阅读说明：本技术 一种基于无重叠视场的多激光雷达的联合标定方法 (Multi-laser radar combined calibration method based on non-overlapping view field ) 是由 骆嫚 凌家武 王科未 曹恺 于 2021-08-12 设计创作，主要内容包括：本发明提供了一种基于无重叠视场的多激光雷达的联合标定方法,包含：步骤一、激光雷达与GNSS系统的联合标定,可以获取到各激光雷达标定的初始参数；步骤二、各激光雷达之间的联合标定,通过NDT配准其他激光雷达部分精确的点云数据与主激光雷达的建图点云数据完成各激光雷达之间的标定；步骤三、采用低频的在线标定校准方式,在符合标定场景时自动触发步骤一和二的执行。本发明能够实现将不同安装位置的且没有视场交叉或者点云密度较低的多激光雷达完成在线坐标系的统一标定,满足了激光雷达对初始参数的高要求,提高了标定精度。(The invention provides a combined calibration method of multiple laser radars based on non-overlapping fields of view, which comprises the following steps: step one, joint calibration of a laser radar and a GNSS system can obtain initial parameters calibrated by the laser radars; step two, joint calibration among all laser radars, namely, registering partial accurate point cloud data of other laser radars and the mapping point cloud data of the main laser radar through NDT to finish calibration among all laser radars; and step three, adopting a low-frequency online calibration and calibration mode, and automatically triggering the execution of the step one and the step two when the calibration scene is met. The invention can realize the unified calibration of the online coordinate system by a plurality of laser radars which have different installation positions and no view field intersection or have lower point cloud density, meets the high requirement of the laser radars on the initial parameters and improves the calibration precision.)

1. A joint calibration method of multiple laser radars based on non-overlapping fields of view is characterized by comprising the following steps:

step one, joint calibration of a laser radar and a GNSS system can obtain initial parameters calibrated by the laser radars;

step two, joint calibration among all laser radars, namely, registering partial accurate point cloud data of other laser radars and the mapping point cloud data of the main laser radar through NDT to finish calibration among all laser radars;

and step three, adopting a low-frequency online calibration and calibration mode, and automatically triggering the execution of the step one and the step two when the calibration scene is met.

2. The method of claim 1, wherein the radar is an embedded lidar.

3. The method of claim 1, wherein the step one is to obtain the lidar tracks and local tracks of the GNSS at that time by using a laser odometer method, and then to match the lidar tracks and the GNSS tracks by optimizing a loss function of the reprojection error to obtain the angle external reference relationship between each lidar and the GNSS.

4. The method as claimed in claim 3, wherein in the second step, SLAM mapping is performed by using the master laser radar, and then NDT registration is performed by selecting partial accurate point cloud data of other laser radars and mapping point cloud data of the master laser radar, so as to complete calibration between the laser radars.

5. The method of claim 4, wherein step three comprises: when the external environment reaches the setting condition of the sensing system on the calibration scene, the system automatically starts the analysis of each laser radar data under the preset frequency, and re-registers the mapping point cloud of the main laser radar and the matching point cloud data of other laser radars on the basis of the previous calibration to obtain a new calibration result; and comparing the two results, and saving the better result as a new calibration result.

6. The method of claim 5, wherein the calibration scenario selects a calibration field in an environment with pronounced vertical features for enabling offline batch calibration and calibration of mass production vehicle lidar.

7. An automatic combined calibration method based on multiple embedded laser radars is characterized by comprising the following steps:

step 1, selecting a main laser radar;

step 2, single laser radar SLAM mapping based on feature point matching is carried out, a Lego-SLAM mode is adopted, and simultaneously odometer pose output at the mapping time and corresponding pose data of combined navigation are recorded

Step 3, calibrating the pose of the laser radar and the GNSS coordinate system based on track matching: re-projecting the pose of each laser radar odometer acquired in the step (2) to a GNSS coordinate system to obtain the pose transformation of each laser radar relative to the GNSS coordinate system so as to form relative position initial parameters among the laser radars;

step 4, performing external parameter fine calibration of the multi-laser radar based on point cloud registration: and (3) performing point cloud registration conversion on the odometer information of each laser radar in the step (2) relative to the main laser radar to obtain the fine calibration of other laser radars relative to the main laser radar.

8. The method of claim 7, further comprising time synchronization between the multi-lidar and the GNSS prior to step 1.

9. The method of claim 7, further comprising, after step 4, step 5, low frequency online adaptive calibration: when the external environment reaches the setting condition of the sensing system on the calibration scene, triggering on-line automatic calibration and calibration under the preset frequency;

the system automatically starts data analysis of each laser radar and GNSS system in the scene, and repeats the steps 2-4 on the basis of the previous calibration result to obtain a new calibration result; comparing the two results, and saving the better result as a new calibration result;

and selecting a calibration field with obvious vertical characteristics in the environment according to the calibration scene.

10. The method of claim 7, wherein the step 2 comprises:

step 2.1, laser radar point cloud data preprocessing and feature extraction: preprocessing each frame of received point cloud, screening abnormal points according to the point curvature, sorting the point cloud, ordering the laser point cloud into a distance map, extracting and distinguishing remarkable laser points into an edge point set Fe and a plane point set Fp on the distance map according to the change characteristics of the local curvature of adjacent laser point clouds, and constructing an accumulated point cloud map containing plane points and edge points;

step 2.2, laser radar external parameter optimization: optimizing the external ginseng by adopting an L-M two-step method;

step 2.3, establishing a graph and carrying out closed-loop detection: when the map is closed-loop, the map optimization of the overall pose is carried out once, the accumulated error of the laser odometer is eliminated, and thus the corresponding pose of the single laser radar point cloud and the odometer are respectively obtained.

Technical Field

The invention relates to the technical field of automatic driving of automobiles, in particular to a multi-laser-radar combined calibration method based on non-overlapping fields of view.

Background

In the existing automatic driving process, a laser radar is generally used for effectively identifying obstacles around a vehicle. However, the solution of the 360-degree all-directional mechanical laser radar on the market is expensive, and the detection range of the laser radar is affected by the installation position and has a blind area. The embedded laser radar has more advantages in performance and service life due to the mechanical structure characteristics of the embedded laser radar. With the popularization of embedded laser radars, the combined splicing and use of multiple laser radars is gradually becoming a new trend in the automatic driving design.

However, each lidar has its own independent coordinate system, and therefore, in the process of combined use, the coordinate systems of a plurality of lidar need to be calibrated in a unified manner. The traditional calibration plate has low calibration efficiency and low precision, and requires that the field of view overlap regions exist among all laser radars. However, in a general automatic calibration method, which acquires a point cloud map of a laser radar and then performs registration on each point cloud, the requirement on initial parameters of the laser radar is high, otherwise, the precision of a calibration result is low.

Therefore, it is highly desirable to improve obstacle recognition in existing autonomous driving to meet the requirements of cost saving, efficient calibration, and accurate precision.

Disclosure of Invention

The invention aims to overcome the defects or the improvement requirements in the prior art, and provides an automatic combined calibration method based on an embedded laser radar.

In order to achieve the above object, the present invention provides a joint calibration method for multiple lidar based on non-overlapping fields of view, which is characterized by comprising the following steps:

step one, joint calibration of a laser radar and a GNSS system can obtain initial parameters calibrated by the laser radars;

step two, joint calibration among all laser radars, namely, registering partial accurate point cloud data of other laser radars and the mapping point cloud data of the main laser radar through NDT to finish calibration among all laser radars;

and step three, adopting a low-frequency online calibration and calibration mode, and automatically triggering the execution of the step one and the step two when the calibration scene is met.

Preferably, the radar is an embedded lidar.

And step one, acquiring a laser radar track and a current local track of the GNSS by adopting a laser odometer method, and then matching the laser radar track and the GNSS track by optimizing a loss function of a reprojection error to obtain an external reference relation of angles between each laser radar and the GNSS.

And secondly, performing SLAM mapping by adopting the main laser radar, and then selecting partial accurate point cloud data of other laser radars and mapping point cloud data of the main laser radar to perform NDT registration to finish calibration of the laser radars.

The third step comprises: when the external environment reaches the setting condition of the sensing system on the calibration scene, the system automatically starts the analysis of each laser radar data under the preset frequency, and re-registers the mapping point cloud of the main laser radar and the matching point cloud data of other laser radars on the basis of the previous calibration to obtain a new calibration result; and comparing the two results, and saving the better result as a new calibration result.

The calibration scene selects a calibration field with obvious vertical characteristics in the environment, and is used for realizing offline batch calibration and calibration of the laser radar of the mass production vehicle.

Further, the invention also provides an automatic combined calibration method based on the multiple embedded laser radars, which comprises the following steps:

step 1, selecting a main laser radar;

step 2, single laser radar SLAM mapping based on feature point matching is carried out, a Lego-SLAM mode is adopted, and simultaneously odometer pose output at the mapping time and corresponding pose data of combined navigation are recorded

Step 3, calibrating the pose of the laser radar and the GNSS coordinate system based on track matching: re-projecting the pose of each laser radar odometer acquired in the step (2) to a GNSS coordinate system to obtain the pose transformation of each laser radar relative to the GNSS coordinate system so as to form relative position initial parameters among the laser radars;

step 4, performing external parameter fine calibration of the multi-laser radar based on point cloud registration: and (3) performing point cloud registration conversion on the odometer information of each laser radar in the step (2) relative to the main laser radar to obtain the fine calibration of other laser radars relative to the main laser radar.

Step 1 is preceded by a time synchronization between the multi-lidar and the GNSS.

Step 5, low-frequency online adaptive calibration and calibration are also included after step 4: and when the external environment reaches the setting condition of the sensing system on the calibration scene, triggering the online automatic calibration under the preset frequency. The system automatically starts data analysis of each laser radar and GNSS system in the scene, and repeats the steps 2-4 on the basis of the previous calibration result to obtain a new calibration result; and comparing the two results, and saving the better result as a new calibration result.

The calibration scene selects a calibration field with obvious vertical characteristics in the environment, and preferably but not limited to, the forward laser radar is selected as the main laser radar.

Further, the step 2 comprises the following sub-steps: step 2.1, laser radar point cloud data preprocessing and feature extraction: preprocessing each frame of received point cloud, screening abnormal points according to the point curvature, sorting the point cloud, ordering the laser point cloud into a distance map, extracting and distinguishing remarkable laser points into an edge point set Fe and a plane point set Fp on the distance map according to the change characteristics of the local curvature of adjacent laser point clouds, and constructing an accumulated point cloud map containing plane points and edge points; step 2.2, laser radar external parameter optimization: optimizing the external ginseng by adopting an L-M two-step method; step 2.3, establishing a graph and carrying out closed-loop detection: when the map is closed-loop, the map optimization of the overall pose is carried out once, the accumulated error of the laser odometer is eliminated, and thus the corresponding pose of the single laser radar point cloud and the odometer are respectively obtained.

The invention designs a multi-laser radar combined calibration method based on non-overlapping view fields, which comprises the steps of carrying out local SLAM mapping on each laser radar under a corresponding environment scene, generating laser mileage poses, matching the laser mileage poses with the poses of GNSS and IMU combined navigation collected at the same time, obtaining initial position parameters of the multi-laser radar relative to a GNSS, estimating accurate external parameters through point cloud registration of the laser radars, and synchronously carrying out display interface visual calibration. The method is suitable for combination splicing of various laser radars, and meets the high requirement of initial parameters with minimum cost by the means, so that the calibration efficiency and accuracy are improved.

Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

Drawings

The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the example serve to explain the principles of the invention and not to limit the invention.

Fig. 1 is a schematic flow chart of joint calibration of multiple lidar according to an embodiment of the present invention;

FIG. 2 is a schematic top view of a vehicle lidar arrangement provided by an embodiment of the invention;

fig. 3 is a schematic flow chart of a single lidar (single laser odometer) map creation according to an embodiment of the present invention.

Detailed Description

For the purpose of making the present invention more comprehensible, and for the purpose of making the present application more comprehensible, embodiments and advantages thereof, the present invention will be further described with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not to be construed as limiting the invention in all respects. It should be noted that the embodiments and features of the embodiments in the present application may be arbitrarily combined with each other without conflict.

The invention provides an automatic combined calibration method of multiple laser radars based on non-overlapping fields of view, in particular to an embedded laser radar. The method mainly comprises two main links of joint calibration of the laser radar and the GNSS system and joint calibration among the laser radars, wherein initial parameters calibrated by the laser radars can be obtained through the joint calibration of the laser radar and the GNSS system; and (4) registering the accurate point cloud data of other laser radars and the mapping point cloud data of the main laser radar through NDT to complete calibration of the laser radars. Described below, respectively:

s1, for the combined calibration of the laser radar and the GNSS system, the laser radar track and the current local track of the GNSS are obtained by adopting a laser odometer method, and then the laser radar track and the GNSS track are matched by optimizing a loss function of a reprojection error, so that the external reference relation of the angles between the laser radars and the GNSS is obtained.

And S2, for the joint calibration among the laser radars, drawing by adopting a main laser radar SLAM, and then selecting partial accurate point cloud data of other laser radars and drawing point cloud data of the main laser radar to carry out NDT registration to finish the calibration among the laser radars.

On the basis of the two links, a low-frequency online calibration mode is further adopted, and when a calibration scene is met, the registration calibration of each laser radar is triggered to be started, and the method comprises the following steps: when the external environment reaches the setting condition of the sensing system on the calibration scene, the system automatically starts the analysis of each laser radar data under a certain frequency, and re-registers the mapping point cloud of the main laser radar and the matching point cloud data of other laser radars on the basis of the previous calibration to obtain a new calibration result; and comparing the two results, and saving the better result as a new calibration result.

The calibration scene selects a calibration field with obvious vertical characteristics in the environment, and offline batch calibration and calibration of the laser radar of mass production vehicles can be realized.

As shown in fig. 1, which is a schematic flow chart of a multi-lidar joint calibration method according to an embodiment of the present invention, before the main calibration flow shown in the figure is performed, time synchronization between the multi-lidar and the GNSS needs to be ensured. After time synchronization, the main calibration flow comprises the following steps:

and step 1, selecting a main laser radar. As an embodiment, the multiple lidar employs a plurality of embedded radars, and a vehicle is used as a carrier, wherein the arrangement effect diagram of the lidar is shown in fig. 2. In fig. 2, the embedded lidar is subject to technical limitations, with field angles typically not exceeding 120 °, where there is no intersection of field angles between front and rear lidar. The position relation among all laser radars belongs to rigid body transformation. Preferably, the calibration method of the present invention preferably, but not limited to, selects the forward lidar as the primary lidar.

Step 2, performing single laser radar SLAM graph building based on feature point matching: when the setting condition of a calibration scene is met, data acquisition of each laser radar and GNSS system is triggered, a local map is built by SLAM for each laser radar data, a Lego-SLAM mode is adopted, and simultaneously odometer pose output at the time of map building and corresponding combined navigation pose data are recorded.

As an embodiment, the flow of mapping by the single laser radar SLAM is shown in fig. 3, and includes the following sub-steps:

step 2.1, laser radar point cloud data preprocessing and feature extraction: preprocessing each frame of received point cloud, screening abnormal points according to the point curvature, sorting the point cloud, ordering the laser point cloud into a distance map, extracting and distinguishing remarkable laser points into an edge point set Fe and a plane point set Fp on the distance map according to the change characteristics of the local curvature of adjacent laser point clouds, and accordingly constructing an accumulated point cloud map containing plane points and edge points.

Step 2.2, laser radar external parameter optimization: optimizing the external ginseng by adopting an L-M two-step method.

Firstly, matching is carried out through feature plane point sets at t and t-1 moments of two continuous frames to obtain (in six degrees of freedom) of optimized laser radar) The calculation formula is as follows:

；

matching is carried out through characteristic edge point sets of two continuous frames at the time of t and t-1 and the determined () constraint to obtain three other freedom degree parameters (a)) The calculation formula is as follows:

；

and obtaining the laser odometer track after the transformation between frames is obtained.

Step 2.3, establishing a graph and carrying out closed-loop detection: when the map is closed-loop, the map optimization of the overall pose is carried out once, the accumulated error of the laser odometer is eliminated, and thus the corresponding pose of the single laser radar point cloud and the odometer are respectively obtained.

Step 3, calibrating the pose of the laser radar and the GNSS coordinate system based on track matching: the pose of each laser radar odometer obtained in the step 2 is based on the unit array starting point of the laser radar at the initial moment, pose deviation exists, the pose of the laser radar at the moment corresponding to the GNSS time synchronization is obtained, and the track output by the laser radar is re-projected to the GNSS coordinate system, so that the pose transformation (namely the external reference transformation between the track of the laser odometer and the track of the GNSS at that time) of each laser radar relative to the GNSS coordinate system can be obtained:

，

therefore, coordinate transformation of each laser radar relative to the same integrated navigation system is obtained, and initial parameters of relative positions among the laser radars are formed.

Step 4, performing external parameter fine calibration of the multi-laser radar based on point cloud registration: according to the parameters of each laser radar relative to the GNSS obtained in the above steps, the external parameters of each laser radar relative to other laser radars can be calculated and obtained, and the external parameters are used as rough parameters (external parameter rough calibration); and (3) performing point cloud registration conversion on the odometer information of each laser radar in the step (2) relative to the main laser radar to obtain the fine calibration of other laser radars relative to the main laser radar.

The method comprises the following steps of (1) acquiring point cloud data of other laser radars according to step 2, carrying out NDT change on the selected point cloud data of about 50m of the laser radars and a point cloud map of a main laser radar, and optimizing an external parameter to minimize the projection error of NDT conversion so as to obtain a relatively accurate external parameter; on the basis of the external parameters, the ICP is registered, so that the external parameter result is more accurate.

Step 5, low-frequency online self-adaptive calibration and calibration: and when the external environment reaches the setting condition of the sensing system on the calibration scene, triggering the online automatic calibration under the preset frequency. The system automatically starts data analysis of each laser radar and GNSS system in the scene, and repeats the steps 2-4 on the basis of the previous calibration result to obtain a new calibration result; and comparing the two results, and saving the better result as a new calibration result.

And selecting a calibration field with obvious vertical characteristics in the environment according to the calibration scene.

Further, the master lidar point cloud may be calibrated manually for accuracy. After the accurate external parameters are obtained, all the laser radars are spliced through the external parameters and visualized to check the accuracy, and then matching and positioning can be carried out through the spliced point cloud.

The method can realize offline batch calibration and calibration of the laser radars of mass production vehicles, can finish unified calibration of an online coordinate system by a plurality of laser radars which have different installation positions and do not have field-of-view intersection or have low point cloud density, is suitable for combination and splicing of various laser radars, meets the high requirement of initial parameters with minimum cost, and further improves the accuracy.

Although the embodiments of the present invention have been described above, the above description is only for the convenience of understanding the present invention, and is not intended to limit the present invention. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the scope of the present invention should be determined by the following claims.