A multi-laser radar external parameter calibration method based on panoramic laser radar assistance

Abstract

The present application relates to a kind of multi-laser radar external parameter calibration method based on panoramic laser radar auxiliary, it is applied to intelligent driving vehicle, the method includes: step 1, the place with obvious structural features is selected as calibration scene;Step 2, the vehicle to be calibrated is placed in calibration scene, records the point cloud data C1 of multiple laser radars needing external parameter calibration on vehicle, and it is stored;Step 3, panoramic laser radar device is placed in the same position of calibration scene, records the point cloud data C2 of panoramic laser radar, and it is stored;Step 4, the point cloud data C1 and C2 recorded are played back simultaneously, each point cloud of C1 is sequentially registered with the point cloud of C2, obtains the external parameter of laser reaching auxiliary radar coordinate system of vehicle to be calibrated;Step 5, each laser radar on vehicle to be calibrated is converted to same coordinate system, the relative external parameter between any two vehicle to be calibrated laser radars is calculated.

Description

Technical Field

[0001] This invention relates to the field of parameter calibration technology, and in particular to a method for calibrating the extrinsic parameters of multiple lidar based on panoramic lidar-assisted calibration. Background Technology

[0002] With the rapid development of autonomous driving technology, LiDAR, as a core sensor for achieving high-precision environmental perception, has been widely used in various autonomous driving systems. To improve the comprehensiveness and robustness of environmental perception, more and more autonomous vehicles are choosing to equip multiple LiDAR sensors to achieve a wider field of view, higher coverage, and stronger redundant perception capabilities. However, in multi-radar systems, the spatial pose relationship (i.e., extrinsic parameters) between various LiDARs becomes a fundamental prerequisite for data fusion, mapping, and decision control. Extrinsic parameter errors will directly affect the point cloud fusion quality, obstacle recognition accuracy, and vehicle positioning stability. Therefore, achieving a high-precision, stable, and universal radar extrinsic parameter calibration method has become a key technical challenge in the research and development of autonomous driving perception systems. Existing multi-radar extrinsic parameter calibration methods mainly rely on the large overlap of the field of view between radars, directly solving the transformation matrix through point cloud registration (such as ICP, NDT, etc.). However, in actual deployment, due to the limitations of vehicle structure and functional layout, the installation positions and orientations of different radars vary greatly, and the field of view often suffers from significant occlusion or excessively small overlap areas, making it difficult for traditional methods to stably and effectively complete extrinsic parameter calibration. In addition, some solutions attempt to use manual auxiliary devices such as calibration boards and feature objects for calibration. Although this improves the calibration accuracy to some extent, it greatly increases the complexity of operation and deployment costs, limiting its promotion and application in mass-produced vehicles or large-scale testing. Summary of the Invention

[0003] In view of the above problems, the present invention provides a multi-lidar extrinsic parameter calibration method based on panoramic lidar assistance, so as to solve the technical problems of existing technologies such as reliance on field of view overlap, complex calibration process, and the need for manual assistance and other calibration tools.

[0004] This invention provides a method for extrinsic parameter calibration of multiple lidars based on panoramic lidar, applicable to intelligent driving vehicles. The method includes: Step 1, selecting a location with obvious structural features as the calibration scene; Step 2, placing the vehicle to be calibrated in the calibration scene, recording and storing the point cloud data C1 of multiple lidars on the vehicle that require extrinsic parameter calibration; Step 3, after the vehicle leaves the calibration scene, placing the panoramic lidar device in the same position in the calibration scene, recording and storing the point cloud data C2 of the panoramic lidar; Step 4, simultaneously replaying the recorded point cloud data C1 and C2, and sequentially registering each point cloud of C1 with the point cloud of C2 using a point cloud registration algorithm to obtain the extrinsic parameters of the lidars on the vehicle to be calibrated to the auxiliary lidar coordinate system; Step 5, based on the extrinsic parameters of the coordinate system, transforming each lidar on the vehicle to be calibrated to the same coordinate system, and calculating the relative extrinsic parameters between any two lidars to be calibrated.

[0005] Furthermore, step 2 also includes: generating and collecting point cloud data of each lidar at the same time through the robot operating system.

[0006] Furthermore, step 3 also includes: acquiring panoramic lidar point cloud data.

[0007] Furthermore, step 4 includes: step 41, preprocessing the point cloud data C1 and C2 to eliminate unstructured differences and retain core structural features; step 42, adjusting the parameters of the NDT point cloud registration algorithm according to the characteristics of the point cloud data C1 and C2; and step 43, outputting a transformation matrix for each LiDAR X on the vehicle to be calibrated using the adjusted point cloud matching algorithm. Its form is: ,in, It is the rotation matrix for converting radar X to panoramic lidar C. It is a translation vector; Step 44, verify the transformation matrix. The registration effect is to ensure the accuracy of the extrinsic parameters of the lidar to be calibrated vehicle to the auxiliary radar coordinate system.

[0008] Furthermore, step 41 includes: step 411, applying statistical filtering to both C1 and C2 to remove outliers caused by radar measurement noise; step 412, reducing point cloud density and computational load by voxel filtering; step 413, manually trimming to retain rigid structural regions and removing edge regions susceptible to installation location; and step 414, performing time synchronization verification on C1 to ensure that its point cloud corresponds to the same state of the scene and is free from interference by dynamic objects.

[0009] Furthermore, step 42 includes: step 421, increasing the grid resolution of C2 to adapt C1 via dynamic grid; step 422, setting the initial transformation T according to the position of the lidar of the vehicle to be calibrated. init To avoid the point cloud registration algorithm getting stuck in local optima; step 423, set the convergence threshold and number of iterations to improve accuracy.

[0010] Furthermore, the convergence threshold is the change in the objective function, and the number of iterations is 500.

[0011] Further, step 44 includes: step 441, calculating the average distance between the transformed point cloud of LiDAR X and the point cloud of panoramic LiDAR C in the overlapping area; step 442, comparing the results of the ICP point cloud registration algorithm with the NDT point cloud registration algorithm, and if the difference between the two is less than a preset value T1, the registration effect passes the consistency verification; step 443, if the registration deviation between the point cloud of each LiDAR X and the panoramic LiDAR C is less than a preset value T2, the registration effect passes the multi-frame verification.

[0012] Furthermore, T1 = 0.01m, T2 = 0.03m.

[0013] Furthermore, step 5 includes: step 51, according to the function... Obtain the rotation matrix from the panoramic LiDAR C to the first LiDAR B of the vehicle to be calibrated, where, This is the rotation matrix from the panoramic lidar C to the second lidar B of the vehicle to be calibrated. The translation vector is used; in step 52, the transformation matrix from the first lidar A to the second lidar B of the vehicle to be calibrated is obtained according to the following function. ,

[0014] ,

[0015] , ,in, Let be the transformation matrix from the first lidar A of the vehicle to be calibrated to the second lidar B of the vehicle to be calibrated. This is the rotation part of the transformation matrix. This refers to the translation component in the transformation matrix; Step 53, based on all the lidar extrinsic parameters of the vehicle to be calibrated. Repeat steps 51 and 52 to obtain the relative extrinsic parameters between any two lidars of the vehicle to be calibrated. Step 54, for relative external parameters Perform reprojection verification, symmetry verification, and consistency verification to avoid error accumulation.

[0016] This invention provides a multi-lidar extrinsic parameter calibration method based on panoramic lidar assistance. This method uses a panoramic lidar as an intermediate coordinate system to achieve extrinsic parameter calibration between two target lidars. This technical solution does not rely on overlapping fields of view between lidars, and is a novel multi-lidar calibration method with stronger versatility and accuracy assurance, meeting the needs of fast, reliable, and high-precision calibration of multi-lidar systems in practical applications. Attached Figure Description

[0017] Figure 1 This invention provides a flowchart of a multi-lidar extrinsic parameter calibration method assisted by panoramic lidar;

[0018] Figure 2 The diagram illustrates the application scenarios provided by this invention.

[0019] Figure 3 This is a flowchart of the method for obtaining the external parameters of the lidar of the vehicle to be calibrated to the coordinate system of the auxiliary radar, provided by the present invention.

[0020] Figure 4 This is a flowchart of the method for preprocessing point cloud data provided by the present invention;

[0021] Figure 5 This is a flowchart of the method for adjusting the parameters of the point cloud registration algorithm provided by the present invention;

[0022] Figure 6 The verification transformation matrix provided by this invention A flowchart illustrating the registration process;

[0023] Figure 7 This invention provides a method for calculating the relative extrinsic parameters between any two lidars to be calibrated. Detailed Implementation

[0024] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0025] Example 1:

[0026] This invention provides a multi-lidar extrinsic parameter calibration method based on panoramic lidar assistance, applicable to intelligent driving vehicles, such as... Figure 1 As shown, the method includes the following steps.

[0027] Step 1: Select a location with obvious structural features as the calibration scene;

[0028] Step 2: Place the vehicle to be calibrated in the calibration scene, record the point cloud data C1 of the multiple lidars on the vehicle that need to be calibrated for external parameters, and store them;

[0029] The lidar is a sensor that measures distance and models data by emitting laser beams and receiving reflected signals; it is commonly used for environmental perception.

[0030] Step 3: After the vehicle leaves the calibration scene, place the panoramic LiDAR device in the same position as the calibration scene, record the point cloud data C2 of the panoramic LiDAR, and store it.

[0031] Step 4: Play back the recorded point cloud data C1 and C2 simultaneously. Using the point cloud registration algorithm, register each point cloud of C1 with the point cloud of C2 in turn to obtain the external parameters of the lidar of the vehicle to be calibrated to the auxiliary radar coordinate system.

[0032] Step 5: Based on the extrinsic parameters of the coordinate system, transform each lidar on the vehicle to be calibrated to the same coordinate system, and calculate the relative extrinsic parameters between any two lidars to be calibrated.

[0033] This invention provides a multi-lidar extrinsic parameter calibration method based on panoramic lidar assistance. This method uses a panoramic lidar as an intermediate coordinate system to achieve extrinsic parameter calibration between two target lidars. This technical solution does not rely on overlapping fields of view between lidars, and is a novel multi-lidar calibration method with stronger versatility and accuracy assurance, meeting the needs of fast, reliable, and high-precision calibration of multi-lidar systems in practical applications.

[0034] Example 2:

[0035] This invention provides a multi-lidar extrinsic parameter calibration method based on panoramic lidar assistance, applicable to intelligent driving vehicles, such as... Figure 1 As shown, the method includes the following steps.

[0036] Step 1: Select a location with obvious structural features as the calibration scene;

[0037] Underground parking garages or other locations with distinctive structural features are typically chosen as the calibration scenario, with vehicles parked in designated spaces. This is because underground parking lots generally have many pillars and prominent structural features such as wall corners. Figure 2 The gray squares (representing columns) and gray lines (representing walls) in the diagram are beneficial for improving calibration accuracy when extrinsic parameters are calibrated, as they exist within the overlapping field of view of two radars.

[0038] Step 2: Place the vehicle to be calibrated in the calibration scene, record the point cloud data C1 of the multiple lidars on the vehicle that need to be calibrated for external parameters, and store them;

[0039] Step 2 further includes: generating and collecting point cloud data of each LiDAR at the same time using the Robot Operating System (ROS system). After all the LiDARs to be calibrated are installed on the target vehicle, ROSBag point cloud data of each LiDAR at the same time are collected. Figure 2 On the left side, use the rosbag record command (the rosbag record command is a command to record the topic messages sent by other ROS nodes in the ROS system) to record the point cloud data of multiple radars that need to be calibrated for external parameters, such as forward view, backward view, left view, and right view, and store it in rosbag format. Recording 10 seconds of data is sufficient.

[0040] Step 3: After the vehicle leaves the calibration scene, place the panoramic LiDAR device in the same position as the calibration scene, record the point cloud data C2 of the panoramic LiDAR, and store it.

[0041] like Figure 2 On the right side, in the same scenario, using a device equipped with a panoramic LiDAR, the `rosbagrecord` command is used to record the point cloud data of the panoramic LiDAR and store it in ROSBAG format. Recording 10 seconds of data is sufficient. Step 3 also includes: acquiring panoramic LiDAR point cloud data. Furthermore, the point cloud data recorded and stored in steps 2 and 3 is in ROSBAG format, which is the file format used in the robot operating system for recording and replaying sensor data. The panoramic LiDAR, also known as a 360-degree mechanical LiDAR, is a type of LiDAR that achieves 360-degree full-field perception through rotation. The point cloud acquired by the LiDAR is a set of three-dimensional points in space, typically used to construct environmental maps or perceive objects.

[0042] Step 4: Play back the recorded point cloud data C1 and C2 simultaneously. Using the point cloud registration algorithm, register each point cloud of C1 with the point cloud of C2 in turn to obtain the external parameters of the lidar of the vehicle to be calibrated to the auxiliary radar coordinate system.

[0043] The `rosbag play` command (in ROS systems, a command used to replay recorded topic data) is used to simultaneously replay two `rosbag` point cloud data packets acquired during the data acquisition phase. The lidar extrinsic parameter calibration algorithm is then run, sequentially registering the point cloud of each lidar to be calibrated with the point cloud of the 360-degree lidar to obtain the extrinsic parameters of the lidar to be calibrated to the auxiliary lidar coordinate system. Specifically, for any lidar to be calibrated (denoted as lidar X), the extrinsic parameter solution logic with a third-party lidar (denoted as lidar C) is as follows: The point cloud of third-party lidar C is used as the reference point cloud (Target): due to its 360-degree coverage, the point cloud contains the complete structure of the scene, enabling the construction of a more robust probability distribution field; the point cloud of lidar X is used as the source point cloud (Source): its point cloud represents a local observation of the scene (limited by the lidar's field of view), but shares some structural features with the reference point cloud; the transformation matrix is ​​solved using the NDT algorithm. This transforms the source point cloud (radar X) to align it with the reference point cloud (radar C). This matrix represents the extrinsic parameters from radar X to the third-party radar C. For example... Figure 3 As shown, step 4 includes the following steps.

[0044] Step 41: Preprocess the point cloud data C1 and C2 to eliminate unstructured differences and retain core structural features;

[0045] Because the panoramic LiDAR and the radar to be calibrated acquire data at different times (after the vehicle leaves, the panoramic LiDAR is placed in the same parking space as the radar to be calibrated at the time of acquisition), preprocessing is required to eliminate unstructured differences (such as noise and redundant points) while preserving core structural features, such as... Figure 4 As shown, step 41 includes the following steps.

[0046] Step 411: Apply statistical filtering to both C1 and C2 to remove outliers caused by radar measurement noise.

[0047] Step 412: Reduce point cloud density by voxel filtering to reduce computational load;

[0048] Step 413: Manually trim and retain the rigid structural area, and remove the edge areas that are easily affected by the installation position;

[0049] Although the third-party radar is placed in the original vehicle location, there may be millimeter-level installation errors. It is necessary to manually trim and retain the "rigid structure area" (such as walls and pillars) and remove the edge areas that are easily affected by the installation position (such as small bumps on the ground).

[0050] Step 414: Perform time synchronization verification on C1 to ensure that its point cloud corresponds to the same state of the scene and is free from interference from dynamic objects.

[0051] Step 42: Adjust the parameters of the point cloud registration algorithm according to the characteristics of point cloud data C1 and C2;

[0052] like Figure 5 As shown, step 42 includes the following steps.

[0053] Step 421: Increase the mesh resolution of C2 in order to adapt C1 through dynamic mesh;

[0054] The reference point cloud (point cloud data of panoramic LiDAR) has a large coverage area, and the grid resolution can be appropriately increased (e.g., 0.5-1.0m) to balance computational efficiency and structure preservation; the source point cloud (point cloud data of the radar to be calibrated) is locally dense and can be adapted through "dynamic grid" (e.g., grid densification in overlapping areas).

[0055] Step 422: Based on the position of the lidar of the vehicle to be calibrated, set the initial transformation T. init This avoids the point cloud registration algorithm getting trapped in local optima;

[0056] Based on prior knowledge of the installation location (such as the approximate location of the radar to be calibrated on the vehicle body), the initial transformation T is set. init (For example, when the radar is in front of the vehicle, the initial translation is along the positive X-axis) to avoid NDT getting trapped in local optima.

[0057] Step 423: Set the convergence threshold and number of iterations to improve accuracy.

[0058] Setting a stricter convergence threshold (e.g., the change in the objective function < 1e-7) allows for more iterations (e.g., 500 times) to improve accuracy, given the static nature of the scenario and the stable structure.

[0059] Step 43: Using the adjusted point cloud matching algorithm, output the transformation matrix for each LiDAR X on the vehicle to be calibrated. Its form is: ,in, It is the rotation matrix for converting radar X to panoramic lidar C. It is a translation vector;

[0060] Step 44, verify the transformation matrix The registration effect is to ensure the accuracy of the extrinsic parameters of the lidar to be calibrated vehicle to the auxiliary radar coordinate system.

[0061] To ensure Reliability requires verification of registration effectiveness, such as... Figure 6 As shown, step 44 includes the following steps.

[0062] Step 441: Calculate the average distance between the transformed point cloud of LiDAR X and the point cloud of panoramic LiDAR C in the overlapping area.

[0063] Calculate the average distance between the radar X-transformed point cloud and the radar C-transformed point cloud in the overlapping area (e.g., points on a wall), with an error of less than 0.05m (depending on radar resolution).

[0064] Step 442: Compare the results of the ICP point cloud registration algorithm with those of the NDT point cloud registration algorithm. If the difference between the two is less than the preset value T1, the registration effect passes the consistency verification.

[0065] The NDT results are refined using the ICP algorithm (NDT+ICP combination). If the difference in transformation before and after refinement is less than 0.01m, the result is considered stable. The point cloud registration algorithm (including the ICP algorithm and the NDT algorithm) refers to a method of registration by matching the probability density of the point cloud distribution.

[0066] Step 443: If the registration deviation between the point cloud of each LiDAR X and the panoramic LiDAR C is less than the preset value T2, then the registration effect is verified through multiple frames.

[0067] If the lidar for the calibrated vehicle collects multiple sets of lidar point clouds to be calibrated, the extrinsic parameter deviation between each set of point clouds and the panoramic lidar must be less than 0.03m to avoid the influence of single-frame noise. Additionally, T1 = 0.01m and T2 = 0.03m.

[0068] The core of this step is to use the NDT algorithm to solve the relative pose (extrinsic parameters) between each LiDAR on the vehicle to be calibrated and the panoramic LiDAR. The panoramic LiDAR, acting as an "intermediate coordinate system," has a complete point cloud coverage of the scene (360 degrees without blind spots) and observes the same static scene (such as an underground parking garage) as the LiDAR to be calibrated. Therefore, their point clouds describe the same static structures in space (such as walls, pillars, parking lines, etc.), providing sufficient overlapping features for each LiDAR to be calibrated. Utilizing this "scene consistency," the transformation relationship from the coordinate system of the LiDAR to the coordinate system of the third-party LiDAR can be found through NDT registration, solving the problem of "insufficient overlapping area" that may exist in traditional pairwise direct calibration.

[0069] Step 5: Based on the extrinsic parameters of the coordinate system, transform each lidar on the vehicle to be calibrated to the same coordinate system, and calculate the relative extrinsic parameters between any two lidars to be calibrated.

[0070] This invention provides a multi-lidar extrinsic parameter calibration method based on panoramic lidar assistance. This method uses a panoramic lidar as an intermediate coordinate system to achieve extrinsic parameter calibration between two target lidars. This technical solution does not rely on overlapping fields of view between lidars, and is a novel multi-lidar calibration method with stronger versatility and accuracy assurance, meeting the needs of fast, reliable, and high-precision calibration of multi-lidar systems in practical applications.

[0071] Example 3:

[0072] This invention provides a multi-lidar extrinsic parameter calibration method based on panoramic lidar assistance, applicable to intelligent driving vehicles, such as... Figure 1 As shown, the method includes the following steps.

[0073] Step 1: Select a location with obvious structural features as the calibration scene;

[0074] Step 2: Place the vehicle to be calibrated in the calibration scene, record the point cloud data C1 of the multiple lidars on the vehicle that need to be calibrated for external parameters, and store them;

[0075] Step 3: After the vehicle leaves the calibration scene, place the panoramic LiDAR device in the same position as the calibration scene, record the point cloud data C2 of the panoramic LiDAR, and store it.

[0076] Step 4: Play back the recorded point cloud data C1 and C2 simultaneously. Using the point cloud registration algorithm, register each point cloud of C1 with the point cloud of C2 in turn to obtain the external parameters of the lidar of the vehicle to be calibrated to the auxiliary radar coordinate system.

[0077] Step 5: Based on the extrinsic parameters of the coordinate system, transform each lidar on the vehicle to be calibrated to the same coordinate system, and calculate the relative extrinsic parameters between any two lidars to be calibrated.

[0078] Step 4 obtains the extrinsic parameters of all the radar to be calibrated, including the panoramic lidar C (such as...). , This step utilizes the transitivity of coordinate system transformations to indirectly calculate the relative extrinsic parameters between any two radars to be calibrated (such as A and B), eliminating the need for direct registration and solving the problem of insufficient overlap between pairs of radars. In three-dimensional space, the transformation between coordinate systems satisfies the "compositeness": if there exist three coordinate systems X, Y, and Z, and the coordinates are known... (Transformation from X to Z) and (Transformation from Y to Z), then the transformation from X to Y can be derived using a formula. In this invention, the panoramic lidar C is the "intermediate coordinate system Z," and the relative extrinsic parameters of any two lidars A and B to be calibrated are... can be and It is derived. For example... Figure 7 As shown, step 5 includes the following steps.

[0079] Step 51, according to the function Obtain the rotation matrix from the panoramic LiDAR C to the first LiDAR B of the vehicle to be calibrated, where, This is the rotation matrix from the panoramic lidar C to the second lidar B of the vehicle to be calibrated. It is a translation vector;

[0080] The inverse transformation represents the transformation from panoramic LiDAR C to LiDAR B, and its calculation rule is: the inverse of the rotation matrix is ​​equal to its transpose. (Because the rotation matrix is ​​an orthogonal matrix); Inverse transformation of the translation vector: .

[0081] Step 52: Based on the following function, obtain the transformation matrix from the first lidar A of the vehicle to be calibrated to the second lidar B of the vehicle to be calibrated. ,

[0082] ,

[0083] , ,

[0084] in, Let be the transformation matrix from the first lidar A of the vehicle to be calibrated to the second lidar B of the vehicle to be calibrated. This is the rotation part of the transformation matrix. This refers to the translation component in the transformation matrix;

[0085] Step 53, based on all lidar extrinsic parameters of the vehicle to be calibrated Repeat steps 51 and 52 to obtain the relative extrinsic parameters between any two lidars. .

[0086] If there are n radars to be calibrated, n external parameters can be obtained through step 4. Using the formulas derived above, the extrinsic parameters between any two pairs can be calculated in batches. This process requires no additional point cloud registration and is completed solely through matrix operations, making it extremely efficient (especially suitable for multi-radar systems, such as the calibration of 5-10 radars in autonomous vehicles).

[0087] Step 54, for relative extrinsic parameters Perform reprojection verification, symmetry verification, and consistency verification to avoid error accumulation.

[0088] Indirectly calculated extrinsic parameters need to be verified through physical meaning to avoid error accumulation, including: reprojection verification: reprojecting the point cloud of radar A through... After transformation, direct registration with the point cloud of radar B (e.g., ICP) is performed. If the average distance between the two is < 0.05m, then TA→B is reliable. Symmetry verification: Calculation and The consistency of the transformation logic must be verified by satisfying TB→A=(TA→B)−1. Multi-mediator consistency: If conditions permit, multiple third-party radars (different brands or locations can be finely adjusted) can be placed in the same parking space to calculate multiple sets of external parameters. If the difference is less than 0.03m, the result is stable.

[0089] This invention provides a multi-lidar extrinsic parameter calibration method based on panoramic lidar assistance. This method uses a panoramic lidar as an intermediate coordinate system to achieve extrinsic parameter calibration between two target lidars. This technical solution does not rely on overlapping fields of view between lidars, and is a novel multi-lidar calibration method with stronger versatility and accuracy assurance, meeting the needs of fast, reliable, and high-precision calibration of multi-lidar systems in practical applications.

[0090] In summary, this invention provides a multi-lidar extrinsic parameter calibration method based on panoramic lidar-assisted calibration. This method is highly versatile, does not require overlapping fields of view between the two lidars, and can be used in various installation locations and orientations. Utilizing the wide-range perception capability of a 360-degree lidar, it can cover a large area of ​​feature points, improving matching robustness and achieving high calibration accuracy. It does not rely on manual settings such as calibration boards or reflectors, reducing deployment complexity. Only two ROSBag data recordings are required, combined with conventional registration algorithms, to complete the calibration.

[0091] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for calibrating the extrinsic parameters of multiple lidar systems based on panoramic lidar assistance, applied to intelligent driving vehicles, characterized in that, The method includes: Step 1: Select a location with obvious structural features as the calibration scene; Step 2: Place the vehicle to be calibrated in the calibration scene, record the point cloud data C1 of the multiple lidars on the vehicle that need to be calibrated for external parameters, and store them; Step 3: After the vehicle leaves the calibration scene, place the panoramic LiDAR device in the same position as the calibration scene, record the point cloud data C2 of the panoramic LiDAR, and store it. Step 4: Play back the recorded point cloud data C1 and C2 simultaneously. Preprocess the point cloud data C1 and C2 to eliminate unstructured differences and retain core structural features. Based on the characteristics of point cloud data C1 and C2, adjust the parameters of the NDT point cloud registration algorithm. Using the adjusted point cloud matching algorithm, output the transformation matrix for each LiDAR X on the vehicle to be calibrated. Its form is: ,in, It is radar Switch to panoramic LiDAR The rotation matrix, It is a translation vector; verify the transformation matrix. The registration effect is to ensure the accuracy of the extrinsic parameters of the lidar of the vehicle to be calibrated to the auxiliary radar coordinate system, thereby obtaining the extrinsic parameters of the lidar of the vehicle to be calibrated to the auxiliary radar coordinate system; Step 5, according to the function Obtain the rotation matrix from the panoramic LiDAR C to the first LiDAR B of the vehicle to be calibrated, where, This is the rotation matrix from the panoramic lidar C to the second lidar B of the vehicle to be calibrated. As the translation vector, the transformation matrix from the first lidar A to the second lidar B of the vehicle to be calibrated is obtained according to the following function. , = , , ,in, Let be the transformation matrix from the first lidar A of the vehicle to be calibrated to the second lidar B of the vehicle to be calibrated. This is the rotation part of the transformation matrix. The translation component in the transformation matrix is ​​based on the extrinsic parameters of all lidar sensors for the vehicle to be calibrated. Repeat this step to obtain the relative extrinsic parameters between any two lidars of the vehicle to be calibrated. Furthermore, it performs reprojection verification, symmetry verification, and consistency verification to avoid error accumulation.

2. The method for calibrating the extrinsic parameters of a multi-laser radar based on panoramic lidar assistance according to claim 1, characterized in that, Step 2 further includes: using a robot operating system to generate and collect point cloud data of each lidar to be calibrated at the same time.

3. The method for calibrating the extrinsic parameters of a multi-laser radar based on panoramic lidar assistance according to claim 1, characterized in that, Step 3 also includes: collecting panoramic lidar point cloud data.

4. The method for calibrating the extrinsic parameters of a multi-laser radar based on panoramic lidar assistance according to claim 1, characterized in that, The method for preprocessing point cloud data C1 and C2 includes: Step 411: Apply statistical filtering to both C1 and C2 to remove outliers caused by radar measurement noise. Step 412: Reduce point cloud density by voxel filtering to reduce computational load; Step 413: Manually trim and retain the rigid structural area, and remove the edge areas that are easily affected by the installation position; Step 414: Perform time synchronization verification on C1 to ensure that its point cloud corresponds to the same state of the scene and is free from interference from dynamic objects.

5. The method for calibrating the extrinsic parameters of a multi-laser radar based on panoramic lidar assistance according to claim 1, characterized in that, The method for adjusting the parameters of the NDT point cloud registration algorithm based on the characteristics of point cloud data C1 and C2 includes: Step 421: Increase the mesh resolution of C2 in order to adapt C1 through dynamic mesh; Step 422: Set the initial transformation based on the position of the lidar of the vehicle to be calibrated. This avoids the point cloud registration algorithm getting trapped in local optima; Step 423: Set the convergence threshold and number of iterations to improve accuracy.

6. The method for calibrating multiple lidar extrinsic parameters based on panoramic lidar-assisted calibration according to claim 5, characterized in that, The convergence threshold is the change in the objective function, and the number of iterations is 500.

7. The method for calibrating the extrinsic parameters of a multi-laser radar based on panoramic lidar assistance according to claim 1, characterized in that, The verification transformation matrix Methods for achieving registration results include: Step 441, calculate the transformed lidar Point cloud and panoramic LiDAR The average distance of the point cloud in the overlapping region; Step 442: The ICP point cloud registration algorithm is used and compared with the NDT point cloud registration algorithm. If the difference between the two is less than the preset value T1, the registration effect passes the consistency verification. Step 443: If the registration deviation between the point cloud of each LiDAR X and the panoramic LiDAR C is less than the preset value T2, then the registration effect is verified through multiple frames.

8. The method for calibrating multiple lidar extrinsic parameters based on panoramic lidar-assisted calibration according to claim 7, characterized in that, The values ​​are T1=0.01m and T2=0.03m.

Images