Front and rear radar external parameter calibration method and device suitable for differential drive AGV

Abstract

The application relates to a front-rear radar external parameter calibration method and equipment suitable for a differential driving AGV. The front-rear radar external parameter calibration method comprises the following steps: acquiring radar data containing line feature information in the AGV movement process, generating a first initial transformation matrix, performing registration processing on the radar data, and solving a plurality of transformation matrices corresponding to different time points; extracting pose information from the transformation matrices to solve a front radar external parameter; obtaining a global point cloud map according to the corresponding pose information of the front radar, registering the rear radar data into the global point cloud map for iterative optimization to obtain a rear radar external parameter; and outputting the front radar external parameter and the rear radar external parameter. The front-rear radar external parameter calibration method and equipment suitable for the differential driving AGV can simultaneously calibrate the front and rear radar external parameters, and improve the multiplexing of the rear radar.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mobile robots, and particularly relates to a front-rear radar external parameter calibration method and device suitable for a differential drive AGV. BACKGROUND

[0002] In the prior art, when calibrating the external parameters of the laser radar of a differential drive AGV (AGV), most of the schemes adopt separate calibration of the front laser radar and the rear laser radar, or directly perform manual external parameter calibration according to mechanical values. In the scheme of separately calibrating the external parameters of the front laser radar and the rear laser radar, the AGV is usually placed in a structured environment, and original laser radar data is obtained through a reference object or a geometric structure set by a person; such a method usually adopts a traditional point-to-point iterative closest point matching algorithm (PP-ICP).

[0003] However, the above two calibration schemes have many limitations. First, separate calibration of the front and rear radars is time-consuming and labor-intensive; separate calibration of the external parameters of the front and rear laser radars not only has low calibration efficiency, but more significantly, it also reduces the consistency of the external parameters of the front and rear laser radars, so that the rear laser radar can only be used as an auxiliary obstacle avoidance laser, and if the front and rear laser radars are used simultaneously for positioning, the positioning accuracy will be reduced. Because of the poor accuracy of the external parameters of the rear radar and the low registration accuracy, the existing system usually only uses the rear radar for obstacle avoidance function, and is not suitable for positioning or mapping tasks. Because the function multiplexing of the rear laser radar is limited, in some large-scale, heavily occluded or highly degraded scenes, a single front radar may not be able to guarantee the stability of positioning, and if the positioning robustness is to be enhanced, a large number of auxiliary facilities such as reflectors need to be added, resulting in complex deployment and rising costs.

[0004] In addition, in order to obtain original laser radar data with obvious features and more accurate matching results, the existing calibration schemes are generally selected to be performed in a structured environment. However, in the registration process of point cloud data, only the Euclidean distance of points is considered, and the line feature information commonly existing in structured environments, such as straight lines and walls, is ignored.

[0005] Therefore, how to improve the consistency of the calibration of the external parameters of the front and rear laser radars under the premise of controlling costs, so that the rear laser radar can fully participate in the positioning and mapping tasks, has become a technical problem that needs to be solved in the field. SUMMARY

[0006] To solve the problems in the prior art, the present application provides a front-rear radar external parameter calibration method and device suitable for a differential drive AGV, which calibrates the front and rear radars simultaneously by fully exploiting and utilizing the line feature information in the environment, so as to ensure the consistency of the external parameters of the front and rear radars.

[0007] According to an embodiment of the present application, in one aspect, a front and rear radar extrinsic parameter calibration method suitable for a differential drive AGV is provided, and the method comprises the following steps:

[0008] S1: In the process of the AGV performing a first action, front radar first laser data containing first line feature information is acquired; a first initial transformation matrix is generated according to the mechanical extrinsic parameters of the front radar and first action parameters; the front radar first laser data is registered based on the first line feature information and the first initial transformation matrix, and a plurality of first transformation matrices corresponding to different time points are solved, wherein the first transformation matrices contain the pose information of the front radar at the corresponding time points;

[0009] S2: In the process of the AGV performing a second action, front radar second laser data containing second line feature information is acquired; a second initial transformation matrix is generated according to the mechanical extrinsic parameters of the front radar and second action parameters; the front radar second laser data is registered based on the second line feature information and the second initial transformation matrix, and a plurality of second transformation matrices corresponding to different time points are solved, wherein the second transformation matrices contain the pose information of the front radar at the corresponding time points;

[0010] S3: In the process of the AGV performing a third action, rear radar laser data containing third line feature information is acquired; a third initial transformation matrix is generated according to the mechanical extrinsic parameters of the rear radar and third action parameters; the rear radar laser data is registered based on the third line feature information and the third initial transformation matrix, and a plurality of third transformation matrices corresponding to different time points are solved, wherein the third transformation matrices contain the pose information of the rear radar at the corresponding time points;

[0011] S4: The pose information is extracted from the first transformation matrix and the second transformation matrix, and the front radar extrinsic parameters are solved;

[0012] S5: The global point cloud map is constructed according to the corresponding pose information of the front radar; the rear radar laser data is registered into the global point cloud map based on the third transformation matrix, and the rear radar mechanical extrinsic parameters are used as initial values for iterative optimization to obtain the rear radar extrinsic parameters;

[0013] S6: The front radar extrinsic parameters and the rear radar extrinsic parameters are output.

[0014] Further, in one or more of steps S1-S3:

[0015] The action parameters include speed information corresponding to the action and time difference information between the current frame and the key frame;

[0016] The initial transformation matrix contains the pose transformation information of the current frame coordinate system relative to the key frame coordinate system.

[0017] In the registration processing of the laser data based on the line feature information and the corresponding initial transformation matrix, a relative transformation matrix applicable to the current frame coordinate system and the key frame coordinate system is obtained first, and then a transformation matrix between the current frame coordinate system and the fixed reference coordinate system is calculated in combination with the key frame transformation matrix and the relative transformation matrix;

[0018] The fixed reference coordinate system is a corresponding coordinate system of the corresponding radar at the start of the action;

[0019] The key frame transformation matrix is used for transformation between the key frame coordinate system and the fixed reference coordinate system;

[0020] The key frame is updated according to a preset threshold condition.

[0021] Further, in one or more of steps S1-S3, in the registration processing of the laser data based on the line feature information and the corresponding initial transformation matrix, the corresponding initial transformation matrix is used as an initial input, the current frame laser data and the key frame laser data are matched and optimized by using a PL-ICP method, and the relative transformation matrix is obtained.

[0022] Further, in one or more of steps S1-S3, in combination with the key frame transformation matrix and the relative transformation matrix, a transformation matrix to be optimized is calculated, and the transformation matrix to be optimized is used as an initial input, the current frame laser data and the laser data corresponding to the fixed reference coordinate system are matched and optimized by using a PL-ICP method, and a final transformation matrix is obtained.

[0023] Further, in step S4, after the pose information is extracted from the transformation matrix, an image is generated according to the pose information, the image is graphically fitted, and the front radar extrinsic parameter is calculated according to the fitting result.

[0024] Further:

[0025] The first action is a rotating action, and the second action is a straight-line action;

[0026] The front radar extrinsic parameter includes a front radar angle extrinsic parameter and a front radar coordinate extrinsic parameter;

[0027] In step S4, after the pose information is extracted from the second transformation matrix, an image containing a straight-line-like graph is generated according to the pose information, a straight-line fitting is performed on the image by using a fitLine function in an OpenCV library to obtain a straight-line tilt angle, and the front radar angle extrinsic parameter is calculated according to the straight-line tilt angle;

[0028] After the pose information is extracted from the first transformation matrix, an image containing a circle-like figure is generated according to the pose information, a rotating rectangle is obtained by performing figure fitting on the image using a fitEllipse function in an OpenCV library, a center coordinate of the rotating rectangle is extracted, and the front radar coordinate extrinsic parameter is calculated according to the center coordinate and the front radar angle extrinsic parameter.

[0029] Further,

[0030] The rear radar extrinsic parameter includes a rear radar angle extrinsic parameter and a rear radar coordinate extrinsic parameter.

[0031] In step S5, the rear radar laser data is registered into the global point cloud map using a PL-ICP method based on a front-rear radar transformation matrix calculated according to the third transformation matrix and the rear radar mechanical extrinsic parameter; when registration is performed, the rear radar angle extrinsic parameter and the rear radar coordinate extrinsic parameter are taken as optimization variables to construct a residual equation, and the rear radar mechanical extrinsic parameter is taken as an initial value, the residual equation is iteratively optimized using a Ceres library to obtain a final rear radar extrinsic parameter.

[0032] Further, in step S5, the front radar first laser data and / or the front radar second laser data are converted from a current frame coordinate system to the fixed reference coordinate system based on the first transformation matrix and / or the second transformation matrix, data points in the front radar first laser data and / or the front radar second laser data are sorted according to angle values in the fixed reference coordinate system, the sorted data points are screened according to a preset step length, and the global point cloud map is generated according to the screened data points.

[0033] Further, the third action is a rotating action, and the first action includes the third action.

[0034] In another aspect, an embodiment of the present application provides a front-rear radar extrinsic parameter calibration device suitable for a differential driving AGV, the front-rear radar extrinsic parameter calibration device including one or more processors, a memory, and one or more programs stored in the memory and configured to be executed by the one or more processors, for executing the front-rear radar extrinsic parameter calibration method described above.

[0035] The technical solution provided by the embodiments of the present application can include the following beneficial effects:

[0036] (a) The front and rear radar extrinsic parameters are simultaneously calibrated by using line features in a calibration environment, the consistency of the front and rear laser radar extrinsic parameters is ensured, the rear radar can participate in positioning, the reuse of the rear radar is improved, and the positioning accuracy is improved.

[0037] (b) The deployment requirement of the calibration environment is low, time-saving and labor-saving; the calibration process is efficient;

[0038] (c) By setting key frames and updating the key frames, the accuracy and convergence speed of the solution are further improved.

[0039] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0041] Figure 1 is a schematic diagram of the steps of the front and rear radar extrinsic parameter calibration method provided by the embodiments of the present application;

[0042] Figure 2 is a schematic diagram of the steps of obtaining the first transformation matrix in the front and rear radar extrinsic parameter calibration method provided by the embodiments of the present application;

[0043] Figure 3 is a schematic diagram of the steps of generating a point cloud map in the front and rear radar extrinsic parameter calibration method provided by the embodiments of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0045] The current differential driving AGV is usually equipped with front and rear laser radars for map construction, autonomous positioning and obstacle avoidance in actual industrial environment. In the prior art, the method of calibrating the front radar and the rear radar respectively is generally adopted. This method is low in efficiency and causes poor consistency of the front and rear radar extrinsic parameters, which further makes the rear radar only used in obstacle avoidance scene. If the rear radar is used in positioning scene, the positioning accuracy will be reduced.

[0046] In order to improve the consistency of the front and rear radar calibration and improve the positioning accuracy, the embodiments of the present application propose a front and rear radar extrinsic parameter calibration method suitable for differential driving AGV. Referring to Figure 1 , the front and rear radar extrinsic parameter calibration method comprises the following steps:

[0047] S1: In the process of the AGV performing a first action, first radar first laser data containing first line feature information is acquired; a first initial transformation matrix is generated according to the mechanical extrinsic parameters of the front radar and first action parameters; the first radar first laser data is registered based on the first line feature information and the first initial transformation matrix, and a plurality of first transformation matrices corresponding to different time points are solved; the first transformation matrix contains the pose information of the front radar at the corresponding time point;

[0048] S2: In the process of the AGV performing a second action, front radar second laser data containing second line feature information is acquired; a second initial transformation matrix is generated according to the mechanical extrinsic parameters of the front radar and second action parameters; the front radar second laser data is registered based on the second line feature information and the second initial transformation matrix, and a plurality of second transformation matrices corresponding to different time points are solved; the second transformation matrix contains the pose information of the front radar at the corresponding time point;

[0049] S3: In the process of the AGV performing a third action, rear radar laser data containing third line feature information is acquired; a third initial transformation matrix is generated according to the mechanical extrinsic parameters of the rear radar and third action parameters; the rear radar laser data is registered based on the third line feature information and the third initial transformation matrix, and a plurality of third transformation matrices corresponding to different time points are solved; the third transformation matrix contains the pose information of the rear radar at the corresponding time point;

[0050] S4: Pose information is extracted from the first transformation matrix and the second transformation matrix to solve the front radar extrinsic parameters;

[0051] S5: The global point cloud map is constructed according to the corresponding pose information of the front radar; the rear radar laser data is registered into the global point cloud map based on the third transformation matrix, and the rear radar mechanical extrinsic parameters are used as initial values for iterative optimization to obtain the rear radar extrinsic parameters;

[0052] S6: The front radar extrinsic parameters and the rear radar extrinsic parameters are output.

[0053] The embodiment can more conveniently and accurately obtain the front radar extrinsic parameters by using the line feature information in the environment, and then obtain the rear radar extrinsic parameters based on the line feature information in the laser data acquired by the front radar, so as to realize the synchronous calibration of the front and rear radars; since the front and rear radar extrinsic parameters are consistent, the rear radar can participate in positioning together, the reuse of the rear radar is improved, good positioning accuracy can still be realized under the premise of controlling the cost, and the calibration efficiency is improved.

[0054] The above various transformation matrices represent the transformation from the current frame coordinate system of the radar to a fixed reference coordinate system, which can be the coordinate system corresponding to the radar at the beginning of the action or before the beginning of the action, or a preset input fixed coordinate system.

[0055] The front and rear radar mechanical parameters are parameters input or set in advance, and are usually related to the mechanical installation structure of the corresponding radar. The mechanical parameters can reflect the installation pose of the radar to a certain extent, but the accuracy is limited, and in some cases can only provide a rough approximation of the pose, or even has little to do with the actual pose. The front and rear radar mechanical parameters are mainly used as initial reference values in the iterative solving process to generate the starting point of iterative optimization, and the front and rear radar parameters finally solved through iterative optimization can accurately represent the actual installation pose of the corresponding radar.

[0056] Referring to Figure 1 and Figure 2 In an embodiment of the present application, in steps S1-S3, to prevent too few overlapping parts of the objects scanned by the two frames of laser during matching, or too large error of the initial relative pose solved, a key frame is introduced as a transition, that is, the transformation from the current frame coordinate system to the key frame coordinate system is solved first, and then the transformation matrix is finally obtained according to the information of the key frame. Specifically, the action parameters include velocity information of the corresponding action and time difference information between the current frame and the key frame; the initial transformation matrix includes pose transformation information of the current frame coordinate system relative to the key frame coordinate system; when the laser data is registered based on the line feature information and the corresponding initial transformation matrix, a relative transformation matrix suitable for the current frame coordinate system and the key frame coordinate system is first obtained, and then the transformation matrix between the current frame coordinate system and the fixed reference coordinate system is calculated by combining the key frame transformation matrix and the relative transformation matrix; the key frame transformation matrix is used for the transformation between the key frame coordinate system and the fixed reference coordinate system; the key frame is updated according to a preset threshold condition, such as frame number counting and screening, or time difference information screening, or reference to the displacement of the current frame relative to the fixed reference coordinate system.

[0057] Obviously, when the fixed reference coordinate system is the coordinate system corresponding to the radar at the beginning of the action, the frame corresponding to the coordinate system is the first key frame, referring to Figure 2 .

[0058] Referring to Figure 1 and Figure 2In an embodiment of the present application, in steps S1-S3, when the laser data is registered based on the line feature information and the corresponding initial transformation matrix, the corresponding initial transformation matrix is used as an initial input, the current frame laser data is matched with the key frame laser data using a point-to-line iterative closest point (PL-ICP) matching method, and the relative transformation matrix is obtained by optimization, and then the required transformation matrix is directly calculated from the relative transformation matrix and the key frame transformation matrix.

[0059] Preferably, after the transformation matrix is calculated by combining the key frame transformation matrix and the relative transformation matrix, the transformation matrix is used as an initial input, the current frame laser data is further matched with the laser data corresponding to the fixed reference coordinate system using the PL-ICP method, and the final output transformation matrix is obtained by optimization.

[0060] The PL-ICP method uses a piecewise linear method to approximate the actual surface represented by the laser data, and uses the distance between the target frame point and the nearest two points in the reference frame to construct an error function. Compared with the PP-ICP method, the error function of the PL-ICP method not only closely approximates the actual situation, but also has higher accuracy and faster convergence speed. The iterative calculation steps of the PL-ICP method are as follows:

[0061] 1. According to the given initial transformation matrix, the target frame data is converted to the reference frame (i.e. the aforementioned key frame) coordinate system;

[0062] 2. For each data point in the target frame, find the two nearest points in the reference frame in the Euclidean distance;

[0063] 3. Remove points with large errors;

[0064] 4. Use point-to-line distance to construct a minimum error equation;

[0065] 5. Continuously iterate and update the transformation matrix;

[0066] By using the PL-ICP method twice, the final output matrix can have excellent accuracy.

[0067] The PL-ICP method is sensitive to the initial value, and the result of the iterative calculation depends on the accuracy of the initial value. In this embodiment, the first, second, and third initial transformation matrices and the first, second, and third transformation matrices are used as initial inputs for the PL-ICP method, which can efficiently obtain accurate calculation results.

[0068] In the embodiment of the present application, a mixed calculation method of PL-ICP and PP-ICP can also be used, or other methods using environmental line features for point cloud data registration.

[0069] In an embodiment of the present application, the first motion is a rotating motion, and the second motion is a straight motion. Obviously, it can also be the opposite, that is, the first motion is a straight motion, and the second motion is a rotating motion, and the rest of the operations can be adjusted accordingly. For the convenience of description, the first motion is a rotating motion, and the second motion is a straight motion in the following description. The first motion includes a third motion, that is, the third motion is the same motion as the first motion, or the third motion is a part of the rotation (generally at least half of it to ensure that the front laser point cloud map can cover the entire calibration environment) in the rotation process of the first motion. The calibration of the rear radar is almost synchronous with the calibration of the front radar, or the front radar is calibrated for a period of time first, and then the front and rear radars are calibrated synchronously. Since the above-mentioned motions all occur in the same environment, the environmental line features corresponding to the first line feature information, the second line feature information, and the third line feature information often have overlapping parts, and in some cases they can even be completely the same.

[0070] The technical solutions of the present application will be described and illustrated in combination with a specific embodiment. It should be noted that the technical features in the specific embodiment should not limit the protection scope of the present application, especially the specific size of the calibration environment, the specific data of the frame number screening, etc., which are only used as examples to help understanding.

[0071] In the calibration process of the present embodiment, the motion mode of rotating one circle in place first and then moving straight for about 50 cm is adopted. When fitting and solving the external parameters of the front laser radar, the rotation center is taken as the center of the vehicle body, so there is a certain requirement for the differential drive of the AGV. The straight distance can be determined according to actual needs. If more data is needed when moving straight, the value can be appropriately increased, and if less data is needed, the value can be appropriately reduced. The value of the distance should not constitute a limitation on the protection scope of the present application.

[0072] The calibration method uses the PL-ICP method for matching. In order to ensure the accuracy of the PL-ICP matching, there are certain requirements for the calibration environment used. The requirements are as follows:

[0073] Calibration environment: AGV must be in a structured environment of appropriate size. On the one hand, selecting an appropriate size of the environment ensures that the AGV will not collide with objects during calibration, and on the other hand, it ensures the accuracy of the original laser radar point cloud data ranging (for laser radar, there is a certain error when the laser radar ranges, and greater errors will occur when the ranging is small or far). Selecting a structured environment is to obtain relatively regular laser radar point clouds, which provides better original data for the PL-ICP method. In this embodiment, the calibration environment used for testing is generally a closed "rectangle" surrounded by flat boards or walls, and the length and width of the rectangle is about 3-4 times the length of the AGV body, and the height of the rectangle is about 2 times the height of the laser radar.

[0074] Differential drive structure: Before calibration, it is necessary to ensure that the AGV can stably move straight, and can stably rotate in place with the center of the vehicle body as the center. That is, during the calibration process, the motion mode of rotating in place for one revolution first and then moving straight for about 50 cm is adopted. Therefore, it is necessary to ensure that the AGV can stably move straight and can stably rotate in place with the center of the vehicle body as the center, so the method of this embodiment is only applicable to AGVs with differential drive.

[0075] The front laser radar external parameter solved by calibration is the transformation of the front laser radar coordinate system (denoted as front_laser) relative to the center of the vehicle body coordinate system (denoted as base), and the rear laser radar external parameter is the transformation of the rear laser radar coordinate system (denoted as back_laser) relative to front_laser. For the front laser radar data in the calibration process, the point-to-line iterative closest point (PL-ICP) is used to solve the relative pose of the current frame and the key frame in rotation and straight movement, respectively. In the rotation action, the rotation angle between the current frame and the key frame is constructed by using the angular velocity and the time difference between frames, and the initial transformation matrix is generated in combination with the mechanical external parameter; in the straight movement action, the translation amount is generated based on the linear velocity, the time difference between frames and the relative position of the front radar and the center of the AGV, and the initial transformation matrix is constructed.

[0076] Referring to Figure 2 , preferably, in order to ensure the coincidence of the laser scanning area in the matching, a key frame selection strategy is set in this embodiment: when the rotation angle of the current frame relative to the key frame exceeds a set threshold (for example, 5°), the current frame is updated to a new key frame.

[0077] Preferably, when using the PL-ICP method for registration, firstly, the current frame laser data is matched with the key frame laser data to obtain a relative transformation matrix of the current frame coordinate system relative to the key frame coordinate system; then, the final transformation matrix of the current frame coordinate system relative to the fixed reference coordinate system is calculated in combination with the transformation relationship of the key frame relative to the fixed reference coordinate system.

[0078] Only the back laser data in the rotation process is processed, and the processing manner is consistent with the processing manner of the front laser data in the rotation process. Still taking front_laser as the fixed reference coordinate system, all the front laser data in the rotation process is converted into this coordinate system, the point cloud data is filtered, and thus a laser radar point cloud map with front_laser as the global coordinate system can be obtained. The transformation of back_laser relative to front_laser is taken as an optimization variable, and the PL-ICP method is used to match and solve the back laser data and the point cloud map, so as to construct a nonlinear optimization, and the result of the nonlinear optimization is the back laser extrinsic parameter.

[0079] The relative pose point set of all the front laser radars obtained in the straight running process is subjected to straight line fitting, and the solved tilt angle is the rotation angle corresponding to the rotation matrix in the transformation matrix from base coordinate system to front_laser coordinate system. The relative pose point set of all the front laser radars obtained in the rotation process is subjected to rotation rectangle solving, and the center coordinate value of the rotation rectangle is the translation corresponding to the transformation matrix from base coordinate system to front_laser coordinate system. The inverse of the transformation matrix can obtain the transformation matrix from front_laser to base coordinate system, and the front laser extrinsic parameter can be obtained.

[0080] More specifically, referring to Figure 2 For the front laser data in the rotation process, the coordinate system corresponding to the front laser at the beginning of the rotation is taken as the fixed reference coordinate system, that is, the front_laser coordinate system. The relative pose of the front laser coordinate system (denoted as front_i) at other time relative to front_laser is matched and solved by using the PL-ICP method. In order to ensure the matching accuracy of PL-ICP, the initial relative pose is solved by means of the front laser coordinate system mechanical extrinsic parameter, the AGV rotation speed and the laser frame time difference. In order to prevent the matching of the two frames of laser from being too small and the initial relative pose solved by using the rotation speed from being too large, the key frame is constantly updated as the reference frame when solving by using PL-ICP, and the coordinate system corresponding to the key frame is denoted as front_key. The initial relative pose transformation matrix relative to front_key is constructed as follows:

[0081] ,

[0082] In the above formula, The angular velocity is the rotational speed in place; AGVs generally rotate counterclockwise. , The time difference between two frames of lidar data. This is the angle value of rotation in place. This represents the mechanical value of the Euclidean distance from the front_laser coordinate system to the base coordinate system. Indicates the first laser The initial transformation matrix from the front_i coordinate system to the front_key key coordinate system. As a solution using PL-ICP Frame laser data and keyframe laser data The initial value for matching results is denoted as the solution result of PL-ICP. This represents the transformation matrix from the front_i coordinate system to the front_key coordinate system. Since keyframes are selected from laser frames, The value is known, and the transformation from front_i to front_laser can be obtained according to the following calculation method:

[0083] ,

[0084] In the above formula The initial transformation from front_i to front_laser, theoretically, should be the final transformation from front_i to front_laser calculated using the above formula. However, this embodiment considers the matching errors between front_key and front_laser, and between front_i and front_key, so the result calculated using the above formula is defined as the new initial transformation. Use this as the initial value for PL-ICP calculation. Laser data corresponding to the front_laser coordinate system The matching result is denoted as The records include All of them And the corresponding laser data.

[0085] For the rotation process, an angle threshold is defined to filter keyframes. In this embodiment, the selected angle threshold is... It is 5°, that is, when the relative transformation matrix of a certain coordinate system relative to the keyframe coordinate system corresponds to... When the angle is greater than 5°, the key frame is updated. The scheme of taking other angles (e.g. 3°, 6°, 10°) as angle thresholds to update the key frame and improve the matching accuracy is also within the protection scope of the technical scheme of the present application.

[0086] For the front laser radar data in the straight running process, the processing idea is basically the same as in the rotating process, and only the initial transformation matrix relative to the key frame coordinate system front_key and the screening method of the key frame are different. For the convenience of understanding, variables with similar meanings are named similarly or identically as in the rotating process. The following is the construction method of the initial relative pose transformation matrix relative to the key frame front_key:

[0087] ;

[0088] In the above formula, is the time difference of two frames of laser radar data, represents the AGV straight running speed, represents the distance of the AGV straight running, represents the mechanical value of the front laser radar angle parameter, which is generally 0, still represents the initial transformation matrix of the jth coordinate system front_j of the front laser to the key frame coordinate system front_key, and front_laser still represents the fixed reference coordinate system of the front laser radar in the straight running process.

[0089] For the straight running process, a distance threshold is defined to screen the key frame. The distance threshold selected by the embodiment of the present application is 0.05 m, that is, when the relative transformation matrix of a certain coordinate system relative to the key frame coordinate system corresponds to a straight running distance greater than 0.05 m, the key frame is updated. In the actual calibration process, the distance threshold is fine-tuned (e.g. 0.1 m, 0.08 m) in combination with the total straight running distance to update the key frame and improve the matching accuracy. For the case of a straight running distance of about 50 cm, the distance threshold value range is generally between 0.02 m and 0.1 m. Too small value may lead to large relative error, and too large value may lead to insufficient data. For different straight running distances, in order to balance error control and sufficient data, the distance threshold value range will also have a corresponding floating. Up to now, all the transformation matrices including

[0090] are recorded. .

[0091] ​​For the back laser data in the rotation process, the coordinate system corresponding to the back laser at the beginning of the rotation is taken as the fixed reference coordinate system, i.e., the back_laser coordinate system. The relative pose of the back laser coordinate system at other time points (denoted as back_i) relative to the back_laser is solved by using the PL-ICP method. In order to ensure the matching accuracy of the PL-ICP, the initial relative pose is solved by means of the Euclidean distance mechanical value of the back_laser relative to the base coordinate system, the rotation speed of the AGV, and the time difference of the laser frame. In order to prevent the matching of the two frames of laser from being too small and the initial relative pose solved by using the rotation speed from being too large, the key frame is constantly updated as the reference frame when solving the PL-ICP, and the coordinate system corresponding to the key frame is denoted as back_key.

[0092] The processing method of the back laser data in the rotation process is similar to the processing method of the front laser data in the rotation process, which will not be described here again. The relative pose of the i-th frame of back laser data relative to the fixed reference coordinate system back_laser in the rotation process is denoted as The relative pose of the i-th frame of back laser data relative to the fixed reference coordinate system back_laser in the rotation process is denoted as and the corresponding laser data.

[0093] In an embodiment of the present application, after the pose information is extracted from the transformation matrix in step S4, an image is generated according to the pose information, the image is graphically fitted, and the front radar external parameter is calculated according to the fitting result.

[0094] The front radar external parameter includes a front radar angle external parameter and a front radar coordinate external parameter.

[0095] Specifically, in a specific embodiment, after the pose information is extracted from the second transformation matrix in step S4, an image containing a straight line-like graph is generated according to the pose information, a straight line fitting is performed on the image using the fitLine function in the OpenCV library to obtain a straight line tilt angle, and the front radar angle external parameter is calculated according to the straight line tilt angle. After the pose information is extracted from the first transformation matrix, an image containing a circle-like graph is generated according to the pose information, a graph fitting is performed on the image using the fitEllipse function in the OpenCV library to obtain a rotated rectangle, the center coordinates of the rotated rectangle are extracted, and the front radar coordinate external parameter is calculated according to the center coordinates and the front radar angle external parameter.

[0096] ​​That is, in the process of actually solving the front radar external parameter, the pose information is extracted from the plurality of transformation matrices obtained from the above two actions. Based on the pose translation amount extracted in the second action (straight running), the point cloud trajectory is drawn in the OpenCV image and the fitLine function is used for linear fitting, so as to solve the front radar angle external parameter; based on the pose translation amount extracted in the first action (rotation), the trajectory ring chart is drawn, and the fitEllipse function is used for fitting to obtain the ellipse center, and then the front laser radar coordinate external parameter is calculated.

[0097] Still in combination with the above specific embodiments, the front laser radar external parameter is solved, and is respectively denoted as , the angle external parameter is solved first , and then the coordinate external parameter is solved . With the aid of OpenCV, a three-channel white image matrix with the same width and height is created. The corresponding translation amount (denoted as ) in all matrices is taken out, all the translation amounts are drawn into the image matrix with 0.005 as the resolution and half of the width as the offset value. When , the identity matrix , so the first point on the image matrix is at the center of the image matrix. The image matrix used in the embodiments of the application is a white image matrix with a width and height of 1000. The data points in the image matrix are close to a straight line, and the fitLine function in the OpenCV library is used to perform linear fitting on the data points on the image matrix, and the inclination angle of the straight line is calculated according to the fitting result, denoted as . Since all are calculated in the fixed reference coordinate system front_laser of the front laser radar, the inclination angle obtained is with the axis of front_laser as the reference, that is, indicates the rotation of the base coordinate system to the front_laser coordinate. Then the rotation of the front_laser coordinate system to the base coordinate system corresponds to the angle :

[0098] .

[0099] Similarly, with the aid of OpenCV, a three-channel white image matrix with the same width and height is created. The corresponding translation amount (denoted as ) in all matrices is taken out, all the translation amounts are drawn into the image matrix with 0.005 as the resolution and half of the width as the offset value. When , the identity matrix is a unit matrix, so the first point on the image matrix is at the center of the image matrix. In the image matrix, all data points form a figure close to a circle. The data points are fitted by using the fitEllipse function in the OpenCV library, and a rotating rectangle calculated by the least square method is returned. Because the differential driving AGV rotates around the center of the vehicle body (i.e., the origin of the base coordinate system), the center of the rotating rectangle can be considered as the origin of the base coordinate system, and the center of the rotating rectangle is denoted as , and the solution of the above is similar, all are calculated in the fixed reference coordinate system of the front laser radar, so the rotating rectangle is also taken as the reference coordinate system of the front laser radar, and represents the translation component of the base coordinate system to the front_laser coordinate. The front laser parameter can be solved according to the following formula:

[0100] .

[0101] In the above formula, represents the transformation matrix of the base coordinate system to the front_laser coordinate system, and its inverse is obtained as , function represents extracting from . The right side of the above formula is the step of solving the external parameter obtained by the left side formula.

[0102] At this point, the external parameter of the front laser radar coordinate system is obtained.

[0103] Referring to Figure 3 , in an embodiment of the present application, the rear radar external parameter includes a rear radar angle external parameter and a rear radar coordinate external parameter; in step S5, based on the front-rear radar transformation matrix calculated according to the third transformation matrix and the rear radar mechanical external parameter, the PL-ICP method is used to register the rear radar laser data to the global point cloud map; when registering, the rear radar angle external parameter and the rear radar coordinate external parameter are taken as optimization variables to construct a residual equation, and the rear radar mechanical external parameter is taken as an initial value, and the Ceres library is used to iteratively optimize the residual equation to obtain the final rear radar external parameter.

[0104] ​In an embodiment of the present application, in step S5, based on the first transformation matrix, the front radar laser data is converted from the current frame coordinate system to the fixed reference coordinate system, and the data points in the front radar laser data are sorted according to the angle values in the fixed reference coordinate system, the sorted data points are filtered according to the preset step length, and the global point cloud map is generated according to the filtered data points.

[0105] That is, in the process of solving the rear radar extrinsic parameter, first, according to the front radar extrinsic parameter, all the rear radar laser data is converted to the front_laser coordinate system, and a point cloud map corresponding to the front radar data is constructed. In this embodiment, the front radar laser data is filtered by indexing the angle, and a high-resolution point cloud map is finally generated.

[0106] Subsequently, based on the combination of the rear radar mechanical extrinsic parameter and the third transformation matrix, the initial transformation relationship of the current frame rear radar coordinate system relative to the front radar fixed reference coordinate system is estimated. Taking this as the initial value, a residual function between the rear radar data and the point cloud map is constructed, and the rear radar extrinsic parameter (angle and position) is used as the optimization variable, and the Ceres nonlinear optimization library is used for iterative optimization, and finally the high-precision rear radar extrinsic parameter is solved.

[0107] Finally, the final extrinsic parameter of the front radar and the rear radar is output.

[0108] In an embodiment, the rear radar mechanical extrinsic parameter can be regarded as a set containing multiple types of parameters, which can be described with the initial state of the rear radar as the reference system, or with the front radar as the reference system. In actual application, the mechanical extrinsic parameters in the above two reference systems can be approximately equivalent under certain conditions, so that the mechanical extrinsic parameters in the set can be used without distinction for extrinsic parameter solving when there is a requirement for calculation speed or implementation simplicity. However, when there is a high demand for precision and robustness, the mechanical extrinsic parameters in different reference systems can be distinguished to achieve more accurate pose estimation and extrinsic parameter optimization.

[0109] Still in combination with the above specific embodiments, the extrinsic parameter of the rear lidar coordinate system is solved, denoted as The reference coordinate system of the rear lidar coordinate system back_laser is the front lidar coordinate system front_laser. First, a point cloud map with front_laser as the global coordinate system is constructed, as shown in Figure 3 The steps are as follows:

[0110] According to the obtained The corresponding front laser data is converted to the front_laser coordinate system.

[0111] According to the coordinate values under front_laser, the distance value and the angle value are calculated and recorded. Assuming that the coordinate value of a point under front_laser is , then the distance value and the angle value can be calculated as follows:

[0112]

[0113] In the above formula, is a function representing the calculation of the signed angle between the line connecting the origin to the point and the positive axis, is a function representing the solution of the arithmetic square root.

[0114] (3) The laser points are reordered according to the angle value from small to large.

[0115] (4) All laser points are screened with a fixed angle step, and a suitable angle value smaller than the laser resolution value is selected as the screening step . The embodiment of the application selects 0.002 radian. The screening step is as follows: the first laser point is retained, and its angle value is recorded as . It is determined whether the angle value of the second laser point is greater than . If it is greater, the second laser point is retained, and its angle value is assigned to . Otherwise, the second laser point is discarded, and it is determined whether the third laser point meets the requirements, and so on.

[0116] (5) Thus, a point cloud map with an angle resolution of about is obtained, and the point cloud map is recorded as .

[0117] The transformation matrix recorded during the rotation process and the transformation matrix corresponding to the mechanical value (recorded as ) of the rear laser radar external parameter (relative to the front laser radar coordinate system) are multiplied, and all the corresponding rear laser radar data (recorded as ) are converted to the front_laser coordinate system. The rear laser radar external parameter is used as a to-be-optimized variable, is used as the initial value of the to-be-optimized variable, the rear laser radar data under front_laser and the point cloud map are used for PL-ICP matching to construct a residual equation, and the residual equation is optimized using the Google nonlinear optimization library Ceres. The result obtained by iterative optimization of the Ceres library is the rear laser radar external parameter . The solution process of the rear laser radar external parameter can be represented by the following formula:​

[0118] ,

[0119] In the above formula represents the transformation matrix of the back_laser coordinate system to the front_laser coordinate system, using mechanical values Assign an initial value to the transformation matrix, represents the transformation matrix of the first back laser coordinate system to the front_laser coordinate system, represents the transformation of the first frame of back laser data to the front_laser coordinate system by means of the transformation matrix , The function represents the PL-ICP matching of laser data and point cloud map, The function represents Take the optimization variable, and optimize the residual function constructed by using the PL-ICP matching of the first frame of laser data and the point cloud map, represents the total number of laser data frames. The function will automatically assign the result of the solution to the variable .

[0120] So far, the front laser radar coordinate system extrinsic parameter and the back laser radar coordinate system extrinsic parameter have been obtained.

[0121] In an embodiment of the present application, a front and rear radar extrinsic parameter calibration device suitable for a differential drive AGV is provided, which comprises one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs are used to execute the above-mentioned front and rear radar extrinsic parameter calibration method.

[0122] The embodiment of the present application utilizes the point cloud matching method in the simultaneous localization and mapping (SLAM, Simultaneous Localization and Mapping) technology, and synchronously calibrates the extrinsic parameters of the 2D front and rear laser radars of the differential drive AGV by means of the nonlinear optimization idea, which can greatly reduce the time required for calibration while ensuring the consistency of the front and rear laser radar extrinsic parameters and improving the accuracy of the front and rear laser radar extrinsic parameters.

[0123] This technology can mainly be applied to the following aspects.

[0124] Most of the differential drive AGVs use the scheme of separately calibrating the front and rear laser radar external parameters, or directly manually calibrate the external parameters according to the mechanical value. The above two calibration schemes not only consume time and effort, but also reduce the consistency of the front and rear laser radar external parameters, causing the rear laser radar to be used only as an auxiliary obstacle avoidance laser. If the front and rear laser radars are used for positioning at the same time, the positioning accuracy will be reduced. In order to obtain characteristic obvious original laser radar data, and then obtain more accurate matching results, the existing calibration scheme generally selects to be carried out in a structured environment. The existing front and rear laser radar external parameter calibration scheme for differential drive AGVs usually separately calibrates the front and rear laser radar external parameters in a structured environment, and uses PP-ICP in the matching processing of laser radar data, which faces the problems of low calibration efficiency, poor consistency of front and rear laser radar external parameters, and poor reusability of rear laser radar. When facing some special environments (such as large area scenes or highly degraded scenes), it is difficult to guarantee the positioning accuracy only by using the front laser radar, at which time more reflective plates and other measures are often selected to help positioning. But this type of scheme is very dependent on the deployment and transformation of the site.

[0125] The rear laser radar external parameters obtained by using the existing scheme have poor consistency. In order to improve the accuracy of laser radar data matching, the existing scheme basically selects to place the AGV in a structured environment constructed by humans (such as a rectangular environment constructed using straight boards, walls, etc.), so as to obtain original laser radar data with obvious straight line features. But in the implementation of the existing scheme, the line features of the structured environment are not fully utilized, and the PL-ICP method with faster convergence speed and higher accuracy in structured environments is not used.

[0126] The embodiments of the present application can simultaneously calibrate the external parameters of the front and rear laser radars of the differential drive AGV, reduce the time consumption of calibration, ensure the consistency of the front and rear laser radar external parameters, and improve the reuse of laser. The embodiments of the present application further use the PL-ICP method with initial value with higher accuracy in structured environments for matching in combination with the characteristics of structured environments, fully utilize the line features in the environment, and construct an error function that can better reflect the actual situation. In the same environment, the embodiments of the present application can obtain more accurate laser radar external parameters.

[0127] The embodiments of the present application involve technologies such as SLAM point cloud matching technology, nonlinear optimization technology, information data processing technology, sensor technology, etc.

[0128] The method proposed by the embodiment of the application is mainly to improve the consistency of the front and rear laser radar external parameters of the differential driving AGV, improve the calibration accuracy and improve the calibration efficiency. During the daily operation of the AGV, the rear laser radar can not only be used as an auxiliary obstacle avoidance laser radar, but also be used as a positioning laser radar in some extreme scenarios, thereby improving the reuse of the sensor, ensuring the positioning accuracy and reducing the on-site modification cost.

[0129] For example, when the AGV faces a large-scale scene or a highly degraded scene, it is difficult to ensure the positioning accuracy by relying only on the front laser radar, and even the positioning may be lost. In view of the above situation, the solution of adding a reflector plate to the front laser radar (the rear laser radar is still used as an auxiliary obstacle avoidance laser radar) is often used, but because the scanning range of a single laser radar is limited, more number and more dense reflector plates need to be deployed at this time, so this solution is very dependent on the deployment and modification of the site.

[0130] In view of the above situation, the positioning solution of the embodiment of the application uses the front and rear dual laser radars for synchronous calibration, which can achieve a scanning range of 360°, thereby ensuring the positioning accuracy of the AGV. If it is necessary to further ensure the positioning accuracy, the solution of adding a reflector plate to the dual laser radars can be used, and only a small number of reflector plates need to be deployed at this time, which saves time and effort.

[0131] The technical solution of the application uses the PL-ICP method to solve the relative pose, and combines the motion characteristics of the differential driving AGV to estimate the initial pose by using the motion speed and time difference, thereby providing an excellent initial estimate for the PL-ICP. For a structured calibration environment, the PL-ICP matching method fully utilizes the line features contained in the laser points, the error function constructed is more in line with the actual situation, and the solving accuracy is higher. In addition, the traditional PP-ICP algorithm with point feature constraint only has a first-order convergence characteristic, and the convergence speed and calculation efficiency are limited, while the PL-ICP algorithm based on line feature constraint introduces a second-order convergence mechanism, which significantly improves the iteration efficiency and convergence speed.

[0132] During the rotation and straight movement actions, the coordinate system corresponding to the initial frame of the front radar is used as a fixed reference coordinate system, and by combining the mechanical external parameters and action parameters (such as rotation angular velocity, time difference and straight line speed), an initial transformation matrix is constructed as the initial value of the PL-ICP matching, which effectively improves the rationality of the initial estimate of the registration and the matching accuracy.

[0133] In the front radar external parameter solving process, the translation and rotation in the transformation matrix are extracted to generate a straight line image and a circle image, respectively. The angle external parameter and coordinate external parameter of the front radar are fitted by using the fitLine and fitEllipse functions in OpenCV, which completely reproduces the characteristics of the circle center offset and angle offset of the differential driving AGV when rotating in place in the front radar coordinate system, and makes the external parameter solving more physically interpretable and stable.

[0134] In the rear radar external parameter solving process, the rear radar laser data is converted to the front radar reference coordinate system, and the PL-ICP method is used to construct a residual function based on the front radar point cloud map. The rear radar mechanical external parameter is used as the initial value, and the iterative optimization is performed in the Ceres nonlinear optimization library to finally solve the angle and coordinate external parameters of the rear radar. The above process realizes the automatic joint calibration of the front and rear radars in the unified reference coordinate system, reduces the calibration time, and ensures the consistency of the front and rear laser radar external parameters and the efficient reuse of the laser.

[0135] In summary, the application can simultaneously calibrate the front and rear laser radar external parameters, ensure the consistency of the external parameters, greatly shorten the calibration time, support the joint positioning of the dual laser radars, improve the system positioning accuracy and sensor reuse rate, and reduce the field modification and maintenance cost, which has a positive significance for improving the flexibility and reliability of AGV application.

[0136] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, and the division of the modules is only a logical function division. In actual implementation, another division mode can be used, for example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the indirect coupling or communication connection between devices or modules can be electrical, mechanical or other forms.

[0137] The modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment.

[0138] In addition, the functional modules in each embodiment of the present application can be integrated in one processing unit, or each module can be physically present alone, or two or more modules can be integrated in one unit. The integrated unit can be realized in the form of hardware or hardware plus software function unit.

[0139] It is to be noted that, in the present text, relational terms such as "first" and "second", and the like, can only be used to distinguish one entity or action from another entity or action, without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

Claims

1. A method for calibrating front and rear radar extrinsic parameters suitable for differentially driven AGVs, characterized in that, The method for calibrating the external parameters of the front and rear radars includes the following steps: S1: During the execution of the first action by the AGV, first laser data of the front radar containing first line feature information is acquired; a first initial transformation matrix is ​​generated based on the first action parameters of the front radar mechanical external participation; the first laser data of the front radar is registered based on the first line feature information and the first initial transformation matrix to obtain multiple first transformation matrices corresponding to different times; the first transformation matrix contains the pose information of the front radar at the corresponding time. S2: During the second action of the AGV, acquire the second laser data of the front radar containing the second line feature information; generate a second initial transformation matrix based on the second action parameters of the front radar mechanical external participation; perform registration processing on the second laser data of the front radar based on the second line feature information and the second initial transformation matrix, and solve for multiple second transformation matrices corresponding to different times; the second transformation matrix contains the pose information of the front radar at the corresponding time. S3: During the third action of the AGV, acquire rear radar laser data containing third line feature information; generate a third initial transformation matrix based on the rear radar mechanical external participation third action parameters; perform registration processing on the rear radar laser data based on the third line feature information and the third initial transformation matrix to obtain multiple third transformation matrices corresponding to different times; the third transformation matrix contains the pose information of the rear radar at the corresponding time. S4: Extract pose information from the first transformation matrix and the second transformation matrix and solve for it to obtain the front radar extrinsic parameters; S5: Based on the pose information corresponding to the front radar, a global point cloud map is constructed; the rear radar laser data is registered to the global point cloud map based on the third transformation matrix, and the rear radar mechanical extrinsic parameters are used as initial values ​​for iterative optimization to obtain the rear radar extrinsic parameters; The rear radar extrinsic parameters include the rear radar angle extrinsic parameters and the rear radar coordinate extrinsic parameters; In step S5, based on the front and rear radar transformation matrices calculated according to the third transformation matrix and the rear radar mechanical extrinsic parameters, the PL-ICP method is used to register the rear radar laser data to the global point cloud map. During registration, the rear radar angle extrinsic parameters and the rear radar coordinate extrinsic parameters are used as variables to be optimized to construct a residual equation. The rear radar mechanical extrinsic parameters are used as initial values, and the Ceres library is used to iteratively optimize the residual equation to obtain the final rear radar extrinsic parameters. S6: Output the front radar extrinsic parameters and the rear radar extrinsic parameters.

2. The method for calibrating front and rear radar extrinsic parameters as described in claim 1, characterized in that, In one or more of steps S1 to S3: Motion parameters include the speed information of the corresponding motion and the time difference information between the current frame and the keyframe; The initial transformation matrix contains pose transformation information of the current frame coordinate system relative to the keyframe coordinate system; When registering laser data based on line feature information and the corresponding initial transformation matrix, the relative transformation matrix applicable to the current frame coordinate system and the key frame coordinate system is first obtained, and then the transformation matrix between the current frame coordinate system and the fixed reference coordinate system is calculated by combining the key frame transformation matrix and the relative transformation matrix. The fixed reference coordinate system is the corresponding coordinate system of the radar at the start of the operation; The keyframe transformation matrix is ​​used for the transformation between the keyframe coordinate system and the fixed reference coordinate system; The keyframes are updated according to preset threshold conditions.

3. The method for calibrating front and rear radar extrinsic parameters as described in claim 2, characterized in that, In one or more steps of steps S1 to S3, when registering the laser data based on the line feature information and the corresponding initial transformation matrix, the corresponding initial transformation matrix is ​​used as the initial input, and the PL-ICP method is used to match and optimize the laser data of the current frame with the laser data of the key frame to obtain the relative transformation matrix.

4. The method for calibrating front and rear radar extrinsic parameters as described in claim 3, characterized in that, In one or more of steps S1 to S3, the transformation matrix to be optimized is calculated by combining the key frame transformation matrix and the relative transformation matrix. The transformation matrix to be optimized is used as the initial input, and the PL-ICP method is used to match and optimize the laser data of the current frame with the laser data corresponding to the fixed reference coordinate system to obtain the final transformation matrix.

5. The method for calibrating front and rear radar extrinsic parameters as described in claim 1, characterized in that, In step S4, after extracting the pose information from the transformation matrix, an image is generated based on the pose information, the image is fitted, and the extrinsic parameters of the front radar are calculated based on the fitting results.

6. The method for calibrating front and rear radar extrinsic parameters as described in claim 5, characterized in that: The first action is a rotational action, and the second action is a linear action; The front radar extrinsic parameters include front radar angle extrinsic parameters and front radar coordinate extrinsic parameters; In step S4, after extracting the pose information from the second transformation matrix, an image containing a straight line-like graphic is generated based on the pose information. The fitLine function in the OpenCV library is used to perform straight line fitting on the image to obtain the straight line tilt angle, and the front radar angle extrinsic parameter is calculated based on the straight line tilt angle. After extracting the pose information from the first transformation matrix, an image containing a circular shape is generated based on the pose information. The fitEllipse function in the OpenCV library is used to perform graphic fitting on the image to obtain a rotating rectangle. The center coordinates of the rotating rectangle are extracted, and the front radar coordinate extrinsic parameters are calculated based on the center coordinates and the front radar angle extrinsic parameters.

7. The method for calibrating front and rear radar extrinsic parameters as described in claim 2, characterized in that, In step S5, based on the first transformation matrix and / or the second transformation matrix, the first laser data and / or the second laser data of the front radar are transformed from the current frame coordinate system to the fixed reference coordinate system. The data points in the first laser data and / or the second laser data of the front radar are sorted according to the angle values ​​in the fixed reference coordinate system. The sorted data points are filtered according to a preset step size. The global point cloud map is generated based on the filtered data points.

8. The method for calibrating front and rear radar extrinsic parameters as described in claim 6, characterized in that, The third action is a rotational action, and the first action includes the third action.

9. A front and rear radar extrinsic parameter calibration device suitable for differential drive AGVs, characterized in that, The front and rear radar extrinsic parameter calibration device includes: one or more processors, a memory, one or more programs stored in the memory, and configured to be executed by the one or more processors for performing the front and rear radar extrinsic parameter calibration method according to any one of claims 1 to 8.

Images