Radar calibration method
By using principal component analysis and a two-stage discrete search method, the yaw angle of the radar is extracted, which solves the problem of limited overlap between the main radar and the blind spot radar, and improves the accuracy of radar calibration and subsequent use.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-19
- Publication Date
- 2026-03-10
AI Technical Summary
The effective overlap area between the main radar and the blind spot radar in space is limited. Point cloud registration cannot be used in the non-overlapping areas, which affects the accuracy of the radar.
Principal component analysis is used to extract the principal direction vector of the original point cloud, and the yaw angle is calculated by combining the angle between the reference plane and the radar. A two-stage discrete search is used to optimize the yaw angle, and finally the extrinsic parameters of the blind spot radar are obtained.
It improves the accuracy of radar calibration, reduces interference from invalid points and outliers in the point cloud, and enhances the accuracy of the radar in subsequent use, with the error controlled within 0.3° to 0.5°.
Smart Images
Figure CN121634015A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radar calibration technology, and more specifically to a radar calibration method. Background Technology
[0002] In related technologies, to ensure vehicle driving safety, the vehicle's radar perception system can usually adopt a configuration of main radar and blind spot radar. The main radar is usually installed at a higher position on the vehicle to achieve long-range detection and to detect distant obstacles in advance, while the blind spot radar is installed on the sides or at a lower position on the vehicle to cover the blind spots of the main radar in the near field. Although this layout strategy improves the overall driving safety, it also results in a limited effective overlap area between the main radar and the blind spot radar in space. The non-overlapping areas cannot be used for radar calibration by point cloud registration, thus affecting the accuracy of the radar. Summary of the Invention
[0003] To address the aforementioned technical problems, this invention provides a radar calibration method to improve the accuracy of radar calibration and at least partially solve the above-mentioned technical problems.
[0004] This invention provides a radar calibration method, comprising: acquiring an original point cloud in a target scene to be calibrated; extracting the principal direction vector of the original point cloud through principal component analysis; and acquiring the radar yaw angle through the principal direction vector.
[0005] Optionally, obtaining the original point cloud in the target scene to be calibrated includes: providing a main radar and a blind spot radar; the target scene to be calibrated has a reference plane, and the original point cloud is obtained by synchronously and continuously scanning the reference plane by the main radar and the blind spot radar.
[0006] Optionally, the target scene to be calibrated has a reference plane, and obtaining the radar yaw angle through the main direction vector includes: obtaining a first angle between the reference plane and the vehicle body; obtaining a second angle between the reference plane and the coordinate system of the blind spot radar; and calculating an initial yaw angle through the first angle and the second angle.
[0007] Optionally, after calculating the initial yaw angle using the first included angle and the second included angle, the method further includes: setting a first search range for the initial yaw angle; performing a discrete search on the first search range to obtain a first yaw angle; setting a second search range for the first yaw angle; and performing a discrete search on the second search range to obtain a second yaw angle.
[0008] Optionally, the step of performing a discrete search on the first search range to obtain the first yaw angle includes: searching within the first search range at a frequency of step length and obtaining a first normal residual; taking the yaw angle with the minimum value of the first normal residual as the first yaw angle.
[0009] Optionally, the step of performing a discrete search on the second search range to obtain the second yaw angle includes: searching within the second search range at a frequency of a second step size and obtaining a second normal residual; taking the yaw angle of the minimum value of the second normal residual as the second yaw angle.
[0010] Optionally, after obtaining the original point cloud in the target scene to be calibrated, and before extracting the principal direction vector of the original point cloud through principal component analysis, the method further includes: removing invalid points and outliers from the original point cloud.
[0011] Optionally, setting the first search range for the initial yaw angle includes setting the angle of the first search range to deviate from the initial yaw angle by ±2°.
[0012] Optionally, setting a second search range for the first yaw angle includes setting the angle of the second search range to deviate from the first yaw angle by ±0.2°.
[0013] Optionally, after taking the yaw angle of the minimum value of the second normal residual as the second yaw angle, the method further includes: taking the second yaw angle as the final extrinsic parameter of the blind spot radar.
[0014] Through the above technical solution, namely the radar calibration method provided by this invention, the radar can extract the principal direction vector of the original point cloud in the target scene to be calibrated by principal component analysis, and obtain the radar yaw angle based on the principal direction vector. That is to say, by using principal component analysis, confusing and unimportant point features in the original point cloud can be eliminated more quickly, thereby obtaining the principal direction vector more quickly. Furthermore, obtaining the radar yaw angle through the principal direction vector makes the radar more accurate in the calibration process and improves the accuracy of the radar in subsequent use. Attached Figure Description
[0015] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0016] Figure 1This is a flowchart of a radar calibration method provided in an exemplary embodiment of the present invention; Figure 2 This is a schematic diagram showing the layout of the vehicle body, the reference plane, and the first and second included angles provided in an exemplary embodiment of the present invention.
[0017] Explanation of reference numerals in the attached figures: 1. Reference plane; 2. Vehicle body; a1. First included angle; a2. Second included angle. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] In related technologies, to ensure vehicle driving safety, the vehicle's radar perception system can typically be configured with a main radar and a blind spot radar. The main radar is usually installed at a higher position on the vehicle to achieve long-range detection and perceive distant obstacles, while the blind spot radar is installed on the sides or lower part of the vehicle to cover the blind spots of the main radar in the near field. Although this layout strategy improves overall driving safety, it also results in a limited effective overlap area between the main radar and the blind spot radar in space. The non-overlapping areas cannot be calibrated by point cloud registration, thus affecting the accuracy of the radar.
[0020] To address the aforementioned technical problems, this invention provides a radar calibration method, with reference to... Figure 1 and Figure 2 As shown, the radar calibration method mainly includes the following steps.
[0021] S1, Obtain the original point cloud in the target scene to be calibrated.
[0022] S2 extracts the principal direction vector of the original point cloud through principal component analysis.
[0023] S3 obtains the radar's yaw angle through the main direction vector.
[0024] Through the above technical solution, namely the radar calibration method provided by this invention, the radar can extract the principal direction vector of the original point cloud in the target scene to be calibrated by principal component analysis, and obtain the radar yaw angle based on the principal direction vector. That is to say, by using principal component analysis, confusing and unimportant point features in the original point cloud can be eliminated more quickly, thereby obtaining the principal direction vector more quickly. Furthermore, obtaining the radar yaw angle through the principal direction vector makes the radar more accurate in the calibration process and improves the accuracy of the radar in subsequent use.
[0025] Further, refer to Figure 1 and Figure 2 As shown, in step S1 above, the method further includes the following steps.
[0026] S11 provides main radar and blind spot radar.
[0027] S12. The target scene to be calibrated has a reference plane 1. The reference plane 1 is scanned synchronously and continuously by the main radar and the blind spot radar to obtain the original point cloud.
[0028] Through the above steps, the reference plane 1 can be a guardrail, a wall, or any structure with a vertical plane. The reference plane 1 is continuously scanned by the main radar and the blind spot radar. Compared with the radar calibration method in related technologies, the present invention does not require complicated operations such as manually placing boxes or other objects. That is, the reference plane 1 is continuously scanned by the main radar and the blind spot radar to obtain the original point cloud, which improves the portability in the process of obtaining the original point cloud.
[0029] Specifically, refer to Figure 1 and Figure 2 As shown, when the target scene to be calibrated has a reference plane 1, the above step S3 may specifically include the following steps.
[0030] S31, obtain the first included angle a1 between the reference plane 1 and the vehicle body 2.
[0031] S32, obtain the second included angle a2 between the reference plane 1 and the coordinate system of the blind spot radar.
[0032] S33, calculate the initial yaw angle using the first included angle a1 and the second included angle a2.
[0033] The unique algorithm described above, which analyzes the first included angle a1 and the second included angle a2 scanned by the main radar and the blind spot radar to obtain the initial yaw angle, can further improve the accuracy of radar calibration. Specifically, when calculating the initial yaw angle using the first included angle a1 and the second included angle a2, the initial yaw angle can be the difference between the first included angle a1 and the second included angle a2.
[0034] Furthermore, after step S33 above, the radar calibration method provided by the present invention further includes the following steps.
[0035] S34 sets the first search range for the initial yaw angle.
[0036] S35, perform a discrete search on the first search range to obtain the first yaw angle.
[0037] S36, set a second search range for the first yaw angle.
[0038] S37, perform a discrete search on the second search range to obtain the second yaw angle.
[0039] The above steps, namely two discrete searches, can further improve the accuracy of the final yaw angle value. When performing a discrete search on the first search range, it can be understood as being in the coarse-grained search stage, i.e., the coarse sweep stage. After the coarse sweep is completed, when performing a discrete search on the second search range, it can be understood as being in the fine-grained search stage, i.e., the fine sweep stage.
[0040] The purpose of performing a discrete search on the first search range is to address the issues of large initial yaw angle errors and non-linear point cloud distribution, thereby effectively locking in the correct solution.
[0041] The purpose of performing a discrete search on the second search range is to further improve the resolution of the first yaw angle so that the accuracy of the second yaw angle can be further improved.
[0042] Specifically, refer to Figure 1 and Figure 2 As shown, step S35 above specifically includes the following steps.
[0043] S351, within the first search range, perform a search at a frequency of one step length and obtain the first normal phase residual.
[0044] S352, take the yaw angle of the minimum value of the first normal phase residual as the first yaw angle.
[0045] Through the above steps, during the coarse scan, a larger search range and a larger step size can be set for discrete searching. For example, the angle of the first search range can be set to deviate from the initial yaw angle by ±2°, and the first step size can be set to 0.05°. In this arrangement, a search scan is performed every 0.05°, and the value of the yaw angle is recorded each time. The point cloud of the blind spot radar is then transformed onto the vehicle body 2 coordinate system to calculate its first normal phase residual relative to the main radar.
[0046] Specifically, refer to Figure 1 and Figure 2As shown, step S37 above specifically includes the following steps.
[0047] S371, within the second search range, a search is performed at a frequency of the second step size to obtain the second normal phase residual.
[0048] S372, take the yaw angle of the minimum value of the second normal phase residual as the second yaw angle.
[0049] Through the above steps, during the fine scan process, a smaller search range and a smaller step size can be set for discrete searching. For example, the angle of the second search range can be set to deviate from the first yaw angle by ±0.2°, and the second step size can be set to 0.01°. In this arrangement, a search scan is performed every 0.01°, and the value of the yaw angle is recorded each time. Similarly, the point cloud of the blind spot radar is transformed to the vehicle body 2 coordinate system in the same way to calculate the second normal phase residual.
[0050] Furthermore, refer to Figure 1 and Figure 2 As shown, after step S1 and before step S2, the method further includes the following steps.
[0051] Remove invalid and outlier points from the original point cloud.
[0052] By removing invalid and outlier points from the original point cloud using the above method, the proportion of valid data in the original point cloud can be further increased. This reduces the interference of invalid and outlier points on the subsequent extraction of the main direction vector, thereby improving the accuracy of the extracted main direction vector.
[0053] In some embodiments, reference Figure 1 and Figure 2 As shown, after step S372 above, the method further includes the following steps.
[0054] S373 uses the second yaw angle as the final extrinsic parameter for the blind spot radar.
[0055] By using the second yaw angle as the final external parameter of the blind spot filling radar in the above manner, the accuracy of radar calibration can be further improved. When the second yaw angle is used as the final external parameter of the blind spot filling radar, the accuracy error of the final yaw angle can be controlled within the range of 0.3° to 0.5°.
[0056] To facilitate understanding of the above method by those skilled in the art, the radar calibration process is described in detail below in this embodiment.
[0057] The first step, taking a vehicle as an example, is to turn on the main radar and blind spot radar installed on the vehicle, park the vehicle next to a fence or wall, and simultaneously collect point cloud data from the main radar and lidar.
[0058] The second step is to preprocess the point cloud data collected by the main radar and the blind spot radar respectively. The preprocessing includes the above-mentioned operations of removing invalid points and outliers.
[0059] The third step is to extract ground point clouds from the point cloud data, perform plane fitting based on the ground point clouds, calculate the ground normal vector, and determine the initial or calibration values of the pitch and roll angles of the lidar based on the ground normal vector.
[0060] The fourth step is to extract the point cloud of the fence or guardrail in the lateral environment of the vehicle from the point cloud data.
[0061] The fifth step involves performing linear feature extraction on the fence point clouds extracted by the main radar and the blind spot radar respectively, to obtain the direction information or fitted line information of the corresponding fence.
[0062] The sixth step involves constructing a residual function related to the yaw angle based on the fence direction information or fitted line information obtained from the main radar and the blind spot radar. The residual function is used to characterize the directional differences or geometric inconsistencies between the fence linear features observed by the main radar and the blind spot radar.
[0063] The seventh step involves optimizing the residual function using the yaw angle of the blind-spot radar relative to the main radar as the variable to be optimized, thereby obtaining the yaw angle extrinsic parameters of the blind-spot radar.
[0064] The eighth step involves combining the pitch angle, roll angle, and yaw angle mentioned above to output the external parameter calibration results of the blind spot radar relative to the main radar.
[0065] It is understood that when performing radar calibration using the above-described radar calibration method, the present invention can use structures such as guardrails or walls without placing additional obstacles or requiring the vehicle to be parallel to the guardrail or wall. The accuracy of radar calibration can be further improved by using principal component analysis to fit whether the vector of the guardrail direction is consistent.
[0066] Furthermore, by employing the aforementioned two-stage one-dimensional search yaw angle optimization method—namely, the operation of coarse scanning followed by fine scanning—the accuracy of radar calibration can be further improved.
[0067] Although embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope of protection claimed by the present invention.
Claims
1. A radar calibration method, characterized by, The method comprises: acquiring an original point cloud in a target scene to be calibrated; extracting a principal direction vector of the original point cloud by principal component analysis; acquiring a yaw angle of a radar by the principal direction vector.
2. The radar calibration method of claim 1, wherein, The acquiring of the original point cloud in the target scene to be calibrated comprises: providing a main radar and a blind-filling radar; the target scene to be calibrated has a reference plane (1), and the reference plane (1) is synchronously and continuously scanned by the main radar and the blind-filling radar to obtain the original point cloud.
3. The radar calibration method of claim 1, wherein, The target scene to be calibrated has a reference plane (1), and the acquiring of the yaw angle of the radar by the principal direction vector comprises: acquiring a first included angle (a1) between the reference plane (1) and a vehicle body (2); acquiring a second included angle (a2) between the reference plane (1) and a coordinate system of the blind-filling radar; calculating an initial yaw angle by the first included angle (a1) and the second included angle (a2).
4. The radar calibration method of claim 3, wherein, After the calculating of the initial yaw angle by the first included angle (a1) and the second included angle (a2), the method further comprises: setting a first search range for the initial yaw angle; discretely searching the first search range to obtain a first yaw angle; setting a second search range for the first yaw angle; discretely searching the second search range to obtain a second yaw angle.
5. The radar calibration method of claim 4, wherein, The discretely searching the first search range to obtain the first yaw angle comprises: searching in the first search range at a frequency of a first step length and acquiring a first normal residual; taking a yaw angle at which the first normal residual is minimum as the first yaw angle.
6. The radar calibration method of claim 4, wherein, The discretely searching the second search range to obtain the second yaw angle comprises: searching in the second search range at a frequency of a second step length and acquiring a second normal residual; taking a yaw angle at which the second normal residual is minimum as the second yaw angle.
7. The radar calibration method of claim 1, wherein, After the acquiring of the original point cloud in the target scene to be calibrated, and before the extracting of the principal direction vector of the original point cloud by the principal component analysis, the method further comprises: removing invalid points and outliers in the original point cloud.
8. The radar calibration method of claim 4, wherein, The setting of the first search range for the initial yaw angle comprises: setting an angle of the first search range to be deviated from the initial yaw angle by ±2°.
9. The radar calibration method of claim 4, wherein, The setting of the second search range for the first yaw angle comprises: setting an angle of the second search range to be deviated from the first yaw angle by ±0.2°.
10. The radar calibration method of claim 6, wherein, After the taking of the yaw angle at which the second normal residual is minimum as the second yaw angle, the method further comprises: taking the second yaw angle as a final external parameter of the blind-filling radar.