IMU extrinsic parameter calibration method for laser SLAM measurement system

Abstract

The present invention provides a method for calibrating the external parameters of the IMU of a laser SLAM measurement system, which utilizes a high-precision scanner: by using a high-precision stand-type three-dimensional laser scanner with tilt correction to scan a specific site, high-precision point cloud data C0 can be obtained. This high-precision reference benchmark point cloud provides accurate basic data for the subsequent calibration process, which helps to improve the accuracy of the calibration. Static acquisition method: data is collected by locking the SLAM system device on a tripod for static scanning. Unlike the prior art in which the point cloud coordinates may be affected by motion distortion due to device movement (such as the operator shaking the device), this static acquisition method fundamentally avoids the distortion problem caused by motion, so that the collected data can more accurately reflect the actual geometric relationship, thereby improving the accuracy of the calibration.

Description

Technical Field

[0001] The present invention relates to the field of laser radar, and in particular to an IMU extrinsic parameter calibration method for a laser SLAM measurement system. Background Art

[0002] Common laser SLAM measurement systems (hereafter referred to as SLAM systems) typically include two key sensors: a lidar (LiDAR) and an IMU (Inertial Measurement Unit). IMU extrinsic parameters typically refer to the rotational and translational transformations required to convert point coordinates from the lidar coordinate system to the IMU coordinate system. This transformation can be represented by three Euler angles plus a three-axis translation, or as a 4×4 matrix. Due to precision limitations in structural processing and installation, IMU extrinsic parameters can vary from SLAM system to SLAM system of the same model. Rotational transformations, in particular, can have a significant impact on SLAM system measurement accuracy. Therefore, calibrating the IMU extrinsic parameters to obtain relatively accurate values ​​is crucial.

[0003] Existing publicly available calibration methods are generally categorized as online and offline. Online calibration involves optimizing the initial values ​​of IMU extrinsic parameters during the SLAM measurement and mapping process, estimating the IMU extrinsic parameters in real time. The mathematical models of these methods couple the system's motion state (position, attitude, IMU intrinsic parameters, etc.) with the IMU extrinsic parameters and calculate them using filtering or optimization methods. These methods assume that the IMU extrinsic parameters are constantly changing, which is inconsistent with reality. Consequently, the calibrated IMU extrinsic parameters are not fixed, resulting in unstable calibration accuracy. Offline calibration methods perform calibration before SLAM measurement and mapping, fixing the IMU extrinsic parameters during the measurement and mapping process. These mathematical models are more realistic. Existing offline calibration methods typically require a relatively open area with buildings. The operator stands still and shakes the SLAM system to scan the scene. The angular velocity and acceleration information calculated by the SLAM system's pure laser odometry are then matched with the IMU measurements to calibrate the extrinsic parameters. Because each point in each frame of LiDAR data is not acquired simultaneously, the point cloud coordinates are subject to motion distortion. The pure laser odometry used here requires using a uniform motion assumption or a B-spline-based continuous-time trajectory to fit the motion state to offset motion distortion, which inevitably results in fitting errors. Furthermore, the stability of the calibration results is affected by the amplitude and speed of the device during acquisition, as well as the spatial distribution of objects on the site. Summary of the Invention

[0004] The main purpose of the present invention is to provide an IMU extrinsic parameter calibration method for a laser SLAM measurement system. The proposed method adopts a static data acquisition method, which fundamentally avoids the point cloud coordinate distortion caused by the movement of the SLAM system and performs multiple data acquisitions to improve the repeatability stability.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: a method for calibrating the IMU extrinsic parameters of a laser SLAM measurement system, the method comprising:

[0006] S1. Select a relatively open site with multiple-facing tall building facades. Use a high-precision stand-mounted 3D laser scanner with tilt correction to scan the site to obtain a high-precision point cloud C0. Extract at least two facades with different orientations from C0 and calculate the facade normal vectors. 、 、…、 ;

[0007] S2. Lock the SLAM system device on a tripod with adjustable height, and make the SLAM system statically scan the site for a period of time. Adjust the scanning time according to the scanning mode of the laser radar. Adjust the height of the tripod legs to make the device at different tilt angles. Repeat the acquisition multiple times to obtain point cloud data C. j and IMU data I j ;

[0008] S3. Use ICP algorithm to find C j The transformation relationship between C0 and IMU data I j Processing is performed to calculate the angle between the IMU coordinate system and the gravity acceleration based on the accelerometer data, and the IMU inclination angle at different tilt angles is obtained;

[0009] S4. Express the IMU external parameters as , according to the formula, all point cloud data C j Transform to the IMU coordinate system, transform the vertical normal vector to the IMU coordinate system, and then transform the vertical normal vector to the horizontal coordinate system according to the IMU inclination angle. With the angle between the normal vector and the Z axis equal as the constraint condition, use the nonlinear least squares method to calculate the rotation transformation parameters of the MLI, and directly use the design value for the translation transformation parameters;

[0010] S5. Normal vector obtained by high-precision 3D scanner It can be considered to be in the horizontal plane coordinate system, and the coordinates of the normal vector in the horizontal plane coordinate system are obtained based on the IMU observation value ,by and The angles between them and the Z axis should be equal as a constraint condition. The rotation transformation parameters of MLI are obtained using the nonlinear least squares method, and the translation transformation parameters of MLI are directly used as the designed values.

[0011] In a preferred embodiment, the intrinsic parameters of the IMU are considered to have been calibrated before calibration using this method.

[0012] In the preferred embodiment, the specific method of step S2 is:

[0013] Lock the SLAM system device on a tripod with adjustable height, and allow the SLAM system to scan the site statically for a period of time;

[0014] If the LiDAR of the SLAM system is in non-repeated scanning mode, or the LiDAR is rotated by an external motor, a denser point cloud can be obtained as the scanning time accumulates;

[0015] If the LiDAR is in mechanical multi-line scanning mode and there is no external motor rotation, only multiple scan lines will be obtained. The single scan time does not need to be too long, about one minute;

[0016] By adjusting the height of the tripod legs to make the device at different tilt angles, the C j and I j , respectively represent the point cloud data and IMU data collected for the jth time.

[0017] In the preferred embodiment, the specific method of step S3 is:

[0018] Use the ICP algorithm to find C i The transformation relationship between and C0 , which means that the point in the C0 coordinate system is transformed to C j The transformation in the coordinate system, that is , and the normal vector of the vertical surface in step S1 should also satisfy this formula, and the normal vector in C j Coordinates in the coordinate system 、 、…、 ;

[0019]

[0020] IMU data I j Processing is mainly carried out by using static conditions to calculate the angle between the IMU coordinate system and the gravity acceleration based on the accelerometer data, so as to obtain the IMU inclination angle at different inclination angles, that is, the angle with the horizontal plane, A j Is a transformation matrix, indicating that the IMU coordinate system passes through A jAfter the transformation, the XY plane is parallel to the horizontal plane, and the Z axis of the IMU is vertically upward. The following formula is the acceleration error model:

[0021]

[0022] in represents the actual observed value of acceleration, represents the corrected acceleration, represents the acceleration deviation, represents the acceleration measurement noise, represents the acceleration scale coefficient matrix;

[0023] is the rotation transformation matrix;

[0024] When the parameters in the IMU have been calibrated, take the average of the acceleration observations for a period of static time , ignoring Gaussian white noise , after correction of the above error model, we can get , and then use the Rodrigues formula to find Rotate the matrix in the opposite direction of gravity (0,0,g) to get A j .

[0025] In the preferred embodiment, the specific method of step S4 is:

[0026] The IMU external parameters are expressed as , C j The point in the lidar coordinate system is represented as , the point in the IMU coordinate system is expressed as ,but If the external parameters of IMU are fixed, all collected data will satisfy this formula, and all point cloud data C j Transformed to the IMU coordinate system, the vertical normal vectors of all point clouds can also be transformed to the IMU coordinate system. Then, according to the IMU inclination angle calculated in step S3, the vertical normal vector can be transformed to the horizontal coordinate system. The formula is:

[0027]

[0028] In the preferred embodiment, the specific method of step S5 is:

[0029] In step S1, the high-precision 3D scanner has the function of tilt correction, and the normal vector obtained is It can be considered to be in the horizontal plane coordinate system, and in step S4, the coordinates of the normal vector in the horizontal plane coordinate system can be obtained according to the IMU observation value. ,but and The angles between them and the Z axis should be equal, which can be used as a constraint to obtain the value using the nonlinear least squares method. The rotation transformation parameters are The translation transformation parameters use the designed values ​​directly;

[0030] According to the vector angle formula:

[0031]

[0032] Restore the normal vector from homogeneous coordinate representation to three-dimensional column vector representation. The three-dimensional normal vector obtained in step 5 is expressed as , the three-dimensional normal vector corresponding to step 1 is , the Z axis is represented by , then the error of the kth normal vector measured at the jth time is expressed as:

[0033]

[0034] The goal of the least squares method is to find an IMU external parameter that minimizes the sum of squares of the errors of multiple normal vectors measured multiple times.

[0035] This paper provides a method for calibrating the extrinsic parameters of an IMU (Instrumental Measurement Unit) for a laser SLAM measurement system. This method utilizes a high-precision scanner: By scanning a specific site using a high-precision, frame-mounted 3D laser scanner with tilt correction, high-precision point cloud data (C0) is acquired. This high-precision reference point cloud provides accurate baseline data for subsequent calibration, helping to improve calibration accuracy.

[0036] Static acquisition: Data is collected by locking the SLAM system device on a tripod and performing static scanning. Unlike existing technologies, where point cloud coordinates may be affected by motion distortion due to device movement (such as operator shaking the device), this static acquisition method fundamentally avoids motion-induced distortion, ensuring that the collected data more accurately reflects the actual geometric relationships, thereby improving calibration accuracy.

[0037] Repeated acquisition: Repeated acquisition of data at different tilt angles, including point cloud data C j and IMU data I j This repeated acquisition method effectively reduces the potential errors associated with a single acquisition, ensuring the stability of the calibration results. Data collected at different times can verify and complement each other, making the final calibration results more reliable and unaffected by factors such as the spatial distribution of field targets, the amplitude and speed of device movement, and other factors that are difficult to avoid with existing offline calibration methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The present invention will be further described below with reference to the accompanying drawings and examples:

[0039] Figure 1 It is a top view schematic diagram of the calibration site of the present invention;

[0040] Figure 2 It is a schematic side view of the calibration site of the present invention. DETAILED DESCRIPTION

[0041] Example 1

[0042] like Figures 1 and 2 As shown, a method for calibrating IMU extrinsic parameters of a laser SLAM measurement system includes:

[0043] S1. First, select a relatively open site with multiple-facing tall building facades as the data collection site. Use a high-precision stand-mounted 3D laser scanner with tilt correction (with an accuracy of 1 mm) to scan the site and obtain a high-precision point cloud C0. Extract at least two facades with different orientations from C0 and calculate the facade normal vectors. 、 、…、 ;

[0044] S2. Lock the SLAM system device on a tripod with adjustable height, and make the SLAM system statically scan the site for a period of time. Adjust the scanning time according to the scanning mode of the laser radar. Adjust the height of the tripod legs to make the device at different tilt angles. Repeat the acquisition multiple times to obtain point cloud data C. j and IMU data I j ;

[0045] S3. Use ICP algorithm to find C j The transformation relationship between C0 and IMU data I j Processing is performed to calculate the angle between the IMU coordinate system and the gravity acceleration based on the accelerometer data, and the IMU inclination angle at different tilt angles is obtained;

[0046] S4. Express the IMU external parameters as , according to the formula, all point cloud data C j Transform to the IMU coordinate system, transform the vertical normal vector to the IMU coordinate system, and then transform the vertical normal vector to the horizontal coordinate system according to the IMU inclination angle. With the angle between the normal vector and the Z axis equal as the constraint condition, use the nonlinear least squares method to calculate the rotation transformation parameters of the MLI, and directly use the design value for the translation transformation parameters;

[0047] S5. Normal vector obtained by high-precision 3D scanner It can be considered to be in the horizontal plane coordinate system, and the coordinates of the normal vector in the horizontal plane coordinate system are obtained based on the IMU observation value ,by and The angles between them and the Z axis should be equal as a constraint condition. The rotation transformation parameters of MLI are obtained using the nonlinear least squares method, and the translation transformation parameters of MLI are directly used as the designed values.

[0048] In a preferred embodiment, the intrinsic parameters of the IMU are considered to have been calibrated before calibration using this method.

[0049] The actual values ​​of the rotational transformation of the extrinsic parameters usually differ significantly from the designed values. However, the translational transformation has little impact on the mapping accuracy of the SLAM system because the actual values ​​differ from the designed values ​​at the millimeter level. Therefore, the proposed method only calibrates the rotational transformation of the extrinsic parameters. Before using this method for calibration, it is assumed that the intrinsic parameters of the IMU have been calibrated.

[0050] Symbolic convention: The proposed method uses homogeneous coordinates to represent 3D coordinates, for example, the coordinates of a 3D point or a normal vector are represented as

[0051]

[0052] The rotation and translation transformation is expressed as:

[0053]

[0054] in is the translation transformation parameter, is the rotation transformation parameter.

[0055] In the preferred embodiment, the specific method of step S2 is:

[0056] Lock the SLAM system device on a tripod with adjustable height, and allow the SLAM system to scan the site statically for a period of time;

[0057] If the SLAM system's lidar is in non-repeating scanning mode, or the lidar is rotated by an external motor, a denser point cloud can be obtained as the scanning time accumulates (usually three to five minutes);

[0058] If the LiDAR is in mechanical multi-line scanning mode and there is no external motor rotation, only multiple scan lines will be obtained. The single scan time does not need to be too long, about one minute;

[0059] By adjusting the height of the tripod legs to make the device at different tilt angles, repeat the acquisition multiple times (three times or more) to obtain C j and I j , respectively represent the point cloud data and IMU data collected for the jth time.

[0060] Example 2

[0061] Further illustrate with reference to Example 1, Figure 1-2 The structure shown in FIG. 1 , the specific method of step S3 is as follows:

[0062] Use ICP (Iterative Closest Point) algorithm to find C j The transformation relationship between and C0 , which means that the point in the C0 coordinate system is transformed to C j The transformation in the coordinate system, that is , and the vertical normal vector in step S1 should also satisfy this formula, and the normal vector in C j Coordinates in the coordinate system 、 、…、 ;

[0063]

[0064] IMU data I j Processing is mainly carried out by using static conditions to calculate the angle between the IMU coordinate system and the gravity acceleration based on the accelerometer data, so as to obtain the IMU inclination angle at different inclination angles, that is, the angle with the horizontal plane, A j Is a transformation matrix, indicating that the IMU coordinate system passes through A j After the transformation, the XY plane is parallel to the horizontal plane, and the Z axis of the IMU is vertically upward. The following formula is the acceleration error model:

[0065]

[0066] in represents the actual observed value of acceleration, represents the corrected acceleration, represents the acceleration deviation, represents the acceleration measurement noise, represents the acceleration scale coefficient matrix;

[0067] is the rotation transformation matrix;

[0068] When the parameters in the IMU have been calibrated, take the average of the acceleration observations for a period of static time , ignoring Gaussian white noise , after correction of the above error model, we can get , and then use the Rodrigues formula to find Rotate the matrix in the opposite direction of gravity (0,0,g) to get A j .

[0069] in:

[0070] Represents the actual observed value of acceleration.

[0071] Indicates the corrected acceleration.

[0072] Indicates acceleration deviation.

[0073] represents the acceleration measurement noise.

[0074] is the acceleration scaling coefficient matrix.

[0075] is the rotation transformation matrix.

[0076] Calculate the IMU inclination angle under static conditions:

[0077] When the IMU is in a stationary state, collect the average acceleration observation value for a period of time .

[0078] Ignore Gaussian white noise , using the above error model Correction is performed to obtain .

[0079] Using the Rodrigues formula, we can calculate The rotation matrix required to rotate to the opposite direction of gravity (0,0,g) is the rotation matrix A of the IMU coordinate system relative to the ground coordinate system. j , this matrix ensures that the XY plane of the IMU is parallel to the horizontal plane and the Z axis points to the center of the earth.

[0080] Calculate the transformation matrix A j : Transformation matrix A j It is used to convert data in the IMU coordinate system into a new coordinate system parallel to the ground to ensure the consistency and accuracy of data processing.

[0081] Through the above steps, not only can the relative transformation relationship between point cloud data under different perspectives be accurately calculated, but the IMU data can also be effectively corrected, thereby improving the positioning accuracy and stability of the entire system.

[0082] Example 3

[0083] Further illustrate with reference to Example 1, Figure 1-2 The structure shown in FIG. 4 is as follows:

[0084] The IMU external parameters are expressed as , C jThe point in the laser radar coordinate system is represented as , the point in the IMU coordinate system is expressed as ,but If the external parameters of IMU are fixed, all collected data will satisfy this formula, and all point cloud data C j Transformed to the IMU coordinate system, the vertical normal vectors of all point clouds can also be transformed to the IMU coordinate system. Then, according to the IMU inclination angle calculated in step S3, the vertical normal vector can be transformed to the horizontal coordinate system. The formula is:

[0085]

[0086] To make step S4 clearer and more detailed, the following is a tidied and improved description of the steps:

[0087] Convert the point cloud data and vertical normal vector in the lidar coordinate system to the IMU coordinate system.

[0088] According to the inclination angle of the IMU, the point cloud data and the vertical normal vector are further converted to the horizontal coordinate system.

[0089] Specific steps:

[0090] 1. Define IMU external parameters:

[0091] Set the IMU external parameters to , which is a fixed transformation matrix used to transform the laser radar coordinate system The points under are transformed into the IMU coordinate system I.

[0092] The point in the laser radar coordinate system L is expressed as , the point under the IMU coordinate system I is expressed as .

[0093] The conversion relationship is: .

[0094] 2. Convert the point cloud data to the IMU coordinate system:

[0095] For each collected point cloud data , using the transformation matrix obtained in step S2 and IMU external parameters , convert the point cloud data from the lidar coordinate system L to the IMU coordinate system I.

[0096] The conversion formula is: .

[0097] here Indicates the representation of the j-th collected point cloud data in the IMU coordinate system.

[0098] 3. Convert the vertical normal vector to the IMU coordinate system:

[0099] For each facade normal vector , also using the transformation matrix obtained in step S2 and IMU external parameters , transform the normal vector from the lidar coordinate system L to the IMU coordinate system I.

[0100] The conversion formula is:

[0101] here It represents the kth vertical normal vector acquired at the jth time in the IMU coordinate system.

[0102] 4. Convert the data to the horizontal coordinate system based on the IMU inclination:

[0103] In step S3, we have calculated the transformation matrix of the IMU at different tilt angles , this matrix is ​​used to transform the data in the IMU coordinate system into the horizontal coordinate system.

[0104] The point cloud data and vertical normal vector in the IMU coordinate system are further converted to the horizontal coordinate system.

[0105] The conversion formula is: ; .

[0106] here Indicates the representation of the point cloud data collected for the jth time in the horizontal coordinate system, It represents the kth vertical normal vector collected for the jth time in the horizontal coordinate system.

[0107] Through the above steps, we can convert the point cloud data and vertical normal vectors in the lidar coordinate system to the IMU coordinate system, and then further convert them to the horizontal coordinate system based on the IMU's inclination angle. This ensures that all data is in the same reference system, which facilitates subsequent analysis and application.

[0108] Example 4

[0109] Further illustrate with reference to Example 1, Figure 1-2 The structure shown in FIG. 5 is as follows:

[0110] In step S1, the high-precision 3D scanner has the function of tilt correction, and the normal vector obtained is It can be considered to be in the horizontal plane coordinate system, and in step S4, the coordinates of the normal vector in the horizontal plane coordinate system can be obtained according to the IMU observation value. ,but and The angles between them and the Z axis should be equal, which can be used as a constraint to obtain the value using the nonlinear least squares method. The rotation transformation parameters are The translation transformation parameters use the designed values ​​directly;

[0111] According to the vector angle formula:

[0112]

[0113] Restore the normal vector from homogeneous coordinate representation to three-dimensional column vector representation. The three-dimensional normal vector obtained in step 5 is expressed as , the three-dimensional normal vector corresponding to step 1 is , the Z axis is represented by , then the error of the kth normal vector measured at the jth time is expressed as:

[0114]

[0115] The goal of the least squares method is to find an IMU external parameter that minimizes the sum of squares of the errors of multiple normal vectors measured multiple times.

[0116] To make the steps more complete and clear, the following are the detailed step-by-step instructions:

[0117] Using high-precision 3D scanner and IMU data, the IMU external parameters are solved by nonlinear least squares method. The rotation transformation parameters.

[0118] The translation transformation parameters directly use the designed values.

[0119] 1. Preparation

[0120] High-precision 3D scanner data: Assuming that the high-precision 3D scanner has an inclination correction function, the normal vector obtained It can be considered as being in a horizontal coordinate system.

[0121] IMU data: Through step S4, the normal vector has been Converted to the horizontal coordinate system, it is recorded as .

[0122] 2. Define the objective function

[0123] The goal is to minimize the sum of squares of the errors of multiple normal vectors measured multiple times.

[0124] The error is defined as the angle difference between the normal vector and the Z axis.

[0125] 3. Calculate the angle between the normal vector and the Z axis

[0126] The normal vector is converted from homogeneous coordinate representation to three-dimensional column vector representation.

[0127] The three-dimensional normal vector obtained in step S5 is expressed as .

[0128] The three-dimensional normal vector corresponding to step S1 is .

[0129] The Z axis is represented by .

[0130] 4. Calculate the error

[0131] Use the vector angle formula to calculate the error: .

[0132] The error of the kth normal vector measured at the jth time is expressed as: .

[0133] 5. Constructing the least squares problem

[0134] Define the error function is the sum of squares of all measured errors: ;

[0135] Where J is the number of measurements and K is the number of normal vectors in each measurement.

[0136] 6. Solve the least squares problem

[0137] Use nonlinear least squares methods (such as the Levenberg-Marquardt algorithm) to solve The rotation transformation parameters are Reach minimum.

[0138] The translation transformation parameters directly use the designed values ​​without optimization.

[0139] Implementation steps:

[0140] 1. Initialization parameters

[0141] Initialize rotation transformation parameters and translation transformation parameters .

[0142] Translation transformation parameters Use the design value.

[0143] 2. Constructing the error function

[0144] Define the error function :

[0145] .

[0146] 3. Optimize rotation transformation parameters

[0147] Optimize the rotation transformation parameters using nonlinear least squares methods (such as the Levenberg-Marquardt algorithm) ,make Reach minimum.

[0148] 4. Verify the results

[0149] Verify the optimized rotation transformation parameters Whether it is reasonable can be verified by substituting part of the test data into the verification error to see whether it is significantly reduced.

[0150] Through the above steps, the nonlinear least squares method can be effectively used to solve the IMU external parameters The rotation transformation parameters ensure the conversion accuracy of point cloud data and normal vectors in different coordinate systems.

[0151] The above embodiments are merely preferred technical solutions of the present invention and should not be construed as limiting the present invention. The scope of protection of the present invention shall be the technical solutions set forth in the claims, including equivalent alternatives to the technical features of the technical solutions set forth in the claims. In other words, equivalent alternatives and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for calibrating the external parameters of an IMU in a laser SLAM measurement system, characterized by: The method includes: S1. Select a relatively open site with multiple-facing tall building facades. Use a high-precision stand-mounted 3D laser scanner with tilt correction to scan the site to obtain a high-precision point cloud C0. Extract at least two facades with different orientations from C0 and calculate the facade normal vectors. 、 、…、 ; S2. Lock the SLAM system device on a tripod with adjustable height, and make the SLAM system statically scan the site for a period of time. Adjust the scanning time according to the scanning mode of the laser radar. Adjust the height of the tripod legs to make the device at different tilt angles. Repeat the acquisition multiple times to obtain point cloud data C. j and IMU data I j ; S3. Use ICP algorithm to find C j The transformation relationship between C0 and IMU data I j Processing is performed to calculate the angle between the IMU coordinate system and the gravity acceleration based on the accelerometer data, and the IMU inclination angle at different tilt angles is obtained; S4. Express the IMU external parameters as , according to the formula, all point cloud data C j Transform to the IMU coordinate system, transform the vertical normal vector to the IMU coordinate system, and then transform the vertical normal vector to the horizontal coordinate system according to the IMU inclination angle. Take the angle between the normal vector and the Z axis as the constraint condition, and use the nonlinear least squares method to find The rotation transformation parameters and translation transformation parameters use the designed values ​​directly; S5. Normal vector obtained by high-precision 3D scanner It can be considered to be in the horizontal plane coordinate system, and the coordinates of the normal vector in the horizontal plane coordinate system are obtained based on the IMU observation value ,by and The angles between them and the Z axis should be equal as a constraint condition, and the nonlinear least squares method is used to find The rotation transformation parameters, The translation transformation parameters use the designed values ​​directly.

2. The IMU extrinsic parameter calibration method of a laser SLAM measurement system according to claim 1, characterized in that: Before using this method for calibration, it is assumed that the intrinsic parameters of the IMU have been calibrated.

3. The IMU extrinsic parameter calibration method of a laser SLAM measurement system according to claim 1, wherein: The specific method of step S2 is: Lock the SLAM system device on a tripod with adjustable height, and let the SLAM system scan the site statically for a period of time; If the LiDAR of the SLAM system is in non-repeated scanning mode, or the LiDAR is rotated by an external motor, a denser point cloud can be obtained as the scanning time accumulates; If the LiDAR is in mechanical multi-line scanning mode and there is no external motor rotation, only multiple scan lines will be obtained. The single scan time does not need to be too long, about one minute; By adjusting the height of the tripod legs to make the device at different tilt angles, the C j and I j , respectively represent the point cloud data and IMU data collected for the jth time.

4. The IMU extrinsic parameter calibration method of a laser SLAM measurement system according to claim 1, characterized in that: The specific method of step S3 is: Use the ICP algorithm to find C i The transformation relationship between and C0 , which means that the point in the C0 coordinate system is transformed to C j The transformation in the coordinate system, that is , and the vertical normal vector in step S1 should also satisfy this formula, and the normal vector in C j Coordinates in the coordinate system 、 、…、 ; IMU data I j Processing is mainly carried out by using static conditions to calculate the angle between the IMU coordinate system and the gravity acceleration based on the accelerometer data, so as to obtain the IMU inclination angle at different inclination angles, that is, the angle with the horizontal plane, A j Is a transformation matrix, indicating that the IMU coordinate system passes through A j After the transformation, the XY plane is parallel to the horizontal plane, and the Z axis of the IMU is vertically upward. The following formula is the acceleration error model: in represents the actual observed value of acceleration, represents the corrected acceleration, represents the acceleration deviation, represents the acceleration measurement noise, represents the acceleration scale coefficient matrix; is the rotation transformation matrix; When the parameters in the IMU have been calibrated, take the average of the acceleration observations for a period of static time , ignoring Gaussian white noise , after correction of the above error model, we can get , and then use the Rodrigues formula to find Rotate the matrix in the opposite direction of gravity (0,0,g) to get A j .

5. The IMU extrinsic parameter calibration method of a laser SLAM measurement system according to claim 1, characterized in that: The specific method of step S4 is: The IMU external parameters are expressed as , C j The point in the lidar coordinate system is represented as , the point in the IMU coordinate system is expressed as ,but ; If the external parameters of IMU are fixed, all collected data satisfy this formula, and all point cloud data C j Transformed to the IMU coordinate system, the vertical normal vectors of all point clouds can also be transformed to the IMU coordinate system. Then, according to the IMU inclination angle calculated in step S3, the vertical normal vector can be transformed to the horizontal coordinate system. The formula is: in, Calculated by ICP algorithm and The transformation relationship between The point in the coordinate system is transformed to Transformation in coordinate system; Is the transformation matrix, indicating that the IMU coordinate system passes through After the transformation, the XY plane is parallel to the horizontal plane, and the Z axis of the IMU is vertically upward.

6. The IMU extrinsic parameter calibration method of a laser SLAM measurement system according to claim 1, characterized in that: The specific method of step S5 is: In step S1, the high-precision 3D scanner has the function of tilt correction, and the normal vector obtained is It can be considered to be in the horizontal plane coordinate system, and in step S4, the coordinates of the normal vector in the horizontal plane coordinate system can be obtained according to the IMU observation value. ,but and The angles between them and the Z axis should be equal, which can be used as a constraint to obtain the value using the nonlinear least squares method. The rotation transformation parameters are The translation transformation parameters use the designed values ​​directly; According to the vector angle formula: Restore the normal vector from homogeneous coordinate representation to three-dimensional column vector representation. The three-dimensional normal vector obtained in step 5 is expressed as , the three-dimensional normal vector corresponding to step 1 is , the Z axis is represented by , then the error of the kth normal vector measured at the jth time is expressed as: The goal of the least squares method is to find an IMU external parameter that minimizes the sum of squares of the errors of multiple normal vectors measured multiple times.

Images