A method for testing quality of laser radar point cloud in rainfall weather

Abstract

The present application relates to a kind of laser radar point cloud quality test method under rainfall weather, belong to the field of perception test.The present application is divided into three parts: for the problem of distance attenuation of laser beam in the propagation process, design and propose a kind of multi-collaborative target distance information test working condition method based on BEV perspective, to obtain the attenuation degree of target distance under different rainfall conditions;For the problem of excessive data amount of laser radar point cloud map data saved under seven conditions of no rainfall, light rain, moderate rain, heavy rain, heavy rain, heavy rain, design and propose a kind of polar coordinate grid segmentation mechanism based on region space, it is divided into different size region space for the problem of laser radar point cloud near dense far sparse, to solve the problem of excessive data amount;For the voxel grid divided by the polar coordinate grid segmentation mechanism based on region space, design and propose a kind of point cloud matching method based on iterative voting mechanism, to obtain the proportion of point cloud noise under different rainfall conditions.

Description

Technical Field

[0001] This invention belongs to the field of perception testing and relates to a method for testing the quality of lidar point clouds under rainy weather. Background Technology

[0002] LiDAR (Light Detection and Ranging) is a commonly used 3D sensing device that uses emitted laser beams and received echoes to acquire 3D point cloud data of the surrounding environment. LiDAR offers advantages such as high resolution, high precision, and long-range measurement, and is widely used in fields such as autonomous driving.

[0003] However, lidar faces a series of challenges in rainy weather. Raindrops interfere with the propagation and reception of the laser beam, thus affecting the quality of point cloud data. This mainly includes two aspects: first, raindrop scattering causes noise and invalid points in the point cloud, making the edges of target objects blurry; second, the laser beam attenuates during propagation, causing errors in the distance information in the point cloud.

[0004] Currently, research focuses primarily on reflection intensity and point cloud density, leading to the development of various research methods. Since raindrops scatter laser beams, researchers analyze the reflection intensity of received lidar echo signals to distinguish whether noise is caused by raindrops. For example, by designing appropriate thresholds and filtering algorithms, effective differentiation can be achieved. Raindrop scattering blurs the edges of target objects in point cloud data, affecting the extraction and analysis of their geometric features. Researchers assess the quality of point cloud data during rainy weather by analyzing its density, distribution uniformity, shape, and other geometric characteristics. For instance, by modeling and matching the shape and structure of specific target objects, the accuracy and reliability of the point cloud data can be further assessed.

[0005] Research on testing methods for lidar point cloud quality under rainy weather is crucial for improving the application performance of lidar in complex weather conditions. However, to date, there is no comprehensive testing scheme to address this issue. Effective testing and improvement of lidar point cloud quality under rainy weather can enhance lidar perception capabilities and ensure the safety and reliability of autonomous driving systems. This is significant for promoting the development of autonomous driving technology, realizing the widespread adoption of intelligent transportation systems, and improving the accuracy of urban planning and design. Furthermore, research on lidar point cloud quality testing methods under rainy weather provides valuable insights for academic research and engineering practice in related fields. By gaining a deeper understanding of raindrop interference mechanisms, analyzing the characteristics of point cloud data, and developing corresponding algorithms and models, theoretical and methodological support can be provided for optimizing and improving lidar performance under rainy weather. Simultaneously, accumulating and sharing lidar datasets and test results under rainy weather will help build more comprehensive datasets and evaluation standards, promoting further development in this field. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a method for testing the quality of lidar point clouds under rainy weather, which solves the problem of automatic waste sorting in common places such as train stations and improves the accuracy of waste sorting and recycling.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for testing the quality of lidar point clouds under rainy weather, the method comprising the following steps:

[0009] S1: Establish a quality evaluation index for lidar point clouds, and analyze the proportion of noise in lidar point clouds and the attenuation of the distance information corresponding to the target under different rainfall intensities.

[0010] The lidar point cloud quality evaluation indicators include point cloud noise and target distance information point cloud noise.

[0011] Point cloud noise includes the proportion of noisy points and invalid points affected by different rainfall; target distance information point cloud noise includes the degree of attenuation of the target distance measured by LiDAR for different targets;

[0012] S2: Construct a LiDAR point cloud quality test field. This test field adheres to the principle of variable consistency. Based on the rainfall intensity, the rainfall is divided into six levels according to the rainfall level classification published by the meteorological bureau: light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. According to the purpose of the test and the scale of the target being tested, the test field is designed in a closed environment with a length greater than 500m and a width greater than 20m, and the rainfall intensity remains equal at any location. The LiDAR is installed in the perception system device of the intelligent connected vehicle and waterproofed accordingly. Test software equipment is installed in the test scene to record the LiDAR 3D point cloud map for each frame.

[0013] S3: To address the issue of distance attenuation during laser beam propagation, a test method for multi-cooperative target distance information based on the BEV perspective is designed.

[0014] S4: To address the problem of excessively large amounts of lidar point cloud image data stored under seven conditions—no rainfall, light rain, moderate rain, heavy rain, rainstorm, torrential rain, and extremely heavy rain—a polar coordinate grid segmentation method based on regional space is designed.

[0015] S5: For voxel meshes divided by a polar coordinate mesh segmentation mechanism based on regional space, design a point cloud matching method based on an iterative voting mechanism.

[0016] Optionally, S3 specifically includes the following steps:

[0017] S301: Under different rainfall conditions, cars, trucks, bicycles, and people are placed in different scenarios in front, behind, left, and right of the tested intelligent connected vehicle, so that they move according to a certain motion law.

[0018] S302: Keep the test vehicle SV stationary and apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. Set up test vehicles around SV. Test vehicles on the left and right enter the perception range of SV at speeds of 20km / h, 60km / h, and 80km / h, respectively. Test vehicles behind and in front remain stationary. The test vehicles exist within the perception range for a period of time, and the perception equipment records the BEV point cloud map of each frame of the lidar point cloud during this period.

[0019] S303: Keep the test vehicle SV stationary and apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. Set up test trucks around the SV. Test trucks on the left and right enter the SV's sensing range at speeds of 10 km / h, 30 km / h, and 60 km / h, respectively, while test trucks behind and in front remain stationary. The test trucks remain within the sensing range for a period of time, and the sensing equipment records the BEV point cloud map of each frame of the lidar point cloud during that period.

[0020] S304: Keep the test vehicle SV stationary and apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. Place test pedestrians around SV and have them enter the perception range of SV at speeds of 4 km / h, 10 km / h, and 18 km / h, and remain within the perception range for a period of time. The perception device records the BEV point cloud map of each frame of the lidar point cloud during this period of time.

[0021] S305: Keep the test vehicle SV stationary, apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. Set up test bicycles around SV, and drive the test bicycles into the perception range of SV at speeds of 10km / h, 20km / h, and 30km / h, and remain within the perception range for a period of time. The perception device records the BEV point cloud map of each frame of the lidar point cloud during this period of time.

[0022] S306: Based on the target distance information type in the lidar point cloud quality evaluation index, test the distance information of different types of targets under different rainfall conditions, and test the degree of distance attenuation under rainfall conditions;

[0023]

[0024] Where d r d represents the distance to the target under rainfall conditions, β represents the distance to the target under no rainfall conditions, and β is the distance information evaluation index.

[0025] Optionally, S4 specifically includes the following steps:

[0026] S401: Due to the characteristic of lidar point clouds being denser in the near and sparser in the far, the lidar point cloud is divided on the XOY plane, and the model region is defined as follows:

[0027]

[0028] Where C represents the point cloud region of the lidar, and Z... m Let N be the m-th region, and N be the number of lidar point cloud regions divided.

[0029] S402: For a sensing area greater than 80m, divide it into a single space; for a sensing area less than 5m, also divide it into a single area.

[0030] S403: For sensing areas greater than 5m and less than 80m, divide them into grids of different sizes, each grid S i，j，m The definition is as follows:

[0031]

[0032] Where L m，min and L m，max Z m The minimum and maximum radial boundaries, ρ m Let θ be the radial radius range of the m-th region. m Let N be the sector angle of the m-th region. θ，m The number of sectors is the m-th region.

[0033] S404: In a sensing area greater than 5m and less than 80m, the size of each grid in the outermost and innermost layers is set to be larger to solve the problem of denser grids near the edge and sparser grids far away.

[0034] S405: The lidar point cloud is divided into grids of equal size using the principle of equal spacing in the z-axis direction;

[0035] S406: Divides the 3D LiDAR point cloud into voxel grids of different sizes in different regions within the entire LiDAR point cloud space, and processes each small voxel grid separately, thus solving the problem of excessive LiDAR point cloud data volume.

[0036] Optionally, S5 specifically includes the following steps:

[0037] S501: For the six matching mechanisms of no rainfall and light rain, no rainfall and moderate rain, no rainfall and heavy rain, no rainfall and rainstorm, no rainfall and heavy rainstorm, and no rainfall and extremely heavy rainstorm, the lidar point cloud map saved in each case will be matched frame by frame for each region in the manner designed in S4.

[0038] S502: For each matched region, Principal Component Analysis (PCA) is used to reduce the dimensionality of the three-dimensional LiDAR points to one dimension. Specifically, the three-dimensional LiDAR points in each matched region are formed into a three-dimensional matrix. The covariance matrix of the three-dimensional matrix is ​​calculated, and then the eigenvalues ​​and eigenvectors are calculated. The data is then mapped onto the principal components, and the data matrix is ​​multiplied by the principal component transformation matrix to reduce it to one dimension.

[0039] S503: For each dimension-reduced matching region, arrange its values ​​in descending order. First, in the case of rainfall, vote to select the largest value each time. Compare the selected value with the unmarked maximum value in the case of no rainfall. If a certain threshold is met, the point is considered a match and the selected point in both cases is marked as matched. If the threshold is not met and the selected maximum value is greater than the maximum value in the case of rainfall, select the next largest value for comparison. If the threshold is not met and the selected maximum value is less than the maximum value in the case of rainfall, the point is considered noise and is marked as noise. Continue until all points are marked.

[0040] S504: Perform noise evaluation on each segmented region separately, and calculate the proportion of noise to the entire segmented region, as shown in the following formula:

[0041]

[0042] Where N e N represents the total number of regions within a 40m radius. d N represents the total number of regions beyond 40m. noise，i N represents the number of noise points in the i-th partitioned region. l，i γ is the number of lidar points in the i-th partitioned region, γ1 is the proportional coefficient of the partitioned region within 40m, γ2 is the proportional coefficient of the partitioned region beyond 40m, and α is the point cloud noise evaluation index.

[0043] The beneficial effects of this invention are as follows:

[0044] (1) To address the problem of distance attenuation during laser beam propagation, a multi-cooperative target distance information testing method based on BEV perspective is proposed to obtain the degree of target distance attenuation under different rainfall conditions.

[0045] (2) To address the problem of excessive data volume in lidar point cloud images stored under seven conditions: no rainfall, light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm, a polar coordinate grid segmentation mechanism based on regional space is designed and proposed. This mechanism divides lidar point clouds into regional spaces of different sizes to address the problem of excessive data volume.

[0046] (3) For voxel grids divided by the polar coordinate grid segmentation mechanism based on regional space, a point cloud matching method based on iterative voting mechanism is designed and proposed to obtain the proportion of point cloud noise under different rainfall conditions.

[0047] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0048] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0049] Figure 1 This is a system framework diagram of the present invention;

[0050] Figure 2 This is a schematic diagram of the lidar point cloud quality test field of the present invention;

[0051] Figure 3 This is a schematic diagram of the lidar point cloud quality system of the present invention;

[0052] Figure 4 This is a test vehicle operating condition diagram for the present invention;

[0053] Figure 5 This is a test truck operating condition diagram for the present invention;

[0054] Figure 6 This is a test case diagram for the present invention;

[0055] Figure 7 This is a diagram showing the working conditions of the bicycle tested in this invention;

[0056] Figure 8 This is an XOY plane partitioning diagram of the lidar point cloud of the present invention;

[0057] Figure 9 This is a flowchart of the iterative voting mechanism of the present invention. Detailed Implementation

[0058] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0059] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0060] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0061] A method for testing the quality of lidar point clouds under rainy weather includes: a multi-cooperative target range information testing method based on the BEV (Browser-Earning-Earth) perspective, a polar coordinate grid segmentation mechanism based on regional space, and a point cloud matching method based on an iterative voting mechanism. The multi-cooperative target range information testing method based on the BEV perspective designs testing conditions for multiple cooperative targets under different rainfall conditions to obtain target range information. The polar coordinate grid segmentation mechanism based on regional space divides the lidar point cloud region into multiple voxel grids with regular intervals in the radial and azimuth directions. The point cloud matching method based on the iterative voting mechanism designs an iterative voting mechanism to exchange and match frames based on the lidar point cloud images saved under different conditions.

[0062] In the implementation method, in order to address the problem of distance attenuation of the laser beam during propagation, a multi-cooperative target distance information testing method based on the BEV perspective is designed and proposed. Under different rainfall conditions, cars, trucks, bicycles and people are placed in different scenarios in front, behind, left and right of the intelligent connected vehicle under test, and made to move forward according to a certain motion law, and the distance of the target in each case is saved.

[0063] In the implementation, to address the problem of excessively large amounts of LiDAR point cloud image data stored under seven conditions—no rainfall, light rain, moderate rain, heavy rain, torrential rain, heavy downpour, and extremely heavy downpour—a polar coordinate grid segmentation mechanism based on regional space is proposed. Due to the near-dense and far-sparse characteristics of LiDAR point clouds, the LiDAR point cloud region is divided into multiple voxel grids with regular intervals in the radial and azimuth directions. The entire LiDAR point cloud image is divided into multiple blocks for processing, which greatly reduces the computational complexity and can reasonably detect point cloud noise.

[0064] In the implementation, for voxel grids divided by the polar coordinate grid segmentation mechanism based on regional space, a point cloud matching method based on an iterative voting mechanism is proposed. The LiDAR point cloud image frames saved under different conditions are exchanged and matched. The matched point cloud image is used to perform dimensionality reduction operation on the LiDAR point cloud in the z-axis direction using principal component analysis on the corresponding voxel grid. Then, the LiDAR points selected by each vote are matched for the nearest distance using an iterative voting mechanism to find the point cloud noise caused by rainfall.

[0065] like Figure 1 As shown, this invention is a method for testing the quality of lidar point clouds under rainy weather. Specifically, this invention includes the following steps:

[0066] S1. A lidar point cloud quality evaluation index is proposed to analyze the proportion of noise points and the attenuation of target distance information in lidar point clouds under different rainfall intensities. As shown in Table 1, the lidar point cloud quality evaluation index includes point cloud noise and target distance information. The evaluation parameters for point cloud noise include the proportion of noise points and invalid points affected by different rainfall; the target distance information includes the attenuation of the target distance measured by lidar for different targets.

[0067] Table 1. LiDAR Point Cloud Quality Evaluation Indicators

[0068]

[0069] S2. Set up a LiDAR point cloud quality test field, such as Figure 2 As shown; this quality test field adheres to the principle of variable consistency. Excluding other influencing factors, rainfall is classified into six levels—light rain, moderate rain, heavy rain, rainstorm, torrential rain, and extremely heavy rain—based on the rainfall intensity levels published by the meteorological bureau. According to the test objective and the scale of the target, the test field is designed in a closed environment with a length greater than 500m and a width greater than 20m, ensuring equal rainfall intensity at any location. LiDAR is installed in the perception system of the intelligent connected vehicle and waterproofed accordingly. Test software and equipment are installed in the test environment. The test system is as follows: Figure 3 As shown, the three-dimensional point cloud map of the lidar is recorded for each frame.

[0070] S3. To address the issue of laser beam attenuation during propagation, a method for testing the range information of multiple cooperative targets based on the BEV perspective is designed and proposed. The method is as follows:

[0071] S301. Under different rainfall conditions, cars, trucks, bicycles, and people are placed in different scenarios in front, behind, left, and right of the tested intelligent connected vehicle, so that they move according to a certain motion law.

[0072] S302. Keep the test vehicle SV stationary and apply rainfall of seven different intensities: no rain, light rain, moderate rain, heavy rain, torrential rain, extremely heavy rain, and extremely heavy rain. Set up [something] around SV. Figure 4 The test vehicles shown are positioned as follows: the test vehicles on the left and right enter the SV's sensing range at speeds of 20 km / h, 60 km / h, and 80 km / h, respectively, while the test vehicles behind and in front remain stationary. The test vehicles remain within the sensing range for a period of time, and the sensing device records the BEV point cloud map for each frame of the lidar point cloud during that period.

[0073] S303. Keep the test vehicle SV stationary and apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, torrential rain, extremely heavy rain, and extremely heavy rain. Set up [something] around SV. Figure 5 The test trucks shown are shown on the left and right. The test trucks enter the SV's sensing range at speeds of 10 km / h, 30 km / h, and 60 km / h, respectively. The test trucks behind and in front remain stationary. The test trucks exist within the sensing range for a period of time, and the sensing device records the BEV point cloud map of each frame of the lidar point cloud during that period.

[0074] S304. Keep the test vehicle SV stationary and apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, torrential rain, extremely heavy rain, and extremely heavy rain. Set up [something] around SV. Figure 6 The test pedestrians shown entered the SV's sensing range at speeds of 4km / h, 10km / h, and 18km / h and remained within the sensing range for a period of time. The sensing device recorded the BEV point cloud map for each frame of the lidar point cloud during that period.

[0075] S305. Keep the test vehicle SV stationary and apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, torrential rain, extremely heavy rain, and extremely heavy rain. Set up [something] around SV. Figure 7 The test bicycle shown was driven into the sensing range of SV at speeds of 10km / h, 20km / h, and 30km / h, and remained within the sensing range for a period of time. The sensing device recorded the BEV point cloud map of each frame of the lidar point cloud during that period of time.

[0076] S306. Based on the target distance information type in the lidar point cloud quality evaluation index, test the distance information of different types of targets under different rainfall conditions, and test the degree of distance attenuation under rainfall conditions.

[0077]

[0078] Where d r d represents the distance to the target under rainfall conditions, β represents the distance to the target under no rainfall conditions, and β is the distance information evaluation index.

[0079] S4. To address the issue of excessively large amounts of lidar point cloud image data stored under seven scenarios—no rainfall, light rain, moderate rain, heavy rain, torrential rain, extremely heavy rain, and exceptionally heavy rain—a polar coordinate grid segmentation mechanism based on regional space is proposed. The specific method is as follows:

[0080] S401. Due to the characteristic of lidar point clouds being denser in the near and sparser in the far, the lidar point cloud is divided on the XOY plane, such as... Figure 8 As shown. The model region is defined as follows:

[0081]

[0082] Where C represents the point cloud region of the lidar, and Z... m Let N be the m-th region, and N be the number of lidar point cloud regions divided.

[0083] S402. For a sensing area greater than 80m, divide it into a single space; for a sensing area less than 5m, also divide it into a single region.

[0084] S403. For sensing areas greater than 5m and less than 80m, divide them into grids of different sizes, each grid S i，j，m The definition is as follows:

[0085]

[0086] Where L m，min and L m，max Z m The minimum and maximum radial boundaries, ρ m Let θ be the radial radius range of the m-th region. m Let N be the sector angle of the m-th region. θ，m This represents the number of sectors divided into the m-th region.

[0087] S404. In a sensing area greater than 5m and less than 80m, the size of each grid in the outermost and innermost layers is set to be larger to solve the problem of denser grids near the edge and sparser grids far away.

[0088] S405. Using the principle of equal-distance division in the z-axis direction, the lidar point cloud is divided into grids of equal size.

[0089] S406: The three-dimensional LiDAR point cloud is divided into voxel grids of different sizes in different regions within the entire LiDAR point cloud space. Each small voxel grid is processed separately, which solves the problem of excessive data volume in the LiDAR point cloud map.

[0090] S5. For voxel meshes partitioned using a polar coordinate meshing mechanism based on regional space, a point cloud matching method based on an iterative voting mechanism is proposed. The method is as follows:

[0091] S501. For the six matching mechanisms of no rainfall and light rain, no rainfall and moderate rain, no rainfall and heavy rain, no rainfall and rainstorm, no rainfall and torrential rain, and no rainfall and extremely heavy rain, the lidar point cloud map saved in each case is matched frame by frame for each region in the manner designed in S4.

[0092] S502. For each matched region, PCA (Principal Component Analysis) is used to reduce the dimensionality of the three-dimensional LiDAR points to one dimension. Specifically, the three-dimensional LiDAR points of each matched region are formed into a three-dimensional matrix. The covariance matrix of the three-dimensional matrix is ​​calculated, and then the eigenvalues ​​and eigenvectors are calculated. The data is then mapped onto the principal components, and the data matrix is ​​multiplied by the principal component transformation matrix to reduce it to one dimension.

[0093] S503. For each dimension-reduced matching region, arrange its values ​​in descending order. First, in the case of rainfall, vote to select the largest value each time. Compare the selected value with the unmarked maximum value in the case of no rainfall. If a certain threshold is met, the point is considered a match, and the selected point in both cases is marked as matched. If the threshold is not met and the selected maximum value is greater than the maximum value in the case of rainfall, select the next largest value for comparison. If the threshold is not met and the selected maximum value is less than the maximum value in the case of rainfall, the point is considered noise and marked as noise. Continue until all points are marked. The flowchart is as follows. Figure 9 As shown.

[0094] S504. Perform noise evaluation on each divided region separately and calculate the proportion of noise in the entire divided region, as shown in the following formula.

[0095]

[0096] Where N e N represents the total number of regions within a 40m radius. d N represents the total number of regions beyond 40m. noise，i N represents the number of noise points in the i-th partitioned region. l，i γ is the number of lidar points in the i-th partitioned region, γ1 is the proportional coefficient of the partitioned region within 40m, γ2 is the proportional coefficient of the partitioned region beyond 40m, and α is the point cloud noise evaluation index.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for testing the quality of lidar point clouds under rainy weather, characterized in that: The method includes the following steps: S1: Establish a quality evaluation index for lidar point clouds, and analyze the proportion of noise in lidar point clouds and the attenuation of the distance information corresponding to the target under different rainfall intensities. The lidar point cloud quality evaluation indicators include point cloud noise and target distance information point cloud noise. Point cloud noise includes the proportion of noisy points and invalid points affected by different rainfall; target distance information point cloud noise includes the degree of attenuation of the target distance measured by LiDAR for different targets; S2: Construct a LiDAR point cloud quality test field. This test field adheres to the principle of variable consistency. Based on the rainfall intensity, the rainfall is divided into six levels according to the rainfall level classification published by the meteorological bureau: light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. According to the purpose of the test and the scale of the target being tested, the test field is designed in a closed environment with a length greater than 500m and a width greater than 20m, and the rainfall intensity remains equal at any location. The LiDAR is installed in the perception system device of the intelligent connected vehicle and waterproofed accordingly. Test software equipment is installed in the test scene to record the LiDAR 3D point cloud map for each frame. S3: To address the issue of distance attenuation during laser beam propagation, a test method for multi-cooperative target distance information based on the BEV perspective is designed. S4: To address the problem of excessively large amounts of lidar point cloud image data stored under seven conditions—no rainfall, light rain, moderate rain, heavy rain, rainstorm, torrential rain, and extremely heavy rain—a polar coordinate grid segmentation method based on regional space is designed. S5: For voxel meshes partitioned using a polar coordinate meshing mechanism based on region space, a point cloud matching method based on an iterative voting mechanism is designed; specifically, the following steps are included: S501: For the six matching mechanisms of no rainfall and light rain, no rainfall and moderate rain, no rainfall and heavy rain, no rainfall and rainstorm, no rainfall and heavy rainstorm, and no rainfall and extremely heavy rainstorm, the lidar point cloud map saved in each case will be matched frame by frame for each region in the manner designed in S4. S502: For each matched region, Principal Component Analysis (PCA) is used to reduce the dimensionality of the three-dimensional LiDAR points to one dimension. Specifically, the three-dimensional LiDAR points in each matched region are formed into a three-dimensional matrix. The covariance matrix of the three-dimensional matrix is ​​calculated, and then the eigenvalues ​​and eigenvectors are calculated. The data is then mapped onto the principal components, and the data matrix is ​​multiplied by the principal component transformation matrix to reduce it to one dimension. S503: For each dimension-reduced matching region, arrange its values ​​in descending order. First, in the case of rainfall, vote to select the largest value each time. Compare the selected value with the unmarked maximum value in the case of no rainfall. If a certain threshold is met, the point is considered a match and the selected point in both cases is marked as matched. If the threshold is not met and the selected maximum value is greater than the maximum value in the case of rainfall, select the next largest value for comparison. If the threshold is not met and the selected maximum value is less than the maximum value in the case of rainfall, the point is considered noise and is marked as noise. Continue until all points are marked. S504: Perform noise evaluation on each segmented region separately, and calculate the proportion of noise to the entire segmented region, as shown in the following formula: in The total number of areas within 40m. The total number of areas beyond 40m. Let be the number of noise points in the i-th partitioned region. Let be the number of LiDAR points in the i-th partitioned region. The proportional coefficient for dividing the area within 40m. The proportional coefficient for dividing the area beyond 40m. This is a metric for evaluating point cloud noise.

2. The method for testing the quality of lidar point clouds under rainy weather according to claim 1, characterized in that: S3 specifically includes the following steps: S301: Under different rainfall conditions, cars, trucks, bicycles, and people are placed in different scenarios in front, behind, left, and right of the tested intelligent connected vehicle, so that they move according to a certain motion law. S302: Keep the test vehicle SV stationary and apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. Set up test vehicles around SV. Test vehicles on the left and right enter the perception range of SV at speeds of 20km / h, 60km / h, and 80km / h, respectively. Test vehicles behind and in front remain stationary. The test vehicles exist within the perception range for a period of time, and the perception equipment records the BEV point cloud map of each frame of the lidar point cloud during this period. S303: Keep the test vehicle SV stationary and apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. Set up test trucks around the SV. Test trucks on the left and right enter the SV's sensing range at speeds of 10 km / h, 30 km / h, and 60 km / h, respectively, while test trucks behind and in front remain stationary. The test trucks remain within the sensing range for a period of time, and the sensing equipment records the BEV point cloud map of each frame of the lidar point cloud during that period. S304: Keep the test vehicle SV stationary and apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. Place test pedestrians around SV and have them enter the perception range of SV at speeds of 4 km / h, 10 km / h, and 18 km / h, and remain within the perception range for a period of time. The perception device records the BEV point cloud map of each frame of the lidar point cloud during this period of time. S305: Keep the test vehicle SV stationary, apply seven different intensities of rainfall: no rain, light rain, moderate rain, heavy rain, rainstorm, heavy rainstorm, and extremely heavy rainstorm. Set up test bicycles around SV, and drive the test bicycles into the perception range of SV at speeds of 10km / h, 20km / h, and 30km / h, and remain within the perception range for a period of time. The perception device records the BEV point cloud map of each frame of the lidar point cloud during this period of time. S306: Based on the target distance information type in the lidar point cloud quality evaluation index, test the distance information of different types of targets under different rainfall conditions, and test the degree of distance attenuation under rainfall conditions; in The distance to the target under rainfall conditions. The distance to the target in the absence of rainfall. It serves as an evaluation index for distance information.

3. The method for testing the quality of lidar point clouds under rainy weather according to claim 2, characterized in that: S4 specifically includes the following steps: S401: Due to the characteristic of lidar point clouds being denser in the near and sparser in the far, the lidar point cloud is divided on the XOY plane, and the model region is defined as follows: in This represents the point cloud area for lidar. For the m-th region, The number of lidar point cloud regions to be divided; S402: For a sensing area greater than 80m, divide it into a single space; for a sensing area less than 5m, also divide it into a single area. S403: For sensing areas greater than 5m and less than 80m, divide them into grids of different sizes, each grid... The definition is as follows: in and They represent The minimum and maximum radial boundaries, Let m be the radial radius range of the m-th region. Let m be the sector angle of the m-th region. The number of sectors is the m-th region. S404: In a sensing area greater than 5m and less than 80m, the size of each grid in the outermost and innermost layers is set to be larger to solve the problem of denser grids near the edge and sparser grids far away. S405: The lidar point cloud is divided into grids of equal size using the principle of equal spacing in the z-axis direction; S406: Divides the 3D LiDAR point cloud into voxel grids of different sizes in different regions within the entire LiDAR point cloud space, and processes each small voxel grid separately, thus solving the problem of excessive LiDAR point cloud data volume.

Images