Roadside laser radar calibration verification method based on intelligent cone barrel

Abstract

The invention discloses a roadside laser radar calibration verification method based on an intelligent cone barrel, and relates to the technical field of roadside equipment calibration, and the method comprises the steps: obtaining a 3D model of the intelligent cone barrel; identifying the 3D point cloud of the intelligent conical barrel; designing a point cloud registration algorithm and performing point cloud registration on the 3D model and the 3D point cloud of the intelligent cone barrel to obtain calibration parameters and calibration precision indexes of the roadside laser radar; setting a distance threshold value and a direction threshold value; calculating a distance activation value and a direction activation value; according to the distance activation value and the direction activation value, calculating to obtain a corresponding matching score; according to the matching score, obtaining sensing data of the intelligent cone barrel; and calculating a precision index of the roadside sensing data. According to the invention, the current sensor calibration error is output in real time based on the positioning information and the laser radar sensor data uploaded by the intelligent cone buckets, the road side laser radar calibration parameters can be calculated by using the plurality of intelligent cone buckets, and the device can also be used as a facility for monitoring road side sensing data.

Description

Technical Field

[0001] The present invention relates to the technical field of roadside equipment calibration, and in particular to a roadside laser radar calibration and verification method based on an intelligent cone bucket. Background Art

[0002] my country's intelligent connected vehicles have officially entered a new stage of large-scale promotion at the city level and exploration of new business models. At present, the perception method of laser radar plus camera is usually used for key intersections. Due to the lack of convenient calibration and verification methods, the problem of poor perception data quality caused by inaccurate calibration is often reflected during project acceptance.

[0003] In the prior art, the calibration and verification of roadside perception systems are usually carried out in the following three ways: one is to select points on high-precision maps. This method relies on high-precision maps, raises the calibration threshold, and is not universal; the second is to use RTK or total stations to collect the coordinates of static target feature points. This method consumes a lot of manpower and is time-consuming; the third is to collect trajectory data through a true value vehicle equipped with a combined inertial navigation. This method has high requirements for the combined inertial navigation calibration of the true value vehicle itself, cannot be verified online, and needs to be driven according to a certain distinguishable trajectory. In addition, due to traffic regulations, it is not convenient to perform error verification of stationary targets.

[0004] As the most common traffic safety facility, cones are small in size and highly mobile, and do not hinder traffic, so they can be placed in the test area for a long time for calibration and verification. Based on this, the present invention designs an online calibration and verification method for roadside LiDAR based on smart cones, which can output the current sensor calibration error in real time based on the positioning information and LiDAR sensor data uploaded by the smart cones, and can use multiple smart cones to calculate the calibration parameters of roadside LiDAR, and can also be used as a facility for monitoring roadside perception data. Summary of the invention

[0005] The present invention provides a roadside laser radar calibration and verification method based on an intelligent cone bucket, so as to overcome at least one technical problem existing in the prior art.

[0006] The embodiment of the present invention provides a roadside laser radar calibration verification method based on an intelligent cone bucket, comprising:

[0007] Collecting smart cone positioning data and roadside perception system data; the smart cone positioning data includes the spatial positioning coordinates and posture positioning information of the smart cone, the roadside perception system data includes point cloud data and roadside perception data, and the roadside perception data includes at least the perception data of the smart cone;

[0008] Acquire a 3D model of the smart cone bucket according to the spatial positioning coordinates and posture positioning information of the smart cone bucket;

[0009] Identify the 3D point cloud of the smart cone according to the point cloud data;

[0010] Design point cloud registration algorithm;

[0011] Performing point cloud registration on the 3D model of the smart cone and the 3D point cloud using the point cloud registration algorithm to obtain calibration parameters of the roadside laser radar;

[0012] Calculating a calibration accuracy index according to the registration result; the calibration accuracy index at least includes position accuracy and attitude accuracy;

[0013] Set distance threshold and direction threshold;

[0014] Calculating the distance difference and direction difference between the roadside sensing data and the smart cone positioning data;

[0015] Calculating a distance activation value according to the distance threshold and the distance difference;

[0016] Calculating a direction activation value according to the direction threshold and the direction difference;

[0017] Calculate the corresponding matching score according to the distance activation value and the direction activation value;

[0018] Acquiring the perception data of the smart cone bucket according to the matching score;

[0019] Based on the perception data of the smart cone, and the spatial positioning coordinates and posture positioning information of the smart cone, the accuracy index of the roadside perception data is calculated.

[0020] Optionally, the point cloud registration algorithm includes a point cloud registration formula and constraints; wherein,

[0021] The point cloud registration formula is expressed as p i represents the i-th point in the 3D point cloud, q i represents the corresponding point in the 3D model, R represents the rotation matrix, R includes the roll angle α, pitch angle β and yaw angle γ, t represents the translation vector, and N represents the number of points; w i Represents the weight coefficient of the i-th point cloud;

[0022] The constraints include attitude change angle constraints and convergence condition constraints;

[0023] The attitude change angle constraint is expressed as α min ≤α≤α max , β min ≤β≤β max , γ min ≤γ≤γmax ; Among them, α max , α min Respectively represent the upper and lower limits of the roll angle change, β max , β min Respectively represent the upper and lower limits of the pitch angle change, γ max , γ min Respectively represent the upper and lower limits of the yaw angle change;

[0024] The convergence condition constraints include dynamic constraints and static constraints. The dynamic constraints are expressed as ∈ dynamic =∈0+α·v, where ∈0 represents the basic error threshold in a static scene, α represents the weight factor adjusted according to the dynamic changes of the scene, v represents the movement speed of the cone barrel, ∈ dynamic represents the convergence error threshold; the static constraint is represented by p error <∈ position , r error <∈ rotation ; Among them, p error represents the position error threshold, ∈ position represents the position error threshold, t i,current Represents the translation vector of the i-th point cloud in this iteration, t i,previous represents the translation vector of the last iteration of the i-th point cloud, r error represents the attitude error threshold, r error =|R current -R previous |,∈ rotation represents the attitude error threshold, R current Represents the rotation matrix of this iteration, R previous Represents the rotation matrix of the previous iteration.

[0025] Optionally, the weight factor w of the point cloud i Includes multi-scale error weighting or sensor fusion weights;

[0026] The multi-scale error weight refers to the weight coefficient w of the point cloud i It shrinks as the number of point clouds decreases;

[0027] The sensor fusion weight refers to the weight w of the point cloud data collected by the roadside perception system vision The weight w of the smart cone positioning data imu The sum of i =w vision +w imu .

[0028] Optionally, the point cloud registration algorithm is used to perform point cloud registration on the 3D model and 3D point cloud of the smart cone barrel, specifically:

[0029] Determine whether the smart cone is in a stationary state. If so, perform point cloud registration on the 3D model and 3D point cloud of the smart cone based on the posture change angle constraint and the static constraint using the point cloud registration formula, and remove abnormal points during the registration process; if not, perform point cloud registration on the 3D model and 3D point cloud of the smart cone based on the posture change angle constraint and the dynamic constraint using the point cloud registration formula, and remove abnormal points during the registration process.

[0030] Optionally, outliers are removed during the registration process, specifically:

[0031] Set the anomaly threshold d threshold ;

[0032] Calculate the point cloud registration error according to the point cloud registration formula;

[0033] It is determined whether the point cloud registration error is greater than the abnormal threshold. If so, it indicates that the corresponding point cloud is an abnormal point and the point cloud is removed; if not, it indicates that the corresponding point cloud is a normal point and the point cloud is retained.

[0034] Optionally, the distance activation value is expressed as Among them, D t represents the distance activation value, δ d represents the distance threshold, d i represents the distance difference between the roadside perception data and the smart cone positioning data obtained by the i-th detection, n represents the number of detections, and λ d Represents the distance conversion factor.

[0035] Optionally, the direction activation value is expressed as Among them, H t represents the direction activation value, δ h represents the direction threshold, h i represents the direction difference between the roadside perception data and the smart cone positioning data obtained by the i-th detection, n represents the number of detections, and λ h Represents the direction conversion factor.

[0036] Optionally, the matching score is expressed as Where M represents the matching score and S(·) represents the Sigmoid activation function.

[0037] Optionally, the perception data of the smart cone barrel is obtained according to the matching score, specifically:

[0038] Determine whether the matching score is greater than the matching threshold. If so, it indicates that the match is successful, and the corresponding roadside perception data is the perception data of the smart cone; if less than or equal to, it indicates that the match is unsuccessful, and the corresponding roadside perception data is not the perception data of the smart cone.

[0039] Optionally, the perception data of the smart cone barrel includes the perception coordinates and the perception posture of the smart cone barrel; the accuracy index includes at least a position error and a posture error;

[0040] According to the perception data of the smart cone, and the spatial positioning coordinates and posture positioning information of the smart cone, the accuracy index of the roadside perception data is calculated, specifically:

[0041] Calculating the position error according to the sensing coordinates and spatial positioning coordinates of the smart cone barrel;

[0042] The posture error is calculated according to the sensed posture and posture positioning information of the smart cone barrel.

[0043] The innovative features of the embodiments of the present invention include:

[0044] 1. In this embodiment, the calibration parameters of the roadside LiDAR are calculated based on the positioning information and LiDAR sensor data uploaded by the smart cones, which avoids the high cost of high-precision maps and is not restricted by the map range. Since the smart cones themselves are feature objects in the LiDAR point cloud, it is more convenient than searching for feature points in the high-precision map in the LiDAR point cloud, and there is no loss in calibration accuracy, which is one of the innovations of the embodiment of the present invention.

[0045] 2. In this embodiment, based on the positioning information and lidar sensor data uploaded by the smart cone, the current sensor calibration error is output in real time to complete the calibration verification. In addition, since the smart cone can send its own coordinates in real time, the time of heterogeneous data is synchronized, which avoids the problem of inaccurate calibration results due to the movement of feature points in the existing calibration verification method. This is one of the innovations of the embodiment of the present invention.

[0046] 3. In this embodiment, the roadside perception data is monitored based on the positioning information and lidar sensor data uploaded by the smart cones. Since the smart cones can be placed stationary at the test site for a long time, they are not likely to affect the traffic flow, and it is easy to detect the errors caused by the exposure time of the mechanical lidar. In terms of application scenarios, it is also more suitable for long-term monitoring and early warning of roadside perception data than the real value vehicle solution, which is one of the innovations of the embodiment of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A flow chart of a calibration verification method provided by an embodiment of the present invention;

[0049] Figure 2 A flow chart for removing outliers provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] It should be noted that the terms "including" and "having" and any variations thereof in the embodiments of the present invention and the accompanying drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0052] The embodiment of the present invention discloses a roadside laser radar calibration verification method based on an intelligent cone bucket, which is described in detail below.

[0053] Figure 1 For a flow chart of the calibration verification method provided by the embodiment of the present invention, please refer to Figure 1 The method for calibrating and verifying a roadside laser radar based on an intelligent cone bucket provided in an embodiment of the present invention includes:

[0054] Step 1: Collect smart cone positioning data and roadside perception system data; the smart cone positioning data includes the spatial positioning coordinates and posture positioning information of the smart cone, and the roadside perception system data includes point cloud data and roadside perception data, and the roadside perception data at least includes the perception data of the smart cone;

[0055] Step 2: Obtain a 3D model of the smart cone bucket according to the spatial positioning coordinates and posture positioning information of the smart cone bucket;

[0056] Step 3: Identify the 3D point cloud of the smart cone bucket based on the point cloud data;

[0057] Step 4: Design point cloud registration algorithm;

[0058] Step 5: Use the point cloud registration algorithm to perform point cloud registration on the 3D model of the smart cone and the 3D point cloud to obtain the calibration parameters of the roadside lidar;

[0059] Step 6: Calculate the calibration accuracy index according to the registration result; the calibration accuracy index at least includes position accuracy and attitude accuracy;

[0060] Step 7: Set the distance threshold and direction threshold;

[0061] Step 8: Calculate the distance difference and direction difference between the roadside perception data and the smart cone positioning data;

[0062] Step 9: Calculate the distance activation value based on the distance threshold and the distance difference;

[0063] Step 10: Calculate the direction activation value based on the direction threshold and the direction difference;

[0064] Step 11: Calculate the corresponding matching score based on the distance activation value and the direction activation value;

[0065] Step 12: Obtain the perception data of the smart cone bucket according to the matching score;

[0066] Step 13: Calculate the accuracy index of the roadside perception data based on the perception data of the smart cone and the spatial positioning coordinates and posture positioning information of the smart cone.

[0067] Specifically, please refer to Figure 1 The roadside laser radar calibration and verification method based on the smart cone provided by the embodiment of the present invention performs calibration and calibration error verification based on the positioning information uploaded by the smart cone and the laser radar sensor data. Therefore, it is first necessary to collect the smart cone positioning data and the roadside perception system data through step 1. The smart cone positioning data includes the spatial positioning coordinates and posture positioning information of the smart cone. The roadside perception system data includes point cloud data and roadside perception data. The roadside perception data at least includes the perception data of the smart cone.

[0068] The core function of the smart cone system is to achieve precise positioning through the RTK (Real-time kinematic) module, which includes GNSS (Global Navigation Satellite System) and IMU (Inertial Measurement Unit). The GNSS antenna is used to receive satellite signals, and the position is corrected through the IMU to achieve centimeter-level positioning accuracy for the smart cone. The real-time position data of the smart cone is sent to the embedded computing module, and the data is fused and processed by combining other sensors (such as accelerometers, gyroscopes, etc.) to calculate the position and posture of the cone in real time. At the same time, the embedded computing module is also responsible for uploading processed data through the communication module and receiving remote commands.

[0069] In this embodiment, the smart cone barrel system is powered by a 12V DC battery to ensure continuous operation in outdoor environments. The battery is charged from the AC mains through a charging module to ensure that the battery can provide power for a long time.

[0070] After obtaining the positioning data of the smart cone, a 3D model of the smart cone can be constructed in step 2 according to the spatial positioning coordinates and posture positioning information of the smart cone. The 3D point cloud obtained by the roadside perception system includes not only the 3D point cloud of the smart cone, but also the 3D point clouds of other traffic participants such as people and vehicles. Therefore, it is necessary to identify the 3D point cloud of the smart cone from all the point cloud data in step 3.

[0071] After obtaining the 3D point cloud of the smart cone, the 3D model of the smart cone can be used to align the 3D point cloud of the smart cone, thereby realizing the calibration of the roadside laser radar. When performing point cloud registration, it is necessary to perform registration according to a point cloud registration algorithm. Therefore, the present invention designs a point cloud registration algorithm through step 4.

[0072] The present invention is based on the ICP (iterative closest point) point cloud matching algorithm, combines the unique geometric features of the smart cone bucket and multi-source sensor information, and designs a point cloud registration formula and constraint conditions suitable for the smart cone bucket.

[0073] When designing the point cloud registration formula, the weight coefficient of the point cloud is added on the basis of ICP. Therefore, the point cloud registration formula is expressed as Among them, p i represents the i-th point in the 3D point cloud, q i represents the corresponding points in the 3D model, R represents the rotation matrix, R includes the roll angle α, pitch angle β and yaw angle γ, t represents the translation vector, N represents the number of points, w iRepresents the weight coefficient of the i-th point cloud. Solved by SVD (singular value decomposition), when E is the smallest, it corresponds to the optimal R and t.

[0074] In the process of point cloud registration, changing the number of point clouds and the point cloud weight coefficient will affect the registration result. Therefore, in this embodiment, the point cloud weight coefficient w is set i The multi-scale error weight can be weighted. When set to the multi-scale error weight, the corresponding relationship of the point cloud at different scales is set to different weights, so that the weight coefficient w of the point cloud i It shrinks as the number of point clouds decreases.

[0075] In addition, the traditional ICP algorithm only considers the point coordinates in the point cloud, but does not consider the posture data measured by the IMU, which may affect the registration accuracy. Therefore, the present invention sets the weight coefficient w of the point cloud i You can also set the sensor fusion weight, that is, the weight coefficient w of the point cloud i is the weight of the point cloud data collected by the roadside perception system vision and the weight w of the smart cone positioning data imu The sum of the points, the weight coefficient of the point cloud is w i =w vision +w imu .

[0076] In addition to setting the weight coefficients, constraints are also added, such as posture change angle constraints and convergence condition constraints, to improve the accuracy and stability of the iteration.

[0077] The attitude change angle constraint is the restriction on the rotation matrix R. Due to the physical characteristics of the smart cone barrel, its attitude will only change within a small range. Therefore, in the process of point cloud registration, the attitude change angle constraint is set to narrow the traversed attitude range so that the rotation angle change range is within a reasonable range, thereby improving the operation efficiency of the algorithm.

[0078] Assume that the rotation angle in the current ICP iteration is expressed in Euler angles. The Euler angles (α, β, γ) correspond to the roll angle, pitch angle, and yaw angle, respectively. In order to ensure the rationality of the attitude, the attitude angle constraint is set to α min ≤α≤α max , β min ≤β≤β max , γ min ≤γ≤γ max , where α max , α min Respectively represent the upper and lower limits of the roll angle change, which can be plus or minus 30 degrees; β max , β min Respectively represent the upper and lower limits of the pitch angle change, which can be plus or minus 30 degrees; γ max, γ min They represent the upper and lower limits of the yaw angle change, and can take values ​​of positive and negative 120 degrees.

[0079] Normally, the smart cone is in a stationary state, so the convergence condition can be set as a static constraint. However, when strong winds or other extreme conditions cause the position of the smart cone to change, the static constraint is not applicable. Therefore, in order to adapt to different scenarios, the present invention sets the convergence condition to include static constraints and dynamic constraints. During the registration process, according to the state of the smart cone, dynamic constraints or static constraints are selected to ensure that the point cloud registration is continuously effective.

[0080] Among them, the dynamic constraint is expressed as ∈ dynamic =∈0+α·v, where ∈0 represents the basic error threshold in a static scene, α represents the weight factor adjusted according to the dynamic changes of the scene, v represents the movement speed of the cone barrel, ∈ dynamic Represents the convergence error threshold.

[0081] When the smart cone is in a stationary state, in order to comprehensively consider the changes in position and posture, the static constraint is defined as a joint convergence condition. The static constraint is expressed as p error <∈ position , r error <∈ rotation ; Among them, p error represents the position error threshold, ∈ position represents the position error threshold, t i,current Represents the translation vector of the i-th point cloud in this iteration, t i,previous represents the translation vector of the last iteration of the i-th point cloud, r error represents the attitude error threshold, r error =|R current -R previous |,∈ rotation represents the attitude error threshold, R current Represents the rotation matrix of this iteration, R previous Represents the rotation matrix of the previous iteration. When both the position error and the attitude error meet the requirements, the registration is considered complete and the iteration is terminated.

[0082] After designing the point cloud registration algorithm, the 3D model of the smart cone and the 3D point cloud can be registered in step 5. After the registration is completed, the calibration parameters of the roadside lidar can be obtained, which means that the calculation of the calibration parameters of the roadside lidar is completed.

[0083] When performing point cloud registration, it is necessary to first determine the state of the smart cone. When the smart cone is in a stationary state, the point cloud registration formula is used for registration, and attitude change angle constraints and static constraints are used during the registration process to ensure the registration accuracy and effectiveness. When the smart cone is in a dynamic state, attitude change angle constraints and dynamic constraints are used during the registration process.

[0084] In addition, in order to improve the robustness of the registration, noise and outliers need to be removed in each iterative registration. In this embodiment, the random sampling consistency RANSAC method is used to filter outliers.

[0085] After obtaining the registration result, in step 6, the calibration accuracy index can be calculated according to the registration result to obtain the calibration error of the current sensor. Here, the calibration accuracy index includes at least position accuracy and attitude accuracy. Among them, position accuracy and calibration accuracy refer to the Euclidean distance and Euler angle error before and after the registration of the point cloud to be measured (3D point cloud of the smart cone bucket).

[0086] In addition to calculating calibration parameters and calibration errors, the present invention can also monitor the accuracy of roadside perception data through the positioning information uploaded by the smart cones and the lidar sensor data. In addition to the perception data of the smart cones, the roadside perception data also includes the perception data of other traffic participants such as vehicles and people. The present invention uses the positioning data of the smart cones to monitor the roadside perception data, thereby judging the accuracy of the sensor. Therefore, it is necessary to identify the perception data of the smart cone data from all the roadside perception data, that is, to perform target matching. This embodiment completes the perception data accuracy monitoring by calculating the joint probability to determine the point-to-point matching relationship, and the probability synthesis rule is based on the DS evidence theory.

[0087] When monitoring the accuracy of perception data, first set the distance threshold δ through step 7. d and direction threshold δ h In step 8, the distance difference and direction difference between the roadside sensing data and the smart cone positioning data are calculated. When calculating the distance difference and direction difference, in order to improve the calculation accuracy, the average value of the distance difference and direction difference over a period of time is taken as the final distance difference and direction difference. Here, the distance difference is expressed as The direction difference is expressed as Among them, d i represents the distance difference between the roadside perception data and the smart cone positioning data obtained by the i-th detection, h i It represents the direction difference between the roadside perception data and the smart cone positioning data obtained in the i-th detection, and n represents the number of detections.

[0088] In step 9, the distance activation value can be calculated based on the distance threshold and the distance difference. Among them, Dt represents the distance activation value, λ d Represents the distance conversion factor. By setting the distance conversion factor, the data can be made more dispersed to avoid the situation where the data is too concentrated, which affects the judgment results or causes errors.

[0089] In step 10, the direction activation value can be calculated based on the direction threshold and the direction difference. The direction activation value is expressed as Among them, H t represents the direction activation value, λ h Indicates the direction conversion factor. Similarly, by setting the direction conversion factor, the data can be made more dispersed to avoid the situation where the data is too concentrated, which affects the judgment result or causes errors.

[0090] In step 11, according to the distance activation value and the direction activation value, the corresponding matching score is calculated as Where M represents the matching score, S(·) represents the Sigmoid activation function, and the activation function is used to map all data to between 0 and 1.

[0091] After obtaining the matching score of each point, in step 12, the perception data of the smart cone is identified using the matching score. When identifying the perception data of the smart cone, it can be done by comparing the size relationship between the matching score and the matching threshold. When the matching score is greater than the matching threshold, the match is successful, indicating that the current roadside perception data is the perception data of the smart cone; otherwise, the match is unsuccessful, indicating that the current roadside perception data is not the perception data of the smart cone. It should be noted that the matching threshold here can be set based on experience, and the present invention does not make specific limitations on this.

[0092] After the perception data of the smart cone is identified by the matching algorithm, the accuracy index of the roadside perception data can be calculated in step 13. The accuracy index here includes the accuracy of both position and attitude, that is, position error and attitude error. Among them, the position error can be calculated using the perception coordinates and spatial positioning coordinates of the smart cone; and the attitude error can be calculated based on the perception attitude and attitude positioning information of the smart cone.

[0093] The roadside lidar calibration and verification method based on smart cones provided by the present invention calculates the calibration parameters of the roadside lidar based on the positioning information and lidar sensor data uploaded by the smart cones, thereby eliminating the high cost of high-precision maps and being not constrained by the map range. Since the smart cones themselves are feature objects in the lidar point cloud, it is more convenient than searching for feature points in the high-precision map in the lidar point cloud, and there is no loss in calibration accuracy.

[0094] The present invention outputs the current sensor calibration error in real time based on the positioning information and lidar sensor data uploaded by the smart cone bucket to complete the calibration verification. In addition, since the smart cone bucket can send its own coordinates in real time, the time of heterogeneous data is guaranteed to be synchronized, thus avoiding the problem of inaccurate calibration results caused by the movement of feature points in the existing calibration verification method.

[0095] In addition, the roadside lidar calibration and verification method based on smart cones provided by the present invention can also monitor roadside perception data based on the positioning information and lidar sensor data uploaded by the smart cones. Since the smart cones can be placed stationary at the test site for a long time, they are not easy to affect traffic flow, and are convenient for troubleshooting errors caused by exposure time of mechanical lidar. In terms of application scenarios, they are also more suitable for long-term monitoring and early warning of roadside perception data than the true value vehicle solution.

[0096] Optionally, Figure 2 For a flowchart of removing outliers provided by an embodiment of the present invention, please refer to Figure 1 and Figure 2 In step 5, outliers are removed during the registration process. Specifically, step 51 sets an outlier threshold d threshold ; Step 52, calculate the point cloud registration error according to the point cloud registration formula; Step 53, determine whether the point cloud registration error is greater than the abnormal threshold. If so, it means that the corresponding point cloud is an abnormal point and the point cloud is removed; if not, it means that the corresponding point cloud is a normal point and the point cloud is retained.

[0097] Specifically, please refer to Figure 2 When removing abnormal points, first define the consistency threshold through step 51, that is, set the abnormal threshold d threshold , when registering the point cloud, the point cloud whose registration error exceeds the threshold is regarded as an abnormal point and is removed. Therefore, in step 52, the point cloud registration error is calculated according to the point cloud registration formula. It should be noted that the point cloud registration error here is the registration result obtained by each iteration calculation in step 5.

[0098] After obtaining the threshold and the registration error, step 53 can be used to determine whether the point cloud registration error is greater than the abnormal threshold. If it is greater than, it means that the corresponding point cloud is an abnormal point and the point cloud is removed; otherwise, it means that the corresponding point cloud is a normal point and the point cloud is retained. threshold The setting may be performed based on the noise level and prior knowledge, and the present invention does not impose any specific limitation on this.

[0099] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0100] Those skilled in the art can understand that the modules in the device in the embodiment can be distributed in the device in the embodiment according to the description of the embodiment, or can be changed accordingly and located in one or more devices different from the embodiment. The modules in the above embodiment can be combined into one module, or can be further divided into multiple sub-modules.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A roadside laser radar calibration and verification method based on intelligent cone bucket, characterized in that: include: Collecting smart cone positioning data and roadside perception system data; the smart cone positioning data includes the spatial positioning coordinates and posture positioning information of the smart cone, the roadside perception system data includes point cloud data and roadside perception data, and the roadside perception data includes at least the perception data of the smart cone; Acquire a 3D model of the smart cone bucket according to the spatial positioning coordinates and posture positioning information of the smart cone bucket; Identify the 3D point cloud of the smart cone according to the point cloud data; Design point cloud registration algorithm; Performing point cloud registration on the 3D model of the smart cone and the 3D point cloud using the point cloud registration algorithm to obtain calibration parameters of the roadside laser radar; Calculating a calibration accuracy index according to the registration result; the calibration accuracy index at least includes position accuracy and attitude accuracy; Set distance threshold and direction threshold; Calculating the distance difference and direction difference between the roadside sensing data and the smart cone positioning data; Calculating a distance activation value according to the distance threshold and the distance difference; Calculating a direction activation value according to the direction threshold and the direction difference; Calculate the corresponding matching score according to the distance activation value and the direction activation value; Acquiring the perception data of the smart cone bucket according to the matching score; Based on the perception data of the smart cone, and the spatial positioning coordinates and posture positioning information of the smart cone, the accuracy index of the roadside perception data is calculated.

2. The roadside laser radar calibration verification method based on the intelligent cone bucket according to claim 1 is characterized in that: The point cloud registration algorithm includes a point cloud registration formula and constraints; wherein, The point cloud registration formula is expressed as p i represents the i-th point in the 3D point cloud, q i represents the corresponding point in the 3D model, R represents the rotation matrix, R includes the roll angle α, pitch angle β and yaw angle γ, t represents the translation vector, and N represents the number of points; w i Represents the weight coefficient of the i-th point cloud; The constraints include attitude change angle constraints and convergence condition constraints; The attitude change angle constraint is expressed as α min ≤α≤α max , β min ≤β≤β max , γ min ≤γ≤γ max ; Among them, α max , α min Respectively represent the upper and lower limits of the roll angle change, β max , β min Respectively represent the upper and lower limits of the pitch angle change, γ max , γ min Respectively represent the upper and lower limits of the yaw angle change; The convergence condition constraints include dynamic constraints and static constraints. The dynamic constraints are expressed as ∈ dynamic =∈0+αv, where ∈0 represents the basic error threshold in a static scene, α represents the weight factor adjusted according to the dynamic changes of the scene, v represents the movement speed of the cone barrel, ∈ dynamic represents the convergence error threshold; the static constraint is represented by p error <∈ position , r error <∈ rotation ; Among them, p error represents the position error threshold, ∈ position represents the position error threshold, t i,current represents the translation vector of the i-th point cloud in this iteration, t i,previous represents the translation vector of the last iteration of the i-th point cloud, r error represents the attitude error threshold, r error =|R current -R previous |,∈ rotation represents the attitude error threshold, R current Represents the rotation matrix of this iteration, R previous Represents the rotation matrix of the previous iteration.

3. The roadside laser radar calibration verification method based on the intelligent cone bucket according to claim 2 is characterized in that: The weight coefficient w of the point cloud i Includes multi-scale error weighting or sensor fusion weights; The multi-scale error weight refers to the weight coefficient w of the point cloud i It shrinks as the number of point clouds decreases; The sensor fusion weight refers to the weight w of the point cloud data collected by the roadside perception system vision The weight w of the smart cone positioning data imu The sum of i =w vision +w imu .

4. The roadside laser radar calibration verification method based on the intelligent cone bucket according to claim 3 is characterized in that: The point cloud registration algorithm is used to perform point cloud registration on the 3D model and 3D point cloud of the smart cone barrel, specifically: Determine whether the smart cone is in a stationary state. If so, perform point cloud registration on the 3D model and 3D point cloud of the smart cone based on the posture change angle constraint and the static constraint using the point cloud registration formula, and remove abnormal points during the registration process; if not, perform point cloud registration on the 3D model and 3D point cloud of the smart cone based on the posture change angle constraint and the dynamic constraint using the point cloud registration formula, and remove abnormal points during the registration process.

5. The roadside laser radar calibration verification method based on the intelligent cone bucket according to claim 4 is characterized in that: In the registration process, outliers are removed, specifically: Set the anomaly threshold d threshold ; Calculate the point cloud registration error according to the point cloud registration formula; It is determined whether the point cloud registration error is greater than the abnormal threshold. If so, it indicates that the corresponding point cloud is an abnormal point and the point cloud is removed; if not, it indicates that the corresponding point cloud is a normal point and the point cloud is retained.

6. The roadside laser radar calibration verification method based on the intelligent cone bucket according to claim 1 is characterized in that: The distance activation value is expressed as Among them, D t represents the distance activation value, δ d represents the distance threshold, d i represents the distance difference between the roadside perception data and the smart cone positioning data obtained by the i-th detection, n represents the number of detections, and λ d Represents the distance conversion factor.

7. The roadside laser radar calibration verification method based on the intelligent cone bucket according to claim 6 is characterized in that: The direction activation value is expressed as Among them, H t represents the direction activation value, δ h represents the direction threshold, h i represents the direction difference between the roadside perception data and the smart cone positioning data obtained by the i-th detection, n represents the number of detections, and λ h Represents the direction conversion factor.

8. The roadside laser radar calibration verification method based on the intelligent cone bucket according to claim 7 is characterized in that: The matching score is expressed as Where M represents the matching score and S(·) represents the Sigmoid activation function.

9. The roadside laser radar calibration verification method based on the intelligent cone bucket according to claim 1 is characterized in that: According to the matching score, the perception data of the smart cone barrel is obtained, specifically: Determine whether the matching score is greater than the matching threshold. If so, it indicates that the match is successful, and the corresponding roadside perception data is the perception data of the smart cone; if less than or equal to, it indicates that the match is unsuccessful, and the corresponding roadside perception data is not the perception data of the smart cone.

10. The roadside laser radar calibration verification method based on intelligent cone bucket according to claim 1 is characterized in that: The perception data of the smart cone bucket includes the perception coordinates and the perception posture of the smart cone bucket; the accuracy index includes at least the position error and the posture error; According to the perception data of the smart cone, and the spatial positioning coordinates and posture positioning information of the smart cone, the accuracy index of the roadside perception data is calculated, specifically: Calculating the position error according to the sensing coordinates and spatial positioning coordinates of the smart cone barrel; The posture error is calculated according to the sensed posture and posture positioning information of the smart cone barrel.

Images