A laser point cloud ground filtering method and device

Abstract

The present invention discloses a laser point cloud ground filtering method and device, which performs cluster analysis on the terrain image of the area to be measured to remove noise and construct a regular grid; interpolates the blank grids to obtain an initial ground model; merges the initial grids based on the initial ground model to obtain a large-scale grid; uses a sliding window to traverse the multiple grids to generate a feature grid; classifies the terrain into sloping terrain and flat terrain; for sloping terrain, obtains a threshold value for filtering according to a two-dimensional profile curve fitting method; for flat terrain, obtains a threshold value for filtering according to a three-dimensional cloth simulation method. The scheme of the present invention integrates the differences in terrain details at different levels as a basis for weighting, and performs weighted calculation on the multi-scale grid feature values, so that it accurately reflects the degree of terrain change. The terrain is classified into sloping and flat land, and the two types of terrain are filtered separately, achieving a high-precision ground filtering effect for complex road terrain.

Description

Technical Field

[0001] The present invention relates to the technical field of road information processing, and in particular to a laser point cloud ground filtering method and device. Background Art

[0002] In recent years, China has vigorously promoted the development of digital and smart cities. Traditional surveying and mapping methods are no longer able to meet the needs of modern information acquisition. 3D laser scanning technology, with its rapid and proactive advantages, has become an effective means of spatial information acquisition and digital development. After acquiring point cloud data through scanning, it requires post-processing, and filtering, which separates ground points from object points, is a key step in post-processing. The filtered results can be used to generate thematic products such as DEMs and also provide auxiliary information for point cloud classification. Therefore, how to efficiently and accurately acquire ground points has attracted the attention and research of numerous scholars both domestically and internationally.

[0003] The existing technology has the following shortcomings: 3D laser point clouds are discrete and disordered, data interpolation and regularization methods affect filtering accuracy, and methods that directly calculate based on point clouds affect filtering speed. In addition, overly complex threshold parameters require professionals with a certain understanding of point cloud processing and algorithm principles to set them to achieve good filtering effects. Although scholars have developed a variety of ground filtering methods, their applicability has always been insufficient. Most of them are only effective for certain specific terrains and cannot meet the ground filtering needs in areas with complex terrain. Summary of the Invention

[0004] The present invention provides a laser point cloud ground filtering method and device for solving the problem of ground filtering in complex terrain.

[0005] According to one aspect of the present invention, a laser point cloud ground filtering method is provided, comprising the following steps:

[0006] The terrain image of the measured area is clustered and analyzed using the spatial distribution of point clouds to remove noise, and a regular grid is constructed based on the maximum value of the point cloud coordinates. After the point cloud is projected, the blank grid is interpolated using the inverse distance weighted interpolation (IDW) method to obtain the initial ground model.

[0007] Performing small-scale slope calculation based on the initial grid of the initial ground model and recording the number of interpolated grids in the slope direction; merging the initial grids to obtain a large-scale grid; using a sliding window to traverse the large-scale grid and the initial grid, analyzing the distribution characteristics of the grids within the window, and generating a feature grid reflecting terrain changes; calculating a feature threshold based on the maximum inter-class variance principle, and classifying the terrain into sloping terrain and flat terrain;

[0008] For sloping terrain, the threshold is obtained for filtering based on the two-dimensional profile curve fitting method; for flat terrain, the threshold is obtained for filtering based on the three-dimensional cloth simulation method; the filtering of different terrains in the entire test area is completed, and high-precision ground points are extracted.

[0009] The method of performing cluster analysis on the terrain image of the area to be measured by using the spatial distribution law of the point cloud to remove noise includes performing cluster analysis on the terrain image of the area to be measured by using the spatial distribution law of the point cloud to remove noise, wherein:

[0010] Choose any point and create a new point set S P , and search for unclustered points in the neighborhood of the point and add them to S P ;

[0011] Traverse the point set S P , repeat the above operation, if no new points are added, then create the next point set;

[0012] Statistics of each point set S P Number of points included N i , if it is less than the threshold N T , then remove the point set S P Various points in it.

[0013] The initial ground model is constructed as follows:

[0014] According to the maximum value of the point cloud coordinates and the width of the largest feature in the area, a regular grid with a side length of c is constructed;

[0015] Calculate the row and column number of the grid to which the point cloud belongs. The grid value is the elevation value of the lowest point inside.

[0016] The IDW method is used to interpolate the blank areas to obtain a complete initial ground model.

[0017] The generating of a characteristic grid reflecting terrain changes includes:

[0018] Merge the small-scale grids into a large-scale grid, the grid side length of which is C = 3c;

[0019] The relative small-scale grid elevation values ​​are z i 、z i+4 , the spacing is twice the length of the grid side, and the slope values ​​S in the four directions of 0°, 45°, 90°, and 135° are calculated respectively. The maximum slope value is selected as the main slope of the grid, and the number of interpolated grids in the direction is counted:

[0020]

[0021] Determine the weight of the neighborhood grid based on the planar distance The distance ΔC from the neighborhood grid to the central grid and the threshold Decide:

[0022]

[0023] Determine the weights of neighboring grids based on slope differences By the neighborhood grid slope S i The difference from the central grid slope S and the ratio of the latter determine:

[0024]

[0025] Determine the weight of the neighborhood grid based on the number of interpolation grids Determined by the number N of small-scale interpolation grids in the main slope direction of the neighborhood grid:

[0026]

[0027] The sliding window is used to traverse the grid, and the slope value S of the neighboring grid at the relative position is calculated. i 、S i+4 , weighted calculation of the eigenvalue F of the central grid;

[0028]

[0029] The terrain is divided into slope terrain and flat terrain, specifically including:

[0030] Perform characteristic value statistics on the grid, calculate the maximum slope and divide it into multiple intervals, and count the number of grids in each interval;

[0031] The eigenvalue F is divided into two parts, F1 and F2, at a fixed interval. The average value μ1 of F1 and the number of grids ω1, the average value μ2 of F2 and the number of grids ω2 are calculated. The average value of the feature of the entire area is μ. The inter-class variance v is calculated. The threshold F corresponding to the maximum v is t This is the characteristic threshold for terrain classification:

[0032] v=ω1(μ1-μ) 2 +ω2(μ2-μ) 2 .

[0033] For the slope terrain, obtaining a threshold value for filtering according to a two-dimensional profile curve fitting method includes:

[0034] Slope terrain filtering is performed based on the two-dimensional profile curve fitting method;

[0035] Obtain the minimum three-dimensional coordinate value of the point cloud, and obtain the cross-sectional information of the slope point cloud and the initial ground model through the coordinate transformation of the rotation offset;

[0036] Fitting is performed based on the point cloud within the profile. During the fitting process, the fitting accuracy is improved by fitting the data in blocks. The highest order of the fitting polynomial is n=5, and the polynomial coefficient is a. i , fitting curve P n (x):

[0037]

[0038] After the initial fitting, the points with a distance from the curve less than the threshold are used as the basis for the next fitting, and several fittings are performed;

[0039] Calculate the distance from the point cloud in the grid where the curve is located to the curve. If the distance is less than the threshold, the point belongs to the ground point.

[0040] For the flat terrain, the threshold is obtained and filtered according to the three-dimensional cloth simulation method, including:

[0041] Invert the point cloud of the flat terrain and place the cloth composed of regular nodes directly above the point cloud;

[0042] The cloth resolution is set to 3 times the point cloud spacing;

[0043] Calculate the position of the node under the action of gravity and tension, and compare it with the position of the associated point cloud; if the node position is lower, it will be adjusted up to the associated point cloud and will not move further;

[0044] The distance between the point cloud and the cloth after it stops falling is calculated. If it is less than the set threshold, it is confirmed that the point belongs to the ground point.

[0045] According to another aspect of the present invention, a laser point cloud ground filtering device is provided, comprising:

[0046] The initial ground model generation unit is used to perform cluster analysis on the terrain image of the test area using the spatial distribution law of the point cloud to remove noise, and to construct a regular grid based on the maximum value of the point cloud coordinates; after the point cloud is projected, the blank grid is interpolated using the inverse distance weighted interpolation (IDW) method to obtain the initial ground model;

[0047] A terrain classification unit is configured to calculate a small-scale slope based on an initial grid of the initial ground model and record the number of interpolated grids in the slope direction; merge the initial grids to obtain a large-scale grid; use a sliding window to traverse the large-scale grid, analyze the distribution characteristics of the grids within the window, and generate a feature grid reflecting terrain changes; calculate a feature threshold based on the maximum inter-class variance principle to classify the terrain into sloping terrain and flat terrain;

[0048] The ground point extraction unit is used to obtain the threshold value for filtering based on the two-dimensional profile curve fitting method for sloping terrain; for flat terrain, it is used to obtain the threshold value for filtering based on the three-dimensional cloth simulation method; it completes the filtering of different terrains in the entire test area and extracts high-precision ground points.

[0049] The initial ground model generating unit includes:

[0050] Cluster analysis and denoising module, which is used to perform cluster analysis and remove noise points based on the spatial distribution of point clouds;

[0051] Regular grid interpolation module, used to interpolate the blank grid using the inverse distance weighted interpolation (IDW) method to obtain the initial ground model;

[0052] The terrain classification unit includes:

[0053] Multi-scale grid feature calculation module, used to construct multi-scale grids and calculate feature values;

[0054] The maximum inter-class variance classification module is used to classify terrain into two categories: slope terrain and flat terrain based on the maximum inter-class variance principle.

[0055] The ground point extraction unit includes:

[0056] Curve fitting slope filtering module, used to filter slope terrain based on two-dimensional profile curve fitting method;

[0057] The cloth simulation flat terrain filtering module is used to filter flat terrain according to the three-dimensional cloth simulation method.

[0058] The technical solution of the present invention is adopted to propose a laser point cloud ground filtering solution, in which the spatial distribution law of the point cloud is used to perform cluster analysis on the terrain image of the test area to remove noise, and a regular grid is constructed according to the maximum value of the point cloud coordinates; after the point cloud is projected, the blank grid is interpolated using the inverse distance weighted interpolation (IDW) method to obtain an initial ground model; a small-scale slope is calculated based on the initial grid of the initial ground model, and the number of interpolated grids in the slope direction is recorded; the initial grids are merged to obtain a large-scale grid; a sliding window is used to traverse the large-scale grid and the initial grid, and the distribution characteristics of the grids in the window are analyzed to generate a feature grid reflecting the terrain changes; the feature threshold is calculated according to the maximum inter-class variance principle, and the terrain is classified into sloping terrain and flat terrain; for sloping terrain, the threshold is obtained according to the two-dimensional profile curve fitting method for filtering; for flat terrain, the threshold is obtained according to the three-dimensional cloth simulation method for filtering; the filtering of different terrains in the entire test area is completed, and high-precision ground points are extracted.

[0059] Compared to traditional single-scale grid-based slope calculations to obtain local terrain features, the solution presented in this paper constructs a multi-scale feature grid. By integrating the differences in terrain details at different levels as a basis for weighting, the multi-scale grid feature values ​​are weighted and calculated to accurately reflect the degree of terrain variation. Existing filtering methods are often only effective for specific terrain types and struggle to ensure filtering accuracy for road scenes with complex terrain. However, this paper utilizes the principle of maximum inter-class variance to classify terrain into slopes and flat areas. It then filters these two types of terrain using two-dimensional profile curve fitting and three-dimensional fabric simulation, achieving high-precision ground filtering for complex road terrain.

[0060] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0062] Figure 1 This is a flow chart of the principle of laser point cloud ground filtering in an embodiment of the present invention;

[0063] Figure 2 Schematic diagram of the laser point cloud ground filtering data processing process in an embodiment of the present invention;

[0064] Figure 3 Schematic diagram of the structure of the laser point cloud ground filtering device in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0066] Figure 1 This is a flowchart of laser point cloud ground filtering in an embodiment of the present invention. Figure 1 As shown in Figure 1, the laser point cloud ground filtering process includes the following steps:

[0067] Step 101: Perform cluster analysis on the terrain image of the test area using the spatial distribution law of the point cloud to remove noise, and construct a regular grid based on the maximum value of the point cloud coordinates; after projecting the point cloud, use the inverse distance weighted interpolation (IDW) method to interpolate the blank grid to obtain an initial ground model.

[0068] In the embodiment of the present invention, cluster analysis is performed using the spatial distribution law of the point cloud to remove noise points, a regular grid is constructed according to the maximum value of the point cloud coordinates, and after the point cloud is projected, the IDW (Inverse Distance Weighted) method is used to interpolate the blank areas to obtain the initial ground model.

[0069] 1) Cluster analysis denoising.

[0070] First, select a point P as the cluster center and establish a point cloud set S P , search for neighboring points P within a certain radius ε of point P i Join S P , and repeat the clustering operation with the neighborhood points as new cluster centers. P When no new points are added to the point cloud, the clustering of the point set is stopped, and then a point is selected from other unclustered point clouds to cluster again. Finally, the point cloud is clustered into multiple point sets, and the number of points N contained in each point set is counted. i , set the quantity threshold N T , remove N i <N T The point set.

[0071] 2) Project the local minimum point onto a regular grid and perform hole interpolation to construct an initial ground model.

[0072] The grid range is determined by the coordinate maximum value of the horizontal point cloud, and the grid side length c When the initial ground model is constructed, the lowest point inside each grid is retained as the ground seed point. For grids in hollow areas, the IDW method is used for interpolation. This method believes that the closer the ground seed point is to the interpolation point, the greater its influence. To avoid low interpolation accuracy, the number of ground seed points within a certain range of the interpolation point should be greater than 4. Therefore, cyclic interpolation is required until all grids are interpolated.

[0073] Step 102: perform small-scale slope calculation based on the initial grid of the initial ground model and record the number of interpolated grids in the slope direction; merge the initial grids to obtain a large-scale grid; use a sliding window to traverse the large-scale grid and the initial grid, analyze the distribution characteristics of the grids in the window, and generate a feature grid that reflects the terrain changes; calculate the feature threshold according to the maximum inter-class variance principle, and classify the terrain into sloping terrain and flat terrain.

[0074] In an embodiment of the present invention, a small-scale slope calculation is performed based on an initial grid, and the number of interpolated grids in the slope direction is recorded. A large-scale grid is obtained by merging the initial grids. A sliding window is used to traverse the grids, and the distribution characteristics of the grids in the window are analyzed to generate a feature grid reflecting the terrain changes. The feature threshold is calculated using the maximum inter-class variance principle, and the terrain is classified into slopes and flat land.

[0075] 1) Multi-scale grid feature calculation

[0076] On the basis of storing the ground seed point elevation z in the small-scale grid, the central grid and the neighborhood grid are merged into a large-scale grid with a grid side length of C = 3c. The slope values ​​S in the four directions of 0°, 45°, 90°, and 135° are calculated respectively, and the maximum slope value is selected as the main slope of the grid.

[0077]

[0078] Based on the storage of local terrain slope s in large-scale grids, a sliding window is used to traverse the grids, and the slope values ​​of neighboring grids are used to weight the calculation of the eigenvalue F of the central grid. This eigenvalue is determined by the spatial distribution characteristics of the grids in the window (plane distance, slope difference, number of interpolation grids, etc.).

[0079] The distance between the central grid and the neighboring grids is different. The smaller the plane distance is, the more it can reflect the slope of the central grid. Therefore, based on the plane distance, the grid weight is determined in the window containing the 3×3 grid. The distance ΔC from the neighborhood grid to the central grid and the threshold Decide.

[0080]

[0081] The slopes of the central grid and the neighboring grids are different. The smaller the slope difference, the more it can reflect the slope of the central grid. Therefore, the grid weight is determined based on the slope difference. By the neighborhood grid slope S i It is determined by the ratio of the difference between the slope S of the central grid and the latter.

[0082]

[0083] The number of small-scale interpolation grids in the neighborhood grid is different. The fewer interpolation grids, the better they can reflect the slope of the central grid. Therefore, the grid weight is determined based on the number of interpolation grids. It is determined by the number N of small-scale interpolation grids in the main slope direction of the neighborhood grid.

[0084]

[0085] According to the neighborhood grid weight formula defined above, the eigenvalue F of the central grid can be calculated.

[0086]

[0087] 2) The maximum inter-class variance principle is used to classify the terrain into slopes and flat lands.

[0088] Perform eigenvalue statistics on the grid, calculate the maximum slope and divide it into multiple intervals, count the number of grids in each interval, and the overall statistical results have two peaks. The OTSU algorithm can be used for segmentation. The specific segmentation algorithm is: divide the eigenvalue F into two parts, F1 and F2, at a fixed interval. Calculate the average value μ1 of F1 and the number of grids ω1, the average value μ2 of F2 and the number of grids ω2, and the characteristic average value of the entire area is μ. Calculate the inter-class variance v and the threshold F corresponding to the maximum v. t This is the characteristic threshold for terrain classification.

[0089] v=ω1(μ1-μ) 2 +ω2(μ2-μ) 2 .

[0090] Step 103: For slope terrain, a threshold value is obtained based on a two-dimensional profile curve fitting method for filtering; for flat terrain, a threshold value is obtained based on a three-dimensional cloth simulation method for filtering; and filtering is completed for different terrains in the entire test area to extract high-precision ground points.

[0091] In the embodiment of the present invention, for slope terrain, a two-dimensional profile curve fitting method is used to obtain a threshold for filtering. For flat terrain, a three-dimensional cloth simulation method is used to obtain a threshold for filtering, thereby completing the filtering of different terrain areas in the entire region and extracting high-precision ground points.

[0092] 1) The slope terrain is filtered using the two-dimensional profile curve fitting method.

[0093] Get the minimum value of the three-dimensional coordinates of the point cloud, obtain the cross-sectional information of the slope point cloud and the initial ground model through the coordinate transformation of the rotation offset, and use the point cloud in the cross-sectional area for fitting. During the fitting process, the slope data is fitted in blocks to improve the fitting accuracy. The highest order term of the fitting polynomial is n = 5, and the corresponding polynomial coefficient a is obtained. i , we can get the fitting curve P n (x).

[0094]

[0095] Since the fitting process is easily affected by noise, after the initial fitting, the points whose distance to the curve is less than the threshold are used as the basis for the next fitting. Multiple fittings are performed to avoid the influence of noise and obtain a more accurate fitting curve. Finally, the distance from the point cloud in the grid where the curve is located to the curve is calculated. When the distance is less than the threshold, the point is considered to be a ground point.

[0096] 2) Use three-dimensional cloth simulation method to filter flat terrain.

[0097] The point cloud of the flat terrain is inverted, and the cloth composed of regular nodes is placed directly above the point cloud. The cloth resolution is set to 3 times the point cloud spacing to simulate the process of node force displacement. First, the node begins to fall under the action of its own gravity, and then the node is adjusted in the vertical direction according to the height difference between the node and the adjacent node to simulate the process of node displacement under the action of tension. During the iteration process, the node needs to be compared with the associated point cloud. If the node is relatively low, the node is moved up to the position of the associated point cloud and fixed.

[0098] When the number of iterations reaches the set number, or the node displacement is less than a certain distance, the cloth simulation ends. At this time, the cloth shape is approximately the terrain shape. The distance from the point cloud to the cloth is calculated. If the distance is less than the threshold, the point cloud is retained as a ground point, otherwise, high-precision ground points are finally obtained.

[0099] The specific implementation process of the embodiment of the present invention can be found in Figure 2 .

[0100] The technical solution of the present invention proposes a laser point cloud ground filtering solution, which uses the spatial distribution law of the point cloud to perform cluster analysis on the terrain image of the test area to remove noise, and constructs a regular grid according to the maximum value of the point cloud coordinates; after the point cloud is projected, the blank grid is interpolated using the inverse distance weighted interpolation (IDW) method to obtain an initial ground model; a small-scale slope is calculated based on the initial grid of the initial ground model, and the number of interpolated grids in the slope direction is recorded; the initial grids are merged to obtain a large-scale grid; a sliding window is used to traverse the large-scale grid and the initial grid, and the distribution characteristics of the grids in the window are analyzed to generate a feature grid reflecting the terrain changes; the feature threshold is calculated according to the maximum inter-class variance principle, and the terrain is classified into sloping terrain and flat terrain; for sloping terrain, the threshold is obtained according to the two-dimensional profile curve fitting method for filtering; for flat terrain, the threshold is obtained according to the three-dimensional cloth simulation method for filtering; the filtering of different terrains in the entire test area is completed, and high-precision ground points are extracted.

[0101] Compared to traditional single-scale grid-based slope calculations to obtain local terrain features, the solution presented in this paper constructs a multi-scale feature grid. By integrating the differences in terrain details at different levels as a basis for weighting, the multi-scale grid feature values ​​are weighted and calculated to accurately reflect the degree of terrain variation. Existing filtering methods are often only effective for specific terrain types and struggle to ensure filtering accuracy for road scenes with complex terrain. However, this paper utilizes the principle of maximum inter-class variance to classify terrain into slopes and flat areas. It then filters these two types of terrain using two-dimensional profile curve fitting and three-dimensional fabric simulation, achieving high-precision ground filtering for complex road terrain.

[0102] In order to realize the above process, the technical solution of the present invention also provides a laser point cloud ground filtering device, such as Figure 3 As shown, the laser point cloud ground filtering device includes:

[0103] The initial ground model generation unit 21 is used to perform cluster analysis on the terrain image of the test area using the spatial distribution law of the point cloud to remove noise, and to construct a regular grid based on the maximum value of the point cloud coordinates; after projecting the point cloud, the blank grid is interpolated using the inverse distance weighted interpolation (IDW) method to obtain the initial ground model;

[0104] The terrain classification unit 22 is configured to calculate a small-scale slope based on the initial grid of the initial ground model and record the number of interpolated grids in the slope direction; merge the initial grids to obtain a large-scale grid; use a sliding window to traverse the large-scale grid, analyze the distribution characteristics of the grids within the window, and generate a feature grid reflecting the terrain changes; calculate the feature threshold according to the maximum inter-class variance principle, and classify the terrain into sloping terrain and flat terrain;

[0105] The ground point extraction unit 23 is used to obtain a threshold value for filtering based on a two-dimensional profile curve fitting method for sloping terrain; and to obtain a threshold value for filtering based on a three-dimensional cloth simulation method for flat terrain. This completes filtering of different terrains in the entire test area and extracts high-precision ground points.

[0106] The initial ground model generating unit 21 includes:

[0107] Cluster analysis and denoising module, which is used to perform cluster analysis and remove noise points based on the spatial distribution of point clouds;

[0108] Regular grid interpolation module, used to interpolate the blank grid using the inverse distance weighted interpolation (IDW) method to obtain the initial ground model;

[0109] The terrain classification unit 22 includes:

[0110] Multi-scale grid feature calculation module, used to construct multi-scale grids and calculate feature values;

[0111] The maximum inter-class variance classification module is used to classify terrain into two categories: slope terrain and flat terrain based on the maximum inter-class variance principle.

[0112] The ground point extraction unit 23 includes:

[0113] Curve fitting slope filtering module, used to filter slope terrain based on two-dimensional profile curve fitting method;

[0114] The cloth simulation flat terrain filtering module is used to filter flat terrain according to the three-dimensional cloth simulation method.

[0115] In summary, the technical solution of the present invention proposes a laser point cloud ground filtering solution, which uses the spatial distribution law of the point cloud to perform cluster analysis on the terrain image of the test area to remove noise, and constructs a regular grid according to the maximum value of the point cloud coordinates; after the point cloud is projected, the blank grid is interpolated using the inverse distance weighted interpolation IDW method to obtain an initial ground model; a small-scale slope is calculated based on the initial grid of the initial ground model, and the number of interpolated grids in the slope direction is recorded; the initial grids are merged to obtain a large-scale grid; a sliding window is used to traverse the large-scale grid and the initial grid, and the distribution characteristics of the grids in the window are analyzed to generate a feature grid reflecting the terrain changes; the feature threshold is calculated according to the maximum inter-class variance principle, and the terrain is classified into slope terrain and flat terrain; for slope terrain, the threshold is obtained according to the two-dimensional profile curve fitting method for filtering; for flat terrain, the threshold is obtained according to the three-dimensional cloth simulation method for filtering; the filtering of different terrains in the entire test area is completed, and high-precision ground points are extracted.

[0116] Compared to traditional single-scale grid-based slope calculations to obtain local terrain features, the solution presented in this paper constructs a multi-scale feature grid. By integrating the differences in terrain details at different levels as a basis for weighting, the multi-scale grid feature values ​​are weighted and calculated to accurately reflect the degree of terrain variation. Existing filtering methods are often only effective for specific terrain types and struggle to ensure filtering accuracy for road scenes with complex terrain. However, this paper utilizes the principle of maximum inter-class variance to classify terrain into slopes and flat areas. It then filters these two types of terrain using two-dimensional profile curve fitting and three-dimensional fabric simulation, achieving high-precision ground filtering for complex road terrain.

[0117] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0118] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0119] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0121] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A laser point cloud ground filtering method, characterized in that: The following steps are involved: The terrain image of the measured area is clustered and analyzed using the spatial distribution of point clouds to remove noise, and a regular grid is constructed based on the maximum value of the point cloud coordinates. After the point cloud is projected, the blank grid is interpolated using the inverse distance weighted interpolation (IDW) method to obtain the initial ground model. Performing small-scale slope calculation based on the initial grid of the initial ground model and recording the number of interpolated grids in the slope direction; merging the initial grids to obtain a large-scale grid; using a sliding window to traverse the large-scale grid and the initial grid, analyzing the distribution characteristics of the grids within the window, and generating a feature grid reflecting terrain changes; calculating a feature threshold based on the maximum inter-class variance principle, and classifying the terrain into sloping terrain and flat terrain; For slope terrain, the threshold is obtained by two-dimensional profile curve fitting method for filtering; for flat terrain, the threshold is obtained by three-dimensional cloth simulation method for filtering; Complete the filtering of different terrains in the entire test area and extract high-precision ground points; The generating of a characteristic grid reflecting terrain changes includes: Merge the small-scale grids into a large-scale grid, the grid side length of which is C = 3c; The relative small-scale grid elevation values ​​are z i 、z i+4 , the spacing is twice the length of the grid side, and the slope values ​​S in the four directions of 0°, 45°, 90°, and 135° are calculated respectively. The maximum slope value is selected as the main slope of the grid, and the number of interpolated grids in the direction is counted: Determine the weight of the neighborhood grid based on the planar distance The distance ΔC from the neighborhood grid to the central grid and the threshold Decide: Determine the weights of neighboring grids based on slope differences By the neighborhood grid slope S i The difference from the central grid slope S and the ratio of the latter determine: Determine the weight of the neighborhood grid based on the number of interpolation grids Determined by the number N of small-scale interpolation grids in the main slope direction of the neighborhood grid: The sliding window is used to traverse the grid, and the slope value S of the neighboring grid at the relative position is calculated. i 、S i+4 , weighted calculation of the eigenvalue F of the central grid; The terrain is divided into slope terrain and flat terrain, specifically including: Perform eigenvalue statistics on the grid, calculate the maximum slope and divide it into multiple intervals, and count the number of grids in each interval; The eigenvalue F is divided into two parts, F1 and F2, at fixed intervals. The average value μ1 of F1 and the number of grids ω1 are calculated, and the average value μ2 of F2 and the number of grids ω2 are calculated. The characteristic average value of the entire area is μ. The inter-class variance v is calculated. The threshold Ft corresponding to the maximum v is the characteristic threshold of terrain classification: v=ω1(μ1-μ) 2 +ω2(μ2-μ) 2 。 2. A laser point cloud ground filtering method according to claim 1, characterized in that: The method of performing cluster analysis on the terrain image of the area to be measured by using the spatial distribution law of the point cloud to remove noise includes performing cluster analysis on the terrain image of the area to be measured by using the spatial distribution law of the point cloud to remove noise, wherein: Choose any point and create a new point set S P , and search for unclustered points in the neighborhood of the point and add them to S P ; Traverse the point set S P , repeat the above operation, if no new points are added, then create the next point set; Statistics of each point set S P Number of points included N i , if it is less than the threshold N T , then remove the point set S P Various points in it.

3. A laser point cloud ground filtering method according to claim 2, characterized in that: The initial ground model is constructed as follows: According to the maximum value of the point cloud coordinates and the width of the largest feature in the area, a regular grid with a side length of c is constructed; Calculate the row and column number of the grid to which the point cloud belongs. The grid value is the elevation value of the lowest point inside. The IDW method is used to interpolate the blank areas to obtain a complete initial ground model.

4. The laser point cloud ground filtering method according to claim 1, characterized in that: For the slope terrain, obtaining a threshold value for filtering according to a two-dimensional profile curve fitting method includes: Slope terrain filtering is performed based on the two-dimensional profile curve fitting method; Obtain the minimum three-dimensional coordinate value of the point cloud, and obtain the cross-sectional information of the slope point cloud and the initial ground model through the coordinate transformation of the rotation offset; Fitting is performed based on the point cloud within the profile. During the fitting process, the fitting accuracy is improved by fitting the data in blocks. The highest order of the fitting polynomial is n=5, and the polynomial coefficient is a. i , fitting curve P n (x): After the initial fitting, the points with a distance from the curve less than the threshold are used as the basis for the next fitting, and several fittings are performed; Calculate the distance from the point cloud in the grid where the curve is located to the curve. If the distance is less than the threshold, the point belongs to the ground point.

5. A laser point cloud ground filtering method according to claim 4, characterized in that: For the flat terrain, the threshold is obtained according to the three-dimensional cloth simulation method for filtering, including: Invert the point cloud of the flat terrain and place the cloth composed of regular nodes directly above the point cloud; The cloth resolution is set to 3 times the point cloud spacing; Calculate the position of the node under the action of gravity and tension, and compare it with the position of the associated point cloud; if the node position is lower, it will be adjusted up to the associated point cloud and will not move further; The distance between the point cloud and the cloth after it stops falling is calculated. If it is less than the set threshold, it is confirmed that the point belongs to the ground point.

6. A laser point cloud ground filtering device, characterized in that: include: The initial ground model generation unit is used to perform cluster analysis on the terrain image of the test area using the spatial distribution law of the point cloud to remove noise, and to construct a regular grid based on the maximum value of the point cloud coordinates; After point cloud projection, the blank grid is interpolated using the inverse distance weighted interpolation (IDW) method to obtain the initial ground model. A terrain classification unit is configured to calculate a small-scale slope based on an initial grid of the initial ground model and record the number of interpolated grids in the slope direction; merge the initial grids to obtain a large-scale grid; use a sliding window to traverse the large-scale grid, analyze the distribution characteristics of the grids within the window, and generate a characteristic grid reflecting terrain changes; The feature threshold is calculated based on the maximum inter-class variance principle to classify the terrain into slope and flat terrain. The generating of a characteristic grid reflecting terrain changes includes: Merge the small-scale grids into a large-scale grid, the grid side length of which is C = 3c; The relative small-scale grid elevation values ​​are z i 、z i+4 , the spacing is twice the length of the grid side, and the slope values ​​S in the four directions of 0°, 45°, 90°, and 135° are calculated respectively. The maximum slope value is selected as the main slope of the grid, and the number of interpolated grids in the direction is counted: Determine the weight of the neighborhood grid based on the planar distance The distance ΔC from the neighborhood grid to the central grid and the threshold Decide: Determine the weights of neighboring grids based on slope differences By the neighborhood grid slope S i The difference from the central grid slope S and the ratio of the latter determine: Determine the weight of the neighborhood grid based on the number of interpolation grids Determined by the number N of small-scale interpolation grids in the main slope direction of the neighborhood grid: The sliding window is used to traverse the grid, and the slope value S of the neighboring grid at the relative position is calculated. i 、S i+4 , weighted calculation of the eigenvalue F of the central grid; The terrain is divided into slope terrain and flat terrain, specifically including: Perform eigenvalue statistics on the grid, calculate the maximum slope and divide it into multiple intervals, and count the number of grids in each interval; The eigenvalue F is divided into two parts, F1 and F2, at a fixed interval. The average value μ1 of F1 and the number of grids ω1, the average value μ2 of F2 and the number of grids ω2 are calculated. The average value of the feature of the entire area is μ. The inter-class variance v is calculated. The threshold F corresponding to the maximum v is t This is the characteristic threshold for terrain classification: v=ω1(μ1-μ) 2 +ω2(μ2-μ) 2 ; The ground point extraction unit is used to obtain the threshold value for filtering based on the two-dimensional profile curve fitting method for sloping terrain; for flat terrain, it is used to obtain the threshold value for filtering based on the three-dimensional cloth simulation method; it completes the filtering of different terrains in the entire test area and extracts high-precision ground points.

7. The laser point cloud ground filtering device according to claim 6, characterized in that: The initial ground model generating unit includes: Cluster analysis and denoising module, which is used to perform cluster analysis and remove noise points based on the spatial distribution of point clouds; Regular grid interpolation module, used to interpolate the blank grid using the inverse distance weighted interpolation (IDW) method to obtain the initial ground model; The terrain classification unit includes: Multi-scale grid feature calculation module, used to construct multi-scale grids and calculate feature values; The maximum inter-class variance classification module is used to classify terrain into two categories: slope terrain and flat terrain based on the maximum inter-class variance principle.

8. The laser point cloud ground filtering device according to claim 7, characterized in that: The ground point extraction unit includes: Curve fitting slope filtering module, used to filter slope terrain based on two-dimensional profile curve fitting method; The cloth simulation flat terrain filtering module is used to filter flat terrain according to the three-dimensional cloth simulation method.

Images