Ground point cloud segmentation method and device, electronic equipment and storage medium
By rastering and feature extraction of point cloud data, combining with inter-class variance maximization processing, the ground point cloud segmentation threshold is determined, which solves the problem of low accuracy of ground point cloud segmentation in the existing technology, and achieves a more efficient ground point cloud segmentation effect.
Patent Information
- Application Number
- CN202510019444.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-06
- Publication Date
- 2025-05-16
AI Technical Summary
In the prior art, the accuracy of ground point cloud segmentation is not high, and it is difficult to effectively distinguish ground point clouds from non-ground point clouds.
By rastering the point cloud data, the sample point feature information of each raster is extracted, including height features, gradient features and fitting error features, and the ground point cloud segmentation threshold is determined through inter-class variance maximization processing, thereby achieving accurate segmentation of ground point clouds.
It improves the accuracy of ground point cloud segmentation, ensures the applicability and adaptability of ground point cloud segmentation threshold, can more accurately reflect the characteristics of sampling points in the grid, and is suitable for point cloud data segmentation in complex scenarios.
Smart Images
Figure CN120013978A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent driving technology, and more specifically, to a ground point cloud segmentation method, device, electronic device and storage medium. Background Art
[0002] Segmenting the collected point cloud data into ground point clouds is an important part of environmental perception in the field of intelligent driving. The results of ground point cloud segmentation can be used to extract ground information in subsequent processes. Ground point cloud segmentation refers to segmenting the point cloud representing the ground (i.e., ground point cloud) and the point cloud representing other non-ground objects from the point cloud data. A method based on plane fitting is provided in the related art to perform ground point cloud segmentation, which uses a plane fitting method to fit a global plane in a single-frame point cloud, and then segment the point cloud representing the ground (i.e., ground point cloud) and the point cloud representing other non-ground objects. However, the accuracy of ground point cloud segmentation performed according to this method is not high. Therefore, how to improve the accuracy of ground point cloud segmentation is a technical problem that needs to be solved urgently in the related art. Summary of the invention
[0003] In view of the above problems, the embodiments of the present application propose a ground point cloud segmentation method, device, electronic device and storage medium to improve the accuracy of ground point cloud segmentation.
[0004] According to one aspect of an embodiment of the present application, a ground point cloud segmentation method is provided, the method comprising: determining the grid where each sampling point is located according to the position information of each sampling point in the point cloud data and a plurality of preset grids; determining the sampling point feature information of each grid according to the sampling points located in the same grid, the sampling point feature information comprising at least two items of the sampling point height feature, the sampling point gradient feature and the sampling point fitting error feature; performing inter-class variance maximization processing according to the sampling point feature information of the plurality of grids to determine the ground point cloud segmentation threshold; determining the ground point cloud segmentation result corresponding to each grid according to the sampling point feature information of each grid and the ground point cloud segmentation threshold, the ground point cloud segmentation result being used to indicate whether the sampling point in the corresponding grid belongs to the ground point cloud or to the non-ground point cloud.
[0005] According to one aspect of an embodiment of the present application, a ground point cloud segmentation device is provided, comprising: a first determination module, for determining the grid where each sampling point is located according to the position information of each sampling point in the point cloud data and a plurality of preset grids; a feature determination module, for determining the sampling point feature information of each grid according to the sampling points located in the same grid, wherein the sampling point feature information includes at least two items of the sampling point height feature, the sampling point gradient feature and the sampling point fitting error feature; a segmentation threshold determination module, for performing inter-class variance maximization processing according to the sampling point feature information of the plurality of grids to determine the ground point cloud segmentation threshold; a segmentation result determination module, for determining the ground point cloud segmentation result corresponding to each grid according to the sampling point feature information of each grid and the ground point cloud segmentation threshold, wherein the ground point cloud segmentation result is used to indicate whether the sampling point in the corresponding grid belongs to the ground point cloud or to the non-ground point cloud.
[0006] According to one aspect of an embodiment of the present application, an electronic device is provided, comprising: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the above-mentioned ground point cloud segmentation method is implemented.
[0007] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the above-mentioned ground point cloud segmentation method is implemented.
[0008] According to one aspect of an embodiment of the present application, a computer program product is provided, including computer instructions, which implement the above ground point cloud segmentation method when executed by a processor.
[0009] In the present application, after the sampling points in the point cloud data are rasterized, the sampling point feature information of the grid is obtained according to the sampling points located in each grid, and the inter-class variance maximization processing is performed according to the sampling point feature information of the multiple grids to determine the ground point cloud segmentation threshold. Since the determined ground point cloud segmentation threshold can maximize the inter-class variance, the ground point cloud segmentation threshold can be used as the boundary between the ground point cloud and the non-ground point cloud. Moreover, the ground point cloud segmentation threshold suitable for the point cloud data is determined according to the feature information of each sampling point after the point cloud data is rasterized, which ensures the applicability and adaptability of the ground point cloud segmentation threshold to the point cloud data for ground point cloud segmentation. Therefore, the method of the present application can effectively improve the accuracy of subsequent ground point cloud segmentation according to the ground point cloud segmentation threshold and the sampling point feature information of each grid.
[0010] Moreover, based on the ground point cloud segmentation threshold, at least two of the sampling point height features, the sampling point gradient features and the sampling point fitting error features are used to comprehensively perform ground point cloud segmentation. The sampling point height features, the sampling point gradient features and the sampling point fitting error features are all features that are strongly correlated with ground point cloud segmentation. Compared with using only one feature, the present application uses more features and can more accurately reflect the characteristics of the sampling points in the grid. Moreover, the features used are targeted at ground point cloud segmentation and can improve the accuracy of subsequent ground point cloud segmentation. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0012] Figure 1 It is a flowchart of a ground point cloud segmentation method according to an embodiment of the present application.
[0013] Figure 2 FIG. 1 is a flow chart showing step 130 according to an embodiment of the present application.
[0014] Figure 3 FIG. 4 is a flow chart of determining the gradient characteristics of a sampling point according to an embodiment of the present application.
[0015] Figure 4 FIG. 4 is a flow chart showing a method for determining gradient characteristics of sampling points according to another embodiment of the present application.
[0016] Figure 5 It is a flowchart of the steps before step 110 according to an embodiment of the present application.
[0017] Figure 6 It is a flowchart of a ground point cloud segmentation method according to a specific embodiment of the present application.
[0018] Figure 7 It is a block diagram of a ground point cloud segmentation device according to an embodiment of the present application.
[0019] Figure 8 It is a block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION
[0020] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.
[0021] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.
[0022] It should be noted that the "multiple" mentioned in this article refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. In the following description, it involves "some embodiments or some embodiment modes", which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.
[0023] The implementation details of the technical solution of the embodiment of the present application are described in detail below:
[0024] Figure 1 is a flow chart of a ground point cloud segmentation method according to an embodiment of the present application. The method can be executed by an electronic device with processing capabilities, such as a server, etc., and is not specifically limited here. Figure 1 As shown, the method at least includes steps 110 to 140, which are described in detail as follows:
[0025] Step 110, determining the grid where each sampling point is located according to the position information of each sampling point in the point cloud data and a plurality of preset grids.
[0026] The point cloud data includes the location information of multiple sampling points. The point cloud data can be collected by a laser radar. The point cloud data can be obtained by a laser radar deployed on a vehicle facing the driving environment of the vehicle. The point cloud data can also be obtained by standardizing the original point cloud data collected by the laser radar. Please refer to the following description for the process of standardization.
[0027] In some embodiments, the location information of the sampling point may be the Cartesian coordinates of the sampling point in a Cartesian coordinate system, such as the device coordinate system of the device where the laser radar is located, the world coordinate system, etc. Correspondingly, the preset multiple grids may be grids in a Cartesian coordinate space, and the shape of the grids is not limited, and may be a cube. Since the sizes and positions of the multiple grids are preset, it can be determined in which grid the sampling point is located according to the Cartesian coordinates of the sampling point.
[0028] In other embodiments, the location information of the sampling point may also be the polar coordinates of the sampling point obtained by converting the Cartesian coordinates of the sampling point into a polar coordinate system (a three-dimensional polar coordinate system, also called a spherical coordinate system). For example, if the Cartesian coordinates (x, y, z) of a sampling point are transformed into a polar coordinate system, the polar coordinates obtained are expressed as (r, θ, □), where r is the distance from the sampling point to the origin of the polar coordinate system, θ is the horizontal azimuth, and □ is the vertical azimuth (also called the elevation angle). Correspondingly, in this case, the preset multiple grids are also grids in the polar coordinate space, and the size and position of each grid in the polar coordinate space are preset; in step 110, the grid where each sampling point is located can be determined based on the polar coordinates of each sampling point in the point cloud data and the preset multiple grids in the polar coordinate space.
[0029] In some embodiments, the sizes of multiple grids in the polar coordinate space may be different, for example, the closer to the origin of the polar coordinate system, the smaller the size of the grid. Since point cloud data usually shows that the point cloud density is high at close distances and low at long distances, dividing each sampling point in the point cloud data into grids in the polar coordinate space can adapt to the changes in the point cloud density in the point cloud data, and helps to balance the number of point clouds in different distance areas, thereby improving the quality of subsequent feature expression.
[0030] Step 120, determining sampling point feature information of each grid according to the sampling points located in the same grid, wherein the sampling point feature information includes at least two of sampling point height feature, sampling point gradient feature and sampling point fitting error feature.
[0031] The sampling point height feature of a grid is used to reflect the statistical characteristics of the heights of the sampling points in the grid, or in other words, to reflect the distribution of the heights of the sampling points in the grid. The sampling point height feature may include at least one of the average height, height difference, maximum height, minimum height, etc. For example, the sampling point height feature may be the average height and the height difference.
[0032] In some embodiments, for each grid, the height characteristics of the sampling points of the grid can be determined according to the heights of multiple sampling points in the grid. For example, the average height of all the sampling points in the grid can be calculated to obtain the average height; according to the heights of all the sampling points in the grid, the maximum height (i.e., the height of the highest sampling point) and the minimum height (i.e., the height of the lowest sampling point) in the grid can be determined, and the maximum height and the minimum height can be added to obtain the height difference of the grid.
[0033] In some embodiments, for each grid, a gradient calculation may be performed based on position information of a plurality of sampling points in the grid to determine the gradient characteristics of the sampling points of the grid.
[0034] The sampling point gradient feature of a grid is used to reflect the overall distribution of the gradient features of the sampling points in the grid. The gradient feature of the sampling point can be the position gradient of the sampling point or the height gradient of the sampling point. The height gradient of the sampling point can reflect the change of the height of the sampling point. The sampling point gradient feature of a grid can be at least one of the average gradient, gradient difference, maximum gradient, and minimum gradient. The average gradient is obtained by calculating the mean of the gradient features of all the sampling points in a grid; the gradient difference can be the difference between the maximum gradient and the minimum gradient in the gradient features of all the sampling points in a grid.
[0035] The sampling point fitting error characteristics of a grid are used to reflect the deviation of the overall distribution of the sampling points of the grid relative to the fitting model corresponding to the grid. The fitting model corresponding to the grid refers to the model obtained by fitting multiple sampling points located in the grid.
[0036] In some embodiments, the sampling point fitting error characteristics of each grid may be determined in the following manner: multiple sampling points located in the grid are fitted to obtain a fitting model; and the sampling point fitting error characteristics of the grid are determined based on the distance from each sampling point located in the grid to the fitting model.
[0037] In some embodiments, the shape presented by the plurality of sampling points in the grid as a whole can be identified, and a fitting basic model corresponding to the shape can be determined. Then, the plurality of sampling points in the grid are fitted according to the determined fitting basic model to obtain a fitting model corresponding to the grid, wherein the shapes of the fitting model corresponding to the grid and the fitting basic model are the same as the shape presented by the plurality of sampling points in the grid as a whole.
[0038] For example, if the overall shape of multiple sampling points in the grid is a straight line, the fitting basic model corresponding to the straight line is a straight line model. Then, multiple sampling points in the grid are fitted according to the straight line model to obtain a fitted straight line. The obtained fitted straight line is the fitting model corresponding to the grid.
[0039] For another example, if the overall shape of multiple sampling points in a grid is a plane, the fitting basic model corresponding to the plane is a plane basic model. Then, multiple sampling points in the grid are fitted according to the plane basic model to obtain a fitting plane. The obtained fitting plane is the fitting model corresponding to the grid.
[0040] For another example, if the overall shape of multiple sampling points in a grid is a curve, the fitting basic model corresponding to the curve is the curve basic model. Then, multiple sampling points in the grid are fitted according to the curve basic model to obtain a fitting curve. The obtained fitting curve is the fitting model corresponding to the grid.
[0041] For another example, if the overall shape of multiple sampling points in a grid is a curved surface, the fitting basic model corresponding to the curved surface is the curved surface basic model. Then, multiple sampling points in the grid are fitted according to the curved surface basic model to obtain a fitted surface. The obtained fitted surface is the fitting model corresponding to the grid.
[0042] Of course, the overall shape of the multiple sampling points in the grid is not limited to those listed above, and the fitting model determined by the corresponding fitting is also not limited to those listed above. In other embodiments, the fitting model can also be a fitting cylinder, a fitting sphere, etc.
[0043] After fitting to obtain the fitting model, a first sampling point located on the surface or inside of the fitting model corresponding to the grid and a second sampling point located outside the fitting model corresponding to the grid can be determined from among the multiple sampling points located in the grid, and for the first sampling point, the distance from the first sampling point to the fitting model can be set to zero. The distance from the second sampling point to the fitting model is calculated, and the calculated distance can be the shortest distance from the second sampling point to the fitting model.
[0044] In some embodiments, the distances from all sampling points in the grid to the fitting model may be averaged, and the obtained average distance may be used as the sampling point fitting error feature of the grid. Alternatively, the sum of the squares of the distances from all sampling points in the grid to the fitting model may be calculated, and the calculation result may be used as the sampling point fitting error feature of the grid.
[0045] In some embodiments, the distances from all sampling points in the grid to the fitting model can be averaged to obtain an average distance; then, the sum of squares of the deviations of the distances from each sampling point in the grid to the fitting model relative to the average distance is calculated, and the square root of the ratio between the sum of squares and the number of sampling points in the grid is calculated as the sampling point fitting error feature of the grid. In this case, it is equivalent to taking the average distance as the true value and calculating the sampling point fitting error feature of the grid according to the root mean square error formula.
[0046] The sampling point feature information of a grid may include any two of the sampling point height feature, the sampling point gradient feature and the sampling point fitting error feature, or may include the three features of the sampling point height feature, the sampling point gradient feature and the sampling point fitting error feature.
[0047] Step 130 , performing inter-class variance maximization processing based on the sampling point feature information of multiple grids to determine the ground point cloud segmentation threshold.
[0048] The solution of the present application is to classify the sampling points in the point cloud data into two categories, the two categories including ground point cloud and non-ground point cloud. The inter-class variance in step 130 is used to quantify the difference between the means of the two categories (ground point cloud and non-ground point cloud).
[0049] In this application, the ground point cloud segmentation threshold is determined by optimizing the inter-class variance so that when the sampling points in the point cloud data are classified into two categories, ground point cloud and non-ground point cloud, according to the ground point cloud segmentation threshold, the difference between the mean of the ground point cloud and the mean of the non-ground point cloud is maximized, thereby achieving the best ground point cloud segmentation effect.
[0050] The mean value for the ground point cloud refers to the mean value calculated by calculating the feature information of the sampling points of the first grid among multiple grids. Similarly, the mean value for the non-ground point cloud refers to the mean value calculated by calculating the feature information of the sampling points of the second grid among multiple grids. The first grid refers to the grid whose sampling points belong to the ground point cloud; the second grid refers to the grid whose sampling points belong to the non-ground point cloud.
[0051] like Figure 2 As shown, step 130 includes the following steps 210 to 270, which are described in detail as follows:
[0052] Step 210 , weighting the features included in the sampling point feature information of each grid to obtain the target feature value of each grid.
[0053] As described above, the sampling point feature information includes at least two of the sampling point height feature, the sampling point gradient feature and the sampling point fitting error feature. For each grid, all features included in the sampling point feature information of the grid are weighted to obtain the target feature value of the grid.
[0054] For example, if the sampling point feature information includes sampling point height features, sampling point gradient features and sampling point fitting error features, and the sampling point height features include average height and height difference, the four features of average height, height difference, sampling point gradient features and sampling point fitting error features can be weighted calculated to obtain the target feature value of the grid.
[0055] In the weighting process, the weighting coefficients for different features in the sampling point feature information may be the same or different, and may be set according to actual needs. In some embodiments, the weighting coefficients for each feature in the sampling point feature information may all be positive numbers.
[0056] In some embodiments, the weighting coefficient for each feature in the sampling point feature information may satisfy that the determined target feature value is within the range of (0, 1) to simplify subsequent ground point cloud segmentation.
[0057] In some embodiments, the features included in the sampling point feature information of the grid may be weighted first, and then the weighted processing result may be normalized, and the result obtained by the normalization processing may be used as the target feature value of the grid, so that the obtained target feature value may be ensured to be located at (0, 1).
[0058] Step 220 , based on the target feature value of each grid and the candidate segmentation threshold, the sampling points in each grid are divided into ground point clouds to determine the candidate ground point cloud segmentation results of each grid.
[0059] The candidate ground point cloud segmentation result of the grid is used to indicate whether the sampling point in the corresponding grid belongs to the ground point cloud or the non-ground point cloud.
[0060] If the target eigenvalue of the grid is less than the candidate segmentation threshold, the candidate ground point cloud segmentation result corresponding to the grid is determined to be the result indicating that the sampling points in the grid belong to the ground point cloud; if the target eigenvalue of the grid is not less than the candidate segmentation threshold, the candidate ground point cloud segmentation result corresponding to the grid is determined to be the result indicating that the sampling points in the grid belong to the non-ground point cloud.
[0061] It is worth mentioning that since the candidate segmentation threshold may not be the final ground point cloud segmentation threshold, the candidate ground point cloud segmentation result of the grid determined according to the candidate segmentation threshold may not be accurate, and the candidate ground point cloud segmentation result of the grid is used as the ground point cloud segmentation result as an intermediate result in the process of iteratively determining the ground point cloud segmentation threshold. Among them, the candidate segmentation threshold can be determined by initialization and is the initial value for subsequent iterations.
[0062] Step 230, based on the candidate ground point cloud segmentation results of each grid in the multiple grids, determine the first grid and the second grid among the multiple grids; the first grid refers to the grid that the corresponding candidate ground point cloud segmentation result indicates belongs to the ground point cloud; the second grid refers to the grid that the corresponding candidate ground point cloud segmentation result indicates belongs to the non-ground point cloud.
[0063] Step 240 , determining a first proportion of a first grid among the plurality of grids and determining a second proportion of a second grid among the plurality of grids.
[0064] Among them, the first proportion is equal to the ratio of the number of the first grid to the total number of grids; similarly, the second proportion is equal to the ratio of the number of the second grid to the total number of grids. Among them, the total number of grids is equal to the sum of the number of the first grid and the number of the second grid. It can be seen that the sum of the first proportion and the second proportion is equal to 1.
[0065] Step 250, determining a first mean value according to the target feature value of a first grid among the multiple grids; and determining a second mean value according to the target feature value of a second grid among the multiple grids.
[0066] The average value of the target feature values of all first grids in the plurality of grids is calculated to obtain a first average value. Similarly, the average value of the target feature values of all second grids in the plurality of grids is calculated to obtain a second average value.
[0067] Step 260: Determine the inter-class variance for the candidate segmentation threshold according to the first proportion, the second proportion, the first mean, and the second mean.
[0068] The global feature mean (the global mean refers to the mean of the target feature values of all grids in the first grid and the second grid) can be calculated according to the first proportion, the second proportion, the first mean and the second mean. The global feature mean G can be calculated according to the following formula 1:
[0069] G = p1*m1+p2*m2; (Formula 1)
[0070] p1 is the first proportion, m1 is the first mean; p2 is the second proportion, m2 is the second mean.
[0071] After that, the inter-class variance σ can be calculated according to the following formula 2 2 :
[0072] σ 2 =p1*(m1-G) 2 +p2*(m2-G) 2 ; (Formula 2)
[0073] Step 270, with the goal of maximizing the inter-class variance, iteratively update the candidate segmentation threshold until the iteration end condition is reached, and the candidate segmentation threshold that reaches the iteration end condition is used as the ground point cloud segmentation threshold.
[0074] Since the determination of the first grid and the second grid as above depends on the candidate segmentation threshold, correspondingly, the first proportion, the first mean, the second proportion, and the second mean are all affected by the candidate segmentation threshold. If the candidate segmentation threshold changes, the determined first grid and the second grid may change, and the first proportion, the first mean, the second proportion, and the second mean change accordingly, thereby causing the inter-class variance to change. Therefore, the candidate segmentation threshold can be iteratively updated to maximize the inter-class variance.
[0075] After the candidate segmentation threshold is iteratively updated once, the iteratively updated candidate segmentation threshold is used as a new candidate segmentation threshold, and the inter-class variance is recalculated according to the new candidate segmentation threshold according to the process shown in steps 220 to 260 above, until the iteration end condition is met.
[0076] The iteration end condition may be that the inter-class variance reaches a maximum value, or that the number of iterative updates of the candidate segmentation threshold reaches a threshold number, which may be set according to actual needs and is not specifically limited here.
[0077] Step 140, determining the ground point cloud segmentation result corresponding to each grid according to the sampling point feature information of each grid and the ground point cloud segmentation threshold, the ground point cloud segmentation result is used to indicate whether the sampling point in the corresponding grid belongs to the ground point cloud or the non-ground point cloud.
[0078] For each grid, the features in the sampling point feature information of the grid are weighted to obtain the target feature value of the grid; if the target feature value of the grid is less than the ground point cloud segmentation threshold, the ground point cloud segmentation result corresponding to the grid is determined to be the result indicating that the sampling points in the grid belong to the ground point cloud, that is, all the sampling points in the grid belong to the ground point cloud; if the target feature value of the grid is not less than the ground point cloud segmentation threshold, the ground point cloud segmentation result corresponding to the grid is determined to be the result indicating that the sampling points in the grid belong to the non-ground point cloud, that is, all the sampling points in the grid belong to the non-ground point cloud.
[0079] In some embodiments, after step 140, the ground point cloud segmentation result can be used to indicate that the sampling points in the corresponding grid belong to the ground point cloud, and the sampling points in the obtained aggregation result are the ground point cloud representing the ground in the point cloud data. And the ground point cloud segmentation result can be used to indicate that the sampling points in the corresponding grid belong to the ground point cloud, and the sampling points in the obtained aggregation result are the non-ground point cloud representing the non-ground in the point cloud data.
[0080] In the present application, after the sampling points in the point cloud data are rasterized, the sampling point feature information of the grid is extracted based on the sampling points located in each grid, and the inter-class variance maximization processing is performed based on the sampling point feature information of multiple grids to determine the ground point cloud segmentation threshold. Since the determined ground point cloud segmentation threshold can maximize the inter-class variance, the ground point cloud segmentation threshold can be used as the boundary between the ground point cloud and the non-ground point cloud. Moreover, the ground point cloud segmentation threshold suitable for the point cloud data is determined based on the feature information of each sampling point after the point cloud data is rasterized, which ensures the applicability and adaptability of the ground point cloud segmentation threshold to the point cloud data for ground point cloud segmentation. Therefore, the method of the present application can effectively improve the accuracy of subsequent ground point cloud segmentation according to the ground point cloud segmentation threshold and the sampling point feature information of each grid.
[0081] Moreover, based on the ground point cloud segmentation threshold, at least two of the sampling point height features, the sampling point gradient features and the sampling point fitting error features are used to comprehensively perform ground point cloud segmentation. The sampling point height features, the sampling point gradient features and the sampling point fitting error features are all features that are strongly correlated with ground point cloud segmentation. Compared with using only one feature, the present application uses more features and can more accurately reflect the characteristics of the sampling points in the grid. Moreover, the features used are targeted at ground point cloud segmentation and can improve the accuracy of subsequent ground point cloud segmentation.
[0082] In some embodiments, the Figure 3 The process shown determines the gradient characteristics of the sampling points of each grid, which are described in detail as follows:
[0083] Step 310, calculating the position deviation between each sampling point in the grid and the reference point.
[0084] The reference point may be a laser radar from which the point cloud data originates. It may also be the origin of the coordinate system, which is not specifically limited here. The calculated position deviation may be the distance between the sampling point and the reference point, or the angle between the sampling point and the reference point.
[0085] Step 320 , sorting the multiple sampling points located in the grid according to the position deviation to obtain a reference sorting.
[0086] In some embodiments, if the position deviation is the distance between the two, the multiple sampling points in the grid may be sorted in order of distance from large to small or from small to large to obtain a reference sorting.
[0087] In some embodiments, if the position deviation is the angle between the two, the multiple sampling points in the grid may be sorted in order of angle from large to small or from small to large to obtain a reference sorting.
[0088] Step 330: Determine the height difference between the two adjacent sampling points and the distance between the two adjacent sampling points according to the position information of the two adjacent sampling points in the reference sorting.
[0089] If the location information of the sampling points includes the polar coordinates of the sampling points, the distance between two adjacent sampling points may be the radial distance between the two adjacent sampling points in the polar coordinate system, or may be the angular difference between the two adjacent sampling points.
[0090] The location information of the sampling points includes the Cartesian coordinates of the sampling points. The Cartesian coordinates of two adjacent sampling points may be used to perform Euclidean distance calculation to obtain the distance between the two adjacent sampling points.
[0091] Step 340 , performing gradient calculation according to the height difference between two adjacent sampling points and the distance between two adjacent sampling points, to obtain the gradient features corresponding to the two adjacent sampling points.
[0092] The height difference between two adjacent sampling points can be divided by the distance between the two adjacent sampling points to obtain the gradient features corresponding to the two adjacent sampling points. According to the above steps 330-340, the gradient features corresponding to any two adjacent sampling points in the reference order can be determined.
[0093] Step 350 , determining the sampling point gradient features of the grid according to the gradient features corresponding to two adjacent sampling points in the reference sorting.
[0094] The gradient features corresponding to two adjacent sampling points can be regarded as the gradient features of the latter sampling point of the two adjacent sampling points. On this basis, at least one of the average gradient, gradient difference, maximum gradient, and minimum gradient can be determined according to the gradient features of multiple sampling points in the reference sorting as the gradient features of the sampling points of the grid.
[0095] In other embodiments, the Figure 4 The process shown determines the gradient characteristics of the sampling points of each grid, which are described in detail as follows:
[0096] Step 410, obtaining a fitting model corresponding to the grid, wherein the fitting model corresponding to the grid is obtained by fitting a plurality of sampling points located in the grid. The process of determining the fitting model is described above and will not be repeated here.
[0097] Step 420, calculating the normal vector of the surface of the fitting model at each sampling point in the grid according to the position information of each sampling point in the grid.
[0098] Step 430, determining the sampling point gradient characteristics of the grid according to the normal vector of the surface of the fitting model at each sampling point in the grid.
[0099] The normal vector of the surface of the fitted model at each sampling point in the grid reflects the gradient in multiple directions at the sampling point, such as the gradient of each coordinate axis in the Cartesian coordinate system. The gradient along the height direction at the sampling point can be obtained from the normal vector of the surface of the fitted model at each sampling point in the grid.
[0100] Similarly, according to the gradient of all sampling points in the grid along the height direction, at least one of the average gradient, gradient difference, maximum gradient, and minimum gradient can be determined as the gradient feature of the sampling points of the grid.
[0101] In some embodiments, Figure 5 As shown, before step 110, the method further includes the following steps 510 to 540:
[0102] Step 510: Acquire original point cloud data. The original point cloud data may be original point cloud data collected by a laser radar, that is, point cloud data output by the laser radar.
[0103] Step 520 , determining the Gaussian kernel size according to the distribution density of the sampling points in the original point cloud data, wherein the Gaussian kernel size is positively correlated with the distribution density.
[0104] The distribution density of the sampling points in the original point cloud data can be represented by the number of sampling points per unit volume. Correspondingly, the smallest geometric body (e.g., the smallest cuboid) that surrounds all the sampling points in the original point cloud data can be determined, and then the total number of sampling points in the original point cloud data is divided by the volume of the determined smallest geometric body to obtain the distribution density of the sampling points in the original point cloud data.
[0105] The Gaussian kernel size is used to limit the size of the Gaussian kernel, that is, the size of the Gaussian kernel. In the present application, the Gaussian kernel can be a three-dimensional Gaussian kernel, and the size of the Gaussian kernel is an odd-sized cube, such as 3×3×3, 5×5×5, 7×7×7, etc.
[0106] In some embodiments, the correspondence between the point cloud distribution density and the Gaussian kernel size can be pre-set, and the correspondence satisfies the positive correlation between the Gaussian kernel size and the point cloud distribution density; after determining the distribution density of the sampling points in the original point cloud data, the Gaussian kernel size corresponding to the distribution density can be determined.
[0107] Step 530, determining the Gaussian kernel standard deviation according to the noise proportion of the original point cloud data; wherein the Gaussian kernel standard deviation is positively correlated with the noise proportion.
[0108] The Gaussian kernel standard deviation is used to limit the standard deviation of the Gaussian kernel. The noise ratio of the original point cloud data refers to the ratio of the number of noise sampling points in the original point cloud data. The number of noise sampling points in the original point cloud data can be counted to determine the noise ratio.
[0109] Step 540 , filtering the original point cloud data using a Gaussian kernel whose size is the Gaussian kernel size and whose standard deviation is the Gaussian kernel standard deviation, to obtain the position information of each sampling point.
[0110] For the Gaussian kernel, when the size and standard deviation of the Gaussian kernel are determined, the Gaussian expression of the Gaussian kernel (i.e., the Gaussian kernel function) is determined accordingly. The original position information of each sampling point in the original point cloud data can be filtered according to the Gaussian kernel function of the Gaussian kernel whose size is the Gaussian kernel size and whose standard deviation is the Gaussian kernel standard deviation, to obtain the position information of each sampling point.
[0111] In some embodiments, step 540 includes: traversing the sampling points in the original point cloud data, and taking the currently traversed sampling point as the target sampling point; determining the neighborhood range of the target sampling point according to the Gaussian kernel size; calculating the weight of each neighborhood sampling point in the neighborhood range for the target sampling point according to a Gaussian kernel whose size is the Gaussian kernel size and whose standard deviation is the Gaussian kernel standard deviation; performing weighted processing according to the coordinates of each neighborhood sampling point in the neighborhood range and the weight of each neighborhood sampling point for the target sampling point to obtain a reference coordinate; updating the coordinates of the target sampling point to the reference coordinate to obtain the position information of the target sampling point.
[0112] As described above, the area defined by the Gaussian kernel size is a cubic area, and the cubic area centered at the target sampling point and having a size defined by the Gaussian kernel size can be used as the neighborhood range of the target sampling point.
[0113] After determining a Gaussian kernel whose size is the Gaussian kernel size and whose standard deviation is the Gaussian kernel standard deviation, the weights of each neighborhood sampling point in the neighborhood range of the target sampling point for the target sampling point are calculated according to the Gaussian kernel function of the determined Gaussian kernel.
[0114] The size and standard deviation of the Gaussian kernel determine the range of smoothing. A larger Gaussian kernel can provide a wider range of blurring effects and is suitable for large-scale smoothing or noise removal. A smaller Gaussian kernel is suitable for retaining more details and performing light smoothing. The standard deviation of the Gaussian kernel controls the shape of the Gaussian kernel function, which in turn affects the strength of smoothing. The larger the standard deviation, the flatter the peak of the Gaussian kernel function and the more significant the smoothing effect.
[0115] In the above embodiment, according to the distribution density and noise level (noise ratio) of the original point cloud data, the Gaussian kernel used for filtering the original point cloud data is determined according to the principle that the Gaussian kernel size is positively correlated with the distribution density, and the Gaussian kernel standard deviation is positively correlated with the noise ratio. In this way, the Gaussian kernel of the distribution density and noise level of the original point cloud data is used to perform targeted smoothing on the sampling points in the original point cloud data, which can reduce the noise in the point cloud data and smooth the point cloud surface, and realizes the adaptive selection of the Gaussian kernel for filtering according to the characteristics of the original point cloud data itself, thereby ensuring the accuracy of the position information of the sampling points obtained after filtering, and facilitating the subsequent ground point cloud segmentation.
[0116] Figure 6 is a flow chart of a ground point cloud segmentation method according to an embodiment of the present application, such as Figure 6 As shown, the method includes:
[0117] Step 610: Perform Gaussian filtering on the original point cloud data to obtain point cloud data.
[0118] For details on the implementation of Gaussian filtering, see Figure 5 Description of the corresponding embodiment.
[0119] Step 620: Perform polar coordinate rasterization processing on the sampling points in the point cloud data.
[0120] In step 620, the Cartesian coordinates of each sampling point in the point cloud data are transformed to obtain the polar coordinates of each sampling point, and the grid where each sampling point is located is determined according to the polar coordinates of the sampling point and multiple grids preset in the polar coordinate space. All sampling points located in one grid can be regarded as a point cloud cluster.
[0121] Step 630, determining the average height, height difference, sampling point gradient characteristics and sampling point fitting error characteristics of each grid according to the sampling points in each grid.
[0122] Step 640 , adaptively determining the ground point cloud segmentation threshold according to the average height, height difference, gradient and fitting error of the multiple grids.
[0123] Step 650, determining the sampling points belonging to the ground point cloud according to the ground point cloud segmentation threshold and the average height, height difference, gradient and fitting error of each grid.
[0124] The specific implementation details of steps 630-650 are described above and will not be repeated here.
[0125] The present application adaptively determines the ground point cloud segmentation threshold for the current point cloud data through the features of multiple grids in multiple dimensions (i.e., at least two of the sampling point height features, sampling point gradient features, and sampling point fitting error features), and subsequently, performs ground point cloud segmentation based on the ground point cloud segmentation threshold and at least two of the sampling point height features, sampling point gradient features, and sampling point fitting error features of the fused grids. It can be applied to point cloud data collected from complex scene areas such as road edges, higher slopes, and unstructured roads. It can accurately segment ground point clouds and has good scene generalization, providing accurate underlying environmental information for subsequent non-ground target detection, and can improve the environmental perception ability of intelligent vehicles. Fusion of features in multiple dimensions as the basis for ground point cloud segmentation can effectively improve the accuracy of ground point cloud segmentation compared to a single feature.
[0126] The following describes an apparatus embodiment of the present application, which can be used to execute the method in the above-mentioned embodiment of the present application. For details not disclosed in the apparatus embodiment of the present application, please refer to the above-mentioned method embodiment of the present application.
[0127] Figure 7 is a block diagram of a ground point cloud segmentation device according to an embodiment of the present application, such as Figure 7 As shown, the ground point cloud segmentation device includes: a first determination module 710, which is used to determine the grid where each sampling point is located according to the position information of each sampling point in the point cloud data and a plurality of preset grids; a feature determination module 720, which is used to determine the sampling point feature information of each grid according to the sampling points located in the same grid, and the sampling point feature information includes at least two of the sampling point height feature, the sampling point gradient feature and the sampling point fitting error feature; a segmentation threshold determination module 730, which is used to perform inter-class variance maximization processing according to the sampling point feature information of the plurality of grids to determine the ground point cloud segmentation threshold; a segmentation result determination module 740, which is used to determine the ground point cloud segmentation result corresponding to each grid according to the sampling point feature information of each grid and the ground point cloud segmentation threshold, and the ground point cloud segmentation result is used to indicate whether the sampling point in the corresponding grid belongs to the ground point cloud or belongs to the non-ground point cloud.
[0128] In some embodiments, the sampling point feature information includes sampling point height features, sampling point gradient features and sampling point fitting error features; the feature determination module 720 includes: a first feature determination unit, used to determine the sampling point height features of the grid according to the heights of multiple sampling points located in the grid; a second feature determination unit, used to perform gradient calculation according to the position information of multiple sampling points located in the grid, and determine the sampling point gradient features of the grid; a fitting unit, used to fit multiple sampling points located in the grid to obtain a fitting model; a third feature determination unit, used to determine the sampling point fitting error features of the grid according to the distance from each sampling point located in the grid to the fitting model.
[0129] In some embodiments, the second feature determination unit includes: a position deviation determination unit, which is used to calculate the position deviation between each sampling point in the grid and the reference point; a sorting unit, which is used to sort multiple sampling points located in the grid according to the position deviation to obtain a reference sorting; a height difference and distance determination unit, which is used to determine the height difference between two adjacent sampling points and the distance between two adjacent sampling points according to the position information of two adjacent sampling points in the reference sorting; a gradient calculation unit, which is used to perform gradient calculation according to the height difference between two adjacent sampling points and the distance between two adjacent sampling points to obtain the gradient features corresponding to the two adjacent sampling points; the first determination unit is used to determine the gradient features of the sampling points of the grid according to the gradient features corresponding to the two adjacent sampling points in the reference sorting.
[0130] In other embodiments, the second feature determination unit includes: a fitting model acquisition unit, used to acquire the fitting model corresponding to the grid, the fitting model corresponding to the grid is obtained by fitting multiple sampling points located in the grid; a normal vector calculation unit, used to calculate the normal vector of the surface of the fitting model at each sampling point in the grid according to the position information of each sampling point in the grid; and a second determination unit, used to determine the sampling point gradient characteristics of the grid according to the normal vector of the surface of the fitting model at each sampling point in the grid.
[0131] In some embodiments, the segmentation threshold determination module 730 includes: a target feature value determination unit, which is used to perform weighted processing on the features included in the feature information of the sampling points of each grid to obtain the target feature value of each grid; a division unit, which is used to divide the sampling points in each grid into ground point clouds according to the target feature value of each grid and the candidate segmentation threshold, and determine the candidate ground point cloud segmentation results of each grid; a third determination unit, which is used to determine the first grid and the second grid among the multiple grids according to the candidate ground point cloud segmentation results of each grid in the multiple grids; the first grid refers to the grid that the corresponding candidate ground point cloud segmentation result indicates belongs to the ground point cloud; the second grid refers to the grid that the corresponding candidate ground point cloud segmentation result indicates A grid belonging to a non-ground point cloud is indicated; a proportion determining unit is used to determine a first proportion of a first grid among the multiple grids and to determine a second proportion of a second grid among the multiple grids; a mean determining unit is used to determine a first mean according to a target feature value of the first grid among the multiple grids; and to determine a second mean according to the target feature value of the second grid among the multiple grids; an inter-class variance determining unit is used to determine the inter-class variance for a candidate segmentation threshold according to the first proportion, the second proportion, the first mean, and the second mean; an iterative updating unit is used to iteratively update the candidate segmentation threshold with the goal of maximizing the inter-class variance until an iteration end condition is reached, and the candidate segmentation threshold that reaches the iteration end condition is used as the ground point cloud segmentation threshold.
[0132] In some embodiments, the segmentation result determination module 740 includes: a target feature value determination unit, which is used to weight the features in the sampling point feature information of the grid to obtain the target feature value of the grid; a result determination unit, which is used to determine that the ground point cloud segmentation result corresponding to the grid is a result indicating that the sampling points in the grid belong to the ground point cloud if the target feature value of the grid is less than the ground point cloud segmentation threshold; if the target feature value of the grid is not less than the ground point cloud segmentation threshold, the ground point cloud segmentation result corresponding to the grid is a result indicating that the sampling points in the grid belong to the non-ground point cloud.
[0133] In some embodiments, the location information includes polar coordinates; the first determination module 710 is further used to: determine the grid where each sampling point is located according to the polar coordinates of each sampling point in the point cloud data and multiple grids preset in the polar coordinate space.
[0134] In some embodiments, the ground point cloud segmentation device also includes: an acquisition module for acquiring original point cloud data; a Gaussian kernel size determination module for determining the Gaussian kernel size according to the distribution density of the sampling points in the original point cloud data, wherein the Gaussian kernel size is positively correlated with the distribution density; a Gaussian kernel standard deviation determination unit for determining the Gaussian kernel standard deviation according to the noise proportion of the original point cloud data, wherein the Gaussian kernel standard deviation is positively correlated with the noise proportion; a filtering processing unit for filtering the original point cloud data using a Gaussian kernel whose size is the Gaussian kernel size and whose standard deviation is the Gaussian kernel standard deviation, to obtain the position information of each sampling point.
[0135] In some embodiments, the filtering processing unit includes: a traversal unit, which is used to traverse the sampling points in the original point cloud data and use the currently traversed sampling point as the target sampling point; a neighborhood range determination unit, which is used to determine the neighborhood range of the target sampling point according to the Gaussian kernel size; a weight determination unit, which is used to calculate the weight of each neighborhood sampling point in the neighborhood range for the target sampling point according to a Gaussian kernel whose size is the Gaussian kernel size and whose standard deviation is the Gaussian kernel standard deviation; a reference coordinate determination unit, which is used to perform weighted processing according to the coordinates of each neighborhood sampling point in the neighborhood range and the weight of each neighborhood sampling point for the target sampling point to obtain a reference coordinate; and a coordinate updating unit, which is used to update the coordinates of the target sampling point to the reference coordinate to obtain the position information of the target sampling point.
[0136] Figure 8 1 is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. The electronic device may be a device for executing the ground point cloud segmentation method provided by the present application, such as a server.
[0137] like Figure 8As shown, the electronic device may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and optionally, the user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may optionally be a storage device independent of the aforementioned processor 1001. Those skilled in the art will understand that Figure 8 The structure of the electronic device shown in the figure does not constitute a limitation of the electronic device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.
[0138] like Figure 8 As shown, the memory 1005 as a computer-readable storage medium may include an operating system, a network communication module, a user interface module, and a program for implementing a ground point cloud segmentation method.
[0139] exist Figure 8 In the electronic device shown, the network interface 1004 is mainly used to communicate with other devices. The user interface 1003 is mainly used to connect to the client (user end) and communicate data with the client; and the processor 1001 can be used to call the program for implementing the ground point cloud segmentation method stored in the memory 1005, and execute the steps of the ground point cloud segmentation method in any of the above method embodiments.
[0140] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the ground point cloud segmentation method in any of the above method embodiments is implemented.
[0141] According to one aspect of an embodiment of the present application, a computer program product is provided, which includes computer instructions. When the computer instructions are executed by a processor, the ground point cloud segmentation method in any of the above method embodiments is implemented.
[0142] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not, in some cases, constitute limitations on the units themselves.
[0143] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the implementation methods of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation methods of the present application.
[0144] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application.
[0145] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.
Claims
1. A ground point cloud segmentation method, characterized in that: include: Determine the grid where each sampling point is located according to the position information of each sampling point in the point cloud data and a plurality of preset grids; Determine sampling point feature information of each grid according to sampling points located in the same grid, wherein the sampling point feature information includes at least two items of sampling point height feature, sampling point gradient feature and sampling point fitting error feature; Performing inter-class variance maximization processing according to the sampling point feature information of the plurality of grids to determine the ground point cloud segmentation threshold; According to the sampling point feature information of each grid and the ground point cloud segmentation threshold, the ground point cloud segmentation result corresponding to each grid is determined, and the ground point cloud segmentation result is used to indicate whether the sampling point in the corresponding grid belongs to the ground point cloud or the non-ground point cloud.
2. The method according to claim 1, characterized in that The sampling point feature information includes sampling point height feature, sampling point gradient feature and sampling point fitting error feature; The step of determining the sampling point feature information of each grid according to the sampling points located in the same grid includes: The following processing is performed on each grid in the plurality of grids: Determining height characteristics of sampling points of the grid according to the heights of a plurality of sampling points located in the grid; Performing gradient calculation based on position information of a plurality of sampling points in the grid to determine gradient features of the sampling points of the grid; Fitting a plurality of sampling points located in the grid to obtain a fitting model; The sampling point fitting error characteristics of the grid are determined according to the distance from each sampling point in the grid to the fitting model.
3. The method according to claim 2, characterized in that The step of performing gradient calculation according to the position information of a plurality of sampling points in the grid to determine the gradient features of the sampling points of the grid includes: Calculating the position deviation between each sampling point and the reference point in the grid; According to the position deviation, a plurality of sampling points located in the grid are sorted to obtain a reference sorting; Determine, according to the position information of two adjacent sampling points in the reference sorting, a height difference between the two adjacent sampling points and a distance between the two adjacent sampling points; Performing gradient calculation according to the height difference between the two adjacent sampling points and the distance between the two adjacent sampling points to obtain gradient features corresponding to the two adjacent sampling points; The sampling point gradient features of the grid are determined according to the gradient features corresponding to two adjacent sampling points in the reference sorting.
4. The method according to claim 2, characterized in that: The step of performing gradient calculation according to the position information of a plurality of sampling points in the grid to determine the gradient features of the sampling points of the grid includes: Acquire a fitting model corresponding to the grid, where the fitting model corresponding to the grid is obtained by fitting a plurality of sampling points located in the grid; Calculating the normal vector of the surface of the fitting model at each sampling point in the grid according to the position information of each sampling point in the grid; The sampling point gradient features of the grid are determined according to the normal vectors of the surface of the fitting model at each sampling point in the grid.
5. The method according to claim 1, characterized in that The method of performing inter-class variance maximization processing according to the sampling point feature information of multiple grids to determine the ground point cloud segmentation threshold includes: Performing weighted processing on the features included in the sampling point feature information of each grid to obtain a target feature value of each grid; According to the target feature value and the candidate segmentation threshold of each grid, the sampling points in each grid are divided into ground point clouds to determine the candidate ground point cloud segmentation results of each grid; Determine a first grid and a second grid among the multiple grids according to the candidate ground point cloud segmentation result of each of the multiple grids; the first grid refers to a grid that the corresponding candidate ground point cloud segmentation result indicates belongs to a ground point cloud; the second grid refers to a grid that the corresponding candidate ground point cloud segmentation result indicates belongs to a non-ground point cloud; Determine a first proportion of the first grid in the plurality of grids and determine a second proportion of the second grid in the plurality of grids; Determine a first mean value according to a target feature value of a first grid among the plurality of grids; and determine a second mean value according to a target feature value of a second grid among the plurality of grids; Determine an inter-class variance for the candidate segmentation threshold according to the first proportion, the second proportion, the first mean, and the second mean; With the goal of maximizing the inter-class variance, the candidate segmentation threshold is iteratively updated until an iteration end condition is reached, and the candidate segmentation threshold that reaches the iteration end condition is used as the ground point cloud segmentation threshold.
6. The method according to claim 1, characterized in that Determining the ground point cloud segmentation result corresponding to each grid according to the sampling point feature information of each grid and the ground point cloud segmentation threshold comprises: The following processing is performed for each of the grids: Performing weighted processing on the features in the sampling point feature information of the grid to obtain the target feature value of the grid; If the target feature value of the grid is less than the ground point cloud segmentation threshold, determining that the ground point cloud segmentation result corresponding to the grid is a result indicating that the sampling points in the grid belong to the ground point cloud; If the target feature value of the grid is not less than the ground point cloud segmentation threshold, it is determined that the ground point cloud segmentation result corresponding to the grid is a result indicating that the sampling points in the grid belong to a non-ground point cloud.
7. The method according to claim 1, characterized in that The location information includes polar coordinates; Determining the grid where each sampling point is located according to the position information of each sampling point in the point cloud data and a plurality of preset grids includes: The grid where each sampling point is located is determined according to the polar coordinates of each sampling point in the point cloud data and a plurality of grids preset in the polar coordinate space.
8. The method according to claim 1, characterized in that Before determining the grid where each sampling point is located based on the position information of each sampling point in the point cloud data and a plurality of preset grids, the method further includes: Get the original point cloud data; Determine a Gaussian kernel size according to the distribution density of the sampling points in the original point cloud data, wherein the Gaussian kernel size is positively correlated with the distribution density; Determine the Gaussian kernel standard deviation according to the noise proportion of the original point cloud data; wherein the Gaussian kernel standard deviation is positively correlated with the noise proportion; The original point cloud data is filtered by a Gaussian kernel whose size is the Gaussian kernel size and whose standard deviation is the Gaussian kernel standard deviation to obtain the position information of each sampling point.
9. The method according to claim 8, characterized in that The filtering process is performed on the original point cloud data by using a Gaussian kernel whose size is the Gaussian kernel size and whose standard deviation is the Gaussian kernel standard deviation to obtain the position information of each sampling point, including: Traversing the sampling points in the original point cloud data, and taking the currently traversed sampling point as the target sampling point; Determine the neighborhood range of the target sampling point according to the Gaussian kernel size; Calculate the weight of each neighborhood sampling point in the neighborhood range for the target sampling point according to a Gaussian kernel whose size is the Gaussian kernel size and whose standard deviation is the Gaussian kernel standard deviation; Performing weighted processing according to the coordinates of each neighborhood sampling point in the neighborhood range and the weight of each neighborhood sampling point for the target sampling point to obtain a reference coordinate; The coordinates of the target sampling point are updated to the reference coordinates to obtain the position information of the target sampling point.
10. A ground point cloud segmentation device, characterized in that: include: A first determination module, used to determine the grid where each sampling point is located according to the position information of each sampling point in the point cloud data and a plurality of preset grids; A feature determination module, used to determine the sampling point feature information of each grid according to the sampling points located in the same grid, wherein the sampling point feature information includes at least two items of the sampling point height feature, the sampling point gradient feature and the sampling point fitting error feature; The segmentation threshold determination module is used to perform inter-class variance maximization processing based on the sampling point feature information of multiple grids to determine the ground point cloud segmentation threshold; The segmentation result determination module is used to determine the ground point cloud segmentation result corresponding to each grid according to the sampling point feature information of each grid and the ground point cloud segmentation threshold, and the ground point cloud segmentation result is used to indicate whether the sampling point in the corresponding grid belongs to the ground point cloud or the non-ground point cloud.
11. An electronic device, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 9 is implemented.
12. A computer-readable storage medium having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by a processor, the method according to any one of claims 1 to 9 is implemented.