A land bridge extraction method and system

Abstract

The application discloses a land bridge extraction method and system, and comprises the following steps: acquiring bridge original point cloud data; denoising the original point cloud data to obtain first point cloud data; removing the scattered vegetation point cloud in the point cloud data by using a dispersion method to obtain second point cloud data; improving a radius filtering algorithm by using color and elevation characteristics according to the difference between the color and elevation characteristics of the vegetation point cloud and the bridge point cloud; filtering all the vegetation point cloud in the second point cloud data by using the improved radius filtering algorithm to obtain third point cloud data; calculating the elevation information and normal information of the third point cloud data, and filtering the ground point cloud in the third point cloud data by using a constructed normal filtering model to obtain bridge point cloud. The application can more accurately filter the vegetation and ground point cloud, and improves the extraction completeness of the bridge point cloud.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building extraction, in particular to a land bridge extraction method and system. BACKGROUND

[0002] Bridge is one of the important buildings, and the identification of the bridge is of great significance to the global positioning system data acquisition, map updating and auxiliary supervision of bridge construction; due to the complex environment around the road bridge, the buildings, vegetation, road greening and the shadow of the bridge itself generated by light irradiation bring certain difficulties to the extraction of the bridge.

[0003] Most of the current algorithms are based on clustering, statistics, fluctuation, density and other features to extract target point cloud. The filtering algorithm based on intensity information and terrain information uses the progressive triangulation filtering algorithm to set different network priority, but the calculation amount is large and is not suitable for processing large point cloud. The elevation frequency histogram of the ground feature point cloud is drawn based on the elevation statistics to determine the vegetation elevation threshold, and the vegetation point cloud is removed. Although this method is easy to operate, it is only suitable for relatively flat and vegetation point cloud elevation uniformity areas; the point cloud denoising method using density spatial clustering and statistical filtering cannot effectively remove the cluster point cloud with close distance and large density; the normal direction of the sample surface is estimated by using the principal component analysis method to extract the building surface, and the boundary of the vegetation is also extracted, which cannot clearly distinguish the buildings and the vegetation. The progressive morphological filtering algorithm is used to segment the ground points and non-ground points, but this method has poor segmentation effect on inclined ground point cloud; the second derivative of the elevation information and the scanning line direction is used to judge whether the adjacent point cloud is continuous, and then the buildings and the vegetation are distinguished, but the surface of the thick stem is also regular and continuous, so it cannot be accurately distinguished. In view of the problem of incomplete extraction of buildings, the EXG (Excessive Green Index) model is proposed to filter out the vegetation point cloud, but this color model is only suitable for unmanned aerial vehicle point cloud with obvious color difference, and is not suitable for point cloud with mixed color features; the proposed vegetation removal method based on point cloud uniformity can effectively remove the scattered point cloud, but the non-target points with dense and uniform distribution are retained by mistake.

[0004] Therefore, it is urgent to provide a method for more accurate and complete extraction of bridges to solve the problem of incomplete extraction of existing methods for building bridges. SUMMARY

[0005] In order to solve the above technical problems, the technical scheme adopted by the present application is to provide a land bridge extraction method, comprising the following steps:

[0006] The method comprises the following steps: obtaining bridge original point cloud data; taking the bridge original point cloud data as target point cloud data; and denoising the target point cloud data to obtain first point cloud data; removing the randomly distributed vegetation point cloud in the first point cloud data by using a dispersion method to obtain second point cloud data; improving a radius filtering algorithm by using color and height features according to the difference between the color and height features of the vegetation point cloud and the bridge point cloud; filtering all the vegetation point cloud in the second point cloud data by using the improved radius filtering algorithm to obtain third point cloud data; calculating the height information and normal information of the third point cloud data; and filtering the ground point cloud in the third point cloud data by using a normal filtering model to obtain bridge point cloud.

[0007] In the method, the removing of the randomly distributed vegetation point cloud in the first point cloud data by using the dispersion method to obtain the second point cloud data comprises the following steps:

[0008] KD-Tree is constructed according to the first point cloud data, the topological relationship of the point cloud is established, and the k-neighborhood of all points in the point cloud is constructed;

[0009] One of the k-neighborhoods is selected, the maximum coordinate difference in x, y and z directions of the neighborhood is calculated, a maximum bounding volume is constructed with the maximum coordinate difference as the edge length, and the volume V max is calculated;

[0010] The coordinate differences in x, y and z directions are calculated respectively and taken as the length, width and height of a minimum bounding volume respectively, and the volume V min is calculated;

[0011] D is the ratio of the volume of the minimum bounding volume to the volume of the maximum bounding volume; the filtering threshold α is obtained according to the statistical method; if D > α, the point is considered to be a vegetation point and is filtered out; if D < α, the point is considered to be a non-vegetation point and is retained;

[0012] The above steps are repeated until all the point clouds in the first point cloud data are processed, that is, the filtering operation of the randomly distributed vegetation point cloud in the first point cloud data is completed, and the second point cloud data is obtained.

[0013] In the method, the improving of the radius filtering algorithm by using the color and height features according to the difference between the color and height features of the vegetation point cloud and the bridge point cloud, and the filtering of all the vegetation point cloud in the second point cloud data by using the improved radius filtering algorithm to obtain the third point cloud data comprise the following steps:

[0014] The second point cloud data is divided into six part point cloud data based on the height information of the point cloud, the distribution range of each part point cloud is calculated in sequence, and whether each part point cloud contains vegetation point cloud is judged according to the distribution range;

[0015] The part of the point cloud containing only the point cloud of the building bridge is marked as a non-vegetation suspicious point; the part of the data containing both vegetation and bridge point cloud is judged whether it is a vegetation suspicious point by using the color model;

[0016] That is, the color feature I of the vegetation and bridge point cloud is obtained, and the color threshold T is obtained according to the statistical method, when I>T is marked as a vegetation suspicious point, otherwise it is a non-vegetation suspicious point, thus all point cloud data is marked and processed;

[0017] According to the neighboring points of any spherical neighborhood in the marked point cloud, the number of neighborhood point clouds is calculated, if the total number of neighborhood point clouds is greater than the threshold M1 or the proportion of vegetation suspicious points in the neighborhood is less than M2, it is considered to be a non-vegetation point and is retained, otherwise it is considered to be a vegetation point and is filtered out;

[0018] Repeat the above steps until all vegetation points in each part of the point cloud are processed, and finally obtain the third point cloud data.

[0019] In the above method, the height information and normal information of the third point cloud data are calculated, and the ground point cloud in the third point cloud data is filtered out through the constructed normal filtering model to obtain the bridge point cloud, which specifically includes the following steps:

[0020] According to the third point cloud data, a KD-Tree is constructed to establish the topological relationship of the point cloud, and the spherical neighborhood of all points in the point cloud is obtained;

[0021] Optionally, one spherical neighborhood is selected to fit a local plane, and the normal vector is estimated based on the local plane;

[0022] The cosine value of the normal vector of each point and the vector (0, 0, 1) is calculated, and the height value and the cosine value of all third point cloud data are combined, and the threshold is calculated according to the statistical method to filter out the ground point cloud, and the final bridge point cloud is obtained.

[0023] In the above method, after obtaining the third point cloud data, the following steps are further implemented:

[0024] By counting the number of point clouds per unit length, the distribution probability of the corresponding interval point cloud is calculated, the segmentation threshold is determined, and the vegetation point cloud and the building bridge point cloud are segmented.

[0025] In the above method, the step of optionally selecting one spherical neighborhood to fit a local plane and estimating a normal vector based on the local plane specifically includes the following steps:

[0026] Select a point p, search for its k nearest neighbors, and calculate the local plane P in the least square sense of the center point and its neighborhood points. The calculation formula of this plane is:

[0027]

[0028] The normal vector of the local plane P is taken as the normal vector of the current point p, and the normal vector of the plane P can be obtained by principal component analysis, and the normal vector of the plane P is obtained by the operation that the plane P passes through the centroid p0 of the k-neighbor points of p0, and the normal vector Satisfies The covariance matrix M in the following formula is subjected to eigenvalue decomposition, and the eigenvector corresponding to the minimum eigenvalue of M is the normal vector of P

[0029]

[0030] The application also provides a land bridge extraction system, comprising

[0031] The data acquisition unit is configured to acquire bridge original point cloud data, and specifically, the bridge original point cloud data can be acquired by laser radar technology.

[0032] The data processing unit is configured to take the bridge original point cloud data as target point cloud data, and to denoise the target point cloud data to obtain first point cloud data.

[0033] The vegetation point cloud preliminary removal unit is configured to remove the scattered vegetation point cloud in the first point cloud data obtained by the data processing unit by using the dispersion method to obtain second point cloud data.

[0034] The vegetation point cloud removal unit is configured to improve the radius filtering algorithm by using the color and height features according to the differences in the color and height features between the vegetation point cloud and the bridge point cloud, and to filter out all the vegetation point clouds in the second point cloud data by using the improved radius filtering algorithm to obtain third point cloud data.

[0035] The ground point cloud removal unit is configured to calculate the height information and the normal information of the third point cloud data, and to construct a normal filtering model to filter out the ground point cloud in the third point cloud data to obtain bridge point cloud.

[0036] In the above scheme, the vegetation point cloud preliminary removal unit specifically implements the following steps:

[0037] The KD-Tree is constructed according to the first point cloud data of the data processing unit, the topological relationship of the point cloud is established, and the k-neighborhood of all points in the point cloud is constructed.

[0038] Optionally, one of the k-neighborhoods is selected, the maximum coordinate difference in the x, y and z directions of the neighborhood is calculated, a maximum bounding volume is constructed with the maximum coordinate difference as the edge length, and the volume V max is calculated.

[0039] The coordinate differences in the x, y and z directions are calculated again, and the coordinate differences are taken as the length, width and height of the minimum bounding volume respectively, and the volume V min is calculated.

[0040] Let D be the ratio of the volume of the minimum bounding volume and the maximum bounding volume; obtain a filtering threshold value a according to a statistical method, if D > a, it is considered that the point is a vegetation point and is filtered out; if D < a, it is considered that the point is a non-vegetation point and is retained;

[0041] Repeat the above steps until all point clouds in the first point cloud data are processed, that is, the filtering operation of the scattered vegetation point cloud in the first point cloud data is completed, and the second point cloud data is obtained.

[0042] In the above scheme, the vegetation point cloud removal unit specifically implements the following steps:

[0043] Based on the height information of the point cloud, the second point cloud data of the point cloud preliminary removal unit is divided into six partial point cloud data, and the distribution width of each partial point cloud is calculated in turn, and whether each partial point cloud contains vegetation point cloud is judged according to the distribution width;

[0044] Then, all point clouds in each partial point cloud containing only building and bridge point clouds are marked as non-vegetation suspicious points; the part of data containing vegetation and bridge point clouds at the same time is judged whether it is a vegetation suspicious point by using a color model;

[0045] That is, the color feature I of the vegetation and bridge point cloud is obtained, and a color threshold T is obtained according to a statistical method, when I > T is marked as a vegetation suspicious point, otherwise it is a non-vegetation suspicious point, thus all point cloud data are marked and processed;

[0046] According to the neighboring points of any spherical neighborhood in the point cloud, the number of neighborhood point clouds is calculated, if the total number of neighborhood point clouds is greater than a threshold M1 or the proportion of vegetation suspicious points in the neighborhood is less than M2, it is considered to be a non-vegetation point and is retained, otherwise it is considered to be a vegetation point and is filtered out;

[0047] Repeat the above steps until all vegetation points in each partial point cloud are processed, and finally obtain the third point cloud data.

[0048] In the above scheme, the ground point cloud removal unit specifically implements the following steps:

[0049] According to the third point cloud data of the point cloud removal unit, a KD-Tree is constructed to establish the topological relationship of the point cloud, and the spherical neighborhood of all points in the point cloud is obtained;

[0050] Optionally, one spherical neighborhood is selected, a local plane is fitted, and a normal vector is estimated based on the local plane;

[0051] The cosine value of the normal vector of each point and the vector (0, 0, 1) is calculated, the height value is combined with the cosine value, and the threshold is calculated according to a statistical method to filter out the ground point cloud; and the final bridge point cloud is obtained. The present application utilizes the feature that the vegetation point cloud is distributed divergently, and uses the dispersion method to preliminarily filter out the vegetation point cloud; in order to fully utilize the color and height features, an improved radius filtering algorithm is constructed to filter out the trunk vegetation point cloud and the remaining point cloud; finally, the ground point cloud is filtered out by using the normal filtering model; the non-target points with different colors, shapes and sizes are filtered out, the vegetation and ground point clouds can be more accurately filtered out, and the extraction completeness of the bridge point cloud is improved.

[0052] The above description is only a summary of the technical scheme of the present application, in order to more clearly understand the technical means of the present application, and can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application are described below. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the specific embodiments of the present application or the technical scheme in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced below, and obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creating labor.

[0054] Figure 1 The method flowchart provided by the present application is provided;

[0055] Figure 2 The point cloud data provided by the present application is provided;

[0056] Figure 3 The step S3 specific implementation step flowchart provided by the present application is provided;

[0057] Figure 4 The step S4 specific implementation step flowchart provided by the present application is provided;

[0058] Figure 5 The step S5 specific implementation step flowchart provided by the present application is provided;

[0059] Figure 6 The invention method provided by the present application is compared with other cases, and the bridge effect diagram is provided;

[0060] Figure (a) is the effect diagram of the combined filtering method (color model + height model + normal filtering algorithm) of the present application for extracting the bridge;

[0061] Figure (b) is the effect diagram of using radius + color + cloth simulation filtering algorithm to extract the bridge;

[0062] Figure (c) is an effect picture of extracting the bridge using radius + elevation + cloth simulation filtering algorithm;

[0063] Figure (d) is an effect picture of extracting the bridge using radius + color + normal filtering algorithm;

[0064] Figure (e) is an effect picture of extracting the bridge using radius + elevation + normal filtering algorithm;

[0065] Figure 7 The system schematic block diagram provided by the present application is shown in the figure;

[0066] Figure 8 The computer readable storage medium structure schematic block diagram provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0067] The technical solutions of the present application will be described below in conjunction with the embodiments, obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts belong to the scope of protection of the present application.

[0068] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise" and the like indicate the orientation or positional relationship shown in the drawings, which are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements indicated must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0069] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited. In addition, the terms "mounting", "connecting", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0070] The application will be described in detail below with reference to specific embodiments and the accompanying drawings.

[0071] Method embodiments

[0072] According to the embodiments of the application, a land bridge extraction method is provided, as shown in the accompanying drawings. Figure 1 The land bridge extraction method according to the embodiments of the application comprises the following steps:

[0073] S1, obtaining bridge original point cloud data; in this embodiment, the bridge original point cloud data is obtained by laser radar technology.

[0074] S2, obtaining the bridge original point cloud data as target point cloud data; and denoising the target point cloud data to obtain first point cloud data; in this embodiment, the statistical denoising method is used to remove outlier points in the point cloud data.

[0075] S3, using the dispersion method to remove randomly distributed vegetation point clouds in the first point cloud data to obtain second point cloud data;

[0076] S4, according to the differences in color and elevation characteristics between the vegetation and the bridge point clouds, improving the radius filtering algorithm using the color and elevation characteristics, and using the improved radius filtering algorithm to filter out the vegetation point clouds in the second point cloud data to obtain third point cloud data; by statistically analyzing the mean and variance of the red, green and blue color bands of the vegetation and the bridge point clouds, a color model capable of distinguishing the vegetation and the bridge is constructed.

[0077] In this embodiment, the color model is used alone for filtering, which may incorrectly filter out a large number of bridge surface point clouds. Since the height of the vegetation is relatively low compared with the bridge surface, the height information is added to the color model to prevent the bridge surface point clouds from being incorrectly filtered out.

[0078] According to the differences in color and elevation between the vegetation and the bridge point clouds, the vegetation point clouds are filtered out.

[0079] S5, calculating the elevation information and normal information of the third point cloud data, and constructing a normal filtering model to filter out the ground point clouds in the third point cloud data to obtain bridge point clouds.

[0080] In this embodiment, the dispersion method is used to preliminarily filter out the vegetation point clouds according to the feature that the vegetation point clouds are distributed divergently; in order to fully utilize the color and elevation characteristic information, the improved radius filtering algorithm is constructed to filter out the trunk vegetation point clouds and the remaining point clouds; finally, the normal filtering model is used to filter out the ground point clouds; the non-target points with different colors, shapes and sizes are filtered out, the vegetation and ground point clouds can be more accurately filtered out, and the completeness of the bridge point cloud extraction is improved.

[0081] The technical solutions of the embodiments of the application will be described in detail below.

[0082] The bridge original point cloud data obtained in step S1 is taken as the target point cloud data, and the amount of the point cloud data is 11,804,344, as shown in the following table. Figure 2 As shown in the following table, the data is finally output in the format of (x, y, z, R, G, B, I), wherein x, y and z represent the geometric position information of the data; R, G and B represent color information; and I represents echo intensity information. Then, the target point cloud data is removed from outliers, and the specific operation can be as follows:

[0083] Based on the k-neighborhood search of the KD-Tree, the average distance between the query point and the neighborhood point is determined, and then the statistical characteristics of the average distance are used to determine the cutoff threshold to remove the outlier points. After removing the noise, the amount of the point cloud data is 11,256,372, which is the first point cloud data.

[0084] Further preferably, by observing the point cloud data, it is found that the vegetation point cloud can be divided into scattered and linear vegetation point clouds. Therefore, the dispersion method is used to remove the scattered vegetation point cloud in the first point cloud data in step S3 to obtain the second point cloud data, as shown in the following table. Figure 3 The specific operation is as follows:

[0085] S31: KD-Tree is constructed for the first point cloud data to establish the topological relationship of the point cloud and construct the k-neighborhood of all points in the point cloud.

[0086] S32: One of the k-neighborhoods is selected, the maximum coordinate difference in the x, y and z directions of the neighborhood is calculated, a maximum bounding volume is constructed with the maximum coordinate difference as the edge length, and the volume V max is calculated.

[0087] S33: Then, the coordinate differences in the x, y and z directions are calculated as the length, width and height of the minimum bounding volume, respectively, and the volume V min is calculated.

[0088] S34: Let D be the ratio of the volume of the minimum bounding volume to the volume of the maximum bounding volume. The filtering threshold α is obtained according to the statistical method. If D > α, it is considered that the point is a vegetation point and is filtered out. If D < α, it is considered that the point is a non-vegetation point and is retained.

[0089] S35: The above steps S32-S34 are repeated until all points in the point cloud are processed, that is, the filtering operation of the scattered vegetation point cloud in the first point cloud data is completed, and the second point cloud data is obtained.

[0090] Further preferably, the radius filtering algorithm is improved according to the color and height features of the vegetation point cloud and the bridge point cloud, and the improved radius filtering algorithm is used to filter all the vegetation point clouds in the second point cloud data to obtain third point cloud data, which is mainly used to filter the vegetation point clouds such as roadside green belts, branches and trunks. Figure 4 As shown in the following steps:

[0091] S41: Based on the height information of the point cloud, the second point cloud data is divided into six part point cloud data, and the distribution width of each part point cloud is calculated in sequence, so as to determine whether each part point cloud contains vegetation point cloud; the calculation of the distribution width of each part point cloud is to calculate the width of the point cloud in the x direction (that is, the maximum coordinate of the x axis) for the six part point clouds in sequence.

[0092] S42: All point clouds containing only building bridge point clouds in the above-mentioned each part point cloud are marked as non-vegetation suspicious points; the part data containing vegetation and bridge point clouds at the same time is judged by using the color model to determine whether it is a vegetation suspicious point;

[0093] That is, the color feature I of the vegetation and bridge point cloud is obtained, and the color threshold T is obtained according to the statistical method, when I>T is marked as a vegetation suspicious point, otherwise it is a non-vegetation suspicious point, thus all point cloud data are marked and processed.

[0094] S43: According to the neighboring points of any spherical neighborhood in the point cloud marked in step S42, the number of neighborhood point clouds is calculated, if the total number of neighborhood point clouds is greater than the threshold M1 or the proportion of vegetation suspicious points in the neighborhood is less than M2, it is considered to be a non-vegetation point and is retained, otherwise it is considered to be a vegetation point and is filtered out;

[0095] S44: Repeat the above steps S42-S43 until all vegetation points in each part point cloud are processed, and finally obtain the third point cloud data.

[0096] In step S42, the color model is constructed as follows:

[0097] S421: Use K-MEANS clustering algorithm to cluster the original point cloud into 9 categories, and select the point cloud data containing only vegetation and bridge for statistical analysis;

[0098] S422: Statistically analyze the mean and variance of the features of the red, green and blue bands, as shown in Table 1;

[0099] Table 1

[0100]

[0101] S423: Analyze the mean and variance of the statistical analysis, and construct a color model.

[0102] Preferably, the step S4 further comprises the following steps:

[0103] When viewed in the direction of the x-axis, the thick stem point cloud is sparsely distributed on both sides of the building bridge. Based on this feature, the step S4 comprises the following steps:

[0104] The distribution probability of the point cloud in the corresponding interval is calculated by counting the number of third point clouds per unit length, the segmentation threshold is determined, and the vegetation point cloud and the building bridge point cloud are segmented therefrom. This step realizes filtering of the stem vegetation point cloud and the remaining point cloud, so that the filtered point cloud data is more accurate.

[0105] Further preferably, the elevation information and normal information of the third point cloud data are calculated, and a normal filtering model is constructed to filter the ground point cloud in the third point cloud data and obtain the bridge point cloud, as shown in Figure 5 The step S4 comprises the following steps:

[0106] S51: Construct a KD-Tree for the third point cloud data to establish the topological relationship of the point cloud and obtain the spherical neighborhood of all points in the point cloud;

[0107] S52: Select one spherical neighborhood and fit a local plane, and estimate the normal vector based on the local plane;

[0108] S53: Calculate the cosine value of the normal vector of each point and the vector (0, 0, 1), combine the elevation value and the cosine value of all third point cloud data, and calculate the threshold value to filter the ground point cloud according to the statistical method to obtain the final bridge point cloud. Wherein,

[0109] The normal vector estimation in step S52 comprises the following steps:

[0110] S521: Select a point p, search for its k nearest neighbors, and calculate the local plane P in the least squares sense of the center point and its neighborhood points. The calculation formula of this plane is:

[0111]

[0112] S522: Take the normal vector of the local plane P as the normal vector of the current point p. The normal vector of the plane P can be obtained by principal component analysis. It is known from the operation that P passes through the centroid p0 of its k neighborhood points, and the normal vector satisfies The covariance matrix M in the following formula is subjected to eigenvalue decomposition, and the eigenvector corresponding to the smallest eigenvalue of M is the normal vector of P

[0113]

[0114] The above embodiments will be described below through specific comparison cases, as follows:

[0115] The algorithm 1 and the algorithm 2 selected in the case are currently commonly used filtering algorithms, the cloth simulation filtering in the algorithm 1 and the algorithm 2 is replaced by the normal filtering to obtain the algorithm 3 and the algorithm 4, and the combined filtering algorithm constructed by using the color, the height and the normal three features is the algorithm in the paper.

[0116] In the embodiment, the radius+color+cloth simulation filtering algorithm, the radius+height+cloth simulation filtering algorithm, the radius+color+normal filtering algorithm and the radius+height+normal filtering algorithm are compared to extract the bridge in the same scene, and the effect pictures extracted are shown in (a), (b), (c), (d) and (e) in sequence. Figure 6 (a), (b), (c), (d) and (e) in sequence.

[0117] As shown in the comparison, the effect picture of the bridge extracted in the embodiment is compared with the comparison picture of the radius+color+cloth simulation filtering algorithm, and it can be known from the figure that a large number of vegetation point clouds and ground point clouds are retained and a large number of bridge surface point clouds are mis-filtered, the color information is replaced by using the height information, and the comparison picture of the effect is shown in (c), it can be known from the figure that a large number of bridge surface point clouds are retained, but there are still a large number of vegetation and ground point clouds. The cloth simulation filtering model is replaced by using the normal filtering model, and the comparison picture of the effect is shown in (d), it can be known from the figure that a large number of ground point clouds are filtered out. The comparison picture of the effect is shown in (e) by comparing the radius+height+normal filtering algorithm, it can be known from the figure that although a large number of vegetation point clouds are mis-retained, the completeness of the bridge point clouds is relatively good. Figure 6 Figure 6 Figure 6

[0118] In order to quantitatively analyze and evaluate the performance of the filtering algorithm, the original point cloud O is divided into target point cloud O t and non-target point cloud O n , after filtering the point cloud, the retained point cloud includes target point cloud R t and non-target point cloud R n , and the filtered point cloud includes target point cloud D t and non-target point cloud D n . It is defined that

[0119] Type I error: misjudging the non-target point cloud as the target point cloud, that is, E a =R n / O n ×100%.

[0120] Type II error is the total error: misjudging the non-target point cloud as the target point cloud and misjudging the target point cloud as the non-target point cloud, that is, E b =(D t +R n ) / O×100%. The extraction completeness of the bridge point cloud is defined as the ratio of the retained target point cloud to the target point cloud, that is, I=R t / O t .​​​

[0121] The comparison with other methods is shown in Table 2 below. The complete rate of the bridge point cloud extracted by the embodiment is 99.3%, and the error is 0.73%. Compared with the existing filtering method, the combined filtering method can more completely extract the bridge point cloud, and has higher accuracy.

[0122] Table 2

[0123]

[0124] The embodiment method effectively removes the vegetation and ground point cloud in the civil bridge construction scene. The algorithm uses the dispersion, color feature, elevation feature and statistical feature of the point cloud to filter the vegetation point cloud. Then, the distribution characteristics of the ground and bridge point cloud are further analyzed, a normal filtering model is established to filter the ground point cloud, so that the bridge point cloud is more accurately extracted.

[0125] System embodiment

[0126] According to the embodiment of the present application, a land bridge extraction system is provided, as shown in Figure 7 The land bridge extraction system according to the embodiment of the present application comprises:

[0127] The data acquisition unit is used to acquire the original bridge point cloud data, which can be acquired by laser radar technology.

[0128] The data processing unit is used to take the original bridge point cloud data as target point cloud data, and to denoise the target point cloud data to obtain first point cloud data. The statistical denoising method is used to remove outlier noise points in the point cloud data, specifically:

[0129] Based on the k-neighborhood search of KD-Tree, the average distance between the query point and the neighborhood point is determined, and then the statistical feature of the average distance is used to determine the cutoff threshold to remove the outlier noise points. The number of point cloud data after removing noise is 11,256,372, which is the first point cloud data.

[0130] The vegetation point cloud preliminary removal unit uses the dispersion method to remove the scattered vegetation point cloud in the first point cloud data obtained by the data processing unit to obtain second point cloud data. The specific implementation steps of this unit are as follows:

[0131] A1: Construct KD-Tree for the first point cloud data, establish the topological relationship of the point cloud, and construct the k-neighborhood of all points in the point cloud;

[0132] A2: Select one of the k-neighborhoods, calculate the maximum coordinate difference in x, y and z directions, and construct the maximum bounding volume with the maximum coordinate difference as the edge length, and calculate the volume V max ;

[0133] A3: Then calculate the respective coordinate difference in x, y, z three directions as the length, width and height of the minimum bounding box, thus the volume V is calculated min ;

[0134] A4: Let D be the ratio of the volume of the minimum bounding box to the maximum bounding box; obtain the filtering threshold α according to the statistical method, if D > α, it is considered that the point is a vegetation point, and is filtered out; if D < α, it is considered that the point is a non-vegetation point, and is retained;

[0135] A5: Repeat the above steps A2-A4 until all points in the point cloud are processed, that is, the filtering operation of the scattered vegetation point cloud in the first point cloud data is completed, and the second point cloud data is obtained.

[0136] The vegetation point cloud removal unit: according to the difference between the color and elevation characteristics of the vegetation point cloud and the bridge point cloud, the color and elevation characteristics are used to improve the radius filtering algorithm, and the improved radius filtering algorithm is used to filter all vegetation point clouds in the second point cloud data to obtain the third point cloud data; the specific implementation steps of this unit are as follows:

[0137] B1: Based on the height information of the point cloud, the second point cloud data is divided into 6 parts of point cloud data, and the distribution width of each part of point cloud is calculated in turn, according to which it is judged whether each part of point cloud contains vegetation point cloud;

[0138] B2: Mark all point clouds containing only building bridge point clouds in each part of point cloud as non-vegetation suspicious points; use color model to determine whether the part of data containing vegetation and bridge point cloud is vegetation suspicious point;

[0139] That is, the color characteristics I of the vegetation and bridge point cloud are obtained, and the color threshold T is obtained according to the statistical method, when I > T is marked as vegetation suspicious point, otherwise is non-vegetation suspicious point, thus all point cloud data are marked and processed.

[0140] B3: According to the neighboring points of any spherical neighborhood in the point cloud, the number of neighborhood point clouds is calculated, if the total number of neighborhood point clouds is greater than the threshold M1 or the proportion of vegetation suspicious points in the neighborhood is less than M2, it is considered to be a non-vegetation point and retained, otherwise it is considered to be a vegetation point and filtered out;

[0141] B4: Repeat the above steps until all vegetation points in each part of point cloud are processed, and finally obtain the third point cloud data.

[0142] The color model construction in step B2 is as follows:

[0143] B21: Use K-MEANS clustering algorithm to cluster the original point cloud into 9 categories, and select the point cloud data containing only vegetation and bridge for statistical analysis;

[0144] B22: Analyzing the mean and variance of the statistical point cloud data to analyze the feature mean and variance of the red, green and blue three bands;

[0145] B23: Analyzing the mean and variance of the statistical model to construct a color model.

[0146] Preferably, the vegetation point cloud removal unit further implements the following steps:

[0147] By counting the number of point clouds per unit length, the distribution probability of the corresponding interval point cloud is calculated, the segmentation threshold is determined, and the vegetation point cloud and the bridge point cloud are segmented.

[0148] The ground point cloud removal unit is used to calculate the height information and normal information of the third point cloud data, and construct a normal filtering model to filter out the ground point cloud in the third point cloud data to obtain the bridge point cloud.

[0149] C1: Constructing a KD-Tree for the third point cloud data to establish the topological relationship of the point cloud and obtain the spherical neighborhood of all points in the point cloud;

[0150] C2: Selecting one spherical neighborhood to fit a local plane and estimating the normal vector based on the local plane;

[0151] C3: Calculating the cosine value of the normal vector of each point and the vector (0, 0, 1), combining the height value and the cosine value, and calculating the threshold value to filter out the ground point cloud according to the statistical method to obtain the final bridge point cloud. Wherein,

[0152] The normal vector estimation in step C2 specifically includes the following steps:

[0153] C21: Selecting a point p and searching for its nearest k points, calculating the local plane P in the least square sense of the center point and its neighborhood points. The calculation formula of this plane is:

[0154]

[0155] C22: Taking the normal vector of the local plane P as the normal vector of the current point p. The normal vector of the plane P can be obtained by principal component analysis. It is known from the operation that P passes through the centroid p0 of its k neighborhood points, and the normal vector satisfies Eigenvalue decomposition is performed on the covariance matrix M in the following formula, and the eigenvector corresponding to the smallest eigenvalue of M is the normal vector of P

[0156]

[0157] For example Figure 8As shown, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the land bridge extraction method in the above embodiment, or the computer program is executed by the processor to implement the land bridge extraction method in the above embodiment.

[0158] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium, and when executed, the computer program can include the processes of the above-mentioned embodiments of each method. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0159] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, it is described relatively simply, and the relevant part can be referred to the part of the method embodiment. The above-described device and system embodiments are only illustrative, and the units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment scheme. Those skilled in the art can understand and implement without creative labor.

[0160] It has to be noted that, in the present document, relational terms are intended only to convey a possible relationship between elements or

[0161] The foregoing is considered as illustrative only of the principles of the application. Numerous modifications and changes will readily occur to those skilled in the art, which modifications and changes are to be understood as intended to be encompassed by the general scope of the application. Accordingly, the application is not to be limited to the above described or illustrated embodiments, but is intended to encompass all embodiments consistent with the principles of the application.

Claims

1. A land bridge extraction method, characterized by, The method comprises the following steps: Obtain the original point cloud data of the bridge; take the original point cloud data of the bridge as target point cloud data; And denoise the target point cloud data to obtain first point cloud data; Remove the randomly distributed vegetation point cloud in the first point cloud data by using the dispersion method to obtain second point cloud data; improve the radius filtering algorithm according to the difference between the color and elevation characteristics of the vegetation point cloud and the bridge point cloud, filter all the vegetation point cloud in the second point cloud data by using the improved radius filtering algorithm to obtain third point cloud data; calculate the elevation information and normal information of the third point cloud data, and filter the ground point cloud in the third point cloud data through the constructed normal filtering model to obtain the bridge point cloud; The step of improving the radius filtering algorithm according to the difference between the color and elevation characteristics of the vegetation point cloud and the bridge point cloud, filtering all the vegetation point cloud in the second point cloud data by using the improved radius filtering algorithm to obtain the third point cloud data specifically comprises the following steps: Based on the height information of the point cloud, the second point cloud data is divided into six part point cloud data, and the distribution width of each part point cloud is calculated in turn, so as to determine whether each part point cloud contains vegetation point cloud; the calculation of the distribution width of each part point cloud is that the width of the point cloud in the x direction is calculated for the six part point clouds in turn; Then, all the point clouds in each part point cloud containing only building bridge point cloud are marked as non-vegetation suspicious points; the part data containing vegetation and bridge point cloud at the same time is judged whether it is a vegetation suspicious point by using a color model; That is, the color characteristics I of the vegetation and bridge point cloud are obtained, and the color threshold T is obtained according to a statistical method; when I>T, it is marked as a vegetation suspicious point, otherwise it is a non-vegetation suspicious point; thus, all the point cloud data are marked and processed; According to the neighboring points of any spherical neighborhood in the above-mentioned marked point cloud, the number of neighborhood point clouds is calculated; if the total number of neighborhood point clouds is greater than the threshold M1 or the proportion of vegetation suspicious points in the neighborhood is less than M2, it is considered to be a non-vegetation point and is retained, otherwise it is considered to be a vegetation point and is filtered out; Repeat the above steps until all the vegetation points in each part point cloud are processed, and finally obtain the third point cloud data.

2. The land bridge extraction method of claim 1, wherein, The step of removing the randomly distributed vegetation point cloud in the first point cloud data by using the dispersion method to obtain the second point cloud data specifically comprises the following steps: KD-Tree is constructed according to the first point cloud data, the topological relationship of the point cloud is established, and the k-neighborhood of all points in the point cloud is constructed; Optionally, one of the k-neighborhoods is selected, the maximum coordinate difference in the x, y, and z directions of the neighborhood is calculated, a maximum bounding box is constructed with the maximum coordinate difference as the side length, and the volume V is calculated max ; The coordinate differences in x, y, and z directions are recalculated and taken as the length, width, and height of the minimum bounding box, respectively, and the volume is calculated as V min ; Let D be the volume ratio of the minimum bounding volume to the maximum bounding volume; the filtering threshold α is obtained according to a statistical method; if D>α, it is considered that the point cloud is a vegetation point and is filtered out; if D<α, it is considered that the point cloud is a non-vegetation point and is retained; Repeat the above steps until all the point clouds in the first point cloud data are processed, that is, the filtering operation of the randomly distributed vegetation point cloud in the first point cloud data is completed, and the second point cloud data is obtained.

3. The land bridge extraction method of claim 1, wherein, The step of calculating the elevation information and normal information of the third point cloud data, and filtering the ground point cloud in the third point cloud data through the constructed normal filtering model to obtain the bridge point cloud specifically comprises the following steps: According to the third point cloud data, a KD-Tree is constructed, the topological relationship of the point cloud is established, and the spherical neighborhood of all points in the point cloud is obtained; Optionally, one of the spherical neighborhoods is selected, a local plane is fitted, and a normal vector is estimated based on the local plane; The cosine value of the normal vector of each point and the vector (0, 0, 1) is calculated, all third point cloud data elevation values are combined with the cosine values, a threshold value is calculated according to a statistical method to filter out the ground point cloud, and the final bridge point cloud is obtained.

4. The land bridge extraction method of claim 1, wherein, After obtaining the third point cloud data, the following steps are further implemented: By counting the number of third point clouds per unit length, the distribution probability of the point cloud in the corresponding interval is calculated, the segmentation threshold is determined, and the vegetation point cloud and the building bridge point cloud are segmented accordingly.

5. The land bridge extraction method of claim 3, wherein, The steps of selecting one of the spherical neighborhoods, fitting a local plane, and estimating a normal vector based on the local plane specifically include the following steps: A point p is randomly selected, the k nearest neighbors of the point p are searched, and a local plane P in the least square sense is calculated based on the center point and the neighborhood points of the point p. The calculation formula of the plane is as follows: The normal vector of the local plane P is taken as the normal vector of the current point p, the normal vector of the plane P is obtained by principal component analysis, and the normal vector is known by operation that P passes through the centroid p0 of the k-neighbor points of P, and the normal vector satisfies Eigenvalue decomposition is performed on the covariance matrix M in the following formula, and the eigenvector corresponding to the minimum eigenvalue of M is the normal vector of P 6. A land bridge extraction system characterized by, It includes A data acquisition unit is configured to acquire bridge original point cloud data, and the bridge original point cloud data can be acquired by a laser radar technology; A data processing unit is configured to use the bridge original point cloud data as target point cloud data; The target point cloud data is denoised to obtain first point cloud data; A vegetation point cloud preliminary removal unit is configured to remove the scattered vegetation point cloud in the first point cloud data obtained by the data processing unit by using a dispersion method to obtain second point cloud data. The vegetation point cloud removal unit specifically implements the following steps: Based on the height information of the point cloud, the second point cloud data of the point cloud preliminary removal unit is divided into six part point cloud data, the distribution width of each part point cloud is calculated in sequence, and it is determined whether each part point cloud contains vegetation point cloud. The calculation of the distribution width of each part point cloud is as follows: the width of the point cloud in the x direction is calculated for the six part point clouds in sequence. The part point cloud that only contains building bridge point cloud is marked as non-vegetation suspicious point. The part point cloud that contains both vegetation and bridge point cloud is determined to be vegetation suspicious point or non-vegetation suspicious point by using a color model. That is, the color features I of the vegetation and bridge point cloud are obtained, and a color threshold T is obtained according to a statistical method. When I > T, the point is marked as a vegetation suspicious point, otherwise, the point is marked as a non-vegetation suspicious point. Thus, all point cloud data are marked and processed. According to the neighboring points of any spherical neighborhood of the point cloud, the number of neighborhood point clouds is calculated. If the total number of neighborhood point clouds is greater than a threshold M1 or the proportion of vegetation suspicious points in the neighborhood is less than M2, the point is considered to be a non-vegetation point and is retained, otherwise, the point is considered to be a vegetation point and is filtered out. The above steps are repeated until all vegetation points in each part point cloud are processed, and finally, third point cloud data is obtained. A vegetation point cloud removal unit is configured to improve a radius filtering algorithm by using the differences in color and elevation features between the vegetation point cloud and the bridge point cloud, and to filter out all vegetation point clouds in the second point cloud data by using the improved radius filtering algorithm to obtain third point cloud data. The ground point cloud removing unit is configured to calculate the height information and normal information of the third point cloud data, construct a normal filtering model, filter the ground point cloud in the third point cloud data, and obtain the bridge point cloud.

7. A land bridge extraction system as claimed in claim 6, wherein, The vegetation point cloud preliminary removing unit specifically implements the following steps: KD-Tree is constructed according to the first point cloud data of the data processing unit, the topological relationship of the point cloud is established, and the k-neighborhood of all points in the point cloud is constructed; Optionally, one of the k-neighborhoods is selected, the maximum coordinate difference in the x, y, and z directions of the neighborhood is calculated, a maximum bounding box is constructed with the maximum coordinate difference as the side length, and the volume V is calculated max ; The coordinate differences in x, y, and z directions are recalculated and taken as the length, width, and height of the minimum bounding box, respectively, and the volume is calculated as V min ; D is the volume ratio of the minimum bounding volume to the maximum bounding volume; the filtering threshold α is obtained according to a statistical method; if D>α, it is considered that the point cloud is a vegetation point and is filtered out; If D<α, it is considered that the point cloud is a non-vegetation point, and is retained; The above steps are repeated until all point clouds in the first point cloud data are processed, that is, the filtering operation of the scattered vegetation point cloud in the first point cloud data is completed, and the second point cloud data is obtained.

8. A land bridge extraction system as claimed in claim 6, wherein, The ground point cloud removing unit specifically implements the following steps: KD-Tree is constructed according to the third point cloud data of the point cloud removing unit, the topological relationship of the point cloud is established, and the spherical neighborhood of all points in the point cloud is obtained; Optionally, one spherical neighborhood is selected, a local plane is fitted, and a normal vector is estimated based on the local plane; The cosine value of the normal vector of each point and the vector (0, 0, 1) is calculated, the height value and the cosine value are combined, the threshold is calculated according to a statistical method, the ground point cloud is filtered out, and the final bridge point cloud is obtained.

Images