Method and apparatus for testing lidar

Abstract

A method (200) and apparatus (2000) for testing LiDAR. The method (200) comprises: acquiring first point cloud data collected by LiDAR to be tested (S210) when same performs detection on a test environment (150), wherein the test environment (150) comprises a detection target including a high-reflectivity object, and the test environment (150) further comprises at least one of a road surface and a lane; on the basis of the first point cloud data and a ground-truth point cloud map of the test environment (150), determining the distribution of crosstalk points in the first point cloud data (S220), wherein the ground-truth point cloud map indicates a contour boundary of the detection target in the test environment (150); and on the basis of the distribution of the crosstalk points, determining a risk level of high-reflectivity point cloud blooming noise of said LiDAR (S230). The method (200) can be used for testing LiDAR used by a new energy vehicle or an intelligent driving vehicle, can improve the evaluation precision of high-reflectivity point cloud blooming noise of the LiDAR, and can obtain the degree to which the high-reflectivity point cloud blooming noise of the LiDAR to be tested affects a downstream control apparatus thereof.

Description

Test method and test device for lidar

[0001] The present application claims priority to the Chinese patent application No. 202410641017.9, filed on May 22, 2024, and entitled "Test method and test device for lidar", the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0002] The present application relates to the field of intelligent driving, and more particularly, to a test method and test device for lidar. BACKGROUND

[0003] As a perception sensor, lidar is widely used in the fields of intelligent driving, robots, etc. In actual detection scenarios, high-reflectivity objects such as road signs and cone barrels can cause high-reflectivity point cloud inflation, resulting in a deviation between the point cloud formed by the lidar detection and the actual size of the object, and affecting the detection performance of the lidar. Taking autonomous driving as an example, when there is high-reflectivity point cloud inflation noise in the point cloud data, the autonomous driving domain controller and other control devices using the point cloud data may misjudge the environment in which the vehicle is located, and thus control the vehicle to perform unnecessary obstacle avoidance operations, endangering the driving safety of the vehicle.

[0004] Therefore, how to accurately evaluate the risk caused by high-reflectivity point cloud inflation noise of the lidar has become a problem to be solved. SUMMARY

[0005] The present application provides a test method and test device for lidar, which can improve the evaluation accuracy of high-reflectivity point cloud inflation noise of the lidar, and can obtain the influence degree of high-reflectivity point cloud inflation noise of the to-be-tested lidar on the control device downstream thereof.

[0006] In a first aspect, a test method for lidar is provided. The method can be executed by a test device for lidar, or can be executed by a component (such as a chip, a processor or a processing circuit) of the test device, or can be executed by a test system composed of the lidar and the test device.

[0007] The method comprises: acquiring first point cloud data collected by a to-be-tested laser radar in a test environment, the test environment comprising a detection target comprising a high reflectivity object, and the test environment further comprising at least one of a road surface and a lane; determining a distribution of crosstalk points in the first point cloud data according to the first point cloud data and a ground truth point cloud map of the test environment, the ground truth point cloud map indicating a contour boundary of the detection target in the test environment, and the distribution of the crosstalk points comprising at least one of a distance between the crosstalk points and the contour boundary of the high reflectivity object, a distance between the crosstalk points and the road surface, and a lane in which the crosstalk points are located; and determining a risk level of high-reflectivity point cloud blooming noise of the to-be-tested laser radar according to the distribution of the crosstalk points.

[0008] Exemplarily, the risk level of the high-reflectivity point cloud blooming noise can indicate a degree of credibility of point cloud data collected by the laser radar. In other words, the risk level of the high-reflectivity point cloud blooming noise can indicate an influence degree of the high-reflectivity point cloud blooming noise performance of the laser radar on operation of a control device downstream of the laser radar (i.e., a control device directly or indirectly using point cloud data of the laser radar, such as an automatic driving domain controller). Since an external manifestation of the operation of the control device or the automatic driving system can be manifested as control of a driving process of a vehicle by the control device or the automatic driving system, the high-reflectivity point cloud blooming noise can indicate an influence degree of the high-reflectivity point cloud blooming noise performance of the laser radar on the driving process of the vehicle.

[0009] In the present application, since the ground truth point cloud map of the test environment can indicate contour boundaries of objects in the test environment, based on the ground truth point cloud map, crosstalk points corresponding to the high-reflectivity point cloud blooming noise can be accurately screened from point cloud data collected by the laser radar, and the evaluation accuracy of the high-reflectivity point cloud blooming noise performance of the laser radar can be improved. According to the distribution of the crosstalk points, the risk level of the high-reflectivity point cloud blooming noise of the laser radar can be determined, and the influence degree of the high-reflectivity point cloud blooming noise performance of the laser radar on operation of a control device downstream of the laser radar, operation of an automatic driving system using point cloud data of the laser radar, and a driving process of a vehicle can be determined.

[0010] In some implementations of the first aspect, the risk level can include a first risk level and a second risk level, the risk degree of the first risk level being lower than the risk degree of the second risk level. The to-be-tested lidar can be in the first lane when detecting the test environment. According to the distribution of the crosstalk points, determining the risk level of the high-reflection point cloud expansion noise of the to-be-tested lidar can include: when the distance between the crosstalk points and the road surface is greater than a first threshold, determining the risk level of the high-reflection point cloud expansion noise as the first risk level; or, when the crosstalk points are in a lane other than the first lane, determining the risk level of the high-reflection point cloud expansion noise as the first risk level; or, when the number of the crosstalk points is less than or equal to a second threshold, determining the risk level of the high-reflection point cloud expansion noise as the first risk level; or, when the distance between the crosstalk points and the road surface is less than or equal to the first threshold, the crosstalk points are in the first lane, and the number of the crosstalk points is greater than or equal to the second threshold, determining the risk level of the high-reflection point cloud expansion noise as the second risk level.

[0011] In one embodiment, the height between the crosstalk points and the ground is greater than a first threshold. In this scenario, even if the crosstalk points cause the control device downstream of the lidar to misjudge the detected target, the vehicle can be controlled to safely pass through the space between the crosstalk points and the road surface without needing to control the vehicle to avoid obstacles; the driving mode of the vehicle in this scenario can be the same as when there is no high-reflection point cloud expansion noise in the point cloud data. The risk degree of the high-reflection point cloud expansion noise is relatively low, and can be determined as the first risk level.

[0012] In another embodiment, the crosstalk points are in a lane other than the current lane of the vehicle. In this scenario, even if the crosstalk points cause the control device downstream of the lidar to misjudge the detected target, the vehicle can be controlled to safely drive in the current lane without needing to control the vehicle to avoid obstacles. The risk degree of the high-reflection point cloud expansion noise is relatively low, and can be determined as the first risk level.

[0013] In this application, according to the relationship between the distance between the crosstalk points and the road surface, the lane in which the crosstalk points are located, the number of the crosstalk points, and the corresponding threshold, the risk degree caused by the high-reflection point cloud expansion noise performance of the lidar to the driving process of the vehicle can be determined.

[0014] In some implementations of the first aspect, determining the distribution of the crosstalk points in the first point cloud data according to the first point cloud data and the ground truth point cloud map of the test environment can include: determining a local ground truth point cloud map corresponding to the position information of the first point cloud data from the ground truth point cloud map according to the position information; and determining the distribution of the crosstalk points according to the first point cloud data and the local ground truth point cloud map.

[0015] In actual tests, the range of the test environment may be much larger than the detection range of the lidar. In this application, by determining the local ground truth point cloud map, the number of data points required for processing can be reduced, and the test efficiency can be improved.

[0016] In combination with the first aspect, in some implementations of the first aspect, the method can further include: removing the first data points from the first point cloud data according to a white list. The white list can indicate the types of the first data points, and the types of the first data points can indicate the categories of the detection targets corresponding to the first data points. The categories of the detection targets corresponding to the first data points are different from the high-reflectivity objects.

[0017] In addition to the high-reflectivity objects, the test scene often includes other categories of detection targets, such as vegetation, vulnerable road users, walls, etc. In this application, based on the types of the data points in the first point cloud data, in combination with the white list, the data points corresponding to the detection targets irrelevant to the high-reflectivity objects in the first point cloud data can be screened out. In this way, on the one hand, the data points irrelevant to the high-reflectivity object can be excluded from the test results; on the other hand, the number of data points required for processing can be reduced, and the test efficiency can be improved.

[0018] In combination with the first aspect, in some implementations of the first aspect, the method can further include: obtaining a plurality of point cloud data collected by the calibration lidar at a plurality of positions for detecting the test environment; splicing and superimposing the plurality of point cloud data to obtain second point cloud data according to the pose information of the calibration lidar at the plurality of positions; and removing the high-reflectivity point cloud blooming noise from the second point cloud data according to the first artificial annotation information to obtain the ground truth point cloud map of the test environment.

[0019] In actual scenarios, the point cloud data collected by the calibration lidar can also have high-reflectivity point cloud blooming noise. In this application, by removing the high-reflectivity point cloud blooming noise in the point cloud data based on the annotation information, the interference of the invalid points corresponding to the high-reflectivity point cloud blooming noise on the ground truth point cloud map can be avoided, and the accuracy of the test results can be improved.

[0020] In combination with the first aspect, in some implementations of the first aspect, the crosstalk points can include: data points in the first point cloud data that have a difference greater than or equal to a preset threshold with corresponding data points in the ground truth point cloud map and are within a preset range around the high-reflectivity objects.

[0021] In the present application, the data points in the point cloud data collected by the to-be-tested laser radar are taken as the measured values of the detection results in the test, and the data points in the ground truth point cloud map can be taken as the ground truth of the detection results in the test. By matching with the data points in the ground truth point cloud map, the data points in the first point cloud data that have a difference less than a preset threshold with the corresponding data points in the ground truth point cloud map can be taken as valid points in the first point cloud data. The data points in the first point cloud data that have a difference greater than or equal to the preset threshold with the corresponding data points in the ground truth point cloud map and are within a preset range around the high-reflectivity object can be taken as invalid points corresponding to the high-reflectivity point cloud expansion noise, i.e., crosstalk points. In this way, even if the high-reflectivity point cloud expansion noise has the characteristics of a non-fixed shape, a non-fixed distribution, and a non-fixed position, the crosstalk points can still be effectively screened out from the first point cloud data, and the high-reflectivity point cloud expansion noise of the to-be-tested laser radar can be evaluated.

[0022] In a second aspect, a testing device is provided. The testing device includes an obtaining unit and a processing unit. The obtaining unit is configured to obtain first point cloud data collected by a to-be-tested laser radar when detecting a test environment. The test environment includes a detection target including a high-reflectivity object, and the test environment further includes at least one of a road surface and a lane. The processing unit is configured to determine a distribution of crosstalk points in the first point cloud data according to the first point cloud data and a ground truth point cloud map of the test environment, the ground truth point cloud map indicating a contour boundary of the detection target in the test environment, and the distribution of the crosstalk points including at least one of a distance between the crosstalk points and the contour boundary of the high-reflectivity object, a distance between the crosstalk points and the road surface, and a lane in which the crosstalk points are located; and determine a risk level of high-reflectivity point cloud expansion noise of the to-be-tested laser radar according to the distribution of the crosstalk points.

[0023] In combination with the second aspect, in some implementations of the second aspect, the risk level can include a first risk level and a second risk level, and a risk degree of the first risk level is lower than that of the second risk level. The to-be-tested laser radar can be in a first lane when detecting the test environment. The processing unit is configured to: determine that the risk level of the high-reflectivity point cloud expansion noise is the first risk level when the distance between the crosstalk points and the road surface is greater than a first threshold; or determine that the risk level of the high-reflectivity point cloud expansion noise is the first risk level when the crosstalk points are in a lane other than the first lane; or determine that the risk level of the high-reflectivity point cloud expansion noise is the first risk level when a number of the crosstalk points is less than or equal to a second threshold; or determine that the risk level of the high-reflectivity point cloud expansion noise is the second risk level when the distance between the crosstalk points and the road surface is less than or equal to the first threshold, the crosstalk points are in the first lane, and the number of the crosstalk points is greater than or equal to the second threshold.

[0024] In some implementations of the second aspect, the processing unit can be configured to: determine, according to the position information corresponding to the first point cloud data, a local ground truth point cloud map corresponding to the position information from the ground truth point cloud map; and determine the distribution of the crosstalk points according to the first point cloud data and the local ground truth point cloud map.

[0025] In some implementations of the second aspect, the processing unit can be further configured to: remove the first data points from the first point cloud data according to a white list. The white list can indicate a type of the first data points, and the type of the first data points can indicate a category of a detection target corresponding to the first data points. The category of the detection target corresponding to the first data points is different from the high reflectivity object.

[0026] In some implementations of the second aspect, the obtaining unit can be further configured to: obtain a plurality of point cloud data collected by the calibration lidar at a plurality of positions in the test environment. The processing unit can be further configured to: splice and superimpose the plurality of point cloud data to obtain second point cloud data according to pose information of the calibration lidar at the plurality of positions; and remove high reflectivity point cloud inflation noise from the second point cloud data according to the first artificial annotation information to obtain the ground truth point cloud map of the test environment.

[0027] In some implementations of the second aspect, the crosstalk points can include: data points in the first point cloud data that are different from corresponding data points in the ground truth point cloud map by more than or equal to a preset threshold and are within a preset range around the high reflectivity object.

[0028] In a third aspect, a test device is provided. The device includes a memory configured to store a computer program, and a processor configured to execute the computer program stored in the memory to cause the device to perform the method of the first aspect and any possible implementation thereof.

[0029] In a fourth aspect, a computer program product is provided. The computer program product includes computer program code, which, when executed on a computer, causes the computer to perform the method of the first aspect and any possible implementation thereof.

[0030] In a fifth aspect, a computer-readable storage medium is provided. The computer-readable medium stores a computer program, which, when executed on a computer, causes the computer to perform the method of the first aspect and any possible implementation thereof.

[0031] In a sixth aspect, a chip is provided. The chip includes a circuit configured to perform the method of the first aspect and any possible implementation thereof. BRIEF DESCRIPTION OF DRAWINGS

[0032] FIG. 1 is a schematic diagram of a test scene according to an embodiment of the present application;

[0033] FIG. 2 is a flowchart of a test method according to an embodiment of the present application;

[0034] FIG. 3 is a flowchart of a method of constructing a ground truth point cloud map according to an embodiment of the present application;

[0035] FIG. 4 shows a schematic diagram of point cloud data;

[0036] FIG. 5 shows a schematic diagram of a ground truth point cloud map;

[0037] FIG. 6 is a flowchart of a test method according to an embodiment of the present application;

[0038] FIG. 7 is a schematic diagram of a scene for determining an inflation point according to an embodiment of the present application;

[0039] FIG. 8 is a schematic diagram of a test device according to an embodiment of the present application;

[0040] FIG. 9 is a schematic diagram of another test device according to an embodiment of the present application. DETAILED DESCRIPTION

[0041] The technical solutions in the present application will be described below with reference to the accompanying drawings.

[0042] A laser radar can use electromagnetic waves of a certain waveband, such as electromagnetic waves of the ultraviolet waveband to the far infrared waveband, electromagnetic waves with a wavelength in the 250 nanometer (nm) to 11 micrometer (m) waveband, and by detecting the scattering light characteristics of a target object, the position, speed and other related information of the target object can be obtained. Compared with other types of radars (such as millimeter wave radars, etc.), the laser radar has higher measurement accuracy, finer time resolution and spatial resolution, and has broad application prospects in intelligent transportation, autonomous driving, atmospheric environment monitoring, geographic mapping, unmanned aerial vehicles and other fields. Correspondingly, the performance of intelligent transportation, autonomous driving, unmanned aerial vehicles and the like is affected and restricted by the performance of the laser radar used.

[0043] In an actual detection scene, when detecting an object with normal reflectivity / scattering rate, the echo signal of the light beam barycenter can generate an effective point; the echo signal formed by the light beam non-barycenter falling on the object is often difficult to be detected by the detector of the laser radar due to its excessively low energy.

[0044] However, in actual detection scenarios, there can be objects with high reflectivity (such as signs, road cones, road signs, etc.). Compared with objects with normal scattering, due to the strong reflection and scattering effect of objects with high reflectivity on the detection light, the echo signal formed by the objects with high reflectivity will have greater energy. Therefore, when detecting objects with high reflectivity, in addition to the echo signal of the beam center of gravity that can generate an effective point, the echo signal formed by the beam non-center of gravity falling on the object is easily detected by the detector due to its high energy, and a corresponding invalid point is generated. The point cloud region originally composed of effective points is enlarged due to the addition of invalid points. The existence of such invalid points is easy to cause the misjudgment of objects with high reflectivity by the automatic driving domain controller. In the detection result, the existence of invalid points causes the size of the object detected by the automatic driving domain controller to be greater than the actual size of the object, that is, the so-called high-reflection point cloud expansion phenomenon (which can also be referred to as high-reflection expansion or point cloud expansion). The noise caused by such invalid points can be referred to as high-reflection point cloud expansion noise (which can also be referred to as high-reflection expansion noise).

[0045] Before the laser radar is shipped, the point cloud performance of the laser radar in a generalization road (such as high-reflection point cloud expansion noise, spike point cloud expansion noise, and cluster noise) can be evaluated through a generalization road test. The high-reflection point cloud expansion noise is one of the important evaluation items of the point cloud performance of the laser radar.

[0046] However, the high-reflection point cloud expansion noise has the characteristics of non-fixed shape, non-fixed distribution, and non-fixed position. Taking a road sign with high reflectivity as an example, in the detection result, the high-reflection point cloud expansion noise can appear at any position around the point cloud data corresponding to the road sign, and due to the close distance between the high-reflection point cloud expansion noise and the actual effective point, it is difficult to accurately distinguish such noise from the effective point, which leads to the difficulty in effectively detecting the high-reflection point cloud expansion noise and accurately evaluating the high-reflection point cloud noise of the laser radar. In addition, in some use scenarios (such as a rainy environment, dirt on the laser radar window, etc.), the high-reflection point cloud expansion phenomenon will be more serious. Since the high-reflection point cloud expansion noise is easy to cause misjudgment of the detected target, the point cloud data collected by the laser radar will seriously interfere with the downstream processing device (such as a vehicle control unit using the point cloud data, an automatic driving domain controller, etc.), which endangers the safety of vehicle driving. For example, in an automatic driving scenario, the high-reflection point cloud expansion noise can force the vehicle to take unnecessary obstacle avoidance measures.

[0047] In view of this, the embodiments of the present application provide a test method and a test device, which can improve the evaluation accuracy of the high-reflection point cloud expansion noise of the laser radar and can know the influence degree of the high-reflection point cloud expansion noise of the to-be-tested laser radar on the control device downstream thereof.

[0048] Exemplarily, FIG. 1 is a schematic diagram of a test scene according to an embodiment of the present application.

[0049] As shown in FIG. 1, the system 100 comprises a detection device 110. The detection device 110 can be configured to detect the test environment 150 and collect detection data. For example, the detection device 110 can be a calibration lidar. The point cloud data collected by the calibration lidar can be used to construct a ground truth point cloud map of the test environment 150. For another example, the detection device 110 can be a to-be-tested lidar. The point cloud data collected by the to-be-tested lidar can be used to evaluate the high-reflective point cloud inflation noise performance of the to-be-tested lidar.

[0050] The test environment 150 can comprise detection targets, such as targets 151 to 15n (n is a positive integer), as shown in FIG. 1. The targets 151 to 15n can comprise high-reflectivity objects. In some embodiments, the number of high-reflectivity objects comprised in the test environment 150 is greater than or equal to a preset value. For example, the preset value can be 20, 22 or 30. For another example, the preset value can be set according to test requirements.

[0051] In some possible implementations, the system 100 can further comprise a test platform 120. The detection device 110 can be disposed on the test platform 120. The test platform 120 can be fixedly disposed relative to the ground or can be movable relative to the ground. For example, the test platform 120 can comprise a vehicle, and the detection device 110 can be disposed on the vehicle.

[0052] In some embodiments, the test platform 120 can obtain the detection data collected by the detection device 110. For example, the test platform 120 can be further configured to store and process the detection data, and / or send the detection data to the cloud.

[0053] In some embodiments, the test platform 120 can comprise a navigation unit 121. For example, the navigation unit 121 can comprise an inertial measurement unit (IMU), and / or a global positioning system (GPS), a Beidou positioning system or other positioning systems.

[0054] Exemplarily, according to the data collected by the navigation unit 121, the position information corresponding to the detection data can be determined. For example, according to the data collected by the navigation unit 121, the position where the test platform 120 is located can be determined. For another example, in combination with the position and posture of the detection device 110 at the position where the test platform 120 is located, the position and posture of the detection device 110 can be determined. For yet another example, the position information corresponding to the detection data can include the position where the detection device 110 and the test platform 120 are located when the test environment 150 is detected to obtain the detection data.

[0055] In an embodiment, the test platform 120 can drive along the test road of the test environment 150, so that the detection device 110 can detect the test environment at different positions. The obtained single-frame point cloud data can be associated with the position where the test platform 120 and the detection device 110 are located when the single-frame point cloud data is collected. For example, when constructing the ground truth point cloud map of the test environment, the multiple frames of point cloud data collected by the detection device 110 at different positions can be superimposed according to the position information associated with the single-frame point cloud data. For another example, the to-be-tested lidar can be arranged on a vehicle, and by controlling the vehicle to drive along the test road, the detection results of the to-be-tested lidar on the test environment at different positions of the test road can be obtained, and the detection results corresponding to each position can be evaluated individually or the multiple detection results can be evaluated comprehensively.

[0056] Exemplarily, FIG. 2 is a flowchart of a test method provided by an embodiment of the present application. The method 200 can include the following steps:

[0057] S210, obtaining first point cloud data collected by a to-be-tested lidar when detecting a test environment.

[0058] The detection target included in the test environment contains a high-reflectivity object (which can also be referred to as a high-reflective object for short). For example, a road sign, a cone barrel, a construction sign, etc. The test environment can also include at least one of a road surface and a lane.

[0059] In an embodiment, a test road can be arranged in the test environment. For example, when detecting the test environment, the test platform 120 provided with the to-be-tested lidar can drive along the test road to detect the test environment at different positions. In this way, the real use scenario of the to-be-tested lidar can be simulated. For another example, the test road can include a road surface and can be provided with one or more lanes. The high-reflectivity object can be fixedly arranged in the test road (for example, a cone barrel, a construction sign, etc. arranged on a road surface), or can be distributed outside the test road, or can occupy the space above the road (for example, a road sign hung). The specific distribution form of the high-reflective object in the test environment is not limited in the embodiments of the present application.

[0060] In some possible implementation manners, the number of the high reflectivity objects contained in the test environment is greater than or equal to a preset value. For example, the preset value can be 20, 23, 25, or 30, or can be another value.

[0061] S220, determining the distribution of the crosstalk points in the first point cloud data according to the first point cloud data and the ground truth point cloud map of the test environment.

[0062] The ground truth point cloud map can indicate the contour boundary of the test target in the test environment. For example, the boundary of the target 151-15n can be labeled in the ground truth point cloud map by automatic labeling or manual labeling. For another example, by determining the data points corresponding to a certain type of detection target in the ground truth point cloud map, the boundary of the space region corresponding to the data points can be taken as the contour boundary of the detection target or the detection target of this type.

[0063] The crosstalk points can include data points corresponding to high reflectivity point cloud expansion noise in the first point cloud data.

[0064] In some possible implementation manners, since the high reflectivity point cloud expansion phenomenon occurs around the effective points corresponding to the high reflectivity objects in the point cloud data, the crosstalk points can include data points in the first point cloud data that are greater than or equal to a preset threshold from the corresponding data points in the ground truth point cloud map and are within a preset range around the high reflectivity objects.

[0065] For example, the difference between the data points in the first point cloud data and the corresponding data points in the ground truth point cloud map can be greater than or equal to a preset threshold, which can include that the distance between the data points in the first point cloud data and the corresponding data points in the ground truth point cloud map is greater than or equal to a distance threshold.

[0066] In an embodiment, the distance threshold can be a preset value, such as 3 cm, 5 cm, or the like.

[0067] In another embodiment, the detection light beams generated by the laser radar can be scanned by the scanning device to scan the test environment; the distance threshold can be associated with the distance between the data point and the radar. For example, the distance threshold can be determined according to the distance between the data point and the radar and the scanning resolution of the scanning device.

[0068] Exemplarily, the preset range around the high reflectivity object can include a range of a space region with a distance to a boundary contour of the high reflectivity object less than a preset value. For example, assuming that the preset value is 10 cm, a range of a space region outside the high reflectivity object and with a distance to the contour boundary of the high reflectivity object less than or equal to 10 cm can be determined as the preset range around the high reflectivity object, and the crosstalk points can be screened in the range. The preset value can also be other numerical values. For example, 15 cm; for another example, the preset value can be related to at least one of a shape, a size, and a distribution position of the high reflectivity object.

[0069] In the embodiment of the application, the data points in the point cloud data collected by the to-be-tested laser radar are measurement values of the detection results in the test; correspondingly, the data points in the ground truth point cloud map can be regarded as the ground truth of the detection results in the test. By matching with the data points in the ground truth point cloud map, the data points in the first point cloud data with a difference between the corresponding data points in the ground truth point cloud map less than a preset threshold can be regarded as valid points in the first point cloud data; the data points in the first point cloud data with a difference between the corresponding data points in the ground truth point cloud map greater than or equal to the preset threshold and in the preset range around the high reflectivity object can be regarded as invalid points corresponding to the high reflectivity point cloud expansion noise. In this way, even if the high reflectivity point cloud expansion noise has the characteristics of a shape not fixed, a distribution not fixed, and a position not fixed, the invalid points corresponding to the high reflectivity point cloud expansion noise can still be effectively screened from the first point cloud data, and the high reflectivity point cloud expansion noise performance of the to-be-tested laser radar can be evaluated.

[0070] Exemplarily, the distribution of the crosstalk points can include at least one of the following: a distance between the crosstalk points and a contour boundary of the high reflectivity object, a distance between the crosstalk points and a road surface, and a lane in which the crosstalk points are located.

[0071] S230, determining a risk level of the high reflectivity point cloud expansion noise of the to-be-tested laser radar according to the distribution of the crosstalk points.

[0072] Exemplarily, the risk level of the high reflectivity point cloud expansion noise can represent a degree of credibility of the point cloud data collected by the to-be-tested laser radar.

[0073] The laser radar is an important sensor of a vehicle perception system. The control device downstream of the laser radar includes a control device directly or indirectly using the point cloud data collected by the laser radar, such as an automatic driving domain controller, a vehicle control unit, etc. using the point cloud data. The control of the vehicle by these control devices will be affected by the reliability of the point cloud data of the laser radar. For example, for the control device using the point cloud data of the laser radar, even if the function of the control device is normal, the higher the risk level of the high reflection point cloud expansion noise performance of the laser radar, the greater the possibility of misjudgment of the control device, the greater and more frequent the possibility of unnecessary obstacle avoidance operation of the vehicle controlled by the control device. That is, the risk level of the high reflection point cloud expansion noise of the laser radar can represent the degree of harm of the point cloud data of the laser radar to the operation of the control device downstream of the laser radar.

[0074] In the embodiments of the present application, according to the ground truth point cloud map of the test environment, the crosstalk points corresponding to the high reflection point cloud expansion noise can be effectively screened out from the point cloud data collected by the to-be-tested laser radar, and the evaluation accuracy of the high reflection point cloud expansion noise performance of the laser radar can be improved. According to the distribution of the crosstalk points, the risk level of the high reflection point cloud expansion noise of the to-be-tested laser radar can be determined, and the influence of the high reflection point cloud expansion noise performance of the laser radar on the operation of the control device downstream of the laser radar, the operation of the automatic driving system using the point cloud data of the laser radar, and the driving process of the vehicle can be determined.

[0075] In some possible implementations, the risk level can include a first risk level and a second risk level, and the risk level of the first risk level is lower than that of the second risk level. For example, when the risk level is divided into two risk levels, the first risk level can be referred to as a low risk level, and the second risk level can also be referred to as a high risk level. For another example, when the risk level is divided into three or more levels, the second risk level can be any risk level, and the first risk level can be any risk level with a risk level lower than that of the second risk level.

[0076] Exemplarily, the to-be-tested laser radar is in the first lane when detecting the test environment. According to the distribution of the crosstalk points, determining the risk level of the high reflection point cloud expansion noise of the to-be-tested laser radar can include: when the distance between the crosstalk points and the road surface is greater than a first threshold, determining that the risk level of the high reflection point cloud expansion noise is the first risk level; or when the crosstalk points are in a lane other than the first lane, determining that the risk level of the high reflection point cloud expansion noise is the first risk level; or when the number of crosstalk points is less than or equal to a second threshold, determining that the risk level of the high reflection point cloud expansion noise is the first risk level; or when the distance between the crosstalk points and the road surface is less than or equal to the first threshold, the crosstalk points are in the first lane, and the number of crosstalk points is greater than or equal to the second threshold, determining that the risk level of the high reflection point cloud expansion noise is the second risk level.

[0077] In one embodiment, the first threshold value can be 1.8 meters, 2 meters, or other values. For example, the first threshold value can be set according to the height of the vehicle on which the laser radar is installed.

[0078] In the embodiments of the present application, when the height between the crosstalk point and the ground is greater than the first threshold value, it can be considered that the vehicle can pass through the space between the crosstalk point and the ground relatively safely. In this scenario, even if the high-reflective point cloud expansion noise causes false detection of the detected target, the control device downstream of the laser radar will not control the vehicle to perform unnecessary obstacle avoidance operation, and it can be considered that the high-reflective point cloud expansion noise has a low risk level in this scenario.

[0079] In yet another embodiment, the second threshold value can be 5, 7, 8, or other values. For example, the second threshold value can adopt different values in the case that the first point cloud data is single-frame point cloud data and multi-frame point cloud data, respectively. For another example, when the first point cloud data is multi-frame point cloud data, the level of the high-reflective point cloud expansion noise can be determined to be the first risk level or the second risk level according to the total number of crosstalk points in the multi-frame point cloud data and the second threshold value.

[0080] In some possible implementations, determining the distribution of the crosstalk points in the first point cloud data according to the first point cloud data and the ground truth point cloud map of the test environment can include: determining a local ground truth point cloud map corresponding to the position information of the first point cloud data from the ground truth point cloud map; and determining the distribution of the crosstalk points according to the first point cloud data and the local ground truth point cloud map.

[0081] For example, the ground truth point cloud map of the test environment can include point cloud data in a relatively large area range (such as a first area range), and the single-frame point cloud data collected by the to-be-tested laser radar can correspond to a smaller area range (such as a second area range). In order to improve the test efficiency, the ground truth point cloud map of the smaller area range (i.e., the local ground truth point cloud map) can be determined from the ground truth point cloud map of the test environment. Further, the ground truth point cloud map of the smaller area range can be compared with the single-frame point cloud data collected by the to-be-tested laser.

[0082] In one embodiment, when the first point cloud data is single-frame point cloud data, a local ground truth point cloud map can be determined from the ground truth point cloud map according to the position information of the single-frame point cloud data.

[0083] In yet another embodiment, when the first point cloud data is multi-frame point cloud data, one or more local ground truth point cloud maps can be determined from the ground truth point cloud map according to the multiple position information corresponding to the multi-frame point cloud data.

[0084] In actual testing, the range of the test environment may be much larger than the detection range of the lidar, and the range corresponding to the ground truth point cloud map may be much larger than the detection range of the lidar. In the embodiment of the present application, by determining the local ground truth point cloud map, the number of data points required for processing can be reduced, and the testing efficiency can be improved.

[0085] In some possible implementation manners, the method can further include: removing the first data point from the first point cloud data according to a white list. The white list can indicate the type of the first data point; the type of the first data point can indicate the category of the detection target corresponding to the first data point. The category of the detection target corresponding to the first data point is different from the high-reflectivity object.

[0086] Exemplarily, for any data point, the type of the data point can indicate the type of the detection target corresponding to the data point. For example, when the types of the data points are 1 and 2 respectively, the detection targets corresponding to the data points can be vegetation and vulnerable road users respectively.

[0087] In addition to the high-reflectivity object, the test scene often also includes other categories of detection targets, such as vegetation, vulnerable road users, walls, and the like. In the embodiment of the present application, based on the type of the data point in the first point cloud data, in combination with the white list, the data point corresponding to the detection target irrelevant to the high-reflectivity object in the first point cloud data can be screened out. In this way, on the one hand, the data point irrelevant to the high-reflective point cloud inflation noise can be excluded from interfering with the test result; on the other hand, the number of data points required for processing can be reduced, and the testing efficiency can be improved.

[0088] In some possible implementation manners, the method can further include: obtaining a plurality of point cloud data collected by the calibration lidar in a plurality of positions for detecting the test environment; splicing and superimposing the plurality of point cloud data to obtain second point cloud data according to the pose information of the calibration lidar in the plurality of positions; and removing the high-reflective point cloud inflation noise from the second point cloud data according to the first artificial annotation information to obtain the ground truth point cloud map of the test environment.

[0089] Exemplarily, the calibration lidar and the to-be-tested lidar can be radars of the same type / model, or can also be radars of different types / models.

[0090] In an actual scene, the point cloud data collected by the calibration lidar can also have high-reflective point cloud inflation noise. By removing the high-reflective point cloud inflation noise through artificial annotation, the interference of the invalid points corresponding to the high-reflective point cloud inflation noise on the ground truth point cloud map can be avoided, and the accuracy of the test result can be improved.

[0091] The method 200 will be introduced and described below in combination with FIGS. 3 to 7.

[0092] Exemplarily, FIG. 3 is a flowchart of a method for constructing a ground truth point cloud map according to an embodiment of the present application. The method 300 can be understood as an extension or variation of the method 200. The method 300 can include the following steps:

[0093] S310, obtaining detection data collected by the calibration laser radar when detecting the test environment.

[0094] Exemplarily, the detection device (such as the calibration laser radar) can be arranged on the vehicle, such as the front bumper of the vehicle. The vehicle can be provided with a navigation device, and the position information corresponding to the detection data can be obtained through the navigation device. For example, by controlling the vehicle to drive along the test road, the test environment can be detected when the vehicle is at different positions, and the obtained multiple frames of detection data can be superimposed, spliced and fused to form the detection result of the calibration laser radar.

[0095] In one embodiment, the detection device can periodically or non-periodically detect the test environment. For example, the detection device can detect the test environment at time 1, time 2, etc., to obtain detection data 1, detection data 2, etc. Correspondingly, the positions of the vehicle at the time 1, time 2, etc. can be recorded as position 1, position 2, etc. In this scenario, the detection data 1 has a corresponding relationship with the time 1 and the position 1, and the detection data 2 has a corresponding relationship with the time 2 and the position 2. For another example, the range of the test environment can exceed the detection range of the detection device, and there can be occlusion relationships between different targets in the test environment. The single frame of point cloud data collected by the detection device can only include the detection data of part of the test environment. In order to improve the accuracy of the ground truth point cloud map of the test environment, the detection data 1, detection data 2, etc. can be spliced, superimposed and fused in combination with the corresponding position information.

[0096] S320, determining the type of the data points in the detection data.

[0097] Exemplarily, the type of the data points in the detection data can be determined according to a point cloud segmentation model. The type of any data point can represent the category of the detection target corresponding to the data point. For example, the point cloud segmentation model can be trained by deep learning. For another example, the detection targets can be divided into categories such as road, wall, vulnerable road user (VRU), temporary obstacle, vehicle, vegetation, dust, leaf, high reflectivity object, etc. Correspondingly, the types of the data points in the detection data can be determined as road, VRU, vehicle, tree, high reflectivity object, etc. through the point cloud segmentation model.

[0098] In one embodiment, FIG. 4 shows a schematic diagram of point cloud data. As shown in FIG. 4, different colors of data points can represent different types of data points. For example, in FIG. 4, the type of green data points can be road signs; the type of red data points can be road surfaces; the type of yellow data points can be guardrails; and the type of orange data points can be walls.

[0099] S330, splicing and superimposing the multiple frames of detection data to obtain global point cloud data of the test environment.

[0100] Exemplarily, in an actual scene, the range of the test environment can exceed the detection range of the detection device, and different targets in the test environment can have occlusion relationships. The single frame of point cloud data collected by the detection device can only include detection data of part of the test environment. By splicing and superimposing the multiple frames of detection data, global point cloud data of the test environment can be obtained.

[0101] In some possible implementation manners, step S320 can be performed first, that is, the types of data points in each frame of detection data are determined first, and then the multiple frames of detection data are spliced and superimposed; or step S330 can be performed first, that is, the global point cloud data of the test environment is obtained first, and then the types of data points are determined.

[0102] In the embodiments of the present application, by determining the types of data points in the detection data and splicing and superimposing the multiple frames of detection data, global point cloud data with type information of each data point can be obtained.

[0103] S340, obtaining artificial annotation information.

[0104] Since the high-reflectivity object is arranged in the test environment, high-reflectivity point cloud inflation noise can exist in the detection data obtained by detecting the test environment by the calibration laser radar. By using the artificial annotation method, valid points corresponding to the high-reflectivity object and / or crosstalk points in the periphery of the data points of the high-reflectivity object in the detection data of the test environment can be annotated. That is, according to the artificial annotation information, the high-reflectivity point cloud inflation noise in the global point cloud data of the test environment can be determined.

[0105] S350, determining a ground truth point cloud map of the test environment according to the global point cloud data of the test environment and the artificial annotation information.

[0106] Exemplarily, according to the artificial annotation information, the high-reflectivity point cloud inflation noise in the global point cloud data can be removed, and a ground truth point cloud map of the test environment can be obtained. For example, as shown in FIG. 5, FIG. 5 shows a ground truth point cloud map of the test environment. In FIG. 5, different types of data points can be represented by different colors of data points.

[0107] Exemplarily, FIG. 6 is a flowchart of another test method provided by the embodiments of the present application. The method 400 can be understood as an extension or variation of the method 200. The method 400 can include the following steps:

[0108] S410, obtaining detection data of a test environment collected by the to-be-tested laser radar.

[0109] In one embodiment, the to-be-tested laser radar can be arranged on a vehicle. During driving of the vehicle along a test road, the to-be-tested laser radar can periodically or non-periodically detect the test environment. The detection data of the test environment collected by the to-be-tested laser radar can also be referred to as to-be-tested point cloud data.

[0110] Exemplarily, according to the detection data of the test environment collected by the to-be-tested laser radar, in combination with the ground truth point cloud map of the test environment, a difference point cloud between the two can be determined. When there is high reflection point cloud inflation noise in the point cloud data collected by the to-be-tested laser radar, the difference point cloud will include data points related to the high reflection point cloud inflation noise.

[0111] S420, determining the category of each data point in the detection data.

[0112] Exemplarily, the category of each data point in the detection data can be determined according to a point cloud segmentation model.

[0113] Since there can be multiple categories of detection targets in the test environment, a high reflectivity object can only correspond to one of the categories. In order to improve test efficiency, one or more types of data points irrelevant to the high reflectivity object can be indicated by a white list. For example, the categories of data points indicated by the white list can include vehicles, vegetation, dust, VIUs, walls, etc.

[0114] S430, filtering the types of data points in the detection data according to the white list to obtain a point cloud C1.

[0115] Exemplarily, according to the white list, data points of corresponding categories can be filtered out from the detection data of the to-be-tested laser radar to obtain filtered point cloud data (which can be denoted as point cloud C1). For example, data points of the type of vegetation can be in the white list. In this case, the point cloud C1 does not include data points of the type of vegetation. For another example, data points of the type of road can be outside the white list. In this case, the point cloud C1 can include data points of the type of road.

[0116] In the embodiments of the present application, the data points in the to-be-tested point cloud data are filtered according to the white list, which can reduce the number of data points involved in the matching process when matching with the ground truth point cloud map, and can improve the test efficiency.

[0117] S440, determine a local ground truth point cloud map C0 of the test environment according to position information corresponding to the detection data.

[0118] In some possible implementations, the range of the area involved in the ground truth point cloud map of the test environment can be greater than the detection range of the to-be-tested lidar, and the ground truth point cloud map of the test environment will involve data points outside the detection range. In order to improve the test efficiency, a local ground truth point cloud map C0 (also denoted as point cloud C0 and ground truth point cloud C0) of a preset area can be determined from the ground truth point cloud map of the test environment based on position information corresponding to the detection data.

[0119] In one embodiment, it is assumed that the detection data 11 is obtained by the to-be-tested lidar at time 11, and the vehicle provided with the to-be-tested lidar is at position 11 at time 11. For example, a local ground truth point cloud map of an area with a periphery of 100 m*100 m centered at the position 11 can be selected from the ground truth point cloud map of the test environment. In this scenario, the preset area can be an area with a periphery of 100 m*100 m centered at the position. The preset area can also be set based on other manners, such as a circular area centered at the position, and the like, which are not limited in the embodiments of the present application.

[0120] S450, compare the point cloud C1 and the point cloud C0 to determine a difference point cloud C2.

[0121] By comparing the point cloud C1 and the local ground truth point cloud map C0, data points in the point cloud C1 that are different from the ground truth point cloud map can be determined, and this part of data points can be referred to as crosstalk points. The point cloud data formed by the crosstalk points can be used as a difference point cloud (denoted as point cloud C2).

[0122] Since the ground truth point cloud map of the test environment does not involve high-reflective point cloud inflation noise, by comparing the point cloud C1 and the point cloud C0, data points in the to-be-tested point cloud data that are related to the high-reflective point cloud inflation noise can be determined. That is, the difference point cloud C2 can include data points in the to-be-tested point cloud data that are related to the high-reflective point cloud inflation noise.

[0123] Exemplarily, a search tree is constructed based on the point cloud C0, and a data point P1 in the point cloud C1 is used as a starting point to search for a data point P2 closest to the data point in the search tree C0. When the distance between P1 and P2 is greater than or equal to a preset value, it can be considered that the point P1 is a difference point. By screening the difference points in the point cloud C1, the difference point cloud C2 can be determined.

[0124] In one embodiment, P1 and P2 satisfy the formula d > A*B, P1 can be considered as a difference point. Wherein, d represents the distance between P1 and P2, B represents the distance between P1 and the radar, and A represents a conversion coefficient. A can be a preset value (such as 0.01, 0.02, or other values), or can be determined according to the lateral resolution of the radar. For example, A can satisfy the formula A = C*D, where C is a coefficient (such as 1, 2, or other values), and D is the lateral angular resolution of the radar. When the radar uses a scanning device to scan the detection beam, D can also represent the lateral scanning resolution of the scanning device.

[0125] S460, according to the positional relationship between the data points in the difference point cloud C2 and the high reflectivity object, the inflation points are determined.

[0126] For example, the ground truth point cloud map can indicate the position and shape of the detected target in the test environment. According to the local ground truth map, the distance between each data point in the difference point cloud C2 and the corresponding high reflectivity object can be determined. The data points in the point cloud C2 within the preset range around the high reflectivity object are determined as inflation points. The inflation points can correspond to the crosstalk points in method 200.

[0127] In some possible implementations, according to the ground point cloud in the point cloud C1 (denoted as point cloud C3), the distance between the data points in the point cloud C2 and the ground can be determined.

[0128] For example, FIG. 7 is a schematic diagram of a scene for determining inflation points provided by an embodiment of the present application.

[0129] As shown in FIG. 7, taking a road sign with high reflectivity as an example, the range for screening inflation points is within a preset range around the road sign. For example, as shown in (a) of FIG. 7, data points #1-6 are within the range, and data points #1-6 can be considered as inflation points. Data points #7-8 are outside the range, and data points #7-8 can be considered as not belonging to the inflation points.

[0130] In one embodiment, the setting of the range around the road sign for screening inflation points can adopt the manner shown in (b) and (c) of FIG. 7. Characteristic values a-e can represent the positional relationship between the range of the inflation points and the road sign. For example, the values of a, b, c, d, and e can be 1 meter, 1 meter, 0.5 meter, 2.5 meters, and 1 meter respectively. In some possible implementations, the range for screening the path point can be set according to actual needs, and the values of a, b, c, d, and e can be set according to actual needs.

[0131] S470, statistics are performed on multiple frames of detection data to evaluate the risk level of the high reflectivity point cloud inflation noise of the to-be-tested laser radar.

[0132] Exemplarily, during the vehicle provided with the to-be-tested laser radar drives along the test road, the to-be-tested laser radar can periodically or non-periodically detect the test environment, and a plurality of frames of detection data can be acquired. For example, the blooming points in the single frame of detection data can be determined according to steps S420 to S460. The distribution of the blooming points involved in the plurality of frames of detection data can be counted.

[0133] Exemplarily, Table 1 shows the relationship between the distribution of the blooming points and the level of the high-reflection-point-cloud blooming noise.

[0134] Table 1

[0135] In one embodiment, in the longitudinal distribution, when the height of the blooming point from the ground is less than 1.8 meters, the blooming point is in the current lane of the vehicle, the number of blooming points is greater than 2, and the distance between different blooming points is less than 0.2 meters, it can be considered that the high-reflection-point-cloud blooming noise of the to-be-tested laser radar is at a high-risk level.

[0136] In another embodiment, in the longitudinal distribution, when the height of the blooming point from the ground is greater than 1.8 meters, or when the blooming point is in a lane other than the current lane of the vehicle, it can be considered that the high-reflection-point-cloud blooming noise of the to-be-tested laser radar is at a low-risk level.

[0137] In another embodiment, in the lateral distribution, when the blooming point is in a lane other than the current lane of the vehicle, or when the blooming point invades the current lane of the vehicle but the invasion distance is less than an invasion threshold, it can be considered that the high-reflection-point-cloud blooming noise of the to-be-tested laser radar is at a low-risk level. The invasion threshold can be a preset value, such as 3 cm, 5 cm, etc.; or other numerical values, such as a value related to the lateral resolution of the radar. For example, the invasion threshold can satisfy the formula E=F*G*H, where E represents the invasion threshold, F represents a coefficient (such as 1, 2, etc.), G represents the distance between the high-reflectivity object and the radar, and H represents the lateral angular resolution of the radar.

[0138] The above describes the method embodiments provided by the embodiments of the present application in combination with FIGS. 1 to 7. The following describes the device embodiments provided by the embodiments of the present application in combination with FIGS. 8 and 9. The description of the device embodiments corresponds to the description of the method embodiments, and therefore, the content not described in detail can be referred to the above method embodiments.

[0139] Exemplarily, FIG. 8 shows a schematic block diagram of a test device (hereinafter referred to as device 2000) provided by an embodiment of the present application. The device 2000 can include units for performing the method in FIG. 2, and each unit in the device 2000 can be used to perform the corresponding flow in the method embodiments of FIG. 2 described above.

[0140] Exemplarily, the apparatus 2000 can include an obtaining unit 2010 and a processing unit 2020. When the apparatus 2000 is used to execute the method 200 in FIG. 2, the obtaining unit 2010 can be configured to execute step S210 in the method 200, and the processing unit 2020 can be configured to execute steps S220 and S230 in the method 200.

[0141] Specifically, the obtaining unit 2010 can be configured to obtain first point cloud data collected by a to-be-tested laser radar when detecting a test environment. The processing unit 2020 can be configured to determine a distribution of crosstalk points in the first point cloud data according to the first point cloud data and a ground truth point cloud map of the test environment, and determine a risk level of high-reflection point cloud blooming noise of the to-be-tested laser radar according to the distribution of the crosstalk points.

[0142] In some possible implementation manners, the risk level can include a first risk level and a second risk level, and the risk degree of the first risk level is lower than that of the second risk level. The to-be-tested laser radar can be in a first lane when detecting the test environment. The processing unit 2020 can be configured to determine the risk level of the high-reflection point cloud blooming noise as the first risk level or the second risk level according to a number of the crosstalk points, a distance between the crosstalk points and a road surface, and whether the crosstalk points are in the first lane.

[0143] In some possible implementation manners, the processing unit 2020 can be configured to determine a local ground truth point cloud map corresponding to position information of the first point cloud data from the ground truth point cloud map according to the position information, and determine the distribution of the crosstalk points according to the first point cloud data and the local ground truth point cloud map.

[0144] In some possible implementation manners, the processing unit 2020 can be further configured to exclude the first data points from the first point cloud data according to the white list.

[0145] In some possible implementation manners, the processing unit 2020 can be further configured to obtain a plurality of point cloud data collected by a calibration laser radar when detecting the test environment at a plurality of positions, splice and superimpose the plurality of point cloud data to obtain second point cloud data according to pose information of the calibration laser radar at the plurality of positions, and exclude the high-reflection point cloud blooming noise from the second point cloud data according to the first artificial annotation information to obtain the ground truth point cloud map of the test environment.

[0146] It should be understood that the division of units in the above apparatus is only a logical division of functions, and all or part of the units can be integrated into one physical entity, or can be physically separated. All units of the above apparatus can be implemented in the form of calling software by a processor, or in the form of hardware circuit, or partially in the form of calling software by a processor and partially in the form of hardware circuit.

[0147] In the implementation process, the acquisition unit 2010 can be implemented by at least one processor or processor-related circuit, and the processing unit 2020 can be implemented by at least one transceiver or transceiver-related circuit. In an example, the one or more processors can acquire the first point cloud data. In an example, the one or more processors can determine the distribution of the crosstalk points in the first point cloud data according to the first point cloud data and the ground truth point cloud map of the test environment. In an example, the one or more processors can determine the risk level of the high-reflective point cloud expansion noise of the to-be-tested lidar according to the distribution of the crosstalk points. Illustratively, in the implementation process, the device 2000 can be a test platform for testing the to-be-tested lidar, or a chip or processor disposed in the test platform.

[0148] Illustratively, FIG. 9 is a schematic block diagram of another control device 3000 (hereinafter referred to as device 3000) provided by an embodiment of the present application. The device 3000 can include a processor 3010, an interface circuit 3020, and a memory 3030. The processor 3010, the interface circuit 3020, and the memory 3030 are connected through an internal connection path. The memory 3030 is configured to store instructions, and the processor 3010 is configured to execute the instructions stored in the memory 3030 to receive / send part of parameters through the interface circuit 3020. Optionally, the memory 3030 can be coupled to the processor 3010 through an interface, or the memory 3030 and the processor 3010 can be integrated together.

[0149] It should be noted that the interface circuit 3020 can include, but is not limited to, a transceiving device such as an input / output interface to realize the communication between the device 3000 and other devices or communication networks. For example, the first point cloud data or the ground truth point cloud map of the test environment can be acquired through the interface circuit 3020.

[0150] In embodiments of the present application, the processor is a circuit with signal processing capability. In one implementation, the processor can be a circuit with instruction reading and running capability, such as a central processing unit (CPU), a microprocessor, a graphics processing unit (GPU) (which can be understood as a microprocessor), a digital signal processor (DSP), etc. In another implementation, the processor can implement certain functions through a logical relationship of hardware circuit, which is fixed or reconfigurable. For example, the processor is a hardware circuit implemented by an application-specific integrated circuit (ASIC) or a programmable logic device (PLD), such as a field programmable gate array (FPGA). In a reconfigurable hardware circuit, the processor loads a configuration document to implement hardware circuit configuration. It can be understood that the processor loads instructions to implement the functions of the above part or all units. In addition, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as a neural network processing unit (NPU), a tensor processing unit (TPU), a deep learning processing unit (DPU), etc.

[0151] Embodiments of the present application also provide a computer program product, which includes computer program code. When the computer program code is executed on a computer, the computer executes any of the method embodiments of FIGS. 2-7 and any possible implementation thereof.

[0152] Embodiments of the present application also provide a computer readable storage medium, which stores program code or instructions. When the computer program code or instructions are executed by a processor of a computer, the processor implements any of the method embodiments of FIGS. 2-7 and any possible implementation thereof.

[0153] Embodiments of the present application also provide a chip, which includes a processing circuit. The processing circuit can be used to run a computer program, so that the chip executes any of the method embodiments of FIGS. 2-7 and any possible implementation thereof.

[0154] Exemplarily, the intelligent driving device can be a vehicle. The vehicle involved in the embodiments of the present application is a vehicle in a broad sense, which can be a traffic tool (such as a commercial vehicle, a passenger vehicle, a motorcycle, a flying vehicle, a train, etc.), an industrial vehicle (such as a forklift, a trailer, a tractor, etc.), an engineering vehicle (such as an excavator, a bulldozer, a crane, etc.), an agricultural device (such as a mower, a harvester, etc.), a recreational device, a toy vehicle, etc. The embodiments of the present application do not specifically limit the type of the vehicle. For example, the vehicle in the present application can include a pure electric vehicle (pure EV / battery EV), a hybrid electric vehicle (HEV), a range extended electric vehicle (REEV), a plug-in hybrid electric vehicle (PHEV), or a new energy vehicle (NEV), etc.

[0155] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0156] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0157] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0158] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0159] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present alone, or two or more units can be integrated in one unit.

[0160] If the functions are realized in the form of software functional units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various program code storage media.

[0161] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of testing a lidar, characterized by, The method comprises: obtaining first point cloud data collected by a laser radar detection test environment, the test environment comprising a detection target comprising a high reflectivity object, and the test environment further comprising at least one of a road surface and a lane; determining a distribution of crosstalk points in the first point cloud data according to the first point cloud data and a ground truth point cloud map of the test environment, the ground truth point cloud map indicating a contour boundary of the detection target in the test environment, and the distribution of crosstalk points comprising at least one of a distance between the crosstalk points and a contour boundary of the high reflectivity object, a distance between the crosstalk points and the road surface, and a lane in which the crosstalk points are located; determining a risk level of high-reflectivity point cloud blooming noise of the laser radar to be tested according to the distribution of crosstalk points.

2. The method of claim 1, wherein, The risk level comprises a first risk level and a second risk level, the risk level of the first risk level being lower than the risk level of the second risk level, the laser radar to be tested being in a first lane when detecting the test environment, and the determination of the risk level of high-reflectivity point cloud blooming noise of the laser radar to be tested according to the distribution of crosstalk points comprising: when the distance between the crosstalk points and the road surface is greater than a first threshold, determining the risk level of high-reflectivity point cloud blooming noise as the first risk level; or, when the crosstalk points are located in a lane other than the first lane, determining the risk level of high-reflectivity point cloud blooming noise as the first risk level; or, when the number of crosstalk points is less than or equal to a second threshold, determining the risk level of high-reflectivity point cloud blooming noise as the first risk level; or, when the distance between the crosstalk points and the road surface is less than or equal to the first threshold, the crosstalk points are located in the first lane, and the number of crosstalk points is greater than or equal to the second threshold, determining the risk level of high-reflectivity point cloud blooming noise as the second risk level.

3. The method according to claim 1 or 2, characterized in that, The determination of the distribution of crosstalk points in the first point cloud data according to the first point cloud data and the ground truth point cloud map of the test environment comprises: determining a local ground truth point cloud map corresponding to position information of the first point cloud data from the ground truth point cloud map according to the position information; determining the distribution of crosstalk points according to the first point cloud data and the local ground truth point cloud map.

4. The method according to any one of claims 1 to 3, characterized in that, The method further comprises: eliminating a first data point from the first point cloud data according to a white list, the white list indicating a type of the first data point, the type of the first data point indicating a category of a detection target corresponding to the first data point, and the category of the detection target corresponding to the first data point being different from the high reflectivity object.

5. The method according to any one of claims 1 to 4, characterized in that, The method further comprises: obtaining a plurality of point cloud data collected by a calibration laser radar when detecting the test environment at a plurality of positions; splicing and superimposing the plurality of point cloud data to obtain second point cloud data according to pose information of the calibration laser radar at the plurality of positions; eliminating high-reflectivity point cloud blooming noise from the second point cloud data according to first artificial annotation information to obtain a ground truth point cloud map of the test environment.

6. The method according to any one of claims 1 to 5, characterized in that, The crosstalk points include: data points in the first point cloud data that are greater than or equal to a preset threshold and are within a preset range around the high-reflectivity object.

7. A test device characterized by, Comprise: An acquisition unit is configured to acquire first point cloud data collected by a to-be-tested laser radar in a detection test environment, the test environment including a detection target including a high-reflectivity object, and the test environment further including at least one of a road surface and a lane; A processing unit is configured to determine a distribution of crosstalk points in the first point cloud data according to the first point cloud data and a ground truth point cloud map of the test environment, the ground truth point cloud map indicating a contour boundary of the detection target in the test environment, and the distribution of the crosstalk points including at least one of a distance between the crosstalk points and the contour boundary of the high-reflectivity object, a distance between the crosstalk points and the road surface, and a lane in which the crosstalk points are located; and determine a risk level of high-reflectivity point cloud blooming noise of the to-be-tested laser radar according to the distribution of the crosstalk points.

8. The apparatus of claim 7, wherein, The risk level includes a first risk level and a second risk level, the risk level of the first risk level being lower than the risk level of the second risk level, and the to-be-tested laser radar being in a first lane when detecting the test environment; and the processing unit is configured to: determine that the risk level of the high-reflectivity point cloud blooming noise is the first risk level when the distance between the crosstalk points and the road surface is greater than a first threshold value; or, determine that the risk level of the high-reflectivity point cloud blooming noise is the first risk level when the crosstalk points are in a lane other than the first lane; or, determine that the risk level of the high-reflectivity point cloud blooming noise is the first risk level when a number of the crosstalk points is less than or equal to a second threshold value; or, determine that the risk level of the high-reflectivity point cloud blooming noise is the second risk level when the distance between the crosstalk points and the road surface is less than or equal to the first threshold value, the crosstalk points are in the first lane, and the number of the crosstalk points is greater than or equal to the second threshold value.

9. The apparatus of claim 7 or 8, wherein, The processing unit is configured to: determine a local ground truth point cloud map corresponding to position information of the first point cloud data from the ground truth point cloud map; determine the distribution of the crosstalk points according to the first point cloud data and the local ground truth point cloud map.

10. The apparatus of any one of claims 7 to 9, wherein, The processing unit is further configured to: remove a first data point from the first point cloud data according to a white list, the white list indicating a type of the first data point, the type of the first data point indicating a category of a detection target corresponding to the first data point, and the category of the detection target corresponding to the first data point being different from the high-reflectivity object.

11. The apparatus of any one of claims 7 to 10, wherein: the acquisition unit is further configured to acquire a plurality of point cloud data collected by a calibration laser radar in a plurality of positions when detecting the test environment. The processing unit is further configured to: splice and superimpose the plurality of point cloud data to obtain second point cloud data according to the pose information of the calibration lidar at the plurality of positions; and remove high-reflection point cloud inflation noise from the second point cloud data according to the first artificial annotation information to obtain a ground truth point cloud map of the test environment.

12. The apparatus of any one of claims 7-11, wherein, The crosstalk points include data points in the first point cloud data that are greater than or equal to a preset threshold in difference between the data points and corresponding data points in the ground truth point cloud map and are within a preset range around the high-reflection object.

13. A test device, characterized by The computer program product comprises a computer program code, and when the computer program code is executed on a computer, the method of any one of claims 1 to 6 is executed. The memory is configured to store a computer program. The processor is configured to execute the computer program stored in the memory, so that the apparatus executes the method of any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that, The computer program product comprises a computer program code, and when the computer program code is executed on a computer, the method of any one of claims 1 to 6 is executed.

15. A computer program product, characterised in that, ​

Images