the azimuth resolution of the imaging result is low. Therefore, it is of great significance to study the method of improving the azimuth resolution of millimeter-wave radar based on multi-frame reconstruction.

[0003] The paper "Sadjadi, F. Radar beam sharpening using an optimum FIR filter Circuits Systems and Signal Processing, 2000, 19 (2): 121–129" uses Wiener inverse filtering to improve the radar azimuth resolution. However, in practical applications, the super-resolution results of this method are over-smoothed and the resolution improvement is limited. The paper "Q. Zhang, Y. Zhang, Y. Huang, Y. Zhang, W. Li and J. Yang," Total variation superresolution method for radar forward-looking imaging," 2019 6th Asia-Pacific Conference on Synthetic Aperture Radar (APSAR), 2019, pp. 1-4" proposes a TV regularized radar forward-looking azimuth super-resolution method. This method achieves azimuth super-resolution by introducing TV norm constraints under the regularization framework. However, this method is sensitive to noise and has poor resolution performance under low signal-to-noise ratio conditions. The paper "Y. Zhang, Y. Zhang, W. Li, Y. Huang and J. Yang," Angular superresolution for real beam radar with iterative adaptive approach,"2013 IEEE International Geoscience and RemoteSensing Symposium-IGARSS,2013,pp.3100-3103" proposed an iterative adaptive azimuth super-resolution method, which uses the weighted least squares criterion to solve the super-resolution problem and can achieve robust azimuth resolution improvement under low signal-to-noise ratio conditions. However, the resolution improvement ratio is low and it is difficult to meet the needs of practical applications. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a millimeter wave radar azimuth super-resolution method based on multi-frame reconstruction.

[0005] The technical solution adopted by the present invention is: a millimeter wave radar azimuth super-resolution method based on multi-frame reconstruction, and the specific steps are as follows:

[0006] Step 1: Establish a millimeter-wave radar multi-frame super-resolution imaging model;

[0007] In the millimeter wave radar imaging process, a multi-transmit and multi-receive method is used to quickly obtain multiple frames of low-resolution echo data, and an imaging model between multiple frames of low-resolution echo data is established:

[0008] y k =D k M k F k x+n k k=1,2,...,K (1)

[0009] Where K represents the number of low-resolution echo data frames, y k represents the k-th frame of low-resolution echo data after being processed by the traditional azimuth super-resolution method, D k Represents the shift matrix between multiple frames of low-resolution echo data, M k represents the downsampling matrix, F k represents the fuzzy matrix, x represents the original high-resolution scene, n k represents additive noise.

[0010] The super-resolution process based on multi-frame reconstruction is regarded as the inverse process of the above imaging process, that is, solving the following unconstrained optimization problem:

[0011]

[0012] in, Represents high-resolution estimated echo data.

[0013] In order to obtain a stable solution, the convex set projection method is used.

[0014] Step 2: Define the convex set constraints of the super-resolution model;

[0015] For N convex set constraints in Hilbert space H: C1~C N , the solution is obtained by projection, namely:

[0016] f i+1 =P N P N-1 …P1f i (3)

[0017] Among them, P N represents the projection operator of the Nth convex set constraint, f i represents the solution after the i-th iteration.

[0018] The multi-frame super-resolution model is solved by using the convex set projection method. First, the displacement of multiple frames of low-resolution echo data is estimated to obtain the displacement matrix D. Then, a frame of low-resolution echo data y is selected. k As the initial reference frame, bicubic interpolation is performed on it to obtain the initial high-resolution estimated echo data Then define the convex set constraint C:

[0019]

[0020] Among them, W k (m,n) represents high-resolution estimated echo data and the kth frame low-resolution echo data y k The azimuth residual between m and n represents the low-resolution echo data y k The distance dimension coordinates and azimuth dimension coordinates in , η represents the projection threshold.

[0021] Step 3, calculating the azimuth residual between the echo data;

[0022] High resolution estimated echo data and the kth frame low-resolution echo data y k The azimuth residual W k (m,n) can be specifically expressed as:

[0023]

[0024] Among them, y k (m,n) represents the amplitude of the k-th frame low-resolution echo data at point (m,n); (m′,n′) represents the high-resolution estimated echo data The k-th frame low-resolution echo data y k The point corresponding to the midpoint (m,n) of (m,n), m′, n′ respectively represent the high-resolution estimated echo data The distance dimension coordinates and azimuth dimension coordinates in , the mapping relationship between point (m,n) and point (m′,n′) is F(n′,n″) represents the normalized point spread function.

[0025] Step 4: Azimuth projection to achieve super-resolution imaging;

[0026] High resolution estimation echo data Project as follows:

[0027]

[0028] in, Express The projection process.

[0029] Since the amplitude of the echo data should be greater than or equal to 0, an amplitude constraint is introduced in the projection process:

[0030]

[0031] The projection operator P, that is, Equation (6), is used to estimate the high-resolution echo data. Projection correction is performed and through several iterations, the final solution is constrained to be within the intersection of convex sets, thereby achieving azimuthal super-resolution imaging.

[0032] Beneficial effects of the present invention: The method of the present invention first establishes a millimeter-wave radar azimuth multi-frame super-resolution imaging model, then performs displacement estimation and obtains an initial high-resolution estimation frame, then constructs a convex set constraint, and uses the azimuth residual between multiple frames of echo data to perform azimuth projection on the initial high-resolution estimation frame, and finally realizes azimuth super-resolution imaging. The method of the present invention solves the problem of low azimuth resolution of traditional millimeter-wave real beam imaging methods, can make full use of the complementary information between multiple frames of echo data, obtain higher azimuth resolution than existing super-resolution methods, and improve the azimuth imaging resolution of millimeter-wave radars. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The present invention is a flowchart of a millimeter wave radar azimuth super-resolution method based on multi-frame reconstruction.

[0034] Figure 2 It is a simulation result diagram for measured data in an embodiment of the present invention.

[0035] Figure 3 It is a comparison diagram of the azimuth profile of the midpoint target in the simulation results of the measured data in the embodiment of the present invention. DETAILED DESCRIPTION

[0036] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0037] The present invention adopts measured data simulation to demonstrate the effectiveness of the proposed method, and all the steps and conclusions of the present invention are verified on the Matlab2020b simulation platform.

[0038] like Figure 1 As shown, a flow chart of a millimeter wave radar azimuth super-resolution method based on multi-frame reconstruction of the present invention, the specific steps are as follows:

[0039] Step 1: Establish a millimeter-wave radar multi-frame super-resolution imaging model;

[0040] When using millimeter-wave radar to perform multiple imaging of the same scene, due to the influence of factors such as platform jitter and noise interference, even for the same target, there will be large errors in the imaging results each time, resulting in a large amount of complementary difference information between different frames. Therefore, the rational use of the complementary information between multiple frames of echo data can effectively improve the imaging resolution.

[0041] Taking MIMO millimeter-wave radar as an example, during the imaging process, a multi-transmit and multi-receive method can be used to quickly obtain multi-frame echo data, and an imaging model between multi-frame echo data can be established based on this:

[0042] y k =D k M k F k x+n k k=1,2,...,K (8)

[0043] Wherein, K represents the number of low-resolution echo data frames, which is 8 frames in this embodiment, and y k represents the kth frame of low-resolution echo data acquired by the MIMO radar, D k Represents the shift matrix between multiple frames of low-resolution echo data, M k represents the downsampling matrix. In this embodiment, the downsampling factor is 0.5. k The fuzzy matrix is ​​expressed. This embodiment only verifies the azimuth super-resolution imaging performance, so F k Take the peak approximation of the one-dimensional Sinc function, x represents the original high-resolution scene, n k represents additive noise.

[0044] The super-resolution process based on multi-frame reconstruction can be regarded as the inverse process of the above imaging process, that is, solving the following unconstrained optimization problem:

[0045]

[0046] in, Represents high-resolution estimated echo data.

[0047] Due to the existence of noise, this problem is an ill-posed problem, and it is difficult to obtain a stable result by using a direct solution method. In order to obtain a stable solution, a convex set projection method is used in this embodiment to solve it.

[0048] Step 2: Define the convex set constraints of the super-resolution model;

[0049] The convex set projection method uses set theory to define all prior knowledge of the image as convex set constraints, so it can effectively introduce prior information. It is often used to solve the super-resolution problem of multi-frame images. The principle is that for the N convex set constraints in the Hilbert space H: C1~C N , the solution is obtained by projection, namely:

[0050] f i+1 =P N P N-1 …P1f i (10)

[0051] Among them, P N represents the projection operator of the Nth convex set constraint, f i represents the solution after the i-th iteration.

[0052] The super-resolution model is solved by using the convex set projection method. First, the displacement of multiple frames of low-resolution echo data is estimated to obtain the displacement matrix D. Then, a frame of low-resolution echo data y is selected. k As the initial reference frame, bicubic interpolation is performed on it to obtain the initial high-resolution estimated echo data In this embodiment, the interpolation magnification factor is 2, and then the convex set constraint C is defined:

[0053]

[0054] Among them, W k (m,n) represents high-resolution estimated echo data and the kth frame low-resolution echo data y k The azimuth residual between them reflects the difference between the high-resolution estimated echo data and the low-resolution echo data. m and n represent the low-resolution echo data y and k The distance dimension coordinates and the azimuth dimension coordinates in are represented by , and η represents the projection threshold, which is set to 0.001 in this embodiment.

[0055] Step 3, calculating the azimuth residual between the echo data;

[0056] High resolution estimated echo data and the kth frame low-resolution echo data y k The azimuth residual W k (m,n) is as follows:

[0057]

[0058] Among them, y k (m,n) represents the amplitude of the k-th frame low-resolution echo data at point (m,n); (m′,n′) represents the high-resolution estimated echo data The k-th frame low-resolution echo data y k The point corresponding to the point (m,n) in (m,n), m′, n′ respectively represent the distance dimension coordinate and azimuth dimension coordinate in the high-resolution estimated echo data. The mapping relationship between point (m,n) and point (m′,n′) is: F(n′, n″) represents a normalized point spread function. In this embodiment, only azimuth data is processed, and thus it is approximately represented by the peak data of a one-dimensional sinc function.

[0059] Step 4: Azimuth projection to achieve super-resolution imaging;

[0060] According to the threshold η in formula (11) and the azimuth residual W calculated in formula (12): k (m,n), for high-resolution estimated echo data Project as follows:

[0061]

[0062] in, Express The projection process.

[0063] Since the amplitude of the echo data should be greater than or equal to 0, an amplitude constraint is introduced in the projection process:

[0064]

[0065] Using the projection operator P to estimate the high-resolution echo data Projection correction is performed and through several iterations, the final solution is constrained to be within the intersection of convex sets, thereby achieving azimuthal super-resolution imaging.

[0066] In order to prove the effectiveness of the method of the present invention, the measured data simulation was carried out on the Matlab2020b simulation platform. The measured data was recorded by the self-designed 12-transmit 16-receive MIMO millimeter-wave radar system, and finally 8 frames of echo data were collected by multi-transmit and multi-receive mode. The radar system parameter table is shown in Table 1:

[0067] Table 1

[0068] System Parameters Numeric Range 100m Carrier frequency 77GHz bandwidth 1GHz FM slope 28MHz / us ADC sampling points 512

[0069] Figure 2 is the simulation result of measured data, Figure 2 (a) is the original echo signal. From the target point in the center of the scene, we can see that the azimuth echoes overlap seriously and the effective target cannot be distinguished. Figure 2 (b) is the super-resolution result of the traditional IAA method, which can distinguish the approximate target, but the resolution is poor. Figure 2 (c) is the super-resolution result of the present invention. The target scene can be almost completely distinguished, and the azimuth imaging resolution is significantly improved. Figure 3 It is a comparison diagram of the azimuth profile of the target at point A in the simulation results of the measured data. It can be seen from the figure that the super-resolution effect of the method of the present invention is significantly improved compared with the traditional azimuth super-resolution method.

[0070] Engineers in this field can make relevant applications based on the millimeter-wave radar azimuth super-resolution method based on multi-frame reconstruction disclosed in the present invention, and the relevant knowledge is still within the protection scope of the present invention.

Claims

1. A millimeter wave radar azimuth super-resolution method based on multi-frame reconstruction, the specific steps are as follows: Step 1: Establish a millimeter-wave radar multi-frame super-resolution imaging model; In the millimeter wave radar imaging process, a multi-transmit and multi-receive method is used to quickly obtain multiple frames of low-resolution echo data, and an imaging model between multiple frames of low-resolution echo data is established: y k =D k M k F k x+n k k=1,2,...,K (1) in, K represents the number of low-resolution echo data frames, y k represents the k-th frame of low-resolution echo data after being processed by the traditional azimuth super-resolution method, D k Represents the shift matrix between multiple frames of low-resolution echo data, M k represents the downsampling matrix, F k represents the fuzzy matrix, x represents the original high-resolution scene, n k represents additive noise; The super-resolution process based on multi-frame reconstruction is the inverse process of the above imaging process, that is, solving the following unconstrained optimization problem: in, Represents high-resolution estimated echo data; In order to obtain a stable solution, the convex set projection method is used to solve the problem; Step 2: Define the convex set constraints of the super-resolution model; For N convex set constraints in Hilbert space H: C1~C N , the solution is obtained by projection, namely: f i+1 =P N P N-1 …P1f i (3) Among them, P N represents the projection operator of the Nth convex set constraint, f i represents the solution after the i-th iteration; The multi-frame super-resolution model is solved by using the convex set projection method. First, the displacement of multiple frames of low-resolution echo data is estimated to obtain the displacement matrix D. Then, a frame of low-resolution echo data y is selected. k As the initial reference frame, bicubic interpolation is performed on it to obtain the initial high-resolution estimated echo data Then define the convex set constraint C: Among them, W k (m,n) represents high-resolution estimated echo data and the kth frame low-resolution echo data y k The azimuth residual between m and n represents the low-resolution echo data y k The distance dimension coordinates and the azimuth dimension coordinates in , η represents the projection threshold; Step 3, calculating the azimuth residual between the echo data; High resolution estimated echo data and the kth frame low-resolution echo data y k The azimuth residual W k (m,n) can be specifically expressed as: Among them, y k (m,n) represents the amplitude of the k-th frame low-resolution echo data at point (m,n); (m′,n′) represents the high-resolution estimated echo data The k-th frame low-resolution echo data y k The point corresponding to the midpoint (m,n) of (m,n), m′, n′ respectively represent the high-resolution estimated echo data The distance dimension coordinates and azimuth dimension coordinates in , the mapping relationship between point (m,n) and point (m′,n′) is F(n′,n″) represents the normalized point spread function; Step 4: Azimuth projection to achieve super-resolution imaging; High resolution estimation echo data Projection is performed as follows: in, Express The projection process; The amplitude of the echo data should be greater than or equal to 0, and an amplitude constraint is introduced during the projection process: The projection operator P, that is, equation (6), is used to estimate the high-resolution echo data. Projection correction is performed and through several iterations, the final solution is constrained to be within the intersection of convex sets, thereby achieving azimuthal super-resolution imaging.

Images