Conversion device and method for converting radar signals into path-corrected multi-channel sparse images

Abstract

The present application relates to a conversion device (10) for converting a radar signal into a path-corrected multi-channel sparse image, comprising: a radar data processing module (1), configured to generate a position array having at least one attribute based on radar position data; a grid space generation module (2), configured to construct a polar coordinate grid space and a Cartesian coordinate grid space for the same area or overlapping areas; a projection module (3), configured to project the position array into the polar coordinate grid space for each attribute to generate a multi-channel sparse position matrix; a correlation tensor generation module (4), configured to obtain a correlation tensor between the polar coordinate grid space and the Cartesian coordinate grid space; and a cross-correlation module (5), configured to cross-correlate the multi-channel sparse position matrix with the correlation tensor.

Description

Technical Field

[0001] The present application relates to a method for processing radar detection signals and a corresponding device, and more specifically, to a device and method for converting radar signals into a path-corrected multi-channel sparse image. In addition, the present application also relates to a computer device, a computer-readable storage medium and a vehicle. Background Art

[0002] When using radar to detect the surrounding environment of the host vehicle, the information reflected back by the surrounding objects is usually sparse and discrete position information. However, this information data is difficult to use in some image-based information processing methods, such as neural network methods. Summary of the invention

[0003] The purpose of this application is to solve or at least alleviate the problems existing in the prior art.

[0004] According to some aspects of the present application, the present application proposes a conversion device for converting a radar signal into a path-corrected multi-channel sparse image, comprising:

[0005] A radar data processing module configured to generate a position array having at least one attribute based on the radar position data;

[0006] a grid space generation module configured to construct a polar coordinate grid space and a Cartesian coordinate grid space for the same area or overlapping areas, wherein the metric of the Cartesian coordinate grid space corresponds to the metric of the polar coordinate grid space;

[0007] A projection module, configured to project the position array into the polar coordinate grid space for each attribute to generate a multi-channel sparse position matrix;

[0008] a correlation tensor generation module, configured to obtain correlation tensors between the polar coordinate grid space and the Cartesian coordinate grid space; and

[0009] A cross-correlation module is configured to cross-correlate the multi-channel sparse position matrix with the correlation tensor.

[0010] According to other aspects of the present application, the present application proposes a method for converting a radar signal into a path-corrected multi-channel sparse image, comprising the following steps:

[0011] Get an array of positions with at least one attribute in polar coordinates;

[0012] Construct polar coordinate grid space;

[0013] For each attribute, projecting the position array into a polar grid space and generating a multi-channel sparse position matrix;

[0014] Constructing a Cartesian coordinate grid space, wherein the Cartesian coordinate grid space and the polar coordinate grid space are constructed for the same area or an overlapping area;

[0015] Obtaining correlation tensors between the polar coordinate grid space and the Cartesian coordinate grid space; and

[0016] The multi-channel sparse position matrix is ​​cross-correlated with the correlation tensor.

[0017] According to some further aspects of the present application, the present application proposes a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for converting a radar signal into a path-corrected multi-channel sparse image as disclosed in the embodiments of the present application are implemented.

[0018] According to some further aspects of the present application, the present application proposes a computer-readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the steps of the method for converting a radar signal into a path-corrected multi-channel sparse image as disclosed in the various embodiments of the present application.

[0019] According to still other aspects of the present application, the present application provides a vehicle, wherein the vehicle uses the conversion device disclosed in the various embodiments of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In addition, similar numbers in the figures are used to represent similar components, among which:

[0021] Figure 1 The conversion device disclosed in the present application is shown exemplarily;

[0022] Figure 2 A flowchart of a method for converting a radar signal into a path-corrected multi-channel sparse image disclosed in the present application is exemplarily shown;

[0023] Figure 3 The method for converting a radar signal into a path-corrected multi-channel sparse image disclosed in the present application is schematically shown; and

[0024] Figure 4 The processing result of the method for converting radar signals into path-corrected multi-channel sparse images disclosed in the present application is exemplarily shown, wherein the generated image on the left is a velocity image, and the generated image on the right is a reflected energy image. DETAILED DESCRIPTION

[0025] First refer to the attached Figure 1 , which exemplarily shows a conversion device disclosed in the present application. The conversion device is exemplarily used to convert radar signals into a path-corrected multi-channel sparse image. Figure 1 It can be seen that the conversion device 10 includes: a radar data processing module 1, a grid space generation module 2, a projection module 3, a correlation tensor generation module 4 and a cross-correlation module 5. Here, the radar data processing module 1 can receive the position information of the front target detected by the vehicle radar 100, that is, the radar position data, and generate a position array with at least one attribute based on the above radar position data, wherein a position array or a position matrix is ​​assigned to each attribute. Here, the information in the position matrix can optionally include distance and angle information in the sense of polar coordinates. The origin of the polar coordinates in this sense can be selected as the current position of the host vehicle, and the distance in the sense of polar coordinates represents the radial distance relative to the host vehicle, while the angle in the sense of polar coordinates represents the angle relative to the central longitudinal axis of the host vehicle. Of course, it is also feasible to make other settings for the polar coordinates here. The grid space generation module 2 of the conversion device 10 can be configured to construct a polar coordinate grid space and a Cartesian coordinate grid space, wherein the metric of the Cartesian coordinate grid space corresponds to the metric of the polar coordinate grid space. This means that the measurement of a certain length in the Cartesian coordinate grid space is the same as the measurement of the length in the polar coordinate grid space, which is necessary for the correct calculation of the subsequent correlation tensor. In other words, the Cartesian coordinate grid and the polar coordinate grid are different expressions constructed for the same area or two overlapping areas. After obtaining the position array and the polar coordinate grid space and the Cartesian coordinate grid space with at least one attribute, the projection module 3 can project the position array into the polar coordinate grid space for each attribute to generate a multi-channel sparse position matrix. In the multi-channel sparse position matrix, each channel corresponds to an attribute under the position. Since the position data detected by the radar is sparse and discrete, the matrix obtained by the projection module 3 is also a multi-channel sparse position matrix. The correlation tensor generation module 4 can obtain the correlation tensor of the polar coordinate grid space and the Cartesian coordinate grid space. That is to say, the correlation tensor generation module 4 can associate the polar coordinate grid space with the Cartesian coordinate grid space to obtain the correlation tensor. Under the condition of receiving the correlation tensor and the multi-channel sparse position matrix, the cross-correlation module 5 can cross-correlate the multi-channel sparse position matrix with the correlation tensor.

[0026] The conversion device 10 thus obtains a multi-channel sparse image of road correction from the radar position data, which can be used for subsequent image-based neural networks or deep learning applications. Based on the sparse and discrete characteristics of the position data obtained by the radar, the conversion device 10 can quickly convert the radar data into a sparse image. In addition, since the ordinate of the Cartesian coordinate system used by the conversion device 10 is based on the vehicle's driving path, the obtained sparse image can be corrected relative to the traditional method, thereby further reducing the processing complexity and processing time.

[0027] In some embodiments disclosed in the present application, the properties of the position array may include speed and reflected energy. It should be noted that the speed may include absolute speed and relative speed. Exemplarily, the speed here refers to the radial speed of the target relative to the host vehicle, while the reflected energy is determined by the material properties of the target itself. For example, when the radar wave hits the target, the reflected energy of the metal is usually greater than that of the non-metal.

[0028] In some embodiments disclosed in the present application, the radar data processing module 1 may remove noise point position data in the radar position data before generating a position array having at least one attribute in polar coordinates. Here, the noise point position refers to the position information of some interfering objects, such as stones on the road surface, when the vehicle radar 100 detects.

[0029] In some embodiments disclosed in the present application, when multiple position arrays fall into the same polar coordinate grid, the projection module 3 performs position fusion or weighted averaging on the multiple position arrays and then projects them into the polar coordinate grid.

[0030] In some embodiments disclosed in the present application, the grid space generation module 2 pre-constructs the polar coordinate grid space and the Cartesian coordinate grid space. Preferably, the polar coordinate grid space and the Cartesian coordinate grid space can be pre-constructed offline and related tensors can be obtained, which can further save the processing time of the conversion device 10.

[0031] In some embodiments disclosed in the present application, the abscissa of the Cartesian coordinate is constructed along the lateral direction of the current position, and the ordinate is constructed along the direction of the driving route. Compared with the traditional Cartesian coordinate system, the sparse image obtained by the Cartesian coordinate system set in this way is obtained based on the driving route of the vehicle, thereby reducing the complexity of the subsequent algorithm.

[0032] In some embodiments disclosed in the present application, when the correlation tensor generation module 4 obtains the correlation tensor of the polar coordinate grid space and the Cartesian coordinate grid space, it includes: for each grid in the Cartesian coordinate grid space, overlapping the polar coordinate grid space and the Cartesian coordinate grid space; and marking the grid space where the polar coordinate grid overlaps with each grid and the grid space where it does not overlap with each grid with different marks. For example, Figure 3 As shown in the example, for each grid in the Cartesian coordinate grid space, the relevant tensor generation module 4 can check the polar coordinate grid and the grid one by one, and mark 1 or other marks when the polar coordinate grid overlaps with the grid, and mark 0 or other marks when the polar coordinate grid does not overlap with the grid. After all grids are checked, a matrix with the same dimension as the number of grids in the Cartesian coordinate grid space will be formed, and the matrix under each dimension corresponds to the overlap of a grid with the polar coordinate grid space.

[0033] In some embodiments disclosed in the present application, the grid space generation module 2 constructs the polar coordinate grid space and the Cartesian coordinate grid space as follows, so that the grid size of the Cartesian coordinate grid space is larger than the grid size of the polar coordinate grid space. This means that each grid of the Cartesian coordinate grid space may correspond to multiple polar coordinate grids, so that each grid in the Cartesian coordinate grid space may include multiple complete radar data location points, which is conducive to improving the accuracy of the generated image.

[0034] In some embodiments disclosed in the present application, when the cross-correlation module 5 cross-correlates the multi-channel sparse position matrix with the related tensor, it includes: tensor broadcasting the multi-channel sparse position matrix; and cross-correlating the multi-channel sparse position matrix after the tensor broadcast with the related tensor. Here, tensor broadcasting means that when two matrices of different dimensions are operated, the matrix with a smaller dimension needs to be broadcasted to have the same dimension as the matrix with a larger dimension.

[0035] Next reference Figure 2 and Figure 3 ,in, Figure 2 A flowchart of a method for converting a radar signal into a path-corrected multi-channel sparse image disclosed in the present application is schematically shown, and Figure 3 An example of using this method is shown.

[0036] In combination, the method for converting radar signals into path-corrected multi-channel sparse images disclosed in the present application includes the following steps:

[0037] Obtaining a position array S1 having at least one attribute in polar coordinates, this step may include receiving the position information of the front target detected by the vehicle radar 100, that is, the radar position data, and generating a position array having at least one attribute based on the above radar position data, wherein a position array or a position matrix is ​​assigned to each attribute. Here, the information in the position matrix may optionally include distance and angle information in the sense of polar coordinates. The origin of the polar coordinates in this sense can be selected as the current position of the host vehicle, and the distance in the sense of polar coordinates represents the radial distance relative to the host vehicle, while the angle in the sense of polar coordinates represents the angle relative to the central longitudinal axis of the host vehicle. Of course, it is also feasible to make other settings for the polar coordinates here.

[0038] Construct polar coordinate grid space S2;

[0039] For each attribute, the position array is projected into the polar coordinate grid space, and a multi-channel sparse position matrix S3 is generated; after obtaining the position array and polar coordinate grid space having at least one attribute, the position array is projected into the polar coordinate grid space for each attribute to generate a multi-channel sparse position matrix. In the multi-channel sparse position matrix, each channel corresponds to an attribute under the position. Since the position data detected by the radar is sparse and discrete, the resulting matrix is ​​also a multi-channel sparse position matrix.

[0040] Construct a Cartesian coordinate grid space S4, wherein the metric of the Cartesian coordinate grid space corresponds to the metric of the polar coordinate grid space. This means that the metric of a certain length in the Cartesian coordinate grid space is the same as the metric of the length in the polar coordinate grid space, which is necessary for the correct determination of the subsequent related tensors. In other words, the Cartesian coordinate grid and the polar coordinate grid are different expressions constructed for the same area or two overlapping areas.

[0041] Obtaining a correlation tensor S5 between the polar coordinate grid space and the Cartesian coordinate grid space; and

[0042] The multi-channel sparse position matrix is ​​cross-correlated with the correlation tensor S6.

[0043] The method thus obtains a multi-channel sparse image of road correction from the radar position data, which can be used for subsequent image-based neural network or deep learning applications. Based on the sparse and discrete characteristics of the position data obtained by the radar, the method can quickly convert the radar data into a sparse image. In addition, since the ordinate of the Cartesian coordinate system used in the method is based on the vehicle's driving path, the obtained sparse image can be corrected relative to the traditional method, thereby further reducing the processing complexity and processing time.

[0044] In some embodiments disclosed in the present application, the attributes of the position array may include speed and reflected energy. It should be noted that speed refers to the radial speed of the target relative to the host vehicle, while reflected energy is determined by the material properties of the target itself. For example, when radar waves hit the target, the reflected energy of metal is usually greater than that of non-metal.

[0045] In some embodiments disclosed in the present application, before obtaining the position array S1 having at least one attribute in polar coordinates, the noise position data in the radar position data may be removed. Here, the noise position refers to the position information of some interfering objects, such as stones on the road surface, when the vehicle radar 100 detects.

[0046] In some embodiments disclosed in the present application, for the step of projecting the position array into the polar coordinate grid space and generating a multi-channel sparse position matrix S3, when multiple positions fall into the same polar coordinate grid, the multiple position arrays are positionally fused or weighted averaged and then projected into the polar coordinate grid.

[0047] In some embodiments disclosed in the present application, the polar coordinate grid space and the Cartesian coordinate grid space are pre-constructed. Preferably, the polar coordinate grid space and the Cartesian coordinate grid space can be pre-constructed offline and related tensors can be obtained, which can further save processing time.

[0048] In some embodiments disclosed in the present application, the abscissa of the Cartesian coordinate is constructed along the lateral direction of the current position, and the ordinate is constructed along the direction of the driving route. Compared with the traditional Cartesian coordinate system, the sparse image obtained by the Cartesian coordinate system set in this way is obtained based on the driving route of the vehicle, thereby reducing the complexity of the subsequent algorithm.

[0049] In some embodiments disclosed in the present application, when obtaining the related tensor S5 of the polar coordinate grid space and the Cartesian coordinate grid space, it includes: for each grid in the Cartesian coordinate grid space, overlapping the polar coordinate grid space and the Cartesian coordinate grid space; and marking the grid space where the polar coordinate grid overlaps with each grid and the grid space that does not overlap with each other with different marks. For example, for each grid in the Cartesian coordinate grid space, the polar coordinate grid and the Cartesian coordinate grid can be checked one by one, and 1 or other marks are marked when the polar coordinate grid and the Cartesian coordinate grid overlap with each other, and 0 or other marks are marked when the polar coordinate grid and the Cartesian coordinate grid do not overlap with each other. After all grids are checked, a matrix with the same dimension as the number of grids in the Cartesian coordinate grid space will be formed, and the matrix under each dimension corresponds to the overlap of a grid with the polar coordinate grid space.

[0050] In some embodiments disclosed in the present application, the polar coordinate grid space and the Cartesian coordinate grid space are constructed as follows, so that the grid size of the Cartesian coordinate grid space is larger than the grid size of the polar coordinate grid space. This means that each grid of the Cartesian coordinate grid space may correspond to multiple polar coordinate grids, so that each grid in the Cartesian coordinate grid space may include multiple complete radar data location points, which is conducive to improving the accuracy of the generated image.

[0051] In some embodiments disclosed in the present application, when the multi-channel sparse position matrix is ​​cross-correlated with the related tensor S6, it includes: tensor broadcasting the multi-channel sparse position matrix; and cross-correlating the multi-channel sparse position matrix after the tensor broadcast with the related tensor. Here, tensor broadcasting means that when two matrices of different dimensions are operated, the matrix with a smaller dimension needs to be broadcasted to have the same dimension as the matrix with a larger dimension.

[0052] Figure 4 The results of using the conversion device disclosed in this application or implementing the method disclosed in this application are shown. Figure 4 The sparse image on the left is a sparse image of the transformation whose attribute is velocity, while the sparse image on the right is a sparse image of the transformation whose attribute is reflected energy. Figure 4 It can be seen that the converted image can reflect the speed information of the surrounding objects of the host vehicle and the characteristics of the surrounding objects themselves.

[0053] In addition, the present application also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the method for converting a radar signal into a path-corrected multi-channel sparse image described in any embodiment of the present application are implemented.

[0054] In addition, the present application also discloses a computer-readable storage medium on which a computer program is stored. The program is executed by a processor to implement the steps of the method for converting a radar signal into a path-corrected multi-channel sparse image as described in any embodiment of the present application.

[0055] In addition, the present application also discloses a vehicle, which uses the conversion device 10 as described in any embodiment of the present application.

Claims

1. A conversion device (10) for converting a radar signal into a path-corrected multi-channel sparse image, characterized in that: include: A radar data processing module (1) is configured to generate a position array having at least one attribute based on radar position data; A grid space generation module (2) is configured to construct a polar coordinate grid space and a Cartesian coordinate grid space for the same area or overlapping areas; A projection module (3) is configured to project the position array into the polar coordinate grid space for each attribute to generate a multi-channel sparse position matrix; A correlation tensor generation module (4) is configured to obtain correlation tensors between the polar coordinate grid space and the Cartesian coordinate grid space; as well as A cross-correlation module (5) is configured to cross-correlate the multi-channel sparse position matrix with the correlation tensor.

2. The conversion device according to claim 1, characterized in that The properties of the position array include velocity and reflected energy.

3. The conversion device according to claim 1, characterized in that The grid space generation module (2) constructs the polar coordinate grid space and the Cartesian coordinate grid space in advance and / or the related tensor generation module (4) obtains the related tensor in advance.

4. The conversion device according to claim 1, characterized in that When obtaining the correlation tensor between the polar coordinate grid space and the Cartesian coordinate grid space, the correlation tensor generation module (4) comprises: for each grid in the Cartesian coordinate grid space, Overlaying the polar coordinate grid space with the Cartesian coordinate grid space; and The grid spaces where the polar coordinate grid overlaps with each grid and the grid spaces where the polar coordinate grid does not overlap are marked with different marks.

5. The conversion device according to claim 1, characterized in that The grid space generation module (2) constructs the polar coordinate grid space and the Cartesian coordinate grid space as follows, so that the grid size of the Cartesian coordinate grid space is larger than the grid size of the polar coordinate grid space, or the cross-correlation module (5) when cross-correlating the multi-channel sparse position matrix with the correlation tensor comprises: Expanding the dimension of the multi-channel sparse position matrix so that the dimension of the multi-channel sparse position matrix is ​​the same as the dimension of the correlation tensor; and The expanded multi-channel sparse position matrix is ​​cross-correlated with the correlation tensor.

6. The conversion device according to claim 2, characterized in that The radar data processing module (1) removes noise point position data from the radar position data before generating a position array having at least one attribute in polar coordinates.

7. The conversion device according to claim 2, characterized in that When multiple position arrays fall into the same polar coordinate grid, the projection module (3) performs position fusion or weighted averaging on the multiple position arrays and then projects them into the polar coordinate grid.

8. The conversion device according to claim 3, characterized in that The abscissa of the Cartesian coordinates is formed as a transverse direction along the current position, and the ordinate is formed as a direction along the travel route.

9. A method for converting radar signals into path-corrected multi-channel sparse images, characterized in that: The steps include: Get an array of positions (S1) with at least one attribute in polar coordinates; Construct polar coordinate grid space (S2); For each attribute, projecting the position array into a polar grid space and generating a multi-channel sparse position matrix (S3); constructing a Cartesian coordinate grid space, wherein the Cartesian coordinate grid space and the polar coordinate grid space are constructed for the same area or an overlapping area (S4); Obtaining the correlation tensor between the polar coordinate grid space and the Cartesian coordinate grid space (S5); and The multi-channel sparse position matrix is ​​cross-correlated with the correlation tensor (S6).

10. The method for converting radar signals into path-corrected multi-channel sparse images according to claim 9, characterized in that: The attributes of the position array include speed and reflected energy, or, before obtaining the position array (S1) having at least one attribute in polar coordinates, noise point position data in the radar position array is removed.

11. The method for converting radar signals into path-corrected multi-channel sparse images according to claim 9, characterized in that: The polar coordinate grid space and the Cartesian coordinate grid space are constructed in advance and / or the relevant tensor generation module obtains the relevant tensor in advance.

12. The method for converting radar signals into path-corrected multi-channel sparse images according to claim 9, characterized in that: The step of obtaining the correlation tensor between the polar coordinate grid space and the Cartesian coordinate grid space (S5) includes: for each grid in the Cartesian coordinate grid space, Overlaying the polar coordinate grid space with the Cartesian coordinate grid space; and The grid spaces where the polar coordinate grid overlaps with each grid and the grid spaces where the polar coordinate grid does not overlap are marked with different marks.

13. The method for converting radar signals into path-corrected multi-channel sparse images according to claim 9, characterized in that: The grid size of the Cartesian coordinate grid space is greater than the grid size of the polar coordinate grid space.

14. The method for converting radar signals into path-corrected multi-channel sparse images according to claim 10, characterized in that: When multiple position arrays fall into the same polar coordinate grid, the multiple position arrays are positionally fused or weighted averaged and then projected into the polar coordinate grid.

15. The method for converting radar signals into path-corrected multi-channel sparse images according to claim 11, characterized in that: The abscissa of the Cartesian coordinates is formed as a transverse direction along the current position, and the ordinate is formed as a direction along the travel route.

16. The method for converting radar signals into path-corrected multi-channel sparse images according to claim 13, characterized in that: In the step of cross-correlating the multi-channel sparse position matrix with the correlation tensor (S6), it includes: Expanding the dimension of the multi-channel sparse position matrix so that the dimension of the multi-channel sparse position matrix is ​​the same as the dimension of the correlation tensor; and The expanded multi-channel sparse position matrix is ​​cross-correlated with the correlation tensor.

17. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method for converting a radar signal into a path-corrected multi-channel sparse image according to any one of claims 9 to 16 are implemented.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the steps of the method for converting a radar signal into a path-corrected multi-channel sparse image according to any one of claims 9 to 16.

19. A vehicle, characterized in that: Use of a conversion device as claimed in any one of claims 1 to 8.

Images