Method, device, storage medium and electronic device for extracting and segmenting street trees

Abstract

The present application relates to the field of point cloud data processing technology, and provides a method, device, storage medium, and electronic device for extracting and segmenting street trees. The method may include: filtering the original road point cloud to obtain a processed road point cloud; extracting street tree features from the processed road point cloud to obtain a street tree point cloud; extracting trunk features from the street tree point cloud to obtain a trunk point cloud; and obtaining street tree extraction results by analyzing the trunk point cloud. The present application can improve the accuracy of street tree extraction in cities, so as to strongly empower the research and development of fields such as digital twin cities and forestry planning.

Description

Technical Field

[0001] The present application relates to the technical field of point cloud data processing, and in particular to a method, device, storage medium and electronic device for extracting and segmenting street trees. Background Art

[0002] As one of the most important landscape elements in the city, street trees have functions such as dust prevention, noise reduction and air purification. They are a key component of modern production, life and urban ecosystems. Timely, accurate and automatic statistics and inventory of street trees can provide important data support for urban planners and managers. In addition, with the development of smart transportation, the accurate identification and individual fine segmentation of street trees, which are an important component of the urban ecosystem, have become urgent and critical issues. The research on intelligent classification and segmentation of urban three-dimensional large-scene laser point clouds can strongly empower the research and development of fields such as digital twin cities and forestry planning.

[0003] At present, traditional urban green landscape tree statistics mostly use manual field surveys and manual measurements, which are very labor-intensive and inefficient, and it is difficult to achieve dynamic monitoring of large scene areas. The commonly used method in the engineering field is to semi-automatically distinguish street trees from other objects through human-computer interaction. Although this method improves the extraction accuracy, it is labor-intensive, cumbersome, and greatly affected by human intervention, and is not suitable for large-scale complex urban scenes.

[0004] Therefore, how to provide a technical solution for an efficient method of street tree extraction and monomer segmentation has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of some embodiments of the present application is to provide a method, device, storage medium and electronic device for street tree extraction and monomer segmentation. Through the technical solution of the embodiments of the present application, accurate extraction and segmentation of street tree monomers in the landscape can be achieved, with low computational complexity, high universality, and easy to achieve dynamic monitoring of large scene areas.

[0006] In a first aspect, some embodiments of the present application provide a method for street tree extraction and monomer segmentation, including: filtering the original road point cloud to obtain a processed road point cloud; extracting street tree features in the processed road point cloud to obtain a street tree point cloud; extracting trunk features in the street tree point cloud to obtain a trunk point cloud; and obtaining a street tree extraction result by analyzing the trunk point cloud, wherein the street tree extraction result includes: at least one of a street tree monomer segmentation extraction result, a street tree recognition extraction result, and a street tree trunk extraction result.

[0007] Some embodiments of the present application obtain processed road point clouds by filtering the original road point clouds, then extracting street tree features and trunk features from the processed road point clouds to obtain tree trunk point clouds, and finally analyzing the tree trunk point clouds to determine the street tree extraction and segmentation results. Some embodiments of the present application can reduce noise interference and improve extraction and segmentation accuracy by filtering and feature extracting the original road point clouds. In addition, the method is easy to operate, has a low amount of calculation, has high universality, and is easy to realize dynamic monitoring of large scene areas.

[0008] In some embodiments, the filtering of the original road point cloud to obtain the processed road point cloud includes: filtering out invalid point clouds in the original road point cloud to obtain the processed road point cloud, wherein the invalid point clouds include: at least one of ground point clouds, vegetation point clouds and building facade point cloud sets; wherein the vegetation point cloud is determined by a height threshold set along the elevation direction based on the ground point cloud; and the building facade point cloud set is determined by analyzing the component of the normal vector of the point cloud after filtering out the ground point cloud in the elevation direction.

[0009] Some embodiments of the present application obtain a processed road point cloud by filtering out point cloud information irrelevant to roadside trees from the original road point cloud, thereby effectively removing noise from the original point cloud.

[0010] In some embodiments, the extracting of street tree features in the processed road point cloud to obtain the street tree point cloud includes: extracting the processed road point cloud through density constraint to obtain an initial seed point set; clustering the initial seed point set to obtain a clustered point cloud set; and extracting features from the clustered point cloud set through an elevation slice point frequency histogram to obtain the street tree point cloud.

[0011] Some embodiments of the present application perform seed point set extraction, clustering and street tree feature extraction on the processed road point cloud through density constraints to obtain a street tree point cloud, which can improve the accuracy of street tree extraction.

[0012] In some embodiments, the feature extraction of the clustered point cloud set through the elevation slice point frequency histogram to obtain the street tree point cloud includes: after iteratively segmenting the clustered point cloud set using the elevation direction filtering method, counting the number of point clouds in each round of slicing, and drawing the elevation slice point frequency histogram according to the number of point clouds; by counting the point density difference information in the elevation slice point frequency histogram and the cluster posture information of the street trees in space, point cloud clusters of invalid objects are eliminated from the clustered point cloud set to obtain the street tree point cloud.

[0013] Some embodiments of the present application divide and count the clustered point cloud set through a filtering method in the elevation direction, draw an elevation slice point frequency histogram, and then filter the point cloud cluster clusters to obtain the street tree point cloud, which can effectively eliminate the interference of similar rod-shaped object point clouds, overcome the influence of noise, and improve the accuracy of street tree extraction.

[0014] In some embodiments, the extracting of trunk features in the street tree point cloud to obtain the trunk point cloud includes: dividing the regional data of the street tree point cloud into a crown point cloud and a non-crown point cloud by a preset trunk height threshold; obtaining a potential trunk point set in the non-crown point cloud; verifying the trunks in the potential trunk point set by a trunk surface curvature threshold to determine a candidate trunk point cloud; and re-verifying the candidate trunk point cloud to determine the trunk point cloud.

[0015] Some embodiments of the present application divide the street tree point cloud into crown areas to obtain a potential trunk point set in the non-crown point cloud, and then verify the potential trunk point set to obtain a trunk point cloud, which can improve the accuracy of trunk extraction.

[0016] In some embodiments, the candidate trunk point cloud is verified again to determine the trunk point cloud, including: establishing a bounding box for the candidate trunk point cloud, and calculating a reference value of the bounding box in the three-dimensional coordinate axis; if it is confirmed that the reference value is greater than a preset value, the candidate trunk point cloud is used as the trunk point cloud; if it is confirmed that the reference value is less than or equal to the preset value, the candidate trunk point cloud is further verified to obtain the trunk point cloud.

[0017] Some embodiments of the present application verify the validity of the candidate tree trunk point cloud by establishing a bounding box for the candidate tree trunk point cloud to ensure the extraction accuracy.

[0018] In some embodiments, the further verification of the candidate trunk point cloud to obtain the trunk point cloud includes: performing equally spaced elevation pass-through filtering on the candidate trunk point cloud to generate multiple slices; using a random sampling consistency algorithm to fit the hierarchical circle center estimation results of the local circular arcs of the multiple slices projected on the horizontal plane; obtaining the circle radius fitting values ​​of the multiple slices through the hierarchical circle center estimation results; comparing the circle radius fitting values ​​of the multiple slices with a radius threshold to determine the trunk point cloud.

[0019] Some embodiments of the present application adopt a joint constraint of hierarchical circle center estimation for candidate tree trunk point clouds to effectively improve the extraction accuracy of street tree trunks, thereby overcoming the problem of low universality of conventional algorithms.

[0020] In some embodiments, the analysis of the trunk point cloud to obtain the street tree extraction result includes: performing an adjacency analysis on the crown point cloud corresponding to the trunk point cloud, and when it is determined that there is a cross-overlapping area of ​​the crown, solving the connectivity coefficient corresponding to the cross-overlapping area of ​​the crown; when the connectivity coefficient is less than a coefficient threshold, calculating the spatial distance between adjacent crown points in the cross-overlapping area of ​​the crown, and constructing a global graph node; according to the global graph node, segmenting the trees in the cross-overlapping area of ​​the crown to obtain the street tree monomer segmentation extraction result.

[0021] Some embodiments of the present application determine the segmentation results of street tree monomers by combining connectivity coefficients and global graph node analysis, thereby achieving accurate segmentation of street tree monomers.

[0022] In some embodiments, the trees in the cross-overlapping area of ​​the crowns are segmented according to the global graph node to obtain the street tree monomer segmentation result, including: taking a randomly selected overlapping crown point as the starting point, and taking the centroid coordinates of the two trees adjacent to the overlapping crown point as the first end point and the second end point; solving the first path and the second path from the starting point to the first end point and the second end point respectively on the global graph node; determining the target tree to which the overlapping crown point belongs through the distances corresponding to the first path and the second path respectively, so as to obtain the street tree monomer segmentation result for the two trees, wherein the target tree is one of the two trees.

[0023] Some embodiments of the present application use path analysis on the overlapping area of ​​tree crowns to determine the target tree to which the crown point belongs, thereby achieving accurate segmentation of two trees with overlapping crowns.

[0024] In a second aspect, some embodiments of the present application provide a device for street tree extraction and monomer segmentation, including: a preprocessing module, used to filter the original road point cloud to obtain a processed road point cloud; a first extraction module, used to extract street tree features in the processed road point cloud to obtain a street tree point cloud; a second extraction module, used to extract trunk features in the street tree point cloud to obtain a trunk point cloud; a result acquisition module, used to obtain a street tree extraction result by analyzing the trunk point cloud, wherein the street tree extraction result includes: at least one of a street tree monomer segmentation extraction result, a street tree recognition extraction result and a street tree trunk extraction result.

[0025] In a third aspect, some embodiments of the present application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can implement the method described in any embodiment of the first aspect.

[0026] In a fourth aspect, some embodiments of the present application provide an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement a method as described in any embodiment of the first aspect when executing the program.

[0027] In a fifth aspect, some embodiments of the present application provide a computer program product, wherein the computer program product comprises a computer program, wherein the computer program, when executed by a processor, can implement the method described in any embodiment of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of some embodiments of the present application, the drawings required for use in some embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 One of the flow charts of the method for extracting and segmenting street trees provided in some embodiments of the present application;

[0030] Figure 2 A schematic diagram of a street tree point cloud provided for some embodiments of the present application;

[0031] Figure 3 A linear feature map of a street tree trunk point cloud provided in some embodiments of the present application;

[0032] Figure 4 A schematic diagram of a tree trunk point cloud provided for some embodiments of the present application;

[0033] Figure 5 A schematic diagram of path planning provided for some embodiments of the present application;

[0034] Figure 6 A schematic diagram of the segmentation of street trees provided in some embodiments of the present application;

[0035] Figure 7 Flowchart 2 of the method for extracting and segmenting street trees provided in some embodiments of the present application;

[0036] Figure 8 A block diagram of the apparatus for extracting and segmenting street trees provided in some embodiments of the present application;

[0037] Fig. 9 A schematic diagram of an electronic device is provided for some embodiments of the present application. DETAILED DESCRIPTION

[0038] The technical solutions in some embodiments of the present application will be described below in conjunction with the drawings in some embodiments of the present application.

[0039] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0040] In recent years, mobile laser scanning (MLS) technology has played an important role in digital city construction, traffic planning, and road 3D reconstruction by virtue of its ability to obtain high-resolution geospatial information with high precision. Usually, MLS can collect detailed information about objects on both sides of the road during movement by mounting a high-speed laser scanning sensor on a mobile carrier and recording the absolute position and instantaneous attitude of the carrier with the help of a global satellite navigation system and an inertial measurement unit. In addition, the laser point cloud data calculated by the highly integrated MLS scanning system can describe the real size and shape parameters of objects in urban scenes for urban components such as traffic signs, street lights, and buildings. Compared with traditional detection methods, the advantage of MLS point cloud is that it can accurately characterize information such as the trunk structure, crown morphology, and canopy volume of trees; in terms of measuring the vertical height of street trees, the measurement combined with point cloud data is more accurate, which is an extremely effective data primitive for digital city planning and sustainable development.

[0041] However, accurately identifying, extracting and statistically analyzing street trees in urban scenes is a challenging task. Because the urban environment is noisy and the background is complex, dense noise and object occlusion pose certain obstacles to the accurate detection of street trees. In addition, the cross-adhesion of multiple adjacent street trees affects the acquisition of tree morphological parameters.

[0042] At present, the main methods for extracting and segmenting street trees based on laser point clouds are: The geometric feature-based method describes the various features of the tree by simulating an accurate mathematical model according to the spatial morphology and topological structure information presented by the street tree, so as to achieve tree extraction and segmentation. For example, in the existing art, Treeseg method, which extracts tree point clouds from large scenes in a highly automated manner, uses general processing techniques such as principal component analysis, connectivity test and shape fitting to achieve accurate canopy, trunk and stem segmentation results. In addition, methods such as regional growing, watershed analysis, clustering and fitting functions are used to try to optimize the recognition accuracy of street trees, but the cross-interference of the geometric shapes of objects in complex backgrounds will affect the stability of the tree structure, resulting in uncertainty in the quality of tree segmentation. Learning-based methods, such as point-based end-to-end deep learning methods PointNet++, multi-branch networks (MBNet) with fused multi-channel features, and RetinaNet model & PCS algorithms, can accurately classify and segment tree species and improve resource survey efficiency. In some specific scenarios, deep learning-based methods can achieve more robust extraction results than geometric methods. However, learning-based methods often require a large amount of labeled data for training, which is time-consuming and labor-intensive. In addition, occlusion and noise in complex environments may affect the generalization ability of the model, making it less universal.

[0043] From the above-mentioned related technologies, it can be seen that how to automatically and quickly detect roadside trees from road scene laser point clouds and achieve accurate segmentation of single bodies of adhesion tree point clouds is a technical problem that needs to be solved urgently.

[0044] In view of this, some embodiments of the present application provide a method for extracting and segmenting street trees, which can filter the collected original road point cloud to obtain a processed road point cloud; then roughly extract the processed road point cloud to obtain a street tree point cloud, and then finely extract the street tree point cloud to obtain a trunk point cloud. Finally, through a joint constraint method based on connectivity coefficient and shortest path distance planning, the street tree extraction result is obtained by analyzing the trunk point cloud. Among them, the street tree extraction result may include: street tree monomer segmentation extraction result, street tree recognition extraction result and street tree trunk extraction result. Some embodiments of the present application realize more accurate street tree extraction and segmentation, and can accurately realize the extraction, identification, and street tree monomer segmentation of overlapping tree crowns of street trees and trunks, etc., providing reliable data support for urban forestry planning and management; multiple sets of data verification results in actual engineering applications show that the method of the present application effectively improves the accuracy and efficiency of tree point cloud extraction, forming a set of highly automated, highly stable and practical application value integrated technical solutions, and has certain advantages over traditional mainstream methods.

[0045] The following is combined with Figure 1The implementation process of street tree monomer segmentation performed by a terminal device provided in some embodiments of the present application is exemplified.

[0046] Please see attached Figure 1 , Figure 1 A flowchart of a method for extracting and segmenting street trees is provided for some embodiments of the present application. The method for extracting and segmenting street trees may include:

[0047] S110, performing filtering processing on the original road point cloud to obtain a processed road point cloud.

[0048] For example, in some embodiments of the present application, considering that the near-ground and building facade point clouds in the mobile measurement laser data are huge in volume and contain a lot of noise, the accurate extraction of street trees may cause interference. Therefore, in order to improve the accuracy of subsequent street tree monomer segmentation, the original road point cloud needs to be filtered to remove noise and obtain the processed road point cloud.

[0049] In some embodiments of the present application, S110 may include: filtering out invalid point clouds in the original road point cloud to obtain the processed road point cloud, the invalid point cloud including: at least one of a ground point cloud, a vegetation point cloud and a building facade point cloud set; wherein the vegetation point cloud is determined by a height threshold set along the elevation direction based on the ground point cloud; the building facade point cloud set is determined by analyzing the component of the normal vector of the point cloud after filtering out the ground point cloud in the elevation direction.

[0050] For example, in some embodiments of the present application, the essence of invalid point clouds is that they are not helpful for extracting street trees, and may even interfere with the noise or ground object points extracted from street trees, such as ground point clouds, objects, building point clouds, lawns, etc. First, it is necessary to filter out invalid point clouds such as ground point clouds, invalid object point clouds, and building facade point clouds. Specifically, a cloth simulation filtering algorithm is first used to quickly filter out the ground point clouds in the original road point cloud. Based on the ground point cloud, a height threshold is set in the elevation direction to remove invalid object point clouds such as curbstones and low vegetation in the original road point cloud. In addition, it is also necessary to remove the building facade point cloud set. Specifically, according to the spatial distribution relationship of the facades of urban buildings, the normal vector of the ground filtered point cloud is calculated. By analyzing the component of the normal vector in the elevation direction, that is, the component of the normal vector perpendicular to the surface, it is determined whether the facade characteristics are met. In actual scenarios, the facade of a building is usually manifested as the component of the normal vector in the elevation direction approaching zero. Based on this, the component threshold is set to -0.05~0.05. When the point component in the point cloud after filtering out the ground point cloud is within the component threshold, the point set is considered to be a building facade point cloud set and is eliminated.

[0051] It is understandable that the filtering algorithm may also adopt other types of algorithms, and the height threshold and component threshold may be set according to the actual application scenario, and the embodiments of the present application are not limited thereto.

[0052] S120, extracting roadside tree features in the processed road point cloud to obtain a roadside tree point cloud.

[0053] For example, in some embodiments of the present application, in order to obtain accurate street trees and eliminate interference from non-street tree objects, it is necessary to extract street tree point clouds from the processed road point clouds to obtain street tree point clouds.

[0054] In some embodiments of the present application, S120 may include: S121, extracting the processed road point cloud through density constraint to obtain an initial seed point set; S122, clustering the initial seed point set to obtain a clustered point cloud set; S123, extracting features of the clustered point cloud set through an elevation slice point frequency histogram to obtain the street tree point cloud.

[0055] For example, in some embodiments of the present application, potential tree point sets are identified and clustered based on density projection, and interference from non-roadside tree objects is excluded in combination with the elevation slice frequency distribution histogram. Considering that roadside trees are usually arranged vertically or approximately vertically upward along the ground, they have relatively stable column outer contour features. By performing a two-dimensional projection transformation on the processed road point cloud, and based on the characteristic that the projection density of rod-shaped objects in a certain neighborhood is much greater than that of other objects, the point set with too low local density is removed, and then the remaining point cloud that meets the requirements is reversely projected to obtain a point cloud in three-dimensional space (as a specific example of the initial seed point set), so as to achieve rapid identification and rough extraction of potential rod-shaped objects, such as roadside trees and street lamp point clouds. Afterwards, street tree fine extraction based on the elevation slice frequency histogram: the rough extraction method can remove most of the low-density objects, but there are still some abnormal noise points in the potential rod-shaped point set, and street lamps and road sign poles with similar distribution to tree trunks are not extraction targets and need to be removed. Among them, the local density is too low can refer to the local density being less than the set density threshold.

[0056] Specifically, considering the spatial distribution characteristics of street trees, the point set obtained by density constraint is used as the initial seed point set, and the regional growing algorithm is used for clustering to obtain a clustered point cloud set. Since street trees consist of two parts, the trunk and the crown, which are significantly different from other objects in spatial distribution and morphological structure, this step designs a street tree point cloud extraction method based on the elevation slice point frequency histogram to obtain the corresponding street tree point cloud.

[0057] In some embodiments of the present application, S123 may include: after iteratively segmenting the clustered point cloud set using the filtering method in the elevation direction, counting the number of point clouds in each round of slicing, and drawing the elevation slice point frequency histogram based on the number of point clouds; by counting the point density difference information in the elevation slice point frequency histogram and the cluster posture information of the street trees in space, eliminating the point cloud clusters of invalid objects from the clustered point cloud set to obtain the street tree point cloud.

[0058] Considering that the artificially planted street trees in urban areas are basically evergreen trees, the trunk morphology of the middle and lower parts mostly presents a nearly perfect vertical structure, but the crown of the street tree is usually large in size and wide in span, appearing clustered in space, and showing unique characteristics of high density and dynamic changes in the elevation direction. In addition, the density of the trunk and the crown varies greatly at the connection, while street lamps and traffic sign poles generally only have sudden changes in density at the extended lamp heads and sign positions. Therefore, in some embodiments of the present application, for the above-mentioned candidate street tree point set (that is, the cluster point cloud set), the lowest point at the bottom is used as the starting point, and a statistical interval of a certain step length is pre-set, and the cluster point cloud set is iteratively segmented using a straight-through filtering method in the elevation direction, the number of point clouds contained in each round of slicing is counted, and a point cloud number frequency distribution histogram based on elevation slicing (that is, elevation slicing point frequency histogram) is drawn. The information that the maximum elevation in the point cloud frequency distribution histogram is lower than the height threshold (for example, 2m), combined with the point density difference information and the cluster posture information, is used as a discriminant factor to eliminate the point cloud clusters of non-roadside tree objects (as a specific example of invalid objects) and obtain the final street tree point cloud. The point cloud clusters of non-roadside tree objects can be street lights, traffic sign poles, etc.

[0059] This is because, usually, the number of points in the trunk position of the roadside tree point cloud is relatively small and the density change trend is stable. As the slice height increases, the statistical histogram shows an obvious difference in point density at the dividing point between the trunk and the crown (as a specific example of point density difference information); and the distribution characteristics of the street lamp point cloud from bottom to top, from coarse to fine, cause the point density to show a gradually decreasing trend, with only a significant point density mutation at the lamp head position; traffic sign poles show similar characteristics. Therefore, based on the spatial distribution differences reflected by different elevation slices of land objects (as a specific example of clustered posture information), the complete and accurate extraction of street tree targets is achieved on the basis of eliminating the point clouds of objects such as street lamps and traffic signs. For example, by Figure 2 (a) can be extracted by extracting the cluster point cloud set Figure 2 (b) shows the point cloud of the roadside tree. Compared with the existing technology, the statistical analysis based on the layered frequency histogram can effectively eliminate the interference of the point cloud of similar rod-shaped objects, overcome the influence of noise, and improve the accuracy of roadside tree extraction.

[0060] S130, extracting tree trunk features in the roadside tree point cloud to obtain the tree trunk point cloud.

[0061] In some embodiments of the present application, S130 may include: dividing the regional data of the street tree point cloud into a crown point cloud and a non-crown point cloud through a preset trunk height threshold; obtaining a potential trunk point set in the non-crown point cloud; verifying the trunks in the potential trunk point set through a trunk surface curvature threshold to determine a candidate trunk point cloud; and re-verifying the candidate trunk point cloud to determine the trunk point cloud.

[0062] For example, in some embodiments of the present application, linear features are used to identify potential pole-shaped targets, and effective tree trunks are further selected based on the arc curvature features, aspect ratio, and hierarchical circle center fitting estimation method of single-sided scanning to obtain tree trunk point clouds. First, a rough extraction of tree trunks based on linear features is performed: tree trunks are an important part of street trees, and the precise extraction of tree trunk point clouds helps to obtain tree parameter information and reduce the complexity of subsequent individual segmentation of attached street trees.

[0063] Specifically, the corresponding trunk height threshold is set according to the elevation distribution of the trunk, and the regional data is divided into crown point cloud and non-crown point cloud. Considering that the outer contour of a conventional trunk presents a significant columnar feature, the principal component analysis method can be used to obtain potential trunk point sets from the non-crown point cloud.

[0064] Among them, for the three-dimensional non-crown point cloud {Pi (1 = 1, 2, 3...)}, the principal component analysis method uses the random sampling point as the center to find m neighboring points within the range, and calculates the covariance matrix of the decentralized data by the following formula: Among them, x i is the decentralized point data. The covariance matrix describes the relationship between different dimensions. p,i Perform eigenvalue decomposition to obtain eigenvalues ​​λ1, λ2...λn, (λ1>λ2>...>λn) and corresponding eigenvectors. Usually, the principal component direction of the tree trunk point cloud is highly consistent with the linear geometric direction, showing a significant columnar feature. Considering the spatial distribution characteristics of the tree trunk structure, the local linear features Linear=(λ1-λ2) / λ1 in the point set are identified to achieve preliminary extraction of potential tree trunk point sets. The schematic diagram of linear feature extraction of street tree trunk point cloud is shown in the figure. Figure 3 shown.

[0065] Due to the occlusion of dynamic and static objects in the measurement area, the tree trunks near the ground may be missing, and the potential tree trunk point set needs to be further selected. The regular sizes of street tree trunks in real scenes are basically similar (the diameter is generally 10-30cm). Therefore, whether it is a relatively complete or incomplete tree trunk point cloud caused by occlusion, the surface curvature of its local neighborhood is stable in a curvature threshold range (as a specific example of the surface curvature threshold). Based on this, the potential tree trunk point set is verified. Specifically, the surface curvature is calculated as follows:

[0066] For each point Pi(x i ,y i ,z i ), fit a local surface based on the least squares method, and solve these coefficients through the surrounding neighborhood points. The specific function formula is: f(x,y)=ax 2 +by 2 +cxy+dx+ey+f, a, b, c, d, e and f are unknown coefficients. In the tree trunk point cloud, the curvature usually has two principal curvatures: the maximum curvature k1 and the minimum curvature k2. These two principal curvatures can be calculated by fitting the quadratic coefficients of the surface:

[0067]

[0068] In the formula, H is the mean curvature, which indicates the curvature of the surface near a certain point; G is the Gaussian curvature term, and f x and f y is the gradient of the surface in the x and y directions, and the surface curvature K = k1·k2.

[0069] After solving the surface curvature through the above method, it is compared with the curvature threshold interval to obtain the candidate trunk point cloud within the interval. Finally, the candidate trunk point cloud needs to be verified twice before the trunk point cloud can be determined.

[0070] In some embodiments of the present application, a bounding box is established for the candidate trunk point cloud, and a reference value of the bounding box in the three-dimensional coordinate axis is calculated; if it is confirmed that the reference value is greater than a preset value, the candidate trunk point cloud is used as the trunk point cloud; if it is confirmed that the reference value is less than or equal to the preset value, the candidate trunk point cloud is further verified to obtain the trunk point cloud.

[0071] For example, in some embodiments of the present application, due to the significant difference in height and width of the trunks of street trees, the height is much greater than the width, resulting in a large aspect ratio. Therefore, an AABB (Axis-Aligned Bounding Box) bounding box is established for the candidate trunk point cloud, and the difference in the coordinate extreme values ​​of the bounding box in the three-axis direction (that is, the three-dimensional coordinate axis) is calculated to obtain the aspect ratio value (as a specific example of a reference value). When the aspect ratio value exceeds a preset value (for example, the preset value is 3), the point set is considered to conform to the spatial distribution form of the trunk and is identified as a trunk point cloud. When the candidate trunk point set cannot meet the preset value of the aspect ratio, the candidate trunk point set may be missing due to occlusion. At this time, the next step of verification is required to determine the trunk point cloud.

[0072] In some embodiments of the present application, the candidate trunk point cloud is further verified to obtain the trunk point cloud, including: performing equally spaced elevation pass-through filtering on the candidate trunk point cloud to generate multiple slices; using a random sampling consistency algorithm to fit the hierarchical center estimation results of the local arcs of the multiple slices projected on the horizontal plane; obtaining the circle radius fitting values ​​of the multiple slices through the hierarchical center estimation results; comparing the circle radius fitting values ​​of the multiple slices with a radius threshold to determine the trunk point cloud.

[0073] For example, in some embodiments of the present application, multiple slices are generated by selecting candidate tree trunk point clouds for equal-spaced elevation straight-through filtering, and the hierarchical circle center estimation results of the local arcs of the multiple slices projected on the horizontal plane are fitted using a random sampling consistency algorithm. The circle radius fitting results obtained by fitting the circle radii of the multiple slices are compared with the radius threshold (e.g., the radius threshold is limited to ±3cm) to further verify the valid tree trunk point cloud. The verification results are as follows: Figure 4 shown.

[0074] Compared with the prior art, this step of the present application not only relies on a single judgment criterion, but also effectively improves the extraction accuracy of street tree trunks through joint constraints based on curvature, aspect ratio and layered center estimation, overcoming the problem of low universality of conventional algorithms. Even in the face of the slightly tilted distribution or partially missing features that often appear in street tree trunks, the prior art may have reduced accuracy or even fail, but this solution can achieve a higher extraction accuracy.

[0075] S140, obtaining a street tree extraction result by analyzing the tree trunk point cloud, wherein the street tree extraction result includes: at least one of a street tree monomer segmentation extraction result, a street tree recognition extraction result and a street tree trunk extraction result.

[0076] It should be understood that the embodiments of the present application can achieve the technical effects of accurate street tree identification and extraction, tree trunk extraction, and monomer segmentation of overlapping street trees by filtering, feature extracting, and analyzing the original road point cloud to obtain street tree extraction results. The street tree extraction results may include street tree segmentation, street tree identification and extraction, tree trunk identification and extraction, etc., and the embodiments of the present application are not limited thereto.

[0077] In order to facilitate the explanation of the implementation process of this solution, the following description is made by taking the segmentation and extraction of overlapping street trees as an example.

[0078] For example, in some embodiments of the present application, the connectivity coefficient and the shortest path statistical analysis method can be combined to segment trees whose crown point clouds are connected (i.e., overlapping street trees) corresponding to the trunk point cloud, so as to achieve the individualization of the street tree point cloud (i.e., street tree individual segmentation). Among them, the crown point cloud obtained above can be considered to have segmented a single tree, and what needs to be done later is to segment the tree crown point cloud corresponding to the trunk point cloud if there is adhesion.

[0079] In some embodiments of the present application, S140 may include: performing an adjacency analysis on the crown point cloud corresponding to the trunk point cloud, and when it is determined that there is a cross-overlapping area of ​​the crown, solving the connectivity coefficient corresponding to the cross-overlapping area of ​​the crown; when the connectivity coefficient is less than a coefficient threshold, calculating the spatial distance between adjacent crown points in the cross-overlapping area of ​​the crown, and constructing a global graph node; according to the global graph node, segmenting the trees in the cross-overlapping area of ​​the crown to obtain the segmentation result of the street tree monomers.

[0080] For example, in some embodiments of the present application, this step segments the overlapping area of ​​the crown in the trunk point cloud by combining two methods. Among them, the adjacency analysis method is to construct the connectivity function equation of the voxel boundary unit based on the neighborhood distribution, and explore the positional relationship between the canopy seed point and the neighboring points. In order to make the adjacency analysis method applicable to the segmentation of the street tree monomer in the present application, the adjacency analysis method is improved. The improved adjacency analysis algorithm of the present application first establishes voxel units for the crown point cloud, and pre-sets the radius threshold according to the crown width. Within this range, a unit is defined as the top unit (generally the crown vertex), and the position of the seed is defined as the centroid of all points in the unit. Check whether there are other top units outside the radius threshold. If so, it is determined whether the distance between the two centroid points is greater than the preset radius. If it is less than, it indicates that the two trees are close to each other, and when the cluster point set contains more than two trunk point clouds, it is treated as a sticky tree for subsequent processing; otherwise, it belongs to two separate trees.

[0081] Specifically, according to the pre-set hierarchical height threshold, the crown point cloud is divided into multiple vertical layers. The core idea of ​​the adjacency analysis algorithm is to start from the potential seeds at the top of the cluster and assign them downward to the bottom layer. If the neighboring units of the current layer are only connected to one seed unit, these units are assigned to the corresponding tree. Traversing downward step by step, these units inherit the cluster index and form new seeds of the next layer, and the same operation is performed. However, when traversing to a layer where all units are connected together and inherit more seed indexes, that is, there is a crown cross-overlap area in the adjacent crowns, the connectivity coefficient is then used for separation. Among them, the formula of the connectivity coefficient is: con(t,s)=C(t,s)×R(k); where con(t,s) represents the connectivity coefficient of the target unit relative to the source unit; R(k) represents the influence attenuation coefficient of the source unit on its unassigned neighboring units, R(k)=1 / k; where k is defined as the voxel length of the shortest path connecting the source and the target. C(t,s) is the neighbor type of the target point relative to the source point, which is obtained by the following method:

[0082]

[0083] In the formula, if t and s share a face, or all cells on the straight line connecting t and s are face neighbors of s, then the cell t is defined as the face neighbor of the source cell s; in addition, cells with other attributes are all edge neighbors of the source cell s.

[0084] When all the crown point boundary cells in a layer propagate their connectivity coefficients to the adjacent connected unassigned cells of different orders, until the cells in the same layer have the cumulative connectivity coefficients corresponding to each tree. According to the size of the connectivity coefficient, the cells are assigned to the tree with the larger cumulative connectivity value. After all cells are assigned to a single crown, that is, the points in the voxel have a definite affiliation, and finally the segmentation of overlapping crowns is achieved.

[0085] The segmentation method based on adjacency analysis is effective and efficient, but different voxel units often contain point clouds with large density differences. The connectivity coefficients of the overlapping areas of the crowns of two highly similar trees in the same voxel unit may be very close (that is, the connectivity coefficient is less than the coefficient threshold). At this time, judging the ownership of the points in the entire voxel unit based only on the size of the connectivity coefficient is prone to segmentation errors, causing points that should belong to one tree to be incorrectly assigned to another tree. Therefore, based on the adjacency analysis, we focus on the voxel space with extremely close connectivity coefficients (for example, the coefficient threshold is set to 0.3), and by constructing global graph nodes, the Floyd shortest path planning algorithm is used to perform secondary verification of the voxel points in the overlapping areas of the crowns. It can be understood that for the points in the voxel unit whose connectivity coefficient is greater than or equal to the coefficient threshold, their ownership can be directly determined.

[0086] As street trees are artificial landscape trees in the city, the edge points of their crowns and the center of mass of the entire tree have a relatively stable positional relationship. This is mainly reflected in the fact that the physical growth process of the trunk and side branches from the inside to the outside has a strong topological correlation with the end point. Therefore, by comparing the shortest path distance from a certain point to the center of mass of each tree, the ownership relationship can be effectively determined.

[0087] The data volume is reduced by downsampling the overlapping areas of the crowns, and the disordered point cloud in space is adjusted to a uniform distribution. Then, the spatial distance between each two overlapping crown points in the overlapping areas of the crowns is calculated, and the global graph node is constructed using the kd tree neighbor distance search algorithm. Finally, the street tree monomers are segmented using the global graph nodes and the Floyd shortest path planning algorithm.

[0088] In some embodiments of the present application, S140 may include: taking a randomly selected overlapping crown point as the starting point, and taking the centroid coordinates of two trees adjacent to the overlapping crown point as the first end point and the second end point; solving the first path and the second path from the starting point to the first end point and the second end point respectively on the global graph node; determining the target tree to which the overlapping crown point belongs through the distances corresponding to the first path and the second path respectively, so as to obtain the street tree monomer segmentation result for the two trees, wherein the target tree is one of the two trees.

[0089] For example, in some embodiments of the present application, Figure 5 As shown in the figure, the crown points of the randomly overlapping areas in the voxel units of the crown cross-overlapping area are taken as the initial starting point Ps, and the two adjacent trees (i.e. Figure 5 The centroid coordinates T of the left tree and the right tree in L (i.e., the first endpoint) and T R (i.e., the second end point) as the end point, and use the Floyd algorithm on the global graph node to solve the shortest path from Ps to the centroid coordinate point. Specifically, obtain the shortest path from Ps to T L The shortest path (i.e. the first path) A. Get Ps to T R The shortest path (i.e., the second path) B. Take the smaller value of A and B as the tree to which the crown point Ps in the random overlapping area belongs. For example, A = 2, B = 3, when the value of A is smaller, then the crown point Ps in the random overlapping area belongs to Figure 5 The coordinates of the center of mass are T L The left tree of (as a specific example of the target tree) can be obtained through path analysis as follows Figure 6 The segmentation result is shown.

[0090] Compared with the prior art, the present invention optimizes and corrects controversial tree points in all voxel units through the comprehensive determination of the combined connectivity coefficient and the shortest path, thereby improving the defect of insufficient segmentation precision of the single adjacency analysis method and obtaining a more robust monomerized segmentation result.

[0091] The following is combined with Figure 7 The specific process of dividing a street tree monomer provided by some embodiments of the present application is exemplified.

[0092] Please see attached Figure 7 , Figure 7 A flow chart of a method for street tree extraction and monomer segmentation provided for some embodiments of the present application.

[0093] The above process is explained below as an example.

[0094] S701, filtering out invalid point clouds such as ground points, building facades and low vegetation in the original road point cloud to obtain a processed road point cloud.

[0095] S702, identifying and extracting pole-shaped objects such as roadside trees and street lamps from the processed road point cloud through projection and density constraint to obtain an initial seed point set.

[0096] S703, clustering the initial seed point set to obtain a clustered point cloud set.

[0097] S704, extracting features from the clustered point cloud set through the elevation slice point frequency histogram to obtain a street tree point cloud.

[0098] S705, dividing the regional data of the roadside tree point cloud into a crown point cloud and a non-crown point cloud according to a preset tree trunk height threshold.

[0099] S706, obtaining a potential trunk point set in the non-crown point cloud; verifying the trunks in the potential trunk point set by using a trunk surface curvature threshold to determine a candidate trunk point cloud. The potential trunk point set is obtained by using a principal component analysis method.

[0100] S707, establishing a bounding box for the candidate tree trunk point cloud, and calculating the reference value of the bounding box on the three-dimensional coordinate axis.

[0101] S708, determine whether the reference value is greater than the preset value, if so, execute S709, otherwise execute S710.

[0102] S709: taking the candidate tree trunk point cloud as the tree trunk point cloud.

[0103] S710, identifying the candidate tree trunk point cloud by using layered circle center estimation to obtain the tree trunk point cloud.

[0104] S711, determine whether there is an overlapping area between corresponding tree crowns through the tree trunk point cloud, if so, execute S711, otherwise execute S715.

[0105] S712, performing adjacency analysis on the crown point cloud corresponding to the trunk point cloud, and when it is determined that there is a crown cross-overlapping area, solving the connectivity coefficient corresponding to the crown cross-overlapping area.

[0106] S713, when the connectivity coefficient is less than the coefficient threshold, calculate the spatial distance between adjacent crown points in the crown intersection and overlap area, and construct a global graph node.

[0107] S714, segmenting the point cloud of the overlapping area of ​​the tree crowns according to the global graph nodes and the shortest path algorithm to determine the ownership relationship.

[0108] S715, output the street tree monomer segmentation result.

[0109] It is understood that the specific implementation process of S701 to S715 can refer to the method embodiment provided above, and in order to avoid repetition, the detailed description is appropriately omitted here. It should be understood that the method of street tree extraction and monomer segmentation in this application is not subject to Figure 7 The number and order of the execution steps shown are limited, and the embodiments of the present application are not limited thereto.

[0110] It should be noted that the filtering algorithms, feature extraction and other method steps involved above can be replaced by other algorithms with similar functions according to the actual scene requirements, and the embodiments of the present application are not limited to the above embodiments. For example, in the preprocessing of data, there are various methods to filter out ground point clouds and building facade point clouds, such as morphological-based ground filtering, slope-based ground filtering, density-based building facade filtering and other methods may also achieve the same function in the data preprocessing process of this application, so the related filtering method is an alternative solution in the preprocessing. Similarly, there are many methods to remove noise points in the point cloud, such as statistical-based filtering methods, which may also be alternative solutions. However, it should be noted that the purpose of the above-mentioned related filtering methods is to remove invalid point sets such as ground points and building points in advance as much as possible, which does not affect the core content and technical methods of automatic extraction and monomer segmentation of street tree point clouds proposed in this application.

[0111] It can be seen from some embodiments of the present application that the present application has the following advantages:

[0112] 1) This application can automatically extract street tree point clouds from mobile measurement three-dimensional laser data, and quickly identify potential trunk point sets through linear features. It further selects trunk point clouds using surface curvature, aspect ratio and layered circle center radius estimation threshold constraint method, overcoming the problem of false detection caused by slightly tilted or partially missing trunk distribution features that often appear in actual scenes, improving the extraction accuracy of street trees and trunk point clouds, and providing important data support for digital urban development.

[0113] 2) The single tree segmentation process of this application adopts the connectivity coefficient and the shortest path iterative judgment method, which fully considers the topological correlation relationship of the physical growth of trees in the adhesion area from the inside to the outside, improves the defect of insufficient segmentation precision of the single adjacency method, and makes the attribution of the edge point cloud of the voxel unit in the overlapping area of ​​street trees more accurate, thereby improving the segmentation accuracy of multiple adhesion trees and making the statistical information of single tree parameters more accurate.

[0114] 3) In terms of mobile measurement and real-life three-dimensional (3D) measurement, the method of this application can make full use of high-precision laser point cloud data. The highly automated and integrated technical method can greatly reduce the technical difficulty, field work intensity and time cost of personnel, and the extraction results are robust and efficient. This provides a basis for the rapid and accurate statistical inventory of street trees in large-scale urban scenes and has strong scalability.

[0115] Please refer to Figure 8 , Figure 8 The block diagram of the composition of the apparatus for extracting and segmenting roadside trees provided by some embodiments of the present application is shown. It should be understood that the apparatus for extracting and segmenting roadside trees corresponds to the above method embodiment and can execute the various steps involved in the above method embodiment. The specific functions of the apparatus for extracting and segmenting roadside trees can be found in the description above. To avoid repetition, the detailed description is appropriately omitted here.

[0116] Figure 8 The device for street tree extraction and monomer segmentation includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the device for street tree extraction and monomer segmentation, and the device for street tree extraction and monomer segmentation includes: a preprocessing module 810, used to filter the original road point cloud to obtain a processed road point cloud; a first extraction module 820, used to extract street tree features in the processed road point cloud to obtain a street tree point cloud; a second extraction module 830, used to extract trunk features in the street tree point cloud to obtain a trunk point cloud; a result acquisition module 840, used to obtain a street tree extraction result by analyzing the trunk point cloud, wherein the street tree extraction result includes: at least one of a street tree monomer segmentation extraction result, a street tree recognition extraction result and a street tree trunk extraction result.

[0117] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method, and will not be described in detail here.

[0118] Some embodiments of the present application further provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can implement the operations of the method corresponding to any of the above methods provided in the above embodiments.

[0119] Some embodiments of the present application further provide a computer program product, which includes a computer program, wherein when the computer program is executed by a processor, it can implement the operations corresponding to any of the above methods provided in the above embodiments.

[0120] like Fig. 9 As shown, some embodiments of the present application provide an electronic device 900, which includes: a memory 910, a processor 920, and a computer program stored in the memory 910 and executable on the processor 920, wherein the processor 920 can implement a method as described in any of the above embodiments when reading the program from the memory 910 through a bus 930 and executing the program.

[0121] Processor 920 can process digital signals and can include various computing structures, such as complex instruction set computer structure, reduced instruction set computer structure, or a structure that implements a combination of multiple instruction sets. In some examples, processor 920 can be a microprocessor.

[0122] The memory 910 may be used to store instructions executed by the processor 920 or data related to the execution of instructions. These instructions and / or data may include codes for implementing some or all functions of one or more modules described in the embodiments of the present application. The processor 920 of the disclosed embodiment may be used to execute instructions in the memory 910 to implement the method shown above. The memory 910 includes a dynamic random access memory, a static random access memory, a flash memory, an optical memory, or other memory known to those skilled in the art.

[0123] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0124] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0125] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for extracting and segmenting street trees, characterized in that: include: Perform filtering on the original road point cloud to obtain the processed road point cloud; Extracting roadside tree features in the processed road point cloud to obtain a roadside tree point cloud; Extracting trunk features from the roadside tree point cloud to obtain the trunk point cloud; By analyzing the tree trunk point cloud, a roadside tree extraction result is obtained, wherein the roadside tree extraction result includes: at least one of a roadside tree monomer segmentation extraction result, a roadside tree recognition extraction result, and a roadside tree trunk extraction result; The step of extracting the trunk features in the roadside tree point cloud to obtain the trunk point cloud includes: The regional data of the roadside tree point cloud is divided into a crown point cloud and a non-crown point cloud by a preset trunk height threshold; a potential trunk point set in the non-crown point cloud is obtained; the trunks in the potential trunk point set are verified by a trunk surface curvature threshold to determine a candidate trunk point cloud; When it is confirmed that the reference value of the bounding box of the candidate trunk point cloud is less than or equal to a preset value, the candidate trunk point cloud is subjected to equally spaced elevation pass-through filtering to generate multiple slices; a random sampling consistency algorithm is used to fit the hierarchical circle center estimation results of the local circular arcs of the multiple slices projected on the horizontal plane; the circle radius fitting values ​​of the multiple slices are obtained through the hierarchical circle center estimation results; the circle radius fitting values ​​of the multiple slices are compared with the radius threshold to determine the trunk point cloud.

2. The method according to claim 1, characterized in that The filtering process of the original road point cloud to obtain the processed road point cloud includes: Filtering out invalid point clouds in the original road point cloud to obtain the processed road point cloud, wherein the invalid point cloud includes at least one of a ground point cloud, a vegetation point cloud and a building facade point cloud; The vegetation point cloud is determined by setting a height threshold along the elevation direction based on the ground point cloud; the building facade point cloud set is determined by analyzing the component of the normal vector of the point cloud in the elevation direction after filtering out the ground point cloud.

3. The method according to claim 1 or 2, characterized in that The step of extracting the roadside tree features in the processed road point cloud to obtain the roadside tree point cloud includes: Extracting the processed road point cloud through density constraints to obtain an initial seed point set; Clustering the initial seed point set to obtain a clustered point cloud set; The clustered point cloud set is feature extracted through the elevation slice point frequency histogram to obtain the street tree point cloud; the elevation slice point frequency histogram is obtained by iteratively segmenting the clustered point cloud set using an elevation direction filtering method, counting the number of point clouds in each round of slicing, and drawing according to the number of point clouds.

4. The method according to claim 3, characterized in that The step of extracting features from the clustered point cloud set by using the elevation slice point frequency histogram to obtain the street tree point cloud comprises: By counting the point density difference information in the elevation slice point frequency histogram and the cluster posture information of the street trees in space, the point cloud clusters of invalid objects are eliminated from the cluster point cloud set to obtain the street tree point cloud.

5. The method according to claim 1, characterized in that The step of re-verifying the candidate tree trunk point cloud to determine the tree trunk point cloud includes: Establishing a bounding box for the candidate tree trunk point cloud, and calculating a reference value of the bounding box on a three-dimensional coordinate axis; If it is confirmed that the reference value is greater than the preset value, the candidate trunk point cloud is used as the trunk point cloud.

6. The method according to claim 1 or 2, characterized in that: The step of obtaining a street tree extraction result by analyzing the tree trunk point cloud includes: Performing adjacency analysis on the crown point cloud corresponding to the trunk point cloud, and when determining that there is a crown cross-overlap area, solving the connectivity coefficient corresponding to the crown cross-overlap area; When the connectivity coefficient is less than a coefficient threshold, the spatial distance between adjacent crown points in the crown intersection and overlap region is calculated to construct a global graph node; The trees in the overlapping area of ​​the tree crowns are segmented according to the global graph nodes to obtain the street tree monomer segmentation results.

7. The method according to claim 6, characterized in that The step of segmenting the trees in the overlapping area of ​​the tree crowns according to the global graph nodes to obtain the street tree monomer segmentation result includes: Taking the randomly selected overlapping crown point as the starting point, and taking the centroid coordinates of the two trees adjacent to the overlapping crown point as the first end point and the second end point; Solve the first path and the second path from the starting point to the first end point and the second end point respectively on the global graph node; The target tree to which the overlapping crown points belong is determined by the distances respectively corresponding to the first path and the second path, so as to obtain the street tree monomer segmentation and extraction results for the two trees, wherein the target tree is one of the two trees.

8. A device for extracting and segmenting street trees, characterized in that: The device is used to perform the method according to claim 1, comprising: A preprocessing module is used to filter the original road point cloud to obtain the processed road point cloud; A first extraction module is used to extract the roadside tree features in the processed road point cloud to obtain the roadside tree point cloud; The second extraction module is used to extract the trunk features in the roadside tree point cloud to obtain the trunk point cloud; The result acquisition module is used to acquire the street tree extraction result by analyzing the tree trunk point cloud, wherein the street tree extraction result includes: at least one of the street tree monomer segmentation extraction result, the street tree recognition extraction result and the street tree trunk extraction result.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program executes the method according to any one of claims 1 to 7 when executed by a processor.

10. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the computer program executes the method according to any one of claims 1 to 7 when being run by the processor.

11. A computer program product, characterized in that The computer program product comprises a computer program, wherein the computer program executes the method according to any one of claims 1 to 7 when executed by a processor.

Images