Point cloud registration method, point cloud registration quality detection method, electronic equipment and medium

By conducting quality detection of point cloud data, using local map projection and ray tracing algorithms and other technologies, we can directly detect abnormal point cloud registration, solving the problem of insufficient point cloud map quality inspection capabilities in the existing technology, and achieving high efficiency and low Cost-based point cloud registration quality evaluation and adjustment.

CN119006536APending Publication Date: 2024-11-22SHENZHEN DEEPROUTE AI CO LTD
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
CN202411079686.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The prior art point cloud map quality inspection capabilities in point cloud registration and construction are insufficient, making it difficult to accurately evaluate the quality of point cloud registration, and additional sensors often require consistency verification, which increases costs and calculation overhead.

Method used

Provide a point cloud registration quality detection method. By performing quality detection of point cloud data in continuous and discontinuous times, using local map projection, ray tracing algorithm and other technologies, it can directly detect whether there are abnormalities in point cloud registration, and adjust the Registered point cloud map.

Benefits of technology

Accurate evaluation of point cloud registration quality is achieved, cost reduction, avoid the use of additional sensors, and improve the accuracy of point cloud maps.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006536A_ABST
    Figure CN119006536A_ABST
Patent Text Reader

Abstract

The invention discloses a point cloud registration method, a point cloud registration quality detection method, electronic equipment and a medium. The method comprises the following steps: performing quality detection on point cloud data in a first point cloud group under continuous time to obtain a first detection result; and / or, performing quality detection on the point cloud data in the first point cloud group under the discontinuous time to obtain a second detection result; and / or, performing quality detection on the point cloud data between the first point cloud group and the second point cloud group to obtain a third detection result; wherein a part of point cloud data in the second point cloud group is within a preset range of a part of point cloud data of the first point cloud group, and the first point cloud group and the second point cloud group are used for point cloud registration; and performing corresponding quality detection feedback according to the first detection result, the second detection result and / or the third detection result. Through the method, the point cloud registration quality can be evaluated more accurately.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of point cloud registration, and in particular to a point cloud registration method, a point cloud registration quality detection method, an electronic device, and a medium. Background Art

[0002] When building a map, a large number of point cloud data are usually registered through point cloud registration to build an accurate point cloud map. For example, in the simultaneous localization and mapping SLAM (Simultaneous Localization and Mapping) technology, due to the existence of state estimation errors, the pose estimated by the SLAM system and the point cloud map constructed may be significantly different from the actual environment. Therefore, related technologies use verification trajectories or maps to evaluate / determine whether there are any abnormalities in the positioning and mapping structure given by the current SLAM system.

[0003] The disadvantage is that the quality inspection of the trajectory is relatively simple. Generally, the current estimated trajectory can be judged to be smooth and whether there are any anomalies by calculating the vehicle's speed, acceleration, motion curvature, and travel slope. However, there are problems such as insufficient quality inspection capabilities for point cloud registration and the quality inspection of the constructed point cloud map. Summary of the invention

[0004] In order to solve the above problems, the present application provides a point cloud registration method, a point cloud registration quality detection method, an electronic device, and a medium, which can more accurately evaluate the quality of point cloud registration.

[0005] To solve the above technical problems, a technical solution adopted in the present application is: to provide a point cloud registration quality detection method, the method comprising: performing quality detection on the point cloud data in the first point cloud group at continuous time to obtain a first detection result; and / or, performing quality detection on the point cloud data in the first point cloud group at non-continuous time to obtain a second detection result; and / or, performing quality detection on the point cloud data between the first point cloud group and the second point cloud group to obtain a third detection result; wherein, part of the point cloud data in the second point cloud group is within a preset range of part of the point cloud data in the first point cloud group; the first point cloud group and the second point cloud group have been used for point cloud registration; and performing corresponding quality detection feedback according to the first detection result, the second detection result and / or the third detection result.

[0006] Among them, quality inspection is performed on point cloud data in continuous time in the first point cloud group to obtain a first inspection result, including: obtaining a current point cloud frame from the point cloud data, and using the current point cloud frame and several adjacent point cloud frames to form a local map; projecting the current point cloud frame onto a grid of the target plane to obtain a first grid map, and projecting the local map onto a grid of the target plane to obtain a second grid map; wherein the data stored in each grid is the average height of all points in the grid; and obtaining the first inspection result based on the difference between the first grid map and the second grid map.

[0007] Among them, the current point cloud frame is projected onto the grid of the target plane to obtain a first grid map, including: eliminating points in the current point cloud frame that are not within the height threshold range to obtain a first target point cloud frame; projecting the first target point cloud frame onto the grid of the target plane to obtain a first grid map; projecting the local map onto the grid of the target plane to obtain a second grid map, including: eliminating points in the local map that are not within the height threshold range to obtain the first target map; projecting the first target map onto the grid of the target plane to obtain a second grid map.

[0008] Among them, quality detection is performed on the point cloud data between the first point cloud group and the second point cloud group to obtain a third detection result, including: obtaining the nearest neighbor point cloud frame in the second point cloud group within a preset range of the current point cloud frame in the first point cloud group; determining a reference frame from the nearest neighbor point cloud frame; and performing quality detection based on the current point cloud frame and the reference frame to obtain a third detection result.

[0009] Among them, quality detection is performed on the current point cloud frame and the reference frame to obtain a third detection result, including: in response to the height difference between the current point cloud frame and the reference frame being less than the first height threshold, obtaining the angle between the ground point cloud and the ground normal vector in the current point cloud frame and the reference frame, and performing quality detection based on the angle between the ground point cloud and the ground normal vector to obtain the third detection result; in response to the height difference between the current point cloud frame and the reference frame being greater than the first height threshold and less than the second height threshold, using a ray tracing algorithm to perform quality detection on the current point cloud frame and the reference frame to obtain the third detection result; in response to the height difference between the current point cloud frame and the reference frame being greater than the second height threshold, skipping the quality detection of the current point cloud frame and the reference frame.

[0010] Among them, quality detection is performed based on the angle between the ground point cloud and the ground normal vector to obtain a third detection result, including: obtaining the intersection ratio between the ground point cloud in the current point cloud frame and the reference frame; in response to the intersection ratio being greater than the intersection ratio threshold and the angle between the ground normal vector being greater than the angle threshold, obtaining the actual intersection result of the ground point cloud in the current point cloud frame and the reference frame; and obtaining the third detection result based on the actual intersection result.

[0011] Among them, a ray tracing algorithm is used to perform quality detection on the current point cloud frame and the reference frame to obtain a third detection result, including: taking the one with higher height between the current point cloud frame and the reference frame as the second target point cloud frame, and taking the one with lower height between the current point cloud frame and the reference frame as the third target point cloud frame; traversing all data points of the third target point cloud frame, and determining abnormal points in combination with the ground point cloud of the second target point cloud frame; and obtaining the third detection result according to the number of abnormal points.

[0012] Among them, traversing all data points of the third target point cloud frame and determining abnormal points in combination with the ground point cloud of the second target point cloud frame include: performing the following steps for each data point in the third target point cloud frame in a traversal order: determining a line between the current data point and the radar center point corresponding to the third target point cloud frame; obtaining the intersection of the line and the ground plane corresponding to the second target point cloud frame; taking the intersection as the search position, performing a nearest neighbor search in the ground point cloud of the second target point cloud frame; and determining the current data point as an abnormal point in response to the number of searched ground points being greater than a quantity threshold.

[0013] The step of obtaining the third detection result according to the number of abnormal points includes: obtaining the third detection result by using the number of abnormal points and the proportion of the number of abnormal points in all data points.

[0014] Among them, quality detection is performed on point cloud data in the first point cloud group at non-continuous time to obtain a second detection result, including: obtaining the nearest neighbor point cloud frame in the first point cloud group within a preset range of the current point cloud frame in the first point cloud group; removing point cloud frames that are continuous with the current point cloud frame from the nearest neighbor point cloud frames, and using the remaining point cloud frames as reference frames; performing quality detection based on the current point cloud frame and the reference frame to obtain a second detection result.

[0015] To solve the above technical problems, another technical solution adopted in the present application is: to provide a point cloud registration method, the method comprising: using several point cloud groups to perform point cloud registration; using the method provided by the above technical solution to perform registration quality detection and obtain quality detection feedback; deleting several point cloud groups according to the quality detection feedback, and performing point cloud registration again.

[0016] To solve the above technical problems, another technical solution adopted in the present application is: to provide an electronic device, which includes a processor and a memory coupled to the processor; wherein the memory is used to store computer programs, and the processor is used to execute the computer programs to implement a method provided by any of the above technical solutions.

[0017] To solve the above technical problems, another technical solution adopted in the present application is: providing a computer-readable storage medium, which is used to store a computer program, and when the computer program is executed by a processor, it is used to implement the method provided by the above technical solution.

[0018] The beneficial effects of the embodiments of the present application are as follows: Different from the prior art, the point cloud registration method, point cloud registration quality detection method, electronic device, and medium provided by the present application perform quality detection on the point cloud data in the first point cloud group at continuous time after point cloud registration; and / or perform quality detection on the point cloud data in the first point cloud group at non-continuous time; and / or perform quality detection on the point cloud data between the first point cloud group and the second point cloud group, thereby achieving quality detection of point cloud registration within the registered point cloud group and between point cloud groups, at continuous time and non-continuous time, and being able to more accurately evaluate the quality of point cloud registration, and then being able to adjust the registered point cloud and point cloud map according to quality detection feedback. Furthermore, the present application directly detects whether there are abnormalities in the point cloud registration of the registered point cloud group, without the need to use additional sensors to perform consistency verification on the point cloud trajectory and the point cloud map, and is relatively low in cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0020] Figure 1 It is a flow chart of an embodiment of a point cloud registration quality detection method provided by the present application;

[0021] Figure 2 It is a flowchart of an embodiment of step 11 provided in this application;

[0022] Figure 3 It is a flowchart of an embodiment of step 13 provided by the present application;

[0023] Figure 4 This is a schematic diagram of an application scenario of an embodiment of step 13 provided in this application;

[0024] Figure 5 It is a flow chart of another embodiment of the point cloud registration quality detection method provided by the present application;

[0025] Figure 6 It is a flow chart of another embodiment of the point cloud registration quality detection method provided by the present application;

[0026] Figure 7 It is a flow chart of another embodiment of the point cloud registration quality detection method provided by the present application;

[0027] Figure 8 This is a schematic diagram of an application scenario of an embodiment of steps 71 to 74 provided in this application;

[0028] Fig. 9 It is a flowchart of an embodiment of step 12 provided by the present application;

[0029] Fig.10 This is a schematic diagram of an application scenario of an embodiment of steps 91 to 93 provided in this application;

[0030] Fig.11 It is a flow chart of another embodiment of the point cloud registration quality detection method provided by the present application;

[0031] Fig.12 It is a flow chart of another embodiment of the point cloud registration quality detection method provided by the present application;

[0032] Fig.13 It is a flow chart of another embodiment of the point cloud registration quality detection method provided by the present application;

[0033] Fig.14 It is a flow chart of an embodiment of a point cloud registration method provided by the present application;

[0034] Fig.15 It is a structural schematic diagram of an embodiment of an electronic device provided by the present application;

[0035] Fig.16 It is a structural schematic diagram of an embodiment of a computer-readable storage medium provided by the present application. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It will be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some but not all structures related to the present application are shown in the drawings. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0037] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0038] When building a map, a large number of point cloud data are usually registered through point cloud registration to build an accurate point cloud map. For example, in the simultaneous localization and mapping SLAM (Simultaneous Localization and Mapping) technology, due to the existence of state estimation errors, the pose estimated by the SLAM system and the point cloud map constructed may be significantly different from the actual environment. Therefore, related technologies use verification trajectories or maps to evaluate / determine whether there are any abnormalities in the positioning and mapping structure given by the current SLAM system.

[0039] The disadvantage is that the quality inspection of the trajectory is relatively simple. Generally, the current estimated trajectory can be judged to be smooth and whether there are any abnormalities by calculating the vehicle's speed, acceleration, motion curvature, and travel slope. However, there are problems such as insufficient quality inspection capabilities for point cloud registration and the constructed point cloud map. For example, the reliability of the current point cloud registration can be judged by calculating the covariance of the point cloud registration. However, in many scenarios, even if the covariance matrix of the point cloud registration is very small, the accuracy of the point cloud registration is difficult to guarantee. For another example, the consistency of the point cloud registration is checked based on multi-sensor fusion technology. That is, if there is a large difference in the measurement results between different sensors, it can be considered that the reliability of the point cloud registration is low. However, this solution requires the use of multiple sensors, which increases the hardware cost and computing overhead of the system. For another example, the method is to achieve indirect quality inspection of the point cloud registration and point cloud map by checking whether the trajectory is smooth. However, a smooth trajectory does not necessarily mean that the point cloud registration is accurate. The trajectory quality inspection cannot detect slowly drifting poses and maps, and the goal of the mapping algorithm is to output the most accurate pose of the point cloud registration. Another example is manual quality inspection, which involves manually checking the point cloud map to determine whether there are any registration anomalies, mapping anomalies, etc. However, this method is inefficient and difficult to detect potential registration anomalies.

[0040] Based on this, the present application proposes to perform quality inspection on the point cloud data in the first point cloud group at continuous time after point cloud registration; and / or perform quality inspection on the point cloud data in the first point cloud group at discontinuous time; and / or perform quality inspection on the point cloud data between the first point cloud group and the second point cloud group, thereby realizing quality inspection on point cloud registration within the registered point cloud group and between point cloud groups, at continuous time and discontinuous time, and being able to more accurately evaluate the quality of point cloud registration, and then being able to adjust the registered point cloud and point cloud map according to the quality inspection feedback. Furthermore, the present application directly detects whether there are abnormalities in the point cloud registration of the registered point cloud group, without the need to use additional sensors to perform consistency verification on the point cloud trajectory and the point cloud map, and is relatively low in cost. Please refer to any of the following embodiments for details.

[0041] In the present application, the first point cloud group and the second point cloud group used for quality inspection are both used at least for point cloud registration, that is, the quality inspection of the present application is performed after the first point cloud group and the second point cloud group are registered.

[0042] See also Figure 1 , Figure 1 : is a flow chart of an embodiment of a point cloud registration quality detection method provided by the present application. The method comprises:

[0043] Step 11: Perform quality inspection on the point cloud data in the first point cloud group at continuous time to obtain a first inspection result.

[0044] In some embodiments, the first point cloud group may be point cloud data collected by sensors on the vehicle over a continuous period of time. For example, point cloud data collected by the vehicle within 10 minutes. It is understood that these point cloud data also correspond to the driving trajectory of the vehicle. In some embodiments, the first point cloud group may be point cloud data collected by sensors on the vehicle in an autonomous driving mode.

[0045] In other embodiments, the first point cloud group may be point cloud data collected by a sensor on an autonomous mobile device over a continuous period of time.

[0046] Based on this, the point cloud data in continuous time can be quality checked to obtain a first test result. In some embodiments, the current point cloud frame can be obtained from the point cloud data, and a local map can be formed using the current point cloud frame and several adjacent point cloud frames; the current point cloud frame can be projected onto a grid of the target plane to obtain a first grid map, and the local map can be projected onto a grid of the target plane to obtain a second grid map; wherein the data stored in each grid is the average height of all points in the grid; and the first test result can be obtained based on the difference between the first grid map and the second grid map.

[0047] Among them, the point cloud data under continuous time means that these point cloud data have time continuity, such as the first point cloud group includes 10 minutes of point cloud data. And these point cloud data are arranged in time order, so these point cloud data are continuous in time. Therefore, the present application can perform quality inspection on adjacent point cloud data in the point cloud data under continuous time according to the quality inspection method of the present application to obtain the first inspection result.

[0048] Step 12: Perform quality inspection on the point cloud data in the first point cloud group at non-continuous time to obtain a second inspection result.

[0049] In some embodiments, non-adjacent point cloud data in the first point cloud group may also be subjected to quality inspection to obtain a second inspection result. For example, the first point cloud group includes 10 minutes of point cloud data. The point cloud frame at the 8th minute and the point cloud frame at the 9th minute are subjected to quality inspection to obtain a second inspection result.

[0050] In the present application, point cloud data is collected by the sensor at a certain sampling frequency. Therefore, if the collected point cloud data is continuous in time at the sampling frequency, it means that the point cloud data is continuous in time. For example, if the sampling frequency is 60 Hz, it means that 60 point cloud frames are collected per second. These 60 point cloud frames are point cloud data at a continuous time of 1 second. Based on this, the point cloud frame at the 8th minute and the point cloud frame at the 9th minute mentioned above are discontinuous. Furthermore, adjacent point cloud data are continuous at the sampling frequency.

[0051] Step 13: Perform quality inspection on the point cloud data between the first point cloud group and the second point cloud group to obtain a third inspection result.

[0052] In some embodiments, the point cloud data of the first point cloud group may be acquired by a sensor of the first vehicle, and the point cloud data of the second point cloud group may be acquired by a sensor of the second vehicle.

[0053] Part of the point cloud data in the second point cloud group is within the preset range of part of the point cloud data in the first point cloud group. The first point cloud group and the second point cloud group have been used for point cloud registration. That is, the first point cloud group and the second point cloud group have been used for point cloud registration, and a corresponding point cloud map can be constructed based on the registration result.

[0054] In some embodiments, the point cloud data all have location information, so the point cloud data between the first point cloud group and the second point cloud group all have corresponding location information. The point cloud data within the preset range is quality tested to know the registration quality, and the point cloud data outside the preset range does not need quality testing. For example, the preset range can be 10 meters, 15 meters, 20 meters or 30 meters, etc.

[0055] In some embodiments, after the related first point cloud group and second point cloud group are found in a plurality of point cloud groups, quality detection may be performed on the point cloud data between the first point cloud group and the second point cloud group to obtain a third detection result.

[0056] Step 14: Provide corresponding quality detection feedback based on the first detection result, the second detection result and the third detection result.

[0057] In some embodiments, the quality of the first detection result, the second detection result, and the third detection result all indicate whether the corresponding point clouds are accurate and whether they can be used for point cloud registration to form a point cloud map.

[0058] Therefore, after obtaining the quality inspection feedback, the point cloud can be adjusted according to the quality inspection feedback, re-registered, and a new point cloud map can be formed again. For example, if the quality inspection feedback shows that the point cloud data of the first point cloud group is unqualified, the first point cloud group is discarded, and the remaining point cloud groups are used to re-register, and a new point cloud map is formed again. If the quality inspection feedback shows that the point cloud data of the second point cloud group is unqualified, the second point cloud group is discarded, and the remaining point cloud groups are used to re-register, and a new point cloud map is formed again.

[0059] In some embodiments, the point cloud registration quality detection can be performed by selecting step 11.

[0060] In some embodiments, the point cloud registration quality detection can be performed by selecting step 12.

[0061] In some embodiments, the point cloud registration quality detection can be performed by selecting step 13.

[0062] In some embodiments, step 11 and step 12 may be selected to perform point cloud registration quality detection.

[0063] In some embodiments, step 11 and step 13 may be selected to perform point cloud registration quality detection.

[0064] In some embodiments, the point cloud registration quality detection can be performed by selecting step 12 and step 13.

[0065] In this embodiment, after the point cloud is registered, the quality of the point cloud data in the first point cloud group at continuous time is checked; and / or the quality of the point cloud data in the first point cloud group at discontinuous time is checked; and / or the quality of the point cloud data between the first point cloud group and the second point cloud group is checked, thereby achieving the quality check of the point cloud registration within the registered point cloud group and between the point cloud groups, at continuous time and discontinuous time, and being able to more accurately evaluate the quality of the point cloud registration, and then being able to adjust the registered point cloud and point cloud map according to the quality inspection feedback. Furthermore, the present application directly detects whether there are abnormalities in the point cloud registration of the registered point cloud group, without the need to use additional sensors to perform consistency checks on the point cloud trajectory and the point cloud map, and is relatively low in cost.

[0066] In some embodiments, see Figure 2 , the above step 11 can be the following process:

[0067] Step 111: Obtain the current point cloud frame from the point cloud data, and use the current point cloud frame and several adjacent point cloud frames to form a local map.

[0068] In some embodiments, the current point cloud frame is obtained from the point cloud data of the first point cloud group. The current point cloud frame may be point cloud data collected at a certain moment. For example, the point cloud data of the first point cloud group includes point cloud frame A, point cloud frame B, and point cloud frame C in chronological order. Point cloud frame A can be used as the current point cloud frame, point cloud frame B can also be used as the current point cloud frame, and point cloud frame C can also be used as the current point cloud frame. That is, any point cloud frame can be selected from the point cloud data of the first point cloud group as the current point cloud frame, and then the current point cloud frame and several adjacent point cloud frames are used to form a local map. For example, if point cloud frame B is selected as the current point cloud frame, then point cloud frame A and point cloud frame C are point cloud frames adjacent to the current point cloud frame, and then point cloud frame A, point cloud frame B, and point cloud frame C can be used to form a local map.

[0069] In some embodiments, the current point cloud frame may be acquired from the first point cloud group in a traversal manner in chronological order.

[0070] In some embodiments, the current point cloud frame may be obtained from the first point cloud group in an interval manner in a time sequence to reduce the overall computational complexity. For example, the point cloud data in the first point cloud group may be traversed every n frames as the current point cloud frame.

[0071] In some embodiments, after determining the current point cloud frame from the first point cloud group, the local map is formed using the current point cloud frame and a number of point cloud frames adjacent to the current point cloud frame. In some embodiments, a corresponding number of point cloud frames can be collected within a time range before and after the current point cloud frame. For example, 5 frames before and after the current point cloud frame are obtained, and after removing point cloud distortion, a local map is formed.

[0072] Step 112: Project the current point cloud frame onto the grid of the target plane to obtain a first grid image, and project the local map onto the grid of the target plane to obtain a second grid image.

[0073] The data stored in each grid is the average height of all points in the grid.

[0074] In some embodiments, the data points in the point cloud frame have corresponding three-dimensional coordinates. If three dimensions are defined by XYZ, the target plane may be a plane on XY.

[0075] In some embodiments, the grid of the target plane may be rectangular. In some embodiments, the grid of the target plane may be square, such as 1m*1m in size. It is understood that the size of the grid may be set according to actual needs.

[0076] Based on this, the grids in the first grid image more or less have data points corresponding to the point cloud frame. Therefore, the height average of these data points can be used as the data of the grid. When there is no data point in the grid, the data of the grid is 0.

[0077] The grids in the second grid map have more or less data points corresponding to the local map. Therefore, the average height of these data points can be used as the data of the grid. When there is no data point in the grid, the data of the grid is 0.

[0078] Step 113: Obtain a first detection result according to the difference between the first grid image and the second grid image.

[0079] In some embodiments, the difference between the first raster image and the second raster image can be obtained by calculating the sum of the square distances between the first raster image and the second raster image. If there is no abnormality in the current point cloud registration, the sum of the square distances between the two is small; otherwise, the sum of the square distances is large. Therefore, by comparing the sum of the square distances with a given threshold, it can be determined whether there is an abnormality in the current point cloud registration. That is, the comparison result at this time is used as the first detection result.

[0080] In some embodiments, step 111 may be performed in the following manner: points in the current point cloud frame in the first point cloud group that are not within the height threshold range are eliminated to obtain a first target point cloud frame; the first target point cloud frame is projected onto the target plane to obtain a first grid map. The height threshold range may be determined based on the prior installation height of the sensor that collects the point cloud. For example, if the prior installation height is h, the height threshold range may be [h-δ, h+δ]. Then, points in the current point cloud frame that are not within the height threshold range are eliminated to obtain a first target point cloud frame.

[0081] Step 112 may be performed in the following manner: eliminating points in the local map that are not within the height threshold range to obtain a first target map; projecting the first target map onto the target plane to obtain a second grid map. Similarly, the height threshold range may be determined based on the prior installation height of the sensor that collects the point cloud. For example, if the prior installation height is h, the height threshold range may be [h-δ, h+δ]. Then, eliminating points in the local map that are not within the height threshold range to obtain the first target map.

[0082] Furthermore, in order to reduce the overall amount of calculation, before projecting the current point cloud frame and the local map in the first point cloud group or removing data points, the current point cloud frame and the local map are downsampled, respectively. Then, the downsampled current point cloud frame and the local map are projected or data points are removed.

[0083] By means of the above method, the point cloud registration quality detection between consecutive point cloud frames in a point cloud group is realized. That is, each point cloud group can realize the point cloud registration quality detection between consecutive point cloud frames in the point cloud group according to the above method.

[0084] In some embodiments, see Figure 3 , step 13 can be the following process:

[0085] Step 131: Obtain the nearest neighbor point cloud frame in the second point cloud group within a preset range of the current point cloud frame in the first point cloud group.

[0086] In some embodiments, a first KD tree may be constructed for the position of each scan key frame in the second point cloud group, and then the first KD tree is searched for the nearest neighbor point cloud frame within a preset range of the current point cloud frame in the first point cloud group.

[0087] In some embodiments, the location of the current point cloud frame in the first point cloud group can be directly taken as the center, and the preset radius can be used as the search radius to obtain the point cloud trajectories in the second point cloud group within the search radius. Then, the point cloud frames corresponding to these point cloud trajectories are used as the nearest neighbor point cloud frames.

[0088] In some embodiments, the preset range may be 10 meters, 15 meters, 20 meters, 25 meters or 30 meters. The specific preset range may be determined according to actual conditions.

[0089] The nearest neighbor point cloud frame is a point cloud frame in the second point cloud group.

[0090] Step 132: Determine a reference frame from the nearest neighbor point cloud frames.

[0091] In some embodiments, if the number of the nearest neighbor point cloud frame is 1, the nearest neighbor point cloud frame is directly used as the reference frame.

[0092] In some embodiments, if the number of nearest neighbor point cloud frames is greater than 1, the point cloud frame closest to the current point cloud frame can be obtained from these nearest neighbor point cloud frames as a reference frame. If the number of point cloud frames closest to the current point cloud frame is greater than 1, step 133 is performed for each reference frame. In this way, point cloud frames can be screened, and the nearest point cloud frames have a stronger correlation, which can improve the efficiency and accuracy of subsequent quality inspection.

[0093] In one application scenario, combined with Figure 4 To explain:

[0094] Taking the position of the current point cloud frame A in the first point cloud group group1 as the center, the preset radius R is used as the search radius to obtain the point cloud trajectories in the second point cloud group group2 within the search radius R. Then, the point cloud frames corresponding to these point cloud trajectories are used as the nearest neighbor point cloud frames. Figure 4 As shown in FIG. 1 , the point cloud frame corresponding to the point cloud trajectory in the circle in the second point cloud group group2 is taken as a candidate frame, and then the candidate frames are screened to select the candidate frame closest to the current point cloud frame A as the reference frame, as shown in FIG. Figure 4 As shown, the candidate frame corresponding to the * position is selected from the candidate frames as the reference frame.

[0095] Step 133: Perform quality detection based on the current point cloud frame and the reference frame to obtain a third detection result.

[0096] In some embodiments, in response to the height difference between the current point cloud frame and the reference frame being less than a first height threshold, the angle between the ground point cloud and the ground normal vector in the current point cloud frame and the reference frame is obtained, and quality detection is performed based on the angle between the ground point cloud and the ground normal vector to obtain a third detection result.

[0097] It can be understood that the height difference between the current point cloud frame and the reference frame is a relative height difference.

[0098] Furthermore, the height difference may be the absolute value of the height difference between the current point cloud frame and the reference frame.

[0099] In some embodiments, see Figure 5 , quality inspection is performed based on the angle between the ground point cloud and the ground normal vector, and the third inspection result can be obtained by the following process:

[0100] Step 51: Obtain the intersection-over-union ratio between the ground point cloud in the current point cloud frame and the reference frame.

[0101] In some embodiments, the ground point clouds in the current point cloud frame and the reference frame are extracted respectively. For example, the ground point cloud in the current point cloud frame is the first ground point cloud, and the ground point cloud in the reference frame is the second ground point cloud. Then, the intersection-over-union ratio corresponding to the first ground point cloud and the second ground point cloud is calculated.

[0102] In some embodiments, ground fitting may be performed based on the first ground point cloud and the second ground point cloud to obtain a final ground point cloud and a corresponding plane equation.

[0103] In some embodiments, during the ground point cloud extraction process, the point cloud that is obviously not a ground point can be removed using the a priori installation height of the sensor that collects the point cloud.

[0104] Step 52: In response to the intersection-and-union ratio being greater than the intersection-and-union ratio threshold, and the angle of the ground normal vector being greater than the angle threshold, obtaining actual intersection results of the ground point clouds in the current point cloud frame and the reference frame.

[0105] Among them, the ground normal vector can be obtained by fitting the ground point cloud.

[0106] In some embodiments, when the intersection-and-union ratio is less than or equal to the intersection-and-union ratio threshold, or the angle of the ground normal vector is less than or equal to the angle threshold, the current point cloud frame and the reference frame are considered to meet the requirements, and quality detection of the next group of current point cloud frames and reference frames is performed.

[0107] Because the first point cloud group and the second point cloud group are point clouds corresponding to different trajectories, if the intersection ratio is greater than the intersection ratio threshold and the angle of the ground normal vector is greater than the angle threshold, there may be a registration anomaly. Therefore, the actual intersection result of the ground point cloud in the current point cloud frame and the reference frame is obtained.

[0108] For example, to determine whether there is an actual intersection between the ground point cloud of the current point cloud frame and the reference frame, assuming that the ground point cloud of the current point cloud frame is used as a reference, the point cloud in the ground point cloud of the reference frame that is higher than, lower than, and on the ground of the current point cloud frame will be counted. When the number of the three is greater than a certain threshold, it is considered that the two grounds intersect, which is inconsistent with the actual situation. Therefore, it is considered that there is a registration anomaly in this case, and step 53 can be executed.

[0109] Step 53: Obtain a third detection result according to the actual cross-check result.

[0110] In some embodiments, in response to the height difference between the current point cloud frame and the reference frame being greater than the first height threshold and less than the second height threshold, a ray tracing algorithm is used to perform quality detection on the current point cloud frame and the reference frame to obtain a third detection result.

[0111] It can be understood that the height difference between the current point cloud frame and the reference frame is a relative height difference.

[0112] In some embodiments, see Figure 6 , use the ray tracing algorithm to perform quality detection on the current point cloud frame and the reference frame, and obtain the third detection result, which can be the following process:

[0113] Step 61: taking the one with a higher height between the current point cloud frame and the reference frame as the second target point cloud frame, and taking the one with a lower height between the current point cloud frame and the reference frame as the third target point cloud frame.

[0114] In some embodiments, if the reference frame is higher, the reference frame is used as the second target point cloud frame, and the current point cloud frame is used as the third target point cloud frame.

[0115] In some embodiments, if the current point cloud frame is higher, the current point cloud frame is used as the second target point cloud frame, and the reference frame is used as the third target point cloud frame.

[0116] Step 62: traverse all data points of the third target point cloud frame, and determine abnormal points in combination with the ground point cloud of the second target point cloud frame.

[0117] Under normal circumstances, the point cloud corresponding to the third target point cloud frame at a lower position will not pass through the ground corresponding to the second target point cloud frame, so all data points of the third target point cloud frame can be traversed.

[0118] In some embodiments, a second KD tree may be established for the ground point cloud of the second target point cloud frame, and then all data points of the third target point cloud frame are traversed, and the outliers are determined in combination with the second KD tree.

[0119] In some embodiments, before traversing all data points of the third target point cloud frame, the third target point cloud frame is downsampled, and then all data points of the downsampled third target point cloud frame are traversed, so as to improve the overall detection efficiency.

[0120] In some embodiments, see Figure 7 , in the traversal order, perform the following steps for each data point in the third target point cloud frame:

[0121] Step 71: Determine a line between the current data point and the radar center point corresponding to the third target point cloud frame.

[0122] It can be understood that the data points in the point cloud are all collected by the radar sensor. Therefore, the radar center point can be determined based on the prior installation height of the radar sensor. Then, the coordinates of the current data point and the coordinates of the radar center point can be used to perform a line connection operation to obtain the connection between the current data point and the radar center point.

[0123] Furthermore, the line between the current data point and the radar center point can be represented by a straight line equation l = p0 + t * direction, where direction is the direction vector of the straight line between the radar center and the laser point, that is, direction = p-p0, and t is the parameter of the straight line. Assuming that the ground equation of the second target point cloud frame is ax + by + cz + d = 1, then the normal vector of the ground is = (a, b, c), and a point on the plane is p π =(0,0,-d / c).

[0124] Step 72: Obtain the intersection point of the connecting line and the ground plane corresponding to the second target point cloud frame.

[0125] After determining the connection line between the current data point and the radar center point, since the second target point cloud frame is the higher one, the ground plane corresponding to the second target point cloud frame is between the current data point and the radar center point. Therefore, the connection line between the current data point and the radar center point has an intersection with the ground plane corresponding to the second target point cloud frame.

[0126] Step 73: Using the intersection point as the search position, perform a nearest neighbor search in the ground point cloud of the second target point cloud frame.

[0127] It can be understood that if the nearest neighbor search is performed using the intersection as the search position, the searched ground point cloud is essentially the ground point cloud in the second target point cloud frame.

[0128] In some embodiments, if a second KD tree is established for the ground point cloud of the second target point cloud frame, a nearest neighbor search may be performed in the second KD tree using the intersection as the search position.

[0129] In some embodiments, the intersection location can be directly used as the search center, and a preset radius can be used as the search radius to perform a nearest neighbor search in the ground point cloud of the second target point cloud frame to obtain ground points in the ground point cloud of the second target point cloud frame within the search radius.

[0130] Step 74: In response to the number of searched ground points being greater than a quantity threshold, the current data point is determined as an abnormal point.

[0131] If the number of searched ground points is greater than the number threshold, it proves that the line between the current data point and the radar center point does intersect with the ground of the second target point cloud frame. This situation is impossible to happen in practice, so the current data point needs to be determined as an outlier.

[0132] In an application scenario, the current point cloud frame is at the lower position and the reference frame is at the higher position. Figure 8 To explain:

[0133] A kd-tree is built for the higher reference frame ground point cloud for subsequent nearest neighbor search.

[0134] Downsample the current point cloud frame with a lower position to improve quality inspection efficiency.

[0135] Under normal circumstances, the point cloud corresponding to the current point cloud frame at a lower position will not pass through the upper ground (the ground of the reference frame), so the laser points (data points) of the current point cloud frame can be traversed.

[0136] Assume that the line connecting the radar center p0 of the current point cloud frame and a laser point (current data point) p can be represented by a straight line equation l = p0 + t*direction, where direction is the direction vector of the straight line between the radar center and the laser point, that is, direction = p-p0, and t is the parameter of the straight line. Assume that the ground equation of the reference frame is ax + by + cz + d = 1, then the normal vector of the ground is n = (a, b, c), and a point on the plane is p π =(0,0,-d / c).

[0137] Combine the straight line equation l = p0 + t*direction and the ground equation of the reference frame ax + by + cz + d = 1 to calculate the intersection point p of the straight line and the ground plane cross First, we need to get the parameters of the line t = (n.dot (p π -p0)) / n.dot(direction), then p cross=p0+t*direction.

[0138] Get the coordinates p of the intersection of the line and the plane cross After that, in the kd tree established in the reference frame ground point cloud, p cross To search for the position, the preset distance is used as the radius for the nearest neighbor search. If the number of nearest neighbor ground points searched is greater than 0, it proves that the line connecting the radar center and the laser point of the current point cloud frame does intersect with the ground of the reference frame at a higher position. This situation is impossible in practice, so the laser point will be marked as an abnormal point.

[0139] After traversing each laser point of the current point cloud frame, the number of abnormal points and the proportion of the current frame scanning points are counted, and the number and proportion of abnormal points are used to determine whether there is any registration anomaly between the current point cloud frame and the reference frame.

[0140] Step 63: Obtain a third detection result according to the number of abnormal points.

[0141] In some embodiments, the third detection result may be obtained by using the number of outliers and the proportion of the number of outliers to all data points.

[0142] In some embodiments, in response to the height difference between the current point cloud frame and the reference frame being greater than a second height threshold, the quality check of the current point cloud frame and the reference frame is skipped. This is because the scanning height of the sensor that collects the point cloud is limited. When the height difference is greater than the second height threshold, it is impossible to determine whether there is a height drift between the current point cloud frame and the reference frame, or whether the actual environment is such, so the quality check in this case is directly skipped.

[0143] Through the above method, the point cloud registration quality detection between point cloud groups is achieved.

[0144] In some embodiments, see Fig. 9 , step 12 can be the following process:

[0145] Step 91: Obtain the nearest neighbor point cloud frame in the first point cloud group within a preset range of the current point cloud frame in the first point cloud group.

[0146] In some embodiments, a third KD tree may be constructed for the position of each scan key frame in the first point cloud group, and the nearest neighbor point cloud frame within a preset range of the current point cloud frame in the first point cloud group is searched in the third KD tree.

[0147] In some embodiments, the location of the current point cloud frame in the first point cloud group can be directly taken as the center, and the preset radius can be used as the search radius to obtain the point cloud trajectories in the first point cloud group within the search radius, and then the point cloud frames corresponding to these point cloud trajectories are used as the nearest neighbor point cloud frames.

[0148] In some embodiments, the preset range may be 10 meters, 15 meters, 20 meters, 25 meters or 30 meters. The specific preset range may be determined according to actual conditions.

[0149] The nearest neighbor point cloud frame is a point cloud frame in the first point cloud group.

[0150] In some embodiments, in order to reduce the overall amount of calculation, the current point cloud frame can be acquired in the first point cloud group in an intermittent manner.

[0151] Step 92: Remove the point cloud frames that are continuous with the current point cloud frame from the nearest neighbor point cloud frames, and use the remaining point cloud frames as reference frames.

[0152] In some embodiments, the continuous point cloud frames have been quality-checked in the above manner, so in this step, the point cloud frames that are continuous with the current point cloud frame are removed from the nearest neighbor point cloud frames, and the remaining point cloud frames are used as reference frames. That is, the point cloud frames that are discontinuous with the current point cloud frame are used as reference frames.

[0153] In some embodiments, if the number of remaining point cloud frames is 1, the remaining point cloud frames are directly used as reference frames.

[0154] In some embodiments, if the number of remaining point cloud frames is greater than 1, the point cloud frame closest to the current point cloud frame can be obtained from the remaining point cloud frames as a reference frame. If the number of point cloud frames closest to the current point cloud frame is greater than 1, step 93 is performed for each reference frame. In this way, point cloud frames can be screened, and the closest point cloud frames have a stronger correlation, which can improve the efficiency and accuracy of subsequent quality inspection.

[0155] In one application scenario, combined with Fig.10 To explain:

[0156] Taking the position of the current point cloud frame A in the first point cloud group group1 as the center, the preset radius R is used as the search radius to obtain the point cloud trajectory in the first point cloud group group1 within the search radius R. Fig.10 As shown in FIG. 1 , the cloud frames corresponding to the point cloud trajectories in the circle in the first point cloud group group 1 are taken as candidate frames, and then the candidate frames that are continuous with the current point cloud frame A are removed from the candidate frames. Then, the candidate frame closest to the current point cloud frame A is selected from the remaining candidate frames as the reference frame. Fig.10 As shown, the candidate frame corresponding to the ★ position is selected from the candidate frames as the reference frame.

[0157] Step 93: Perform quality detection based on the current point cloud frame and the reference frame to obtain a second detection result.

[0158] In some embodiments, in response to the height difference between the current point cloud frame and the reference frame being less than a third height threshold, the angle between the ground point cloud and the ground normal vector in the current point cloud frame and the reference frame is obtained, and quality detection is performed based on the angle between the ground point cloud and the ground normal vector to obtain a second detection result.

[0159] It can be understood that the height difference between the current point cloud frame and the reference frame is a relative height difference.

[0160] In some embodiments, see Fig.11 , quality inspection is performed based on the angle between the ground point cloud and the ground normal vector, and the second inspection result can be obtained by the following process:

[0161] Step 81: Obtain the intersection-over-union ratio between the ground point cloud in the current point cloud frame and the reference frame.

[0162] In some embodiments, the ground point clouds in the current point cloud frame and the reference frame are extracted respectively. For example, the ground point cloud in the current point cloud frame is the first ground point cloud, and the ground point cloud in the reference frame is the second ground point cloud. Then, the intersection-over-union ratio corresponding to the first ground point cloud and the second ground point cloud is calculated.

[0163] In some embodiments, ground fitting may be performed based on the first ground point cloud and the second ground point cloud to obtain a final ground point cloud and a corresponding plane equation.

[0164] In some embodiments, during the ground point cloud extraction process, the point cloud that is obviously not a ground point can be removed using the a priori installation height of the sensor that collects the point cloud.

[0165] Step 82: In response to the intersection-and-union ratio being greater than the intersection-and-union ratio threshold and the angle of the ground normal vector being greater than the angle threshold, obtaining actual intersection results of the ground point clouds in the current point cloud frame and the reference frame.

[0166] Among them, the ground normal vector can be obtained by fitting the ground point cloud.

[0167] In some embodiments, when the intersection-and-union ratio is less than or equal to the intersection-and-union ratio threshold, or the angle of the ground normal vector is less than or equal to the angle threshold, the current point cloud frame and the reference frame are considered to meet the requirements, and quality detection of the next group of current point cloud frames and reference frames is performed.

[0168] Because the first point cloud group and the second point cloud group are point clouds corresponding to different trajectories, if the intersection ratio is greater than the intersection ratio threshold and the angle of the ground normal vector is greater than the angle threshold, there may be a registration anomaly. Therefore, the actual intersection result of the ground point cloud in the current point cloud frame and the reference frame is obtained.

[0169] For example, to determine whether there is an actual intersection between the ground point cloud of the current point cloud frame and the reference frame, assuming that the ground point cloud of the current point cloud frame is used as a reference, the point cloud in the ground point cloud of the reference frame that is higher than, lower than, and on the ground of the current point cloud frame will be counted. When the number of the three is greater than a certain threshold, it is considered that the two grounds intersect, which is inconsistent with the actual situation. Therefore, it is considered that there is a registration anomaly in this case, and step 83 can be executed.

[0170] Step 83: Obtain a second detection result according to the actual cross-check result.

[0171] In some embodiments, in response to the height difference between the current point cloud frame and the reference frame being greater than a third height threshold and less than a fourth height threshold, a ray tracing algorithm is used to perform quality detection on the current point cloud frame and the reference frame to obtain a second detection result.

[0172] It can be understood that the height difference between the current point cloud frame and the reference frame is a relative height difference.

[0173] In some embodiments, see Fig.12 , use the ray tracing algorithm to perform quality detection on the current point cloud frame and the reference frame, and obtain the second detection result, which can be the following process:

[0174] Step 121: taking the one with a higher height between the current point cloud frame and the reference frame as the fourth target point cloud frame, and taking the one with a lower height between the current point cloud frame and the reference frame as the fifth target point cloud frame.

[0175] In some embodiments, if the reference frame is higher, the reference frame is used as the fourth target point cloud frame, and the current point cloud frame is used as the fifth target point cloud frame.

[0176] In some embodiments, if the current point cloud frame is higher, the current point cloud frame is used as the fourth target point cloud frame, and the reference frame is used as the fifth target point cloud frame.

[0177] Step 122: traverse all data points of the fifth target point cloud frame, and determine abnormal points in combination with the ground point cloud of the fourth target point cloud frame.

[0178] Under normal circumstances, the point cloud corresponding to the fifth target point cloud frame at a lower position will not pass through the ground corresponding to the fourth target point cloud frame, so all data points of the fifth target point cloud frame can be traversed.

[0179] In some embodiments, a third KD tree may be established for the ground point cloud of the fourth target point cloud frame, and then all data points of the fifth target point cloud frame are traversed, and abnormal points are determined in combination with the third KD tree.

[0180] In some embodiments, before traversing all data points of the fifth target point cloud frame, the fifth target point cloud frame is downsampled, and then all data points of the downsampled fifth target point cloud frame are traversed to improve the overall detection efficiency.

[0181] In some embodiments, see Fig.13 , in the traversal order, perform the following steps for each data point in the fifth target point cloud frame:

[0182] Step 101: Determine a line between a current data point and a radar center point corresponding to a fifth target point cloud frame.

[0183] It can be understood that the data points in the point cloud are all collected by the radar sensor. Therefore, the radar center point can be determined based on the prior installation height of the radar sensor. Then, the coordinates of the current data point and the coordinates of the radar center point can be used to perform a line connection operation to obtain the connection between the current data point and the radar center point.

[0184] Furthermore, the line between the current data point and the center point of the radar can be represented by a straight line equation l = p0 + t * direction, where direction is the direction vector of the straight line between the center of the radar and the laser point, that is, direction = p-p0, and t is the parameter of the straight line. Assuming that the ground equation of the second target point cloud frame is ax + by + cz + d = 1, the normal vector of the ground is n = (a, b, c), and a point on the plane is pπ = (0, 0, -d / c).

[0185] Step 102: Obtain the intersection point of the connecting line and the ground plane corresponding to the fourth target point cloud frame.

[0186] After determining the connection line between the current data point and the radar center point, since the fourth target point cloud frame is the higher one, the ground plane corresponding to the fourth target point cloud frame is between the current data point and the radar center point. Therefore, the connection line between the current data point and the radar center point has an intersection with the ground plane corresponding to the fourth target point cloud frame.

[0187] Step 103: Using the intersection point as the search position, perform a nearest neighbor search in the ground point cloud of the fourth target point cloud frame.

[0188] It can be understood that if the nearest neighbor search is performed with the intersection as the search position, the searched ground point cloud is essentially the ground point cloud in the fourth target point cloud frame.

[0189] In some embodiments, if a third KD tree is established for the ground point cloud of the fourth target point cloud frame, a nearest neighbor search may be performed in the third KD tree using the intersection as the search position.

[0190] In some embodiments, the intersection location can be directly used as the search center, and a preset radius can be used as the search radius to perform a nearest neighbor search in the ground point cloud of the fourth target point cloud frame to obtain ground points in the ground point cloud of the fourth target point cloud frame within the search radius.

[0191] Step 104: In response to the number of searched ground points being greater than a quantity threshold, the current data point is determined as an abnormal point.

[0192] If the number of searched ground points is greater than the number threshold, it proves that the line between the current data point and the radar center point does intersect with the ground of the fourth target point cloud frame. This situation is impossible to happen in practice, so the current data point needs to be determined as an abnormal point.

[0193] Step 123: Obtain a second detection result according to the number of abnormal points.

[0194] In some embodiments, the second detection result may be obtained by using the number of outliers and the proportion of the number of outliers to all data points.

[0195] In some embodiments, in response to the height difference between the current point cloud frame and the reference frame being greater than a fourth height threshold, the quality check of the current point cloud frame and the reference frame is skipped. This is because the scanning height of the sensor for collecting the point cloud is limited. When the height difference is greater than the fourth height threshold, it is impossible to determine whether there is a height drift between the current point cloud frame and the reference frame, or whether the actual environment is such, so the quality check in this case is directly skipped.

[0196] Through the above method, the point cloud registration quality detection between non-continuous point cloud frames in the point cloud group is achieved.

[0197] See also Fig.14 , Fig.14 1 is a flow chart of an embodiment of a point cloud registration method provided by the present application. The method comprises:

[0198] Step 141: Perform point cloud registration using a plurality of point cloud groups.

[0199] In some embodiments, several point cloud groups may be used to perform point cloud registration to form a point cloud map.

[0200] Step 142: After the point cloud is registered, a registration quality test is performed to obtain quality test feedback.

[0201] The technical solution of any of the above embodiments can be used to perform point cloud registration quality detection to obtain corresponding quality detection feedback.

[0202] Step 143: Delete several point cloud groups according to the quality inspection feedback, and perform point cloud registration again.

[0203] After obtaining quality inspection feedback, if any point cloud group registration anomalies are found, the point cloud group can be deleted and not used in point cloud registration.

[0204] In this embodiment, after the point cloud is registered, the technical solution of any of the above embodiments is used to perform point cloud registration quality detection, and then the point cloud group is adjusted according to the quality detection feedback, and then the point cloud is registered again to improve the accuracy of the point cloud registration.

[0205] See also Fig.15 , Fig.15 1 is a schematic diagram of the structure of an embodiment of an electronic device provided by the present application. The electronic device 150 includes a processor 151 and a memory 152 coupled to the processor 151; wherein the memory 152 is used to store a computer program, and the processor 151 is used to execute the computer program to implement the following method:

[0206] Performing quality inspection on point cloud data in continuous time in the first point cloud group to obtain a first inspection result; and / or, performing quality inspection on point cloud data in non-continuous time in the first point cloud group to obtain a second inspection result; and / or, performing quality inspection on point cloud data between the first point cloud group and the second point cloud group to obtain a third inspection result; wherein, part of the point cloud data in the second point cloud group is within a preset range of part of the point cloud data in the first point cloud group, and the first point cloud group and the second point cloud group have been used for point cloud registration; performing corresponding quality inspection feedback according to the first inspection result, the second inspection result and / or the third inspection result.

[0207] It can be understood that the processor 151 is also used to execute computer programs to implement the method of any of the above embodiments, which will not be elaborated here.

[0208] See also Fig.16 , Fig.16 1 is a schematic diagram of a computer readable storage medium according to an embodiment of the present application. The computer readable storage medium 160 is used to store a computer program 161. When the computer program 161 is executed by a processor, it is used to implement the following method:

[0209] Performing quality inspection on point cloud data in continuous time in the first point cloud group to obtain a first inspection result; and / or, performing quality inspection on point cloud data in non-continuous time in the first point cloud group to obtain a second inspection result; and / or, performing quality inspection on point cloud data between the first point cloud group and the second point cloud group to obtain a third inspection result; wherein, part of the point cloud data in the second point cloud group is within a preset range of part of the point cloud data in the first point cloud group, and the first point cloud group and the second point cloud group have been used for point cloud registration; performing corresponding quality inspection feedback according to the first inspection result, the second inspection result and / or the third inspection result.

[0210] It can be understood that when the computer program 161 is executed by the processor, it is also used to implement the method of any of the above embodiments, which will not be elaborated here.

[0211] In summary, the present application provides a point cloud registration method, a point cloud registration quality detection method, an electronic device, and a medium. After point cloud registration, quality detection is performed on the point cloud data in the first point cloud group at continuous time; and / or quality detection is performed on the point cloud data in the first point cloud group at discontinuous time; and / or quality detection is performed on the point cloud data between the first point cloud group and the second point cloud group, thereby achieving point cloud registration quality detection within the registered point cloud group and between point cloud groups, at continuous time and discontinuous time, and being able to more accurately evaluate the quality of point cloud registration, and then being able to adjust the registered point cloud and point cloud map according to quality detection feedback. Furthermore, the present application directly detects whether there are any abnormalities in the point cloud registration of the registered point cloud group, without the need to use additional sensors to perform consistency verification on the point cloud trajectory and the point cloud map, and is relatively low in cost.

[0212] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only illustrative, for example, the division of the circuit or unit is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0213] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0214] In addition, each functional unit in each embodiment of the present application may be integrated into a first processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0215] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made according to the description and drawings of the present application, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A point cloud registration quality detection method, characterized in that: The method comprises: Performing quality inspection on point cloud data at continuous time in the first point cloud group to obtain a first inspection result; and / or, performing quality detection on the point cloud data in the first point cloud group at non-continuous times to obtain a second detection result; and / or, performing a quality inspection on the point cloud data between the first point cloud group and the second point cloud group to obtain a third inspection result; wherein, part of the point cloud data in the second point cloud group is within a preset range of part of the point cloud data in the first point cloud group; and the first point cloud group and the second point cloud group have been used for point cloud registration; Corresponding quality detection feedback is performed according to the first detection result, the second detection result and / or the third detection result.

2. The method according to claim 1, characterized in that The step of performing quality inspection on the point cloud data in the first point cloud group at a continuous time to obtain a first inspection result includes: Acquire a current point cloud frame from the point cloud data, and form a local map using the current point cloud frame and a number of adjacent point cloud frames; Projecting the current point cloud frame onto a grid of a target plane to obtain a first grid map, and projecting the local map onto a grid of the target plane to obtain a second grid map; wherein the data stored in each grid is the average value of the heights of all points in the grid; A first detection result is obtained according to a difference between the first grid image and the second grid image.

3. The method according to claim 2, characterized in that The projecting the current point cloud frame onto a grid of a target plane to obtain a first grid map includes: Eliminate points in the current point cloud frame that are not within a height threshold range to obtain a first target point cloud frame; Projecting the first target point cloud frame onto a grid of the target plane to obtain a first grid image; The projecting the local map onto the grid of the target plane to obtain a second grid map includes: Eliminate points in the local map that are not within the height threshold range to obtain a first target map; The first target map is projected onto a grid of the target plane to obtain a second grid map.

4. The method according to claim 1, characterized in that: The step of performing quality inspection on the point cloud data between the first point cloud group and the second point cloud group to obtain a third inspection result includes: Acquire a nearest neighbor point cloud frame in the second point cloud group within a preset range of the current point cloud frame in the first point cloud group; Determine a reference frame from the nearest neighbor point cloud frames; Quality detection is performed based on the current point cloud frame and the reference frame to obtain a third detection result.

5. The method according to claim 4, characterized in that The performing quality detection according to the current point cloud frame and the reference frame to obtain a third detection result includes: In response to the height difference between the current point cloud frame and the reference frame being less than a first height threshold, obtaining an angle between a ground point cloud and a ground normal vector in the current point cloud frame and the reference frame, and performing quality detection according to the angle between the ground point cloud and the ground normal vector to obtain a third detection result; In response to a height difference between the current point cloud frame and the reference frame being greater than a first height threshold and less than a second height threshold, using a ray tracing algorithm to perform quality detection on the current point cloud frame and the reference frame to obtain a third detection result; In response to the height difference between the current point cloud frame and the reference frame being greater than the second height threshold, quality detection of the current point cloud frame and the reference frame is skipped.

6. The method according to claim 5, characterized in that The performing quality detection according to the ground point cloud and the angle between the ground normal vectors to obtain a third detection result includes: Obtaining an intersection-over-union ratio between the current point cloud frame and the ground point cloud in the reference frame; In response to the intersection-and-union ratio being greater than an intersection-and-union ratio threshold and the angle of the ground normal vector being greater than an angle threshold, obtaining an actual intersection result of the ground point cloud in the current point cloud frame and the reference frame; A third detection result is obtained according to the actual cross-detection result.

7. The method according to claim 5, characterized in that The using of a ray tracing algorithm to perform quality detection on the current point cloud frame and the reference frame to obtain a third detection result includes: The one with a higher height between the current point cloud frame and the reference frame is used as a second target point cloud frame, and the one with a lower height between the current point cloud frame and the reference frame is used as a third target point cloud frame; Traversing all data points of the third target point cloud frame, and determining abnormal points in combination with the ground point cloud of the second target point cloud frame; A third detection result is obtained according to the number of abnormal points.

8. The method according to claim 7, characterized in that The traversing all data points of the third target point cloud frame and determining abnormal points in combination with the ground point cloud of the second target point cloud frame includes: In the traversal order, the following steps are performed for each data point in the third target point cloud frame: Determine a line between the current data point and the radar center point corresponding to the third target point cloud frame; Obtaining an intersection point between the connecting line and a ground plane corresponding to the second target point cloud frame; Taking the intersection as a search position, performing a nearest neighbor search in the ground point cloud of the second target point cloud frame; In response to the number of searched ground points being greater than a number threshold, the current data point is determined as an abnormal point.

9. The method according to claim 7, characterized in that: The obtaining of a third detection result according to the number of abnormal points includes: The third detection result is obtained by using the number of abnormal points and the proportion of the number of abnormal points to all data points.

10. The method according to claim 1, characterized in that The step of performing quality inspection on the point cloud data in the first point cloud group at non-continuous time to obtain a second inspection result includes: Acquire a nearest neighbor point cloud frame in the first point cloud group within a preset range of a current point cloud frame in the first point cloud group; Eliminate point cloud frames that are continuous with the current point cloud frame from the nearest neighbor point cloud frames, and use the remaining point cloud frames as reference frames; Quality detection is performed based on the current point cloud frame and the reference frame to obtain a second detection result.

11. A point cloud registration method, characterized in that: The method comprises: Perform point cloud registration using several point cloud groups; Performing registration quality detection using the method described in any one of claims 1 to 10 to obtain quality detection feedback; Several point cloud groups are selected based on the quality detection feedback, and point cloud registration is performed again.

12. An electronic device, characterized in that: The electronic device comprises a processor and a memory coupled to the processor; wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the method according to any one of claims 1-11.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, it is used to implement the method according to any one of claims 1 to 11.

Citation Information

Patent Citations

  • Point clouds ghosting effects detection system for autonomous driving vehicles

    CN111771229A

  • 3D point cloud map consistency detection method and device, medium and equipment

    CN115131309A

  • Point cloud registration result evaluation method and system based on distribution characteristics

    CN115511929A

  • Related method for constructing point cloud annotation data set, electronic equipment and storage medium

    CN115908937A

  • Double image detection method for point cloud map

    CN117670785A