Laser radar reference surface calibration method suitable for airborne pod of unmanned aerial vehicle

Abstract

The invention discloses a method for calibrating a laser radar reference surface suitable for an airborne pod of an unmanned aerial vehicle, and provides a method for calibrating a laser radar reference surface by collecting laser radar point cloud and inertial navigation data of a large-area flat ground to solve the problems of complex algorithm, large error and low efficiency in the existing calibration technology. And traversal calibration parameters and a plane fitting technology are combined to realize full-automatic calibration. The method comprises the following specific steps: selecting a flat ground, collecting data, generating a single-frame file, and screening ground point clouds; calibration parameters are preset, the range of the calibration parameters is traversed, coordinates are converted through an Euler transformation matrix, and plane fitting is carried out in combination with UTM projection; and screening parameters meeting gradient requirements, calculating an average value as a final calibration parameter, and automatically applying the average value to sensor cooperative work. According to the method, the calibration process is simplified, noise interference is reduced by adopting large-area ground data, the calibration precision is high, the whole process is automatically executed by embedded software, and the method can be expanded to a multi-sensor fusion scene.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) and laser radar, and in particular to a laser radar reference surface calibration method suitable for calibrating an onboard pod of an unmanned aerial vehicle (UAV), aiming to improve the accuracy and reliability of the UAV when performing tasks. Background Art

[0002] Since the introduction of LiDAR technology in the 1960s, its functions and uses have been expanding with the continuous development of technology. It has been widely used in aviation, navigation, automobiles, transportation and other fields, and has also shown great application potential in the field of drones in recent years. In order to meet the multi-functional needs of drones, a variety of sensors are usually integrated in their onboard pods, such as LiDAR, infrared cameras, visible cameras, inertial navigation systems (INS), etc. These sensors each generate data based on their own hardware coordinate system. Therefore, when LiDAR works in conjunction with other sensors, it is necessary to ensure that they are in a unified coordinate system or reference plane, otherwise the output results of the collaborative work will be seriously affected, resulting in large errors.

[0003] In order to eliminate the angle deviation between the sensor coordinate system and the unified coordinate system, a variety of calibration methods or technologies have been proposed in the prior art. Among them, a commonly used algorithm is to use laser radar and inertial navigation for joint calibration. This method is based on the feature marker point cloud and calibrates the deviation angle through point-to-point, line-to-line or face-to-face matching algorithms. However, this calibration method has obvious shortcomings:

[0004] High algorithm complexity: This method uses a complex point cloud matching algorithm, which makes the calibration process cumbersome and time-consuming, especially in application scenarios such as drone-mounted pods, which have high requirements for real-time performance and efficiency.

[0005] Limited calibration accuracy: Since the matching algorithm usually uses small targets as calibration features, it is easily affected by noise and interference, resulting in low calibration accuracy and cannot meet some application scenarios with high accuracy requirements. Summary of the invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the present invention proposes a laser radar reference plane calibration method suitable for calibrating the onboard pod of an unmanned aerial vehicle. The method abandons the complex point cloud matching algorithm and adopts the method of traversing the calibration parameters to fit a large area of ​​ground plane point cloud for calibration. The algorithm is simpler and the operation is more convenient. At the same time, it can ensure a higher calibration accuracy and meet the high-precision requirements for the collaborative work of sensors in the fields of unmanned aerial vehicles and so on.

[0007] The technical solution of the present invention is as follows:

[0008] A laser radar reference surface calibration method suitable for calibrating the pod onboard an unmanned aerial vehicle

[0009] A laser radar reference surface calibration method suitable for an unmanned aerial vehicle airborne pod is characterized in that it comprises the following steps:

[0010] Step 1. Data collection: Select a flat ground, fix the lidar and inertial navigation system on the drone's airborne pod, control the drone to hover and scan the ground, and collect lidar point cloud data and inertial navigation data;

[0011] Step 2. Data integration: Import the laser radar point cloud data and inertial navigation data into the data analysis software to generate a single frame file containing the point cloud coordinates (x, y, z) and the corresponding inertial navigation data (longitude, latitude, altitude, pitch angle, roll angle, yaw angle);

[0012] Step 3. Ground point cloud screening: The laser radar ground point cloud that meets the plane characteristic requirements is screened out through the ground point cloud recognition algorithm to generate the ground point cloud file to be calibrated;

[0013] Step 4. Calibration process:

[0014] Preset the initial values ​​of the calibration parameters labelPitch and labelRoll;

[0015] Traverse the calibration parameter combinations and iterate through the following sub-steps:

[0016] 4.1 Determine whether the calibration parameters exceed the preset threshold range, and terminate if so;

[0017] 4.2 Transform the laser radar point cloud coordinates (x, y, z) through the Euler transformation matrix composed of calibration parameters to obtain the calibrated coordinates (x1, y1, z1);

[0018] 4.3 The calibrated coordinates (x1, y1, z1) are transformed into UTM coordinates (x3, y3, z3) in combination with the inertial navigation data, where the xoy plane of the UTM coordinates is parallel to the sea level;

[0019] 4.4 Perform plane fitting on the UTM coordinates and calculate the slope of the fitting plane;

[0020] 4.5 If the plane slope slop is less than 0.2°, record the current calibration parameter combination to the data set labelResult; 4.6 Increase the calibration parameters and repeat the iteration;

[0021] Step 5. Calibration parameter calculation:

[0022] Based on the dataset labelResult, the average values ​​of labelPitch and labelRoll are calculated as the final calibration parameters.

[0023] Furthermore, the flat ground has a slope of less than 0.2° and an area of ​​more than 100m 2 , the hovering time of the UAV is 5 minutes, and the acquisition frequency of the laser radar point cloud data is not less than 10 Hz.

[0024] Furthermore, the data analysis software is one of MATLAB, Python or ROS.

[0025] Furthermore, in step 3, the ground point cloud recognition algorithm is screened based on point cloud density and height distribution characteristics.

[0026] Furthermore, in step 4, the initial calibration parameters labelPitch and labelRoll are both preset to -2.0, and are increased in steps of 0.1° during traversal until labelPitch and labelRoll are both greater than 2.0 and the traversal is terminated.

[0027] Furthermore, the Euler transformation matrix in step 4 is Rz(0)*Ry(labelPitch)*Rx(labelRoll), which is used to convert the laser radar point cloud coordinates (x, y, z) into calibration coordinates (x1, y1, z1), and further combined with the attitude angle (pitch, roll, yaw) and position information (longitude, latitude, height) of the inertial navigation data, and converted into UTM projection coordinates (x3, y3, z3) through the coordinate transformation matrix.

[0028] Furthermore, in step 4, the plane fitting adopts the least square method or the random sampling consistency algorithm, and the calculation of the slope of the fitted plane includes the analysis of the angle between the plane normal vector and the sea level normal vector.

[0029] Furthermore, all data processing and parameter traversal processes in step 4 are fully automated by the embedded software, including parameter adjustment, error threshold judgment and calibration result storage.

[0030] Furthermore, the final calibration parameters in step 5 are generated by weighted averaging or optimal solution screening, and are automatically written into the sensor calibration configuration file of the drone's onboard pod.

[0031] Furthermore, the calculation of the calibration parameters also includes weighted average or median optimization of the parameters in the data set labelResult.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1) By traversing parameters and fitting large-area planes instead of complex point cloud matching, the algorithm efficiency is improved by more than 50%;

[0034] 2) The embedded software automatically performs data acquisition, parameter traversal and calibration without manual intervention;

[0035] 3) Based on large-area ground point cloud, the noise interference of small targets is reduced, and the calibration accuracy is better than 0.2°;

[0036] 4) It can be extended to the collaborative calibration of lidar, infrared camera and visible light camera. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is the lidar calibration flow chart. DETAILED DESCRIPTION

[0038] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments, but this should not limit the protection scope of the present invention.

[0039] This embodiment provides a method for calibrating a laser radar reference surface applicable to a drone pod. Figure 1 As shown, the specific steps include:

[0040] Step 1. Data collection:

[0041] Select a plot with a slope less than 0.2° and an area of ​​more than 100m 2 The laser radar and inertial navigation system are fastened to the pod on the drone, and then the drone is started to hover above the selected ground and scan the ground for 5 minutes to collect a complete data set containing the laser radar point cloud data and inertial navigation data.

[0042] Step 2. Data integration:

[0043] Import the collected LiDAR point cloud data and inertial navigation data into the data analysis software, automatically analyze these data [parse the timestamp of the LiDAR point cloud coordinates (x, y, z), find the longitude, latitude, altitude, pitch, roll and heading angle at the same time in the inertial navigation data, these data form a point cloud data, after parsing all the point cloud data contained in a frame of point cloud, output to the file], and generate a single-frame file. This single-frame file contains the coordinates of the LiDAR point cloud (x, y, z) and the inertial navigation data (longitude, latitude, height, pitch, roll, yaw) corresponding to the time when the point cloud was generated.

[0044] Step 3. Ground point cloud screening:

[0045] Using the ground point cloud recognition algorithm in the data analysis software, the laser radar ground point cloud is automatically identified and screened according to the characteristics of the plane. The screened ground point cloud is saved in a file named "ground point cloud file to be calibrated".

[0046] Step 4. Calibration process:

[0047] The initial values ​​of the calibration parameters labelPitch and labelRoll are preset, for example, labelPitch = -2.0, labelRoll = -2.0. Then, the calibration algorithm is processed on the "ground point cloud file to be calibrated". The calibration algorithm specifically includes the following sub-steps:

[0048] Step 4.1: If labelPitch is greater than 2.0 and labelRoll is greater than 2.0, the calibration algorithm ends because it has exceeded the reasonable calibration parameter range.

[0049] Step 4.2: Multiply the LiDAR point cloud coordinates (x, y, z) by the Euler transformation matrix composed of labelPitch and labelRoll to obtain the calibration coordinates (x1, y1, z1).

[0050] Step 4.3: Multiply the calibration coordinates (x1, y1, z1) by the transformation matrix composed of the inertial navigation data to obtain the UTM coordinates (x3, y3, z3), where x3, y3 are UTM projection coordinates, z3 is the altitude, and the xoy plane of the UTM coordinates is parallel to the sea level.

[0051] Step 4.4: Perform plane fitting on all UTM coordinates, calculate the angle between the normal vector of the fitting plane and the normal vector of the sea level, and obtain the slope.

[0052] Step 4.5: If the slope slop is less than 0.2°, the current labelPitch and labelRoll are recorded in the data set labelResult as a possible calibration parameter combination.

[0053] Step 4.6: Increment the value of labelPitch or labelRoll, and then repeat the above substeps to traverse more calibration parameter combinations.

[0054] Step 5. Calibration parameter calculation:

[0055] When the calibration algorithm is completed, the parameters in labelResult are weighted averaged or the optimal solution is screened to obtain the average value of labelPitch and labelRoll, which is automatically written into the drone sensor configuration file.

[0056] Step 6. Calibration verification: By collecting dynamic scene data in real time, verify whether the calibration error is less than the preset threshold. If it does not meet the standard, recalibrate.

[0057] In this embodiment, the preset threshold is set to 2.0, but this threshold can be adjusted according to actual needs.

[0058] Embodiment 2:

[0059] To carry out the calibration in an empty parking lot, the steps are as follows:

[0060] Select 100m 2 The hardened ground with a slope of 0.1° is used to fix the drone pod;

[0061] Start the laser radar (scanning frequency 20Hz) and inertial navigation, and continue to collect data for 5 minutes;

[0062] The software automatically generates single-frame files and filters ground point clouds;

[0063] Traverse labelPitch and labelRoll with a step size of 0.1°, and record the parameter combinations with slopes less than 0.2°;

[0064] The final calibration parameters are calculated as labelPitch=0.3°, labelRoll=-0.1°, and the error verification is 0.08°, which meets the requirements.

[0065] Embodiment 3:

[0066] In the grassland scene, by adjusting the plane fitting algorithm to RANSAC, the point cloud interfered by vegetation can be effectively removed, and the calibration accuracy is improved to 0.15°.

[0067] The present invention collects and analyzes laser radar point cloud data and inertial navigation data, and uses a calibration algorithm to obtain the calibration parameters of the laser radar calibration reference surface. This method can improve the accuracy and efficiency of laser radar working in coordination with other sensors, and is suitable for application scenarios such as drone-mounted pods.

[0068] Through the description of the above embodiments, those skilled in the art can clearly understand the specific implementation of the method for calibrating the laser radar reference surface applicable to the airborne pod of a drone provided by the present invention, and can make appropriate modifications and changes as needed. These modifications and changes should fall within the protection scope of the present invention.

Claims

1. A laser radar reference surface calibration method suitable for an unmanned aerial vehicle airborne pod, characterized in that: The following steps are involved: Step 1. Data collection: Select a flat ground, fix the lidar and inertial navigation system on the drone's airborne pod, control the drone to hover and scan the ground, and collect lidar point cloud data and inertial navigation data; Step 2. Data integration: Import the laser radar point cloud data and inertial navigation data into the data analysis software to generate a single frame file containing the point cloud coordinates (x, y, z) and the corresponding inertial navigation data (longitude, latitude, altitude, pitch angle, roll angle, yaw angle); Step 3. Ground point cloud screening: The laser radar ground point cloud that meets the plane characteristic requirements is screened out through the ground point cloud recognition algorithm to generate the ground point cloud file to be calibrated; Step 4. Calibration process: Preset the initial values ​​of the calibration parameters labelPitch and labelRoll; Traverse the calibration parameter combinations and iterate through the following sub-steps: 4.1 Determine whether the calibration parameters exceed the preset threshold range, and terminate if so; 4.2 Transform the laser radar point cloud coordinates (x, y, z) through the Euler transformation matrix composed of calibration parameters to obtain the calibrated coordinates (x1, y1, z1); 4.3 The calibrated coordinates (x1, y1, z1) are transformed into UTM coordinates (x3, y3, z3) in combination with the inertial navigation data, where the xoy plane of the UTM coordinates is parallel to the sea level; 4.4 Perform plane fitting on the UTM coordinates and calculate the slope of the fitting plane; 4.5 If the plane slope slop is less than 0.2°, record the current calibration parameter combination to the data set labelResult; 4.6 Increase the calibration parameters and repeat the iteration; Step 5. Calibration parameter calculation: Based on the dataset labelResult, the average values ​​of labelPitch and labelRoll are calculated as the final calibration parameters.

2. The laser radar reference surface calibration method applicable to the pod of an unmanned aerial vehicle according to claim 1 is characterized in that: The flat ground has a slope of less than 0.2° and an area of ​​more than 100m 2 , the hovering time of the UAV is 5 minutes, and the acquisition frequency of the laser radar point cloud data is not less than 10 Hz.

3. The laser radar reference surface calibration method applicable to the pod of an unmanned aerial vehicle according to claim 1 is characterized in that: The data analysis software is one of MATLAB, Python or ROS.

4. The laser radar reference surface calibration method applicable to the pod of an unmanned aerial vehicle according to claim 1 is characterized in that: In step 3, the ground point cloud recognition algorithm is screened based on point cloud density and height distribution characteristics.

5. The laser radar reference surface calibration method applicable to the pod of an unmanned aerial vehicle according to claim 1 is characterized in that: In step 4, the initial calibration parameters labelPitch and labelRoll are both preset to -2.0, and are increased in steps of 0.1° until both labelPitch and labelRoll are greater than 2.

0.

6. The laser radar reference surface calibration method applicable to the pod of an unmanned aerial vehicle according to claim 1 is characterized in that: The Euler transformation matrix in step 4 is Rz(0)*Ry(labelPitch)*Rx(labelRoll), which is used to convert the lidar point cloud coordinates (x, y, z) into calibration coordinates (x1, y1, z1), and further combined with the attitude angle (pitch, roll, yaw) and position information (longitude, latitude, height) of the inertial navigation data, and converted into UTM projection coordinates (x3, y3, z3) through the coordinate transformation matrix.

7. The laser radar reference surface calibration method applicable to the pod of an unmanned aerial vehicle according to claim 1 is characterized in that: In step 4, the plane fitting adopts the least square method or the random sampling consistency algorithm, and the calculation of the slope of the fitted plane includes the analysis of the angle between the plane normal vector and the sea level normal vector.

8. The laser radar reference surface calibration method applicable to the pod of an unmanned aerial vehicle according to claim 1 is characterized in that: All data processing and parameter traversal processes in step 4 are fully automatically performed by the embedded software, including parameter adjustment, error threshold judgment and calibration result storage.

9. The laser radar reference surface calibration method applicable to the pod of an unmanned aerial vehicle according to claim 1, characterized in that: In step 5, the final calibration parameters are generated by weighted averaging or optimal solution screening, and are automatically written into the sensor calibration configuration file of the drone's onboard pod.

10. The laser radar reference surface calibration method applicable to the pod of an unmanned aerial vehicle according to claim 9, characterized in that: The calculation of the calibration parameters also includes weighted average or median optimization of the parameters in the data set labelResult.

Images