A sensor fusion calibration method for a driverless sightseeing vehicle

Abstract

The present application relates to the field of automatic driving sensor calibration, and discloses a sensor fusion calibration method for an unmanned sightseeing vehicle, which comprises the following steps: constructing a micro-displacement model by synchronously collecting a wheel speed meter and IMU data, and calculating an inter-frame pose change quantity in a vehicle coordinate system; filtering dynamic objects in a laser radar point cloud, eliminating dynamic interference in combination with motion prediction of the wheel speed meter and the IMU, and extracting static geometric features; performing motion compensation on the static point cloud based on the pose change quantity, and generating a time-aligned enhanced point cloud map; and minimizing errors between the enhanced point cloud map and a calibration reference, and optimizing an external parameter matrix of the wheel speed meter, the IMU and the laser radar. The algorithm is particularly suitable for a multi-sensor online calibration scene of a low-speed unmanned sightseeing vehicle, effectively solves the problems of tire slip and dynamic interference in a low-speed scene, and significantly improves the perception robustness and positioning accuracy of the unmanned sightseeing vehicle in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving sensor calibration, and more particularly to a sensor fusion calibration method for unmanned sightseeing vehicles. Background Technology

[0002] Against the backdrop of rapid development in intelligent transportation and autonomous driving technologies, driverless sightseeing vehicles have attracted widespread attention due to their advantages in automation, safety, and intelligent services. With continuous technological advancements, the level of intelligence and automation in driverless sightseeing vehicles is increasing, placing higher demands on the accuracy and reliability of sensor systems.

[0003] Currently, the data collected by sensors commonly used in autonomous sightseeing vehicles exists in their own independent coordinate systems. To achieve effective data fusion, the extrinsic parameter matrices between sensors must be accurately calibrated. Sensor fusion is a key technology for environmental perception and decision-making. Most existing calibration methods focus on static calibration or high-speed motion scenarios. However, in low-speed motion scenarios, such as scenic area tours and park patrols, dynamic extrinsic parameter calibration of sensors remains a challenge. At low speeds, the relative position and orientation of sensors may change due to minor vehicle vibrations, uneven road surfaces, or elastic deformation of mechanical structures, leading to drift in calibration parameters.

[0004] Chinese invention CN113050074B discloses a camera and lidar calibration system and method for environmental perception in autonomous driving. It uses an improved iterative nearest-point algorithm, incorporating the maximum correlation entropy criterion, to register two sets of points to calculate the relative pose relationship between the camera and lidar. This method is primarily designed for general autonomous driving scenarios, without specific optimization for low-speed scenarios and without addressing the low-speed tire slippage problem.

[0005] Chinese invention CN116295329A discloses an online differential odometer and lidar parameter calibration method and system, involving the calibration of differential odometers and lidar, which achieves calibration through motion data and point cloud matching. This method relies only on static maps and does not cover the calibration needs of IMU fusion and dynamic scenes.

[0006] Therefore, this invention proposes a sensor fusion calibration method for unmanned sightseeing vehicles. Summary of the Invention

[0007] The purpose of this invention is to propose a sensor fusion calibration method for unmanned sightseeing vehicles to address the problems in existing technologies. This invention synchronously collects data from multiple sensors, constructs a micro-displacement model to calculate inter-frame pose changes, filters dynamic objects, generates a time-aligned enhanced point cloud map, and further optimizes the extrinsic parameter matrix, particularly for low-speed scenarios, thus achieving high-precision and high-efficiency sensor calibration.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A sensor fusion calibration method for an unmanned sightseeing vehicle includes the following steps:

[0010] Step S1: When the sightseeing vehicle speed is ≤10km / h, the wheel speed meter pulse signal, IMU angular velocity and acceleration data are synchronously collected through the CAN bus using the PTP protocol. Based on the Ackerman steering geometry, a tightly coupled micro-displacement model is constructed to calculate the inter-frame pose change in the vehicle coordinate system.

[0011] Step S2: Use motion prior to filter dynamic objects in the continuous multi-frame point cloud of solid-state lidar, use motion prediction provided by wheel speed meter and IMU to remove dynamic point cloud, and extract static geometric features.

[0012] Step S3: Perform motion compensation on the static point cloud based on the pose change to generate a time-aligned enhanced point cloud map.

[0013] Step S4: Optimize the extrinsic parameter matrix of the solid-state lidar, wheel speedometer, and IMU by minimizing the error between the enhanced point cloud map and the preset calibration reference. The extrinsic parameter matrix includes the rotation matrix R and the translation vector t.

[0014] The beneficial effects of the technical solution provided by this invention include at least the following:

[0015] This invention is designed for low-speed sightseeing vehicles with a speed of ≤10km / h, solves the problem of dynamic interference in densely populated tourist areas, and the calibration results support high-precision positioning, ensuring driving safety on narrow paths.

[0016] This invention utilizes tight coupling modeling of longitudinal displacement of wheel speed gauge and lateral compensation of IMU to significantly reduce pose calculation error when vehicle speed is ≤10km / h, providing a reliable benchmark for subsequent point cloud motion compensation.

[0017] This invention constructs a nonlinear optimization problem that includes point cloud-trajectory matching residuals, wheel velocity meter scale consistency constraints, and IMU-LiDAR attitude constraints. This allows for a more comprehensive consideration of the relationships between different sensors, optimization of the extrinsic parameter matrix, and ensures high accuracy of the calibration results.

[0018] This invention generates an enhanced point cloud map by performing motion compensation and time alignment on static point clouds, ensuring temporal consistency between different frames of data and reducing the time and complexity of data processing. Attached Figure Description

[0019] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The flowchart illustrates the sensor fusion calibration method for unmanned sightseeing vehicles provided in this embodiment of the invention. Detailed Implementation

[0021] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a sensor fusion calibration method for unmanned sightseeing vehicles proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0023] The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0024] The following description, in conjunction with the accompanying drawings, details a specific scheme for a sensor fusion calibration method for an unmanned sightseeing vehicle provided by the present invention.

[0025] Please refer to Figure 1 This is a flowchart of a sensor fusion calibration method for an unmanned sightseeing vehicle provided by an embodiment of the present invention. The algorithm includes the following steps:

[0026] Step S1: When the sightseeing vehicle speed is ≤10km / h, the wheel speed meter pulse signal, IMU angular velocity and acceleration data are synchronously collected through the CAN bus using the PTP protocol. Based on the Ackerman steering geometry, a tightly coupled micro-displacement model is constructed to calculate the inter-frame pose change in the vehicle coordinate system.

[0027] Step S1 further includes the following sub-steps:

[0028] S1-1 uses a fixed sampling frequency to synchronously acquire wheel speed meter pulse signals, IMU angular velocity and acceleration data via CAN bus using PTP protocol, achieving microsecond-level time synchronization;

[0029] S1-2, the longitudinal displacement Δx is calculated based on the wheel speed gauge pulse signal. The specific formula is as follows:

[0030] ;

[0031] Where N represents the number of pulses, i.e. the number of pulses recorded by the wheel speed meter within a specific time interval, r represents the wheel radius in meters, n represents the number of pulses output by the wheel speed meter per revolution of the wheel in pulses / revolution, and c represents the wheel speed meter calibration coefficient.

[0032] The online calibration steps for the wheel speed gauge calibration coefficient c include:

[0033] In the iterative calculation, the value range of the forced constraint c is [0.95, 1.05]. If it exceeds the range, the boundary value is taken.

[0034] While the sightseeing vehicle is traveling at a constant speed in a straight line, record the number of pulses N and the actual displacement D measured by the wheel speed meter;

[0035] There are m groups The data was fitted using the least squares method to obtain the wheel speed gauge calibration coefficient c. The specific formula is as follows:

[0036] ;

[0037] Where m represents the number of data points, i.e., the number of samples used for fitting. This represents the number of wheel speedometer pulses corresponding to the i-th data point. This represents the actual driving distance corresponding to the i-th data point;

[0038] When the change in the value of c calculated in three consecutive iterations is less than Calibration will terminate at this time;

[0039] The calibration coefficient c is updated in real time to the calculation module of the micro-displacement model;

[0040] S1-3, perform zero bias compensation on the IMU angular velocity, and collect the average value of the IMU angular velocity when the vehicle is stationary as the zero bias compensation value. The specific formula is as follows:

[0041] ;

[0042] in, Indicates the angular velocity after compensation. This represents the dynamically acquired IMU angular velocity. This represents the zero bias compensation value;

[0043] S1-4, based on the compensated angular velocity, calculates the change in heading angle Δθ and the lateral displacement compensation Δy. The specific formula is as follows:

[0044] ;

[0045] in, Indicates the angular velocity after compensation. The sampling time interval is represented in seconds, and v represents the current vehicle speed in meters per second.

[0046] S1-5, by fusing the longitudinal displacement Δx of the wheel speedometer and the lateral displacement compensation Δy of the IMU through extended Kalman filtering, outputs the optimized inter-frame pose change, which includes the displacement Δx, Δy and the change in heading angle in the vehicle coordinate system. ;

[0047] The process noise covariance matrix of the extended Kalman filter is:

[0048] ;

[0049] in, Corresponding to x-axis displacement noise, in square meters. Corresponding y-axis displacement noise, in square meters. The corresponding heading angle noise is expressed in square radians.

[0050] It should be noted that when the value of the wheel speed gauge calibration coefficient c exceeds [0.95, 1.05] for three consecutive times, reset c=1 and check the tire pressure and wheel speed gauge wiring.

[0051] The vehicle coordinate system is defined with the origin at the center of the rear axle, the x-axis pointing longitudinally towards the front of the vehicle, the y-axis perpendicular to the x-axis pointing to the left, and the z-axis perpendicular to the ground pointing upwards, conforming to the right-hand rule.

[0052] Extended Kalman filtering is a nonlinear filtering algorithm that progressively optimizes state estimation through two steps: prediction and update. The prediction step involves predicting the state at the next moment based on the system's motion model, while the update step involves revising the predicted state by incorporating measurement data to obtain a more accurate estimate.

[0053] Step S2: Use motion prior to filter dynamic objects in the continuous multi-frame point cloud of solid-state lidar, use motion prediction provided by wheel speed meter and IMU to remove dynamic point cloud, and extract static geometric features.

[0054] Step S2 further includes the following sub-steps:

[0055] S2-1, Based on the inter-frame pose change obtained in step S1, a homogeneous transformation matrix is ​​constructed using the rigid body kinematics principle to transform the point cloud coordinates of the k-1th frame to the kth frame, and the theoretical static position of each point in the current frame point cloud is predicted.

[0056] S2-2, calculate the residual distance d between the actual measured position and the theoretical static position of each laser point. The specific formula is as follows:

[0057] ;

[0058] in, These are the coordinates of the actual measured points. To predict static position coordinates;

[0059] S2-3, determine the dynamic point based on the preset dynamic threshold, when... When, it is determined to be a dynamic point. When the time is right, it is determined to be a static point;

[0060] Dynamic threshold The adjustment is made dynamically based on the vehicle speed v, specifically as follows:

[0061]

[0062] S2-4, perform Euclidean clustering on the point cloud that is determined to be a dynamic point, remove noise points with a preset volume threshold, and calculate the volume through the bounding box of the point cloud.

[0063] S2-5, calculate the curvature of the static point cloud, retain planar feature points with curvature less than 0.05, and use the covariance matrix eigenvalue decomposition method for curvature calculation;

[0064] S2-6, Time alignment is performed on the static point cloud after filtering multiple consecutive frames to construct a static environment map. The time alignment adopts the linear interpolation method, and the time alignment is performed on the static point cloud after filtering multiple consecutive frames according to the sampling frequency of the wheel speed meter.

[0065] It should be noted that the curvature feature extraction steps include:

[0066] Construct a KD-tree from the point cloud to search for the 50 nearest neighbors;

[0067] Calculate the eigenvalues ​​of the covariance matrix ;

[0068] Preserving curvature point.

[0069] Linear interpolation is used to perform time alignment on static point clouds after filtering multiple consecutive frames based on the sampling frequency of the wheel speed meter, ensuring the temporal continuity of the point cloud data.

[0070] Step S3: Perform motion compensation on the static point cloud based on the pose change to generate a time-aligned enhanced point cloud map;

[0071] Step S3 further includes the following sub-steps:

[0072] S3-1, Based on the inter-frame pose change calculated in step S1, construct the inter-frame pose transformation matrix:

[0073] ;

[0074] in, This represents the two-dimensional homogeneous transformation matrix from frame k-1 to frame k. This represents the change in heading angle of the k-th frame relative to the (k-1)-th frame. This represents the translation distance of the k-th frame relative to the (k-1)-th frame along the x-axis of the vehicle coordinate system. The translation distance of the k-th frame relative to the (k-1)-th frame along the y-axis of the vehicle coordinate system;

[0075] S3-2, Perform point-by-point motion compensation on the static point cloud, and then move the point cloud of the (k-1)th frame... The specific formula for transforming to the coordinate system of the k-th frame is as follows:

[0076] ;

[0077] in, This represents the point cloud transformed to the k-frame coordinate system. This represents the point cloud coordinates of the (k-1)th frame, where the point cloud coordinates are represented using homogeneous coordinates.

[0078] S3-3, Voxel filtering and fusion are performed on the point cloud after compensation for M consecutive frames to generate a time-aligned enhanced point cloud map, wherein:

[0079] M≥3;

[0080] The voxel resolution is adjusted inversely to the point cloud density.

[0081] For each voxel lattice, the centroid of the point cloud is taken as the representative point.

[0082] It should be noted that selecting 3 frames as the lower limit of the frame number M can form a stable triangular constraint body.

[0083] In step S3-3, the mean and standard deviation of the cloud computing are calculated for each point in each voxel grid, and points that deviate from the centroid by more than twice the standard deviation are removed.

[0084] Step S4: Optimize the extrinsic parameter matrix of the solid-state lidar, wheel speed sensor, and IMU by minimizing the error between the enhanced point cloud map and the preset calibration reference. The extrinsic parameter matrix includes the rotation matrix R and the translation vector t.

[0085] Step S4 further includes the following sub-steps:

[0086] S4-1, Construct a trajectory point set based on wheel speed meter. The trajectory points include position coordinates, timestamps, and mileage information calculated by accumulating wheel speed meter pulses.

[0087] S4-2, establish the initial external parameter guesses for the lidar coordinate system and the wheel speedometer and IMU coordinate systems;

[0088] S4-3, Construct a nonlinear optimization problem with constraints, with the objective function as follows:

[0089]

[0090] in and These are the weighting coefficients of each constraint term. These weighting coefficients are adjusted experimentally to minimize the objective function. The specific constraints are as follows:

[0091] The formula for the residual of point cloud-trajectory matching is:

[0092] in, The point cloud-trajectory matching residual is an indicator that measures the degree of matching between the lidar point cloud after rotation and translation transformation and the wheel speedometer trajectory points. R represents the rotation matrix to be solved, which is used to rotate the lidar point cloud from its own coordinate system to a direction that matches the coordinate system of the wheel speedometer trajectory points. t represents the translation vector to be solved. This represents the coordinates of the a-th lidar point cloud. represents the coordinates of the a-th wheel speedometer trajectory point, and j represents the number of point clouds;

[0093] The formula for the consistency constraint of wheel speed gauge dimensions is:

[0094] ;

[0095] in, This represents the wheel speedometer dimensional consistency constraint, used to measure the consistency between the mileage measured by the wheel speedometer and the actual mileage calculated based on the wheel speedometer trajectory points. 's' represents the wheel speedometer mileage data. This represents the coordinates of the (a+1)th wheel speedometer trajectory point. Represents the scaling factor, scaling factor The initial value is set to 1, and constraints are applied during the iteration process. The tire slip ratio is limited to ±10%.

[0096] The attitude constraint formula for IMU-LiDAR is:

[0097]

[0098] in, This represents the attitude constraint term of the IMU-LiDAR, used to measure the difference between the rotation matrix measured by the IMU and the local rotation matrix of the LiDAR. Represents the rotation matrix of the IMU measurement. This represents the local rotation matrix of the lidar, where F denotes the subscript identifier of the Frobenius norm;

[0099] S4-4 uses the Levenberg-Marquardt algorithm with boundary constraints to solve the optimization problem. The iteration terminates when any of the following conditions are met:

[0100] The relative change in the objective function J value is less than ;

[0101] scale factor The change is less than ;

[0102] The number of iterations exceeded 100;

[0103] scale factor If the value is at the boundary value for three consecutive iterations, the calibration is considered to have failed.

[0104] If the calibration error exceeds the preset calibration error threshold three consecutive times, the following steps will be executed in sequence:

[0105] Reset wheel speedometer parameters;

[0106] Adjust voxel resolution;

[0107] Reinitialize the IMU.

[0108] It should be noted that the calibration reference is a preset reflector calibration field. The reflector calibration field has the characteristics of high reflectivity and easy detection, and is suitable for high-precision calibration. The size of the reflector calibration field is 1m×1m, and it is arranged in multiple positions in the calibration area to ensure coverage of key areas of the vehicle's driving path.

[0109] The calibration error threshold is the maximum allowable value of the optimized residual of the extrinsic parameter matrix, which is determined based on the positioning accuracy requirements of the unmanned sightseeing vehicle.

[0110] In this way, a sensor fusion calibration method for driverless sightseeing vehicles can be achieved.

[0111] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A sensor fusion calibration method for a driverless sightseeing vehicle, characterized in that, The method comprises: Step S1, when the sightseeing vehicle speed is less than or equal to 10 km / h, wheel speed pulse signals, IMU angular velocity and acceleration data are synchronously collected through a CAN bus by using a PTP protocol, a tightly coupled micro-displacement model is constructed based on Ackerman steering geometry, and an inter-frame pose change quantity in a vehicle body coordinate system is calculated; Step S2, dynamic objects in a plurality of continuous frames of point clouds of the solid-state laser radar are filtered by using motion priors, dynamic point clouds are removed by using motion prediction provided by the wheel speed sensor and the IMU, and static geometric features are extracted; Step S3, the static point clouds are motion compensated based on the pose change quantity, and a time-aligned enhanced point cloud map is generated; Step S4, by minimizing errors between the enhanced point cloud map and a preset calibration reference, an external parameter matrix of the solid-state laser radar, the wheel speed sensor and the IMU is optimized, the external parameter matrix comprising a rotation matrix R and a translation vector t; In step S4, the following sub-steps are further included: S4-1, a trajectory point set based on the wheel speed sensor is constructed, the trajectory point comprising a position coordinate, a time stamp and mileage information calculated by wheel speed sensor pulse accumulation; S4-2, initial external parameter guesses of a laser radar coordinate system and wheel speed sensor and IMU coordinate systems are established; S4-3, a nonlinear optimization problem containing constraints is constructed, and an objective function is: ； wherein , and are weight coefficients of each constraint term, which are adjusted by experiment to minimize the optimization objective function, with the following specific constraints: A point cloud-trajectory matching residual formula is: ； wherein, represents a point cloud-track matching residual, which is an index for measuring the matching degree between the laser radar point cloud after rotation and translation transformation and the odometry track point, represents a rotation matrix to be solved, which is used to rotate the laser radar point cloud from its own coordinate system to a direction matched with the coordinate system in which the odometry track point is located, represents a translation vector to be solved, represents the coordinates of the a-th laser radar point cloud, represents the coordinates of the a-th odometry track point, represents the number of point clouds; A wheel speed sensor scale consistency constraint formula is: ； wherein, represents a wheel speed meter scale consistency constraint term for measuring consistency between wheel speed meter measured mileage and actual mileage calculated according to wheel speed meter trajectory points, s represents wheel speed meter mileage data, represents the coordinates of the a+1th wheel speed meter trajectory point, represents a scale factor, the scale factor is set to 1 initially, and is constrained to limit the tire slip ratio within the range of ±10%. An IMU-laser radar attitude constraint formula is: ； wherein, denotes the IMU-lidar pose constraint term, which measures the difference between the rotation matrix of the IMU measurement and the local rotation matrix of the lidar, denotes the rotation matrix of the IMU measurement, denotes the local rotation matrix of the lidar, F denotes the Frobenius norm. S4-4, a Levenberg-Marquardt algorithm with boundary constraints is used to solve the optimization problem, and iteration is terminated when any of the following conditions is met: The relative change in the objective function J value is less than ; Scale factor The amount of change in the scale factor is less than ; The number of iterations exceeds 100 times; the scale factor If the three consecutive iterations are all at the boundary value, then the calibration is determined to have failed.

2. The sensor fusion calibration method for the unmanned sightseeing vehicle according to claim 1, characterized in that: In step S1, the following sub-steps are further included: S1-1, wheel speed sensor pulse signals, IMU angular velocity and acceleration data are synchronously acquired through a CAN bus by using a PTP protocol at a fixed sampling frequency, and microsecond-level time synchronization is achieved; S1-2, calculating the longitudinal displacement based on the wheel speed meter pulse signal The specific formula is: ； N represents the number of pulses, that is, the number of pulses recorded by the wheel speed sensor within a certain time interval, r represents the wheel radius, which is measured in meters, n represents the number of pulses output by the wheel speed sensor per wheel rotation, which is measured in pulses / revolution, and c represents the wheel speed sensor calibration coefficient; The online calibration step of the wheel speed sensor calibration coefficient c comprises: In the iterative calculation, the value range of c is forced to be [0.95, 1.05], and if it exceeds the range, the boundary value is taken; When the sightseeing vehicle travels at a constant speed in a straight line, the number of pulses N and the actual displacement D measured by the wheel speed sensor are recorded; There are m groups Data, by least square method fitting wheel speed meter calibration coefficient c, the specific formula is: ； wherein m represents the number of data points, i.e. the number of samples used for the fitting, represents the number of tachometer pulses corresponding to the i-th data point, represents the actual distance traveled corresponding to the i-th data point; When the variation of c value calculated by 3 times of successive iteration is less than the calibration is terminated; The calibration coefficient c is updated to the calculation module of the micro-displacement model in real time; S1-3, zero offset compensation is performed on the IMU angular velocity, and the average value of the IMU angular velocity when the vehicle is stationary is taken as the zero offset compensation value, and the specific formula is: ； wherein, represents the compensated angular velocity, represents the dynamically acquired IMU angular velocity, represents a zero offset compensation value; S1-4, based on the compensated angular velocity, calculate a heading angle change amount Δθ and a lateral displacement compensation amount The specific formula is: ； wherein, denotes the compensated angular velocity, denotes the sampling time interval in seconds, v denotes the current vehicle speed in meters per second; S1-5, the longitudinal displacement Δx of the wheel speed sensor and the lateral displacement compensation amount Δy of the IMU are fused by using an extended Kalman filter, and the optimized inter-frame pose change quantity is output, the pose change quantity comprising a displacement amount Δx, Δy in the vehicle body coordinate system and a heading angle change amount Δθ; The process noise covariance matrix of the extended Kalman filter is: ； wherein, corresponding to x-axis displacement noise, in square meters, corresponding to y-axis displacement noise, in square meters, corresponding to heading angle noise, in square radians.

3. The unmanned tourist vehicle sensor fusion calibration method according to claim 1, characterized in that: wherein in step S2, the following sub-steps are further included: S2-1, based on the inter-frame pose change obtained in step S1, a homogeneous transformation matrix is constructed using the rigid body kinematics principle to transform the k-1 frame point cloud coordinates to the k frame, and the theoretical static position of each point in the current frame point cloud is predicted; S2-2, the residual distance d of the actual measurement position and the theoretical static position of each laser point is calculated, and the specific formula is: ； wherein, is the actual measured point coordinate, is the predicted static position coordinate; S2-3, determining a dynamic point according to a preset dynamic threshold, when the dynamic point is determined, when the static point is determined. The dynamic threshold The dynamic threshold is adjusted according to the vehicle speed v in a specific manner. ； S2-4, the point cloud determined as dynamic points is subjected to Euclidean clustering, and the noise points in the preset volume threshold are removed, and the volume is calculated by the bounding box of the point cloud; S2-5, curvature calculation is performed on the static point cloud, and the planar feature points with a curvature less than 0.05 are retained, and the curvature calculation adopts the eigenvalue decomposition method of the covariance matrix; S2-6, the filtered static point clouds of continuous multiple frames are time-aligned to construct a static environment map, and the time alignment adopts a linear interpolation method, and the filtered static point clouds of continuous multiple frames are time-aligned according to the sampling frequency of the wheel speed meter.

4. The unmanned tourist vehicle sensor fusion calibration method according to claim 1, characterized in that: wherein in step S3, the following sub-steps are further included: S3-1, based on the inter-frame pose change calculated in step S1, an inter-frame pose transformation matrix is constructed: ； wherein, denotes a two-dimensional homogeneous transformation matrix from the k-1 frame to the k frame, denotes a change in heading angle of the k frame relative to the k-1 frame, denotes a translation distance of the k frame relative to the k-1 frame along the x axis of the vehicle body coordinate system, denotes a translation distance of the k frame relative to the k-1 frame along the y axis of the vehicle body coordinate system; S3-2, Perform point-by-point motion compensation on the static point cloud, and then move the point cloud of the (k-1)th frame... The specific formula for transforming to the coordinate system of the k-th frame is as follows: ； wherein, represents a point cloud transformed to a k-frame coordinate system, represents a point cloud coordinate of the k-1th frame, and the point cloud coordinate is expressed in homogeneous coordinates; S3-3, voxel filtering fusion is performed on the compensated point clouds of continuous M frames to generate a time-aligned enhanced point cloud map, wherein: M≥3; The voxel resolution is inversely proportional to the point cloud density; The centroid of the point cloud in each voxel grid is taken as the representative point.

5. The unmanned tourist vehicle sensor fusion calibration method according to claim 1, characterized in that: When the calibration error exceeds the preset calibration error threshold for three consecutive times, the following steps are sequentially executed: Reset the wheel speed meter parameters; Adjust the voxel resolution; Reinitialize the IMU.

Images