Laser radar data generation method and device, equipment and storage medium

Abstract

The invention relates to the technical field of unmanned driving, automatic driving and unmanned vehicles, and discloses a laser radar data generation method and device, equipment and a storage medium. The method comprises the steps that a first laser radar installed on a first unmanned vehicle is used for collecting first laser radar point cloud data under the severe weather condition; acquiring second laser radar point cloud data under a common weather condition; and synthesizing the first laser radar point cloud data and the second laser radar point cloud data to obtain synthesized laser radar point cloud data under the simulated severe weather condition. According to the method, on the premise of ensuring the authenticity and diversity of the data, efficient generation of the laser radar data under the severe weather condition is realized, so that sufficient high-quality data support is provided for training and verification of an automatic driving perception algorithm in a complex weather environment.

Description

Technical Field

[0001] The present application relates to the fields of unmanned driving, automatic driving, and unmanned vehicles, and in particular to a laser radar data generation method, device, equipment, and storage medium. Background Art

[0002] In the field of autonomous driving, a large amount of real data is needed for the training of perception algorithms. The current mainstream data acquisition methods include two categories: one is to collect real data on actual roads, and the other is to generate synthetic data using simulation tools. In actual applications, mainstream companies build city and vehicle models through high-precision modeling, and use physically based rendering and ray tracing and other technologies to simulate real effects; some companies generate synthetic data for perception training based on neural network reconstruction engines; some companies have proposed weather change methods based on neural radiation fields to simulate the performance of the same static scene in different weather conditions; and some companies have proposed a block solution based on neural radiation fields to achieve high-precision restoration of large and small areas of the city.

[0003] However, the relevant technical solutions have obvious shortcomings: on the one hand, it is very difficult to obtain lidar data in bad weather through actual collection. This is not because bad weather is rare, but because the areas where autonomous driving vehicles collect data usually do not often experience bad weather; on the other hand, the existing synthetic data generation methods either require huge costs for high-precision modeling, which is unaffordable for ordinary companies, and the authenticity is difficult to meet the needs of perception training, or face the problem of low degrees of freedom and difficulty in adding dynamic weather change effects. These problems have led to a serious shortage of lidar data in bad weather, which cannot meet the training needs of autonomous driving systems under bad weather conditions. Summary of the invention

[0004] The present application provides a lidar data generation method, apparatus, equipment and storage medium, which are used to achieve efficient generation of lidar data under severe weather conditions while ensuring the authenticity and diversity of the data, thereby providing sufficient high-quality data support for the training and verification of autonomous driving perception algorithms in complex weather environments.

[0005] In a first aspect, the present application provides a lidar data generation method, which comprises: collecting first lidar point cloud data using a first lidar installed on a first unmanned vehicle under severe weather conditions; collecting second lidar point cloud data under normal weather conditions; and synthesizing the first lidar point cloud data with the second lidar point cloud data to obtain synthetic lidar point cloud data under simulated severe weather conditions.

[0006] In a second aspect, the present application provides a laser radar data generating device, the laser radar data generating device comprising:

[0007] A first acquisition module is used to acquire first laser radar point cloud data using a first laser radar installed on a first unmanned vehicle under severe weather conditions;

[0008] A second acquisition module, used for acquiring second laser radar point cloud data under normal weather conditions;

[0009] The data synthesis module is used to synthesize the first laser radar point cloud data and the second laser radar point cloud data to obtain synthetic laser radar point cloud data under simulated severe weather conditions.

[0010] In a third aspect, the present application provides a computer device comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the lidar data generation method are performed.

[0011] In a fourth aspect, the present application provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, the steps of the lidar data generation method are executed.

[0012] In the technical solution provided by the present application, by using the first laser radar installed on the first unmanned vehicle to collect the first laser radar point cloud data under severe weather conditions, it is ensured that the real severe weather characteristic data is obtained, and the influence of weather factors such as rain, snow, and fog on the laser radar measurement is accurately reflected; at the same time, the second laser radar point cloud data is collected under normal weather conditions, and the complete scene structure information is obtained, which provides reliable benchmark data for subsequent data synthesis; by synthesizing the first laser radar point cloud data and the second laser radar point cloud data, the synthetic laser radar point cloud data under simulated severe weather conditions is obtained, and the efficient generation of severe weather data is realized. This method does not require huge costs for high-precision modeling, nor does it have the problem of low degrees of freedom, and can quickly generate the required severe weather laser radar data for different scenes. The synthesized data not only retains the characteristics of severe weather, such as reduced reflection intensity and sparse point clouds, but also maintains the integrity and accuracy of the scene. In this way, the problem of the difficulty of obtaining laser radar data in severe weather is solved, and sufficient high-quality data support is provided for the training, testing and verification of autonomous driving perception algorithms, which improves the perception ability and safety of autonomous driving systems under severe weather conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0014] Figure 1 This is a schematic diagram of an embodiment of a method for generating laser radar data in an embodiment of the present application;

[0015] Figure 2 This is a schematic diagram of an embodiment of a laser radar data generating device in an embodiment of the present application;

[0016] Figure 3 It is a schematic diagram of an embodiment of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0017] Embodiments of the present application provide a method and device for generating laser radar data. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the laser radar data generation method in the embodiment of the present application includes:

[0019] Step S101, collecting first laser radar point cloud data using a first laser radar installed on a first unmanned vehicle under severe weather conditions;

[0020] Step S102, collecting second laser radar point cloud data under normal weather conditions;

[0021] Step S103: synthesize the first laser radar point cloud data and the second laser radar point cloud data to obtain synthetic laser radar point cloud data under simulated severe weather conditions.

[0022] It is understandable that the execution subject of the present application may be a laser radar data generating device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0023] Optionally, the first laser radar point cloud data is collected under adverse weather conditions. Adverse weather conditions include weather conditions that affect the normal operation of the laser radar, such as heavy rain, heavy snow, and heavy fog. The first laser radar is installed on the first unmanned vehicle and is fixedly installed. The installation position is at the center of the roof, 1.8 meters above the ground, and the laser radar is kept level with the ground. The data collection area is selected in a place where adverse weather often occurs. The ground is required to be flat and open, without other interferences, the area is not less than 100 square meters, and there are complete drainage facilities. At the same time, the second laser radar point cloud data is collected under normal weather conditions. Normal weather refers to sunny weather conditions without rain or snow. The second laser radar is installed on the second unmanned vehicle. The laser radar model is the same as the first laser radar. The installation position and attitude must also be consistent. The position error does not exceed 1 cm, and the attitude angle error does not exceed 0.1 degrees. The data collection area is selected in areas where autonomous driving vehicles often travel, such as urban roads, test sites and other scenes.

[0024] The two sets of collected data are synthesized and processed, and the specific processing process includes data frame correspondence, point cloud point matching, attribute value extraction and point cloud point screening. The synthesized LiDAR point cloud data not only contains the characteristics of severe weather, but also maintains the integrity of the scene. To illustrate with a specific example: in a data collection, the first LiDAR collected 1000 frames of data in heavy rain, and the second LiDAR collected 1000 frames of data of the corresponding scene under sunny conditions, and each frame of data contained about 10,000 point cloud points. After data correspondence and screening, the synthesized data retains the laser reflection characteristics of rainy days, while maintaining the spatial structure characteristics of the scene, so that the synthesized data can truly reflect the measurement characteristics of the LiDAR under severe weather conditions.

[0025] In the embodiment of the present application, by using the first laser radar installed on the first unmanned vehicle to collect the first laser radar point cloud data under bad weather conditions, it is ensured that the real bad weather feature data is obtained, and the influence of weather factors such as rain, snow, and fog on the laser radar measurement is accurately reflected; at the same time, the second laser radar point cloud data is collected under normal weather conditions, and the complete scene structure information is obtained, providing reliable benchmark data for subsequent data synthesis; by synthesizing the first laser radar point cloud data and the second laser radar point cloud data, the synthetic laser radar point cloud data under simulated bad weather conditions is obtained, and the efficient generation of bad weather data is achieved. This method does not require huge costs for high-precision modeling, nor does it have the problem of low degrees of freedom. It can quickly generate the required bad weather laser radar data for different scenes. The synthesized data not only retains the characteristics of bad weather, such as reduced reflection intensity and sparse point clouds, but also maintains the integrity and accuracy of the scene. In this way, the problem of the difficulty of obtaining laser radar data in bad weather is solved, and sufficient high-quality data support is provided for the training, testing and verification of autonomous driving perception algorithms, which improves the perception ability and safety of autonomous driving systems under bad weather conditions.

[0026] In an optional embodiment, the process of executing step S103 may specifically include the following steps:

[0027] (1) determining, for each first data frame in the first laser radar point cloud data, a corresponding second data frame in the second laser radar point cloud data;

[0028] (2) for each first data frame, for each first point cloud point in the first data frame, determining a corresponding second point cloud point from a corresponding second data frame to obtain a plurality of point cloud point pairs;

[0029] (3) for each point cloud pair, based on the preset point cloud attributes, extract the point cloud attribute values ​​of the point cloud pair respectively;

[0030] (4) Based on the attribute value of the point cloud point pair, the point cloud points in the point cloud point pair are screened to obtain synthesized lidar point cloud data.

[0031] Optionally, the frame correspondence between the first laser radar point cloud data and the second laser radar point cloud data is determined. A data frame refers to a collection of all point cloud data acquired by the laser radar in a complete scanning cycle, and each data frame contains attributes such as a timestamp, spatial position information, and a reflection intensity value. By aligning the first frames of the two sets of data and matching each frame in the first laser radar point cloud data with the corresponding frame in the second laser radar point cloud data in the time sequence of data acquisition. After determining the data frame correspondence, for each pair of matched data frames, a correspondence between the point cloud points in the frame needs to be established. The point cloud point is a spatial sampling point obtained by the laser radar scanning, which contains spatial coordinate information and a reflection intensity value. For each first point cloud point in the first data frame, the point with the closest spatial position is found in the second data frame as the corresponding second point cloud point, thereby obtaining a plurality of point cloud point pairs. The spatial position of the point cloud point is determined by its three-dimensional coordinates. When the distance between two point cloud points is less than a preset threshold (such as 0.1 meters), it is considered that the two point cloud points constitute a corresponding relationship. After obtaining the point cloud point pairs, the preset point cloud attribute values ​​are extracted for each point cloud point pair. The preset point cloud attributes include two key attributes: reflection intensity and point cloud depth. The reflection intensity value is directly obtained from the original data, reflecting the magnitude of the laser signal return intensity, and the value range is 0-255. The point cloud depth value is obtained by calculating the distance from the point cloud point to the laser radar, in meters.

[0032] Finally, the point cloud pairs are screened based on the extracted attribute values. When the reflection intensity value of the second point cloud point is less than the intensity threshold corresponding to the preset intensity threshold curve, the first point cloud point is retained and the second point cloud point is deleted; when the reflection intensity value of the second point cloud point is greater than or equal to the preset intensity threshold, the depth values ​​of the two point cloud points are compared, the point cloud point with a smaller depth value is retained, and the point cloud point with a larger depth value is deleted.

[0033] For example: for a pair of corresponding data frames, there is a point cloud point A in the first data frame, with coordinates (5 meters, 3 meters, 1 meter), a reflection intensity value of 45, and a point cloud depth value of 6 meters; in the second data frame, the point cloud point B closest to point A is found, with coordinates (5.05 meters, 3.02 meters, 1.03 meters), and the distance between the two points is 0.06 meters, which is less than the preset threshold of 0.1 meters, so A and B constitute a point cloud point pair. The reflection intensity value of point B is 80, and the point cloud depth value is 6.3 meters. By querying the intensity threshold curve, it is known that the intensity threshold at a distance of 6 meters is 55. Since the reflection intensity value of point B is 80, which is greater than the threshold of 55, the depth value needs to be compared. Since the depth value of point A is 6 meters, which is less than the depth value of point B, 6.3 meters, point A is retained in the synthetic data. This processing method ensures that the synthetic data retains the characteristics of severe weather and maintains the spatial consistency of the point cloud data.

[0034] In an optional embodiment, for each first data frame in the first laser radar point cloud data, determining a corresponding second data frame in the second laser radar point cloud data includes:

[0035] (1) determining a first frame in the first laser radar point cloud data and a second frame in the second laser radar point cloud data;

[0036] (2) Align the first frame with the second frame, and determine the first data frame in the first laser radar point cloud data and the second data frame in the second laser radar point cloud data as corresponding data frames frame by frame according to the acquisition order of the data frames.

[0037] Optionally, determine the first frame of the first laser radar point cloud data and the second laser radar point cloud data. A data frame refers to a collection of all point cloud data acquired by the laser radar in a complete scanning cycle, and each data frame contains attributes such as timestamp, spatial position information, and reflection intensity value. The determination of the first frame is based on timestamp information. The timestamp is the data acquisition moment recorded by the laser radar, accurate to milliseconds. For the first laser radar point cloud data, select the data frame with the earliest timestamp as the first first frame; for the second laser radar point cloud data, also select the data frame with the earliest timestamp as the second first frame.

[0038] After determining the first frame, the first frame and the second frame need to be aligned. The alignment process is essentially to establish a correspondence between the two sets of data in the time dimension. Since the two sets of data are collected at different times, the timestamps of the two first frames need to be used as the starting time points of their respective data sequences. The acquisition order of the data frames refers to the time sequence in which the laser radar collects data in sequence according to the scanning frequency. The scanning frequency is usually 10Hz, that is, 10 frames of data are generated per second. Based on the time correspondence established between the two first frames, each frame in the first laser radar point cloud data is matched one by one with the corresponding frame in the second laser radar point cloud data according to the acquisition order of the data frames to form a complete data frame correspondence.

[0039] For example, in a data collection, the first laser radar collects data in heavy rain, the timestamp of the first frame is 10:30:00:00:000 milliseconds on December 22, 2023, the scanning frequency is 10Hz, and a total of 300 frames of data are collected; the second laser radar collects data under sunny conditions, the timestamp of the first frame is 14:20:00:00:000 milliseconds on December 23, 2023, the scanning frequency is also 10Hz, and a total of 300 frames of data are collected. During the alignment process, the timestamp of the first frame is set to t0, and the timestamp of the second frame is set to t0'. The time interval between adjacent frames in the two sets of data is 100 milliseconds. Based on the time sequence, the nth frame (n from 1 to 300) of the first laser radar forms a corresponding relationship with the nth frame of the second laser radar. In this way, it is ensured that the two sets of data establish a strict one-to-one correspondence in the time dimension, laying the foundation for the subsequent point cloud data synthesis processing. The whole process ensures the accuracy of data correspondence, so that the synthesized data can accurately reflect the impact of bad weather on laser radar measurement.

[0040] In an optional embodiment, the preset point cloud attributes include reflection intensity and point cloud depth. Based on the attribute values ​​of the point cloud point pair, the point cloud points in the point cloud point pair are screened to obtain synthesized laser radar point cloud data, including:

[0041] (1) respectively determining the reflection intensity value and the point cloud depth value of the two point cloud points in the point cloud point pair;

[0042] (2) when the reflection intensity value of the second point cloud point is less than the intensity threshold corresponding to the preset intensity threshold curve, retain the first point cloud point and delete the second point cloud point;

[0043] (3) When the reflection intensity value of the second point cloud point is greater than or equal to the intensity threshold corresponding to the preset intensity threshold curve, the point cloud depth values ​​of the first point cloud point and the second point cloud point are compared, the point cloud point corresponding to the smaller point cloud depth value is retained, and the point cloud point corresponding to the larger point cloud depth value is deleted.

[0044] Optionally, the preset point cloud attributes include two key attributes: reflection intensity and point cloud depth. Reflection intensity is the intensity of the reflected signal received by the laser radar, and the numerical range is usually 0-255. The larger the value, the stronger the reflected signal. Point cloud depth refers to the actual distance from the laser radar emission point to the measured point, in meters. For each point cloud point pair, it is necessary to extract the reflection intensity values ​​of the first point cloud point and the second point cloud point from the original data, and at the same time calculate the corresponding point cloud depth value according to the spatial coordinate information of the point cloud point. After obtaining the attribute value of the point cloud point pair, the reflection intensity value of the second point cloud point is first compared with the intensity threshold corresponding to the preset intensity threshold curve. The intensity threshold curve describes the reflection intensity threshold at different distances. The curve is obtained by collecting and analyzing data under wet ground conditions, dividing the point cloud into 0.5 meters per grid, and taking the point with the smallest intensity in each grid, so as to obtain the minimum reflection intensity curve that changes with distance. If the reflection intensity value of the second point cloud point is less than the preset intensity threshold at the corresponding distance, the first point cloud point is directly retained and the second point cloud point is deleted, because the lower reflection intensity usually indicates that the point cloud point is greatly affected by bad weather.

[0045] When the reflection intensity value of the second point cloud point is greater than or equal to the preset intensity threshold, the depth values ​​of the two point cloud points need to be further compared. In the depth value comparison, the point cloud point with a smaller depth value is selected to be retained, and the point cloud point with a larger depth value is deleted. This processing method is based on actual observations: Under severe weather conditions, for two points with similar distances, the true reflection point is usually the closer point, and the farther point is often a false reflection caused by weather factors.

[0046] For example: In a pair of point cloud points, the first point cloud point is located 8 meters away from the laser radar, with a reflection intensity value of 40 and a point cloud depth value of 8.2 meters; the second point cloud point is located at a similar position, with a reflection intensity value of 90 and a point cloud depth value of 8.5 meters. By querying the preset intensity threshold curve, it is known that the intensity threshold at a distance of 8 meters is 60. Since the reflection intensity value of the second point cloud point of 90 is greater than the threshold of 60, the depth value needs to be further compared. The comparison found that the depth value of the first point cloud point of 8.2 meters is less than the depth value of the second point cloud point of 8.5 meters, so the first point cloud point is retained in the final synthetic data and the second point cloud point is deleted. This processing method ensures that the synthetic data retains the characteristics of severe weather and maintains the spatial consistency of the point cloud data. The entire processing process fully considers the influence of severe weather on laser radar measurement, so that the generated data can truly reflect the scene characteristics under severe weather conditions.

[0047] In an optional embodiment, the intensity threshold curve is generated in the following manner:

[0048] (1) Under wet ground conditions, using a first laser radar installed on a first unmanned vehicle to collect target laser radar point cloud data;

[0049] (2) dividing the target laser radar point cloud data according to a preset distance threshold to obtain a plurality of divided point cloud data;

[0050] (3) extracting the minimum value of reflection intensity for each divided point cloud data respectively to obtain the minimum value of reflection intensity for each divided point cloud data;

[0051] (4) Generate an intensity threshold curve based on the preset distance threshold and the minimum reflection intensity of each divided point cloud data.

[0052] Optionally, in the case of wet ground, the first laser radar installed on the first unmanned vehicle is used to collect the target laser radar point cloud data. The wet ground is obtained by artificially sprinkling water on the test site to make the ground uniformly wet but without water accumulation, and the amount of water sprinkling is controlled at 2 liters of water per square meter. The first laser radar is installed at the center of the roof, 1.8 meters above the ground, and kept level with the ground. During the collection, the vehicle remains stationary, the laser radar scans at a frequency of 10 Hz, the collection time is 60 seconds, and 600 frames of point cloud data are obtained. The collected target laser radar point cloud data is divided by distance. The preset distance threshold is set to 0.5 meters, that is, the measurement range of 0 to 200 meters is divided into 0.5 meters at equal intervals, and 400 distance intervals are obtained. The point cloud data in each distance interval constitutes a divided point cloud data set. For example, the first divided point cloud data contains all point cloud points with a distance range of 0 to 0.5 meters, the second divided point cloud data contains all point cloud points with a distance range of 0.5 to 1.0 meters, and so on.

[0053] For each divided point cloud data, extract the minimum reflection intensity. The reflection intensity value is the intensity of the reflected signal received by the lidar, and the value range is 0-255. Compare the reflection intensity values ​​of all point cloud points in each distance interval, and select the minimum value as the characteristic value of the interval. In actual operation, in order to avoid the influence of outliers, 2000 point cloud points are sampled in each interval to extract the minimum reflection intensity. Finally, generate an intensity threshold curve based on the preset distance threshold and the extracted minimum reflection intensity value.

[0054] For example, the point cloud data in the interval of 0 to 0.5 meters contains 2358 point cloud points, whose reflection intensity values ​​are distributed between 85 and 120, and the extracted minimum reflection intensity value is 85; the point cloud data in the interval of 0.5 to 1.0 meters contains 2156 point cloud points, whose reflection intensity values ​​are distributed between 78 and 110, and the extracted minimum reflection intensity value is 78; the point cloud data in the interval of 1.0 to 1.5 meters contains 1985 point cloud points, whose reflection intensity values ​​are distributed between 72 and 105, and the extracted minimum reflection intensity value is 72; the point cloud data in the interval of 1.5 to 2.0 meters contains 2103 point cloud points, whose reflection intensity values ​​are distributed between 65 and 100, and the extracted minimum reflection intensity value is 65. The minimum reflection intensity values ​​of all 400 distance intervals are connected in order of distance to form a curve describing the relationship between distance and minimum reflection intensity, which is the required intensity threshold curve. This curve reflects the minimum intensity characteristics of laser reflection signals at different distances under wet ground conditions.

[0055] In an optional embodiment, the laser radar data generation method further includes:

[0056] Preprocess the first laser radar point cloud data, including:

[0057] (1) calculating the height value of each point cloud point in the first laser radar point cloud data to obtain a relative ground height data set;

[0058] (2) performing a threshold comparison between the relative ground height data set and a preset height threshold, and determining the point cloud points whose relative ground height is greater than the height threshold as the first laser radar point cloud data after preprocessing;

[0059] (3) synthesizing the first laser radar point cloud data and the second laser radar point cloud data to obtain synthetic laser radar point cloud data under simulated severe weather conditions, including:

[0060] (4) Synthesize the first laser radar point cloud data and the second laser radar point cloud data after preprocessing.

[0061] Optionally, the first laser radar point cloud data is preprocessed. The first step of the preprocessing is to calculate the relative ground height value of each point cloud point. The relative ground height refers to the vertical distance between the point cloud point and the ground, and the laser radar installation height is used as the reference for calculation. Since the first laser radar is installed at the center of the roof of the first unmanned vehicle, the height from the ground is 1.8 meters, so the actual height value of each point cloud point is subtracted by 1.8 meters to obtain the relative ground height value. This calculation is performed on all point cloud points in the first laser radar point cloud data to obtain a relative ground height data set. After obtaining the relative ground height data set, it is necessary to compare it with a pre-set height threshold. The setting of the height threshold is based on the influence of bad weather on the near-ground point cloud, and is usually set to 0.3 meters. The selection of this threshold takes into account the interference of factors such as rain and snow on the measurement of point clouds near the ground. The relative ground height value of each point cloud point is compared with the height threshold of 0.3 meters, and the point cloud points with a height value greater than 0.3 meters are retained as the preprocessed first laser radar point cloud data. This processing method can effectively remove low-position point cloud data affected by ground water and snow.

[0062] Finally, the preprocessed first laser radar point cloud data is synthesized with the second laser radar point cloud data. The synthesis process still follows the aforementioned data frame correspondence, point cloud point matching, attribute value extraction and point cloud point screening process, but uses the first laser radar point cloud data after height threshold screening.

[0063] For example: In a certain data collection, the first laser radar collected a frame of data containing 10,000 point cloud points. First, the relative ground height of these point cloud points is calculated. For example, the original height value of a point cloud point is 2.1 meters. After subtracting the laser radar installation height of 1.8 meters, the relative ground height is 0.3 meters; the original height value of another point cloud point is 1.9 meters, and the calculated relative ground height is 0.1 meters. Comparing the relative ground height values ​​of all point cloud points with the preset 0.3-meter threshold, the screening results show that 6,000 point cloud points have a relative ground height greater than 0.3 meters, and these point cloud points constitute the preprocessed data set. These data mainly come from reflection points of above-ground targets such as buildings, trees, and vehicles, and remove point cloud data near the ground that is easily affected by bad weather. The preprocessed point cloud data reduces the interference caused by factors such as ground water and improves the accuracy of subsequent data synthesis.

[0064] In an optional embodiment, collecting second laser radar point cloud data under normal weather conditions includes: collecting second laser radar point cloud data for a historical driving area or a designated driving area under normal weather conditions using a second laser radar installed on a second unmanned vehicle;

[0065] Among them, the difference between the position and posture of the second laser radar relative to the second unmanned vehicle and the position and posture of the first laser radar relative to the first unmanned vehicle is less than the difference threshold.

[0066] Optionally, the second laser radar point cloud data collection process under normal weather conditions needs to strictly control the collection conditions. Normal weather conditions refer to clear weather conditions without rain or snow, and good visibility. The data collected under such conditions can maximize the accuracy of laser radar measurement. The second laser radar is installed on the second unmanned vehicle, and data collection needs to be carried out for specific areas, including historical driving areas and designated driving areas. The historical driving area refers to the road section where the autonomous driving vehicle often drives, such as urban roads, test sites, etc.; the designated driving area is a data collection area specially selected according to actual needs. The installation position and attitude parameters of the second laser radar need to be highly consistent with those of the first laser radar. The installation position difference threshold and attitude difference threshold are important indicators for measuring the installation accuracy of the two laser radars. Specifically, the installation of the second laser radar on the second unmanned vehicle needs to meet the following requirements: the installation position is selected at the center of the roof, and the deviation from the installation position of the first laser radar does not exceed 1 cm; the installation height is 1.8 meters, and the height deviation does not exceed 5 mm; the installation posture needs to be kept horizontal, and the deviations of the roll angle, pitch angle, and yaw angle do not exceed 0.1 degrees.

[0067] For example, in actual installation, the first laser radar is installed at the center of the roof of the first unmanned vehicle, with the installation coordinates (0m, 0m, 1.8m) and the attitude angle (0 degrees, 0 degrees, 0 degrees); the second laser radar is installed on the roof of the second unmanned vehicle. Through precise measurement and adjustment, the final installation coordinates are (0.005m, 0.003m, 1.798m) and the attitude angle is (0.05 degrees, 0.08 degrees, 0.06 degrees). The calculated position deviation is 0.007m, which is less than the difference threshold of 1cm; the height deviation is 2mm, which is less than the difference threshold of 5mm; the maximum attitude angle deviation is 0.08 degrees, which is less than the difference threshold of 0.1 degrees.

[0068] The above describes the laser radar data generation method in the embodiment of the present application. The following describes the laser radar data generation device in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the laser radar data generation device includes:

[0069] A first acquisition module 201 is used to acquire first laser radar point cloud data using a first laser radar installed on a first unmanned vehicle under severe weather conditions;

[0070] The second acquisition module 202 is used to acquire second laser radar point cloud data under normal weather conditions;

[0071] The data synthesis module 203 is used to synthesize the first laser radar point cloud data and the second laser radar point cloud data to obtain synthetic laser radar point cloud data under simulated severe weather conditions.

[0072] Optionally, the data synthesis module 203 includes:

[0073] a matching submodule, configured to determine, for each first data frame in the first laser radar point cloud data, a corresponding second data frame in the second laser radar point cloud data;

[0074] A screening submodule, for determining, for each first point cloud point in each first data frame, a corresponding second point cloud point from a corresponding second data frame, to obtain a plurality of point cloud point pairs;

[0075] The extraction submodule is used to extract the point cloud attribute values ​​of each point cloud point pair based on the preset point cloud attributes;

[0076] The synthesis submodule is used to filter the point cloud points in the point cloud point pair based on the attribute value of the point cloud point pair to obtain synthesized lidar point cloud data.

[0077] Optionally, the matching submodule is used to:

[0078] Determine a first frame in the first laser radar point cloud data and a second frame in the second laser radar point cloud data;

[0079] The first first frame and the second first frame are aligned, and according to the acquisition order of the data frames, the first data frame in the first lidar point cloud data and the second data frame in the second lidar point cloud data are determined as corresponding data frames frame by frame.

[0080] Optionally, the synthesis submodule is used to:

[0081] Respectively determine the reflection intensity value and the point cloud depth value of two point cloud points in the point cloud point pair;

[0082] When the reflection intensity value of the second point cloud point is less than the intensity threshold corresponding to the preset intensity threshold curve, retaining the first point cloud point and deleting the second point cloud point;

[0083] When the reflection intensity value of the second point cloud point is greater than or equal to the intensity threshold corresponding to the preset intensity threshold curve, the point cloud depth values ​​of the first point cloud point and the second point cloud point are compared, the point cloud point corresponding to the smaller point cloud depth value is retained, and the point cloud point corresponding to the larger point cloud depth value is deleted.

[0084] Optionally, the laser radar data generating device further includes:

[0085] A third acquisition module is used to acquire target laser radar point cloud data using a first laser radar installed on a first unmanned vehicle under wet ground conditions;

[0086] A division module, used for dividing the target laser radar point cloud data according to a preset distance threshold to obtain a plurality of divided point cloud data;

[0087] An extraction module is used to extract the minimum value of reflection intensity of each divided point cloud data respectively to obtain the minimum value of reflection intensity of each divided point cloud data;

[0088] A generating module is used to generate the intensity threshold curve according to the preset distance threshold and the minimum reflection intensity of each divided point cloud data.

[0089] Optionally, the laser radar data generating device further includes:

[0090] The preprocessing module is used to preprocess the first laser radar point cloud data, specifically for:

[0091] Calculating the height value of each point cloud point in the first laser radar point cloud data to obtain a relative ground height data set;

[0092] Performing a threshold comparison between the relative ground height data set and a preset height threshold, and determining the point cloud points whose relative ground height is greater than the height threshold as the first laser radar point cloud data after preprocessing;

[0093] The first laser radar point cloud data and the second laser radar point cloud data are synthesized to obtain synthetic laser radar point cloud data under simulated severe weather conditions, including:

[0094] The first laser radar point cloud data after the preprocessing is synthesized with the second laser radar point cloud data.

[0095] Optionally, the second acquisition module is used to:

[0096] Under normal weather conditions, using a second laser radar installed on a second unmanned vehicle to collect second laser radar point cloud data for a historical driving area or a designated driving area;

[0097] Among them, the difference between the position and posture of the second laser radar relative to the second unmanned vehicle and the position and posture of the first laser radar relative to the first unmanned vehicle is less than a difference threshold.

[0098] Through the cooperation of the above components, the first laser radar point cloud data is collected by using the first laser radar installed on the first unmanned vehicle under severe weather conditions, ensuring that the real severe weather characteristic data is obtained, accurately reflecting the influence of weather factors such as rain, snow, and fog on the laser radar measurement; at the same time, the second laser radar point cloud data is collected under normal weather conditions, and the complete scene structure information is obtained, providing reliable benchmark data for subsequent data synthesis; by synthesizing the first laser radar point cloud data and the second laser radar point cloud data, the synthetic laser radar point cloud data under simulated severe weather conditions is obtained, and the efficient generation of severe weather data is realized. This method does not require huge costs for high-precision modeling, nor does it have the problem of low degrees of freedom. It can quickly generate the required severe weather laser radar data for different scenes. The synthesized data not only retains the characteristics of severe weather, such as reduced reflection intensity and sparse point clouds, but also maintains the integrity and accuracy of the scene. In this way, the problem of the difficulty of obtaining laser radar data in severe weather is solved, and sufficient high-quality data support is provided for the training, testing and verification of autonomous driving perception algorithms, which improves the perception ability and safety of autonomous driving systems under severe weather conditions.

[0099] Based on the same technical concept, the embodiment of the present application also provides an electronic device. Figure 3 As shown, it is a schematic diagram of the structure of the electronic device 300 provided in the embodiment of the present application, including a processor 301, a memory 302, and a bus 303. Among them, the memory 302 is used to store execution instructions, including a memory 3021 and an external memory 3022; the memory 3021 here is also called an internal memory, which is used to temporarily store the operation data in the processor 301, and the data exchanged with the external memory 3022 such as a hard disk. The processor 301 exchanges data with the external memory 3022 through the memory 3021. When the electronic device 300 is running, the processor 301 communicates with the memory 302 through the bus 303, so that the processor 301 executes the steps of the laser radar data generation method.

[0100] The present disclosure also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the laser radar data generation method described in the above method embodiment are executed. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0101] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. 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 application.

Claims

1. A laser radar data generation method, characterized in that: The following steps are involved: collecting first lidar point cloud data using a first lidar mounted on a first unmanned vehicle under adverse weather conditions; Collecting second lidar point cloud data under normal weather conditions; The first laser radar point cloud data and the second laser radar point cloud data are synthesized to obtain synthetic laser radar point cloud data under simulated severe weather conditions.

2. The method according to claim 1, characterized in that The first laser radar point cloud data and the second laser radar point cloud data are synthesized to obtain synthetic laser radar point cloud data under simulated severe weather conditions, including: For each first data frame in the first lidar point cloud data, determine a corresponding second data frame in the second lidar point cloud data; For each first data frame, for each first point cloud point in the first data frame, determining a corresponding second point cloud point from a corresponding second data frame to obtain a plurality of point cloud point pairs; For each point cloud pair, based on the preset point cloud attributes, the point cloud attribute values ​​of the point cloud pair are extracted respectively; Based on the attribute value of the point cloud point pair, the point cloud points in the point cloud point pair are screened to obtain synthesized lidar point cloud data.

3. The method according to claim 2, characterized in that For each first data frame in the first laser radar point cloud data, determining a corresponding second data frame in the second laser radar point cloud data, comprising: Determine a first frame in the first laser radar point cloud data and a second frame in the second laser radar point cloud data; The first first frame and the second first frame are aligned, and according to the acquisition order of the data frames, the first data frame in the first lidar point cloud data and the second data frame in the second lidar point cloud data are determined as corresponding data frames frame by frame.

4. The method according to claim 2, characterized in that: The preset point cloud attributes include reflection intensity and point cloud depth; Based on the attribute value of the point cloud point pair, the point cloud points in the point cloud point pair are screened to obtain the synthesized lidar point cloud data, including: Respectively determine the reflection intensity value and the point cloud depth value of two point cloud points in the point cloud point pair; When the reflection intensity value of the second point cloud point is less than the intensity threshold corresponding to the preset intensity threshold curve, retaining the first point cloud point and deleting the second point cloud point; When the reflection intensity value of the second point cloud point is greater than or equal to the intensity threshold corresponding to the preset intensity threshold curve, the point cloud depth values ​​of the first point cloud point and the second point cloud point are compared, the point cloud point corresponding to the smaller point cloud depth value is retained, and the point cloud point corresponding to the larger point cloud depth value is deleted.

5. The method according to claim 4, characterized in that The intensity threshold curve is generated as follows: Under wet ground conditions, using a first laser radar installed on a first unmanned vehicle to collect target laser radar point cloud data; Dividing the target laser radar point cloud data according to a preset distance threshold to obtain a plurality of divided point cloud data; Extract the minimum reflection intensity of each divided point cloud data respectively to obtain the minimum reflection intensity of each divided point cloud data; The intensity threshold curve is generated according to the preset distance threshold and the minimum reflection intensity of each divided point cloud data.

6. The method according to claim 1, characterized in that Also includes: Preprocess the first laser radar point cloud data, including: Calculating the height value of each point cloud point in the first laser radar point cloud data to obtain a relative ground height data set; Performing a threshold comparison between the relative ground height data set and a preset height threshold, and determining the point cloud points whose relative ground height is greater than the height threshold as the first laser radar point cloud data after preprocessing; The first laser radar point cloud data and the second laser radar point cloud data are synthesized to obtain synthetic laser radar point cloud data under simulated severe weather conditions, including: The first laser radar point cloud data after the preprocessing is synthesized with the second laser radar point cloud data.

7. The method according to claim 1, characterized in that Collect the second lidar point cloud data under normal weather conditions, including: Under normal weather conditions, using a second laser radar installed on a second unmanned vehicle to collect second laser radar point cloud data for a historical driving area or a designated driving area; Among them, the difference between the position and posture of the second laser radar relative to the second unmanned vehicle and the position and posture of the first laser radar relative to the first unmanned vehicle is less than a difference threshold.

8. A laser radar data generating device, characterized in that: include: A first acquisition module is used to acquire first laser radar point cloud data using a first laser radar installed on a first unmanned vehicle under severe weather conditions; A second acquisition module, used for acquiring second laser radar point cloud data under normal weather conditions; The data synthesis module is used to synthesize the first laser radar point cloud data and the second laser radar point cloud data to obtain synthetic laser radar point cloud data under simulated severe weather conditions.

9. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the steps of the lidar data generation method as described in any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the laser radar data generation method as described in any one of claims 1 to 7.

Images