CN112348829B - Method for separating branches and leaves of ground LiDAR point cloud based on modal point evolution - Google Patents

Method for separating branches and leaves of ground LiDAR point cloud based on modal point evolution Download PDF

Info

Publication number
CN112348829B
CN112348829B CN202011205299.6A CN202011205299A CN112348829B CN 112348829 B CN112348829 B CN 112348829B CN 202011205299 A CN202011205299 A CN 202011205299A CN 112348829 B CN112348829 B CN 112348829B
Authority
CN
China
Prior art keywords
node
nodes
point
modal
evolution
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011205299.6A
Other languages
Chinese (zh)
Other versions
CN112348829A (en
Inventor
惠振阳
李大军
刘波
鲁铁定
吴静
余美
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
East China Institute of Technology
Original Assignee
East China Institute of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by East China Institute of Technology filed Critical East China Institute of Technology
Priority to CN202011205299.6A priority Critical patent/CN112348829B/en
Publication of CN112348829A publication Critical patent/CN112348829A/en
Application granted granted Critical
Publication of CN112348829B publication Critical patent/CN112348829B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G06T2207/30188—Vegetation; Agriculture

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Geometry (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a method for separating branches and leaves of a ground LiDAR point cloud based on modal point evolution, which comprises the following steps: s1, segmenting the ground LiDAR point cloud by adopting a Mean Shift segmentation method to obtain modal points; s2, constructing a graph structure by using the modal points and analyzing the shortest path; s3, detecting leaf nodes based on path backtracking and node evolution; and S4, performing branch node detection based on the node access frequency and the node evolution. According to the method for separating branches and leaves of the ground LiDAR point cloud based on modal point evolution, modal points are obtained by a Mean Shift method, each modal point corresponds to a segmentation object, and branch and leaf separation based on the points is converted into branch and leaf separation based on the objects. Compared with the branch and leaf separation based on the geometric characteristics, the object-based method can greatly reduce the calculation amount and improve the separation efficiency.

Description

Method for separating branches and leaves of ground LiDAR point cloud based on modal point evolution
Technical Field
The invention relates to the technical field of branch and leaf separation methods, in particular to a ground LiDAR point cloud branch and leaf separation method based on modal point evolution.
Background
Three-dimensional laser scanning (LiDAR) technology has evolved dramatically in recent years. LiDAR systems can be classified into airborne LiDAR, ground LiDAR, and handheld LiDAR, depending on the operating platform. Ground LiDAR is a LiDAR system that mounts a three-dimensional laser scanner on a tripod. It can actively transmit laser pulses to the ground and receive echo information from ground target objects. Compared with other measurement modes, the ground LiDAR can quickly and accurately acquire dense point cloud data. Therefore, ground LiDAR has been widely used in forest resource surveys, such as vegetation parameter estimation, biomass estimation, and leaf area index calculation.
For most post-processing applications that employ ground LiDAR for forest resource investigation, branch and leaf separation is a prerequisite that must be achieved. For example, when calculating the leaf area index, the leaves first need to be extracted. The presence of branches will overestimate the leaf area index between 3% and 32%. Furthermore, the presence of leaves also affects the estimation when calculating the amount of impoverishment and estimating the amount of surface biomass. Therefore, accurate separation of leaves and branches is required first for subsequent vegetation application treatments. However, branch and leaf separation remains a challenging task. Especially in complex forest environments, accurate branch and leaf separation is still difficult to achieve.
The method based on the geometric features is a common branch and leaf separation method in the prior art, and the method based on the geometric features mainly realizes the branch and leaf separation based on the different geometric features of leaves and branches. For most machine learning methods, when the branches and leaves are separated by using geometric features, sample marking is required, and a large amount of time is usually required in the process, so that the calculation efficiency is influenced. Furthermore, calculating the geometric features of the individual points often requires selecting the appropriate radius of the field. Inaccurate neighborhood radius often results in poor classification. Although better classification results can be obtained by using the multi-scale neighborhood, the calculation amount of the multi-scale features is large, and the time is often multiplied.
Disclosure of Invention
The invention aims to provide a method for separating branches and leaves of a ground LiDAR point cloud based on modal point evolution, so as to solve the problems of large calculated amount and low efficiency in the prior art.
A method for separating branches and leaves of a ground LiDAR point cloud based on modal point evolution comprises the following steps:
s1, segmenting the ground LiDAR point cloud by adopting a Mean Shift segmentation method to obtain modal points;
s2, constructing a graph structure by using the modal points and analyzing the shortest path;
S3, detecting leaf nodes based on path backtracking and node evolution;
and S4, detecting the branch nodes based on the node access frequency and the node evolution.
According to the method for separating branches and leaves of the ground LiDAR point cloud based on modal point evolution, modal points are obtained by adopting a Mean Shift method, each modal point corresponds to one segmentation object, and branch and leaf separation based on points is converted into branch and leaf separation based on the objects. Compared with the branch and leaf separation based on the geometric characteristics, the object-based method can greatly reduce the calculated amount and improve the separation efficiency. And detecting the leaf nodes and the branch seed nodes by backtracking the paths and calculating the access frequency of each node. And finally, acquiring a final branch point cloud based on the branch nodes obtained by evolution and the Mean Shift segmentation result. The method is simple in process and easy to implement, and can provide a good foundation for the application of subsequent ground LiDAR in the forest area.
In addition, the method for separating branches and leaves of a ground LiDAR point cloud based on modal point evolution can also have the following additional technical characteristics:
further, in step S1, the Mean Shift vector is calculated using the following equation:
Figure BDA0002756827640000021
In the formula, Meanshifth(Vp) Representing the MeanShift vector, V for a three-dimensional point cloudpThe three-dimensional coordinates of the p-point are represented,n represents the number of neighboring points, the value of n is determined by the bandwidth h, and g (g) is a gaussian equation.
Further, the value range of the bandwidth h is: h belongs to [0.5,1.0 ].
Further, in step S2, each side length in the graph structure constructed by using the modal points represents a distance between adjacent modal points, and a path exists between each node and a base point, where the base point is the modal point with the smallest height value.
Further, in step S2, the graph structure is constrained using the following equation:
Figure BDA0002756827640000031
in the formula, Edge (p)i,pj) Is a node piAnd pjEdge between, dis (p)i,pj) For the geometric distance between these two nodes, r is the constraint radius.
Further, in step S2, the shortest path analysis is performed using the following equation:
SP(Graph,base,pm)＝{pm,pn,L,base}
wherein SP (g) represents the shortest path, Graph represents the constructed Graph, base represents the base point, pmRepresenting end nodes, pnIs the node through which the shortest path passes.
Further, in step S3, the path backtracking means deleting a series of nodes from the end node to the base point, and the deleted node is determined by the backtracking step number.
Further, in step S3, the nodes passed by the path are referred to as seed points, and for each seed point, the node is referred to as a seed point
Figure BDA0002756827640000032
Its neighboring nodes are calculated according to:
Figure BDA0002756827640000033
SPL(pi,pj)＝SP(Graph,pi,pj,weights)
in the formula, pjIs the node in the graph, n is the number of nodes in the graph, ctd (p)i,pj) Representing a node piTo pjThe weights represent weights between different nodes in Graph, and D is a threshold value of the evolution radius.
Further, in step S3, after finding the neighboring point of each seed node, the evolution is performed according to the seed node, if the node p isiSatisfies the condition of the formula, node piWill be evolved into non-leaf nodes;
Figure BDA0002756827640000034
in the formula (I), the compound is shown in the specification,
Figure BDA0002756827640000035
representing a node piThe path length to the base point is less than that of the seed node
Figure BDA0002756827640000036
The length of the path to the base point.
Further, in step S4, the branch nodes are detected by the following formula:
Figure BDA0002756827640000041
in the formula (I), the compound is shown in the specification,
Figure BDA0002756827640000042
is a node piMax (log (f)) represents the maximum value of the logarithm of the access frequency of the node, n is the number of the nodes, and δ is a constant.
Drawings
The above and/or additional aspects and advantages of embodiments of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a logic flow diagram of a method for modal point evolution-based ground LiDAR point cloud branch and leaf separation provided in accordance with an embodiment of the present invention;
FIG. 2 is a schematic diagram of a clustering process of the Mean Shift method;
FIG. 3 is a schematic view of a modal point acquisition process, (a) an original tree point cloud; (b) mean Shift segmentation results; (c) obtaining a result by the modal point;
FIG. 4 is a diagram illustrating shortest paths from all nodes to a base point;
FIG. 5 is a diagram illustrating a path backtracking result;
FIG. 6 is a graph of Euclidean distance versus commute time distance;
FIG. 7 is a diagram illustrating leaf node detection results after seed node evolution;
FIG. 8 is a schematic diagram of a point cloud modal point display based on node access frequency;
FIG. 9 is a schematic diagram of a branch point detection implementation based on leaf node access frequency;
FIG. 10 is a schematic diagram of tree nodes and tree point clouds;
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In view of the problems in the prior art, the invention provides a ground LiDAR point cloud branch and leaf separation method based on modal point evolution. Obtaining the modal point requires firstly adopting a Mean Shift method to segment ground LiDAR point cloud, wherein the clustering center of each part is the modal point. Thereafter, a graph-like structure is constructed using the modal points to reflect the main structural information of the trees and leaves. All nodes in the graph are composed of modal points, and each side length represents the distance between adjacent modal points. A path exists between each node and a base point, wherein the base point refers to a modal point with the minimum elevation value. Typically, in each path from a node to a base point, a branch node has a higher node access frequency. This is because each path must necessarily pass through a trunk or branch node. According to this feature, a node having a higher access frequency can be identified as a branch node. Meanwhile, leaf nodes are typically the end nodes of each path. Therefore, the nodes at the end of the path can be identified as leaf nodes by the path backtracking step. However, branch nodes often cannot be identified completely accurately. Particularly, when the set access frequency threshold is large, the number of the acquired branch nodes is often small. In order to obtain more branch nodes, the invention provides a method for evolving based on an initial detection branch node. Its evolution process is based on three conditions, namely: (1) the commute time distance is small; (2) the nodes should have similar verticality; (3) the node does not belong to a leaf node detected according to the path backtracking. And after all the branch nodes are accurately detected, fusing the branch nodes with corresponding Mean Shift segmentation results to obtain final branch point cloud.
Referring to fig. 1, the method for separating branches and leaves of a ground LiDAR point cloud based on modal point evolution includes the following steps S1 to S4.
S1, the ground LiDAR point cloud is segmented by adopting a Mean Shift segmentation method to obtain the modal points.
As described above, the modal point plays a crucial role in acquiring the tree point cloud. Using modal points instead of point clouds has two advantages. In one aspect, representing the results of point cloud segmentation with modal points enables point-based branch and leaf separation to be translated into object-based segmentation. Therefore, the method can obviously improve the experimental efficiency of the method. On the other hand, the point cloud data is difficult to construct, and the modal points are used, so that the calculation burden can be greatly reduced.
The Mean Shift method is a non-parametric clustering method. Compared with the traditional K-means method, the Mean Shift method does not need to preset the clustering number. Therefore, the Mean Shift method can be generally applied to clustering and segmentation of unknown scenes. The Mean Shift method is an iterative method. In each iteration, the Mean Shift vector is first calculated, as shown by the arrow in fig. 2. The Mean Shift vector generally points in the direction of increasing probability density. Thus, after a number of iterations, the points are moved to the corresponding modal point locations. Points that have the same or similar modal points will be grouped into a class.
The Mean Shift vector is calculated according to equations (1), (2).
Figure BDA0002756827640000061
In the formula, Meanshifth(Vp) Representing the MeanShift vector. For three-dimensional point clouds, VpRepresenting the three-dimensional coordinates of the p points, n representing the number of neighboring points, the value of n being generally determined by the bandwidth h. G (g) is a Gaussian equation. When in use
Figure BDA0002756827640000062
Above 1, G (g) has a value of 0. Therefore, the bandwidth h will directly affect the clustering result. Larger h will cause more points to be grouped into one class, resulting in an under-segmented final result. Conversely, a smaller h will over-partition the result. The invention aims to cut the trunk or branch into small sections by using a Mean Shift method. Thus, the set bandwidth h only needs to be larger than the diameter of the trunk. In general, h ∈ [0.5,1.0 ]]。
Fig. 3 is a process of acquiring a modal point based on a point cloud. Fig. 3(a) is the original point cloud data of the tree, and fig. 3(b) is the Mean Shift segmentation result, from which it can be seen that the trunk and branch are segmented into several small segments, and the leaf point cloud is segmented into several point sets. For each Mean Shift segmented object, its corresponding modality point is retained, and the final modality point result is obtained, as shown in fig. 3 (c).
And S2, constructing a graph structure by using the modal points and analyzing the shortest path.
After the modality points are acquired, the modality points can be used to construct a graph structure. Compared with a method taking all points as nodes, the method provided by the invention is obviously faster, easy to implement and capable of reducing the calculation burden. The Graph is composed of nodes and edges, and may be represented as Graph (Node, Edge). As described above, the nodes are modal points, and an edge exists between two adjacent nodes. In order to further reduce the complexity of the graph structure, the invention adopts a formula (2) for constraint.
Figure BDA0002756827640000063
In the formula, Edge (p)i,pj) Is a node piAnd pjEdge between, dis (p)i,pj) R is the constraint radius for the geometric distance between these two nodes. The meaning of formula (2) is if pjIs a distance piA neighboring point smaller than the constraint radius r, then the node pjAnd piThere is a side with a weight equal to the geometric distance between two nodes. If dis (p)i,pj) Greater than r, Edge (p)i,pj) It is not present. It is clear that the radius r affects the composition of the map. The parameter values often need to be obtained experimentally. Typically, r should be greater than twice the bandwidth (h). This is because if r is less than 2h, the connectivity of the graph may be degraded, which may make it difficult to perform the subsequent path analysis. However, the radius r may not be too large as well. If r is too large, the complexity of the graph will increase. More importantly, when performing shortest path analysis, the shortest path of each node will not reflect the structure of the tree. In the present invention, r is set to 2h + 0.5.
After the graph is constructed, shortest path analysis is carried out from each end node to the base point. The cardinal point refers to the modal point having the lowest elevation value in the present invention. In other words, the base points represent the roots of the trees, and the other nodes are branches or leaves. Since the graph has connectivity, each end node has a shortest path to the tree root. Each path consists of nodes traversed from the end node to the base point. This process can be expressed by equation (3).
SP(Graph,base,pm)＝{pm,pn,L,base} (3)
Wherein SP (g) represents the shortest path, Graph represents the constructed Graph, base represents the base point, pmRepresenting end nodes, pnIs the node through which the shortest path passes. The invention adopts Dijkstra algorithm to extract the shortest path, as shown in FIG. 4. As can be seen from fig. 4, these shortest paths can substantially reflect the structural characteristics of the tree.
And S3, performing leaf node detection based on path backtracking and node evolution.
As shown in fig. 4, the leaf nodes are typically at the end of each path. According to the characteristic, part of leaf nodes can be detected by adopting a path backtracking method. In the present invention, path backtracking means deleting a series of nodes from an end node to a base point. The deleted node is typically determined by the number of backtracking steps. Since leaf nodes are typically end nodes, only one step back is needed in each path. For example, the node p in equation (3)mThe result of the trace-through path one-step is { pnL, base }, i.e., the end node p is deletedm}。
Fig. 5 shows the result of the path backtracking of fig. 4. As can be seen from fig. 5, the leaf nodes at the end of the path have been excluded from the path. However, some nodes not at the end of the path are not included in the path. This is because the shortest path method attempts to get the shortest way to reach the base point, resulting in part of the nodes not being visited. In order to accurately detect leaf nodes, nodes that are not at the end need to be identified.
Since leaf nodes generally have a larger path length, if the path length from an unvisited node to a base point is smaller than the path length of the corresponding node in the path, the unvisited node can be evolved into a non-leaf node. The invention refers to the nodes passed by the path as seed points. For each seed point
Figure BDA0002756827640000071
And obtaining the adjacent nodes according to the formulas (4) and (5).
Figure BDA0002756827640000081
SPL(pi,pj)＝SP(Graph,pi,pj,weights) (5)
In the formula, pjIs the node in the graph, and n is the number of nodes in the graph. ctd (p)i,pj) Can be calculated by equation (5) and represents the node piTo pjCommute time distance. Wherein, weights represent weights between different nodes in Graph, D is a threshold value of an evolution radius, and D is set to be 2m in the invention.
It should be noted that the commute time distance is used instead of the geometric distance between the nodes to realize the acquisition of the adjacent nodes. This is because the commute time distance can better reflect the positional relationship between nodes than the geometric distance. As shown in fig. 6, a and B are two nodes in a tree, the line segment of the double-headed arrow is the geometric distance between two points, and the line segment of the single-headed arrow represents their commute time distance. The commute time distance here refers to the path length between nodes. Node a goes to node B and needs to go through other nodes. Since the node evolution method proposed by the present invention is to try to evolve adjacent nodes located on the same path to the base point, such as nodes on the same branch or the same trunk, nodes located on different paths should not be considered as neighboring nodes. For example, in fig. 6, if the geometric distance is used as the basis for the determination of the proximity point, the node B is a proximity point of the node a. However, if the commute time distance is chosen, node B will no longer be a neighbor of node a. Obviously, the latter is true because node a and node B are located on different branches.
After the neighboring points of each seed node are found, evolution can be performed according to the seed nodes. If node piSatisfies the condition described in the formula (6), node piWill be evolved into non-leaf nodes.
Figure BDA0002756827640000082
In the formula (I), the compound is shown in the specification,
Figure BDA0002756827640000083
representing a node piThe path length to the base point is less than that of the seed node
Figure BDA0002756827640000084
The length of the path to the base point. Due to the fact that
Figure BDA0002756827640000085
Is a non-leaf node, then piAlso a non-leaf node. And when all the non-leaf seed nodes complete the evolution, the remaining nodes are the leaf nodes. Fig. 7 is a leaf node detection result obtained by using seed node evolution. As can be seen in the figure, the leaf node at the end is successfully detected after evolution. It should be noted that the leaf nodes detected here are not all the leaf nodes in the graph.
And S4, performing branch node detection based on the node access frequency and the node evolution.
As can be seen from fig. 4, each node has a shortest path to the base point. Each path contains all the nodes through which the path is calculated by equation (3). Obviously, nodes on a trunk or branch have a higher frequency of access. This is because most paths need to reach the base point through these nodes. Conversely, the access frequency of the leaf nodes is low. Fig. 8(a) shows the result of coloring the mode points according to the node access frequency. As can be seen from the figure, a fraction of nodes are visited more than 600 times, whereas some nodes are visited only 1 time. In order to narrow the span range of the access frequency, the invention performs logarithmic calculation on the access frequency. The calculated result is shown in fig. 8 (b). It can be seen that the contrast of the access frequencies of different nodes is more obvious. Nodes located at branches and trunks generally have a greater frequency of access, while nodes located at the ends have a lesser frequency of access. However, it can be found that some nodes located on the trunk also have a smaller access frequency. Even a portion of the nodes are not visited. This is because the shortest path method attempts to find a shortest path to the base point. Therefore, a part of the nodes may not be passed through. This is also the reason why further evolution of the branch nodes is required to obtain the final detection result.
As described above, the nodes with higher access frequency are tree nodes. This process can be expressed by equation (7).
Figure BDA0002756827640000091
In the formula (I), the compound is shown in the specification,
Figure BDA0002756827640000092
is a node piMax (log (f)) represents the maximum value of the logarithm of the access frequency of the node, n is the number of nodes, and δ is a constant. A smaller δ may result in more nodes being detected as tree nodes; and when delta is larger, the number of nodes detected as tree nodes will be reduced. The present invention sets δ to 0.4.
Fig. 9 shows the tree node detection result obtained according to equation (7). It can be seen in fig. 9 that nodes located on the trunk or branches have all been successfully probed. However, not all the branch nodes are detected, which results in a larger false error of the tree point. The method reduces the false-rejection error by evolving the branch nodes.
The present invention refers to the branch nodes detected using equation (7) as branch seed nodes. In the evolution process, the branch nodes obtained through evolution should meet the following three conditions: (1) the path length to the base point is small; (2) the node should belong to a non-leaf node detected in section 2.3 (equation (6)); (3) the verticality of the nodes should be similar. Condition (1) is based on the principle that leaf nodes typically have a larger path length than branch nodes. In other words, if p jIs a branch seed node piIf and only if SPL (base, p)j) Less than SPL (base, p)i) When is, pjIs evolved into a branch node. The condition (2) limits the range of evolution. Condition (3) specifies the basic criteria for evolution, i.e. evolvedThe verticality of the node of the arriving branch is similar to the verticality of the seed node of the branch. This is because only nodes located in the same trunk or same branches can be evolved. The verticality of the node can be calculated according to equation (8).
Figure BDA0002756827640000101
In the formula, normalz(pi) Is the normal vector normal (p)i) The z-coordinate component of (a). The normal vector may be based on the minimum eigenvalue λ3The corresponding feature vector. The covariance matrix can be constructed using equation (9), and the eigenvectors and eigenvalues are calculated by Principal Component Analysis (PCA).
Figure BDA0002756827640000102
In the formula (I), the compound is shown in the specification,
Figure BDA0002756827640000103
is piThe center of the set of neighboring points, n is the number of neighboring points. The invention sets n equal to 10.
The evolution steps of the branch nodes are shown in table 1. In fig. 10(a), for successfully detected branch nodes after evolution, it can be seen that most of the undetected branch nodes in fig. 9 are re-identified. The nodes in the graph are modal points obtained by Mean Shift segmentation. Each modality point corresponds to a segmentation result. Therefore, the point cloud segmentation results obtained by using modal point detection can be fused to obtain the final tree point cloud. The final tree point cloud is shown in fig. 10 (b).
TABLE 1 evolution of branch nodes
Figure BDA0002756827640000104
In summary, according to the ground LiDAR point cloud branch and leaf separation method based on modal point evolution provided by the invention, a Mean Shift method is firstly adopted to obtain modal points, each modal point corresponds to one segmentation object, and branch and leaf separation based on points is converted into branch and leaf separation based on objects. Compared with the branch and leaf separation based on the geometric characteristics, the object-based method can greatly reduce the calculated amount and improve the separation efficiency. And detecting the leaf nodes and the branch seed nodes by backtracking the path and calculating the access frequency of each node. And finally, acquiring a final branch point cloud based on the branch nodes obtained by evolution and the Mean Shift segmentation result. The method is simple in process and easy to implement, and can provide a good foundation for the application of subsequent ground LiDAR in the forest area.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims (3)

1. A method for separating branches and leaves of a ground LiDAR point cloud based on modal point evolution is characterized by comprising the following steps:
s1, segmenting the ground LiDAR point cloud by adopting a Mean Shift segmentation method to obtain modal points;
s2, constructing a graph structure by using the modal points and analyzing the shortest path;
s3, detecting leaf nodes based on path backtracking and node evolution;
s4, detecting branch nodes based on the node access frequency and node evolution;
in step S2, each side length in the graph structure constructed by the modal points represents a distance between adjacent modal points, and a path exists between each node and a base point, where the base point is the modal point with the smallest height value;
in step S2, the graph structure is constrained using the following equation:
Figure FDA0003642198600000011
in the formula, Edge (p)i,pj) Is a node piAnd pjEdge between, dis (p)i,pj) R is the constraint radius, which is the geometric distance between the two nodes;
In step S2, the shortest path analysis is performed by using the following equation:
SP(Graph,base,pm)＝{pm,pn,L,base}
wherein SP (g) represents the shortest path, Graph represents the constructed Graph, base represents the base point, pmRepresenting end nodes, pnIs the node through which the shortest path passes;
in step S3, the path backtracking indicates deletion of a series of nodes from the end node to the base point, and the deleted nodes are determined by the number of backtracking steps;
in step S3, the nodes that the path passes through are called seed points, and for each seed point
Figure FDA0003642198600000012
Its neighboring nodes are calculated according to:
Figure FDA0003642198600000013
SPL(pi,pj)＝SP(Graph,pi,pj,weights)
in the formula, pjAre nodes in the graph, n is a section in the graphNumber of dots, ctd (p)i,pj) Representing a node piTo pjThe commuting time distance of (2), weight represents the weight between different nodes in Graph, D is the threshold value of the evolution radius;
in step S3, after finding the neighboring point of each seed node, performing evolution according to the seed node, if the node p isiSatisfies the condition of the formula, node piWill be evolved into non-leaf nodes;
Figure FDA0003642198600000021
in the formula (I), the compound is shown in the specification,
Figure FDA0003642198600000022
representing a node piThe path length to the base point is less than that of the seed node
Figure FDA0003642198600000023
A path length to a base point;
in step S4, the branch nodes are detected using the following formula:
Figure FDA0003642198600000024
in the formula (I), the compound is shown in the specification,
Figure FDA0003642198600000025
is a node piMax (log (f)) represents the maximum value of the logarithm of the access frequency of the node, n is the number of the nodes, and δ is a constant.
2. The method for separating branches and leaves of a ground LiDAR point cloud based on modal point evolution of claim 1, wherein in step S1, a Mean Shift vector is calculated using the following formula:
Figure FDA0003642198600000026
in the formula, Meanshifth(Vp) Represents the MeanShift vector, V for a three-dimensional point cloudpRepresents the three-dimensional coordinates of p points, n represents the number of adjacent points, the value of n is determined by the bandwidth h, and G () is a gaussian equation.
3. The method for separating branches and leaves of a ground LiDAR point cloud based on modal point evolution of claim 2, wherein the bandwidth h has a value range of: h belongs to [0.5,1.0 ].
CN202011205299.6A 2020-11-02 2020-11-02 Method for separating branches and leaves of ground LiDAR point cloud based on modal point evolution Active CN112348829B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011205299.6A CN112348829B (en) 2020-11-02 2020-11-02 Method for separating branches and leaves of ground LiDAR point cloud based on modal point evolution

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011205299.6A CN112348829B (en) 2020-11-02 2020-11-02 Method for separating branches and leaves of ground LiDAR point cloud based on modal point evolution

Publications (2)

Publication Number Publication Date
CN112348829A CN112348829A (en) 2021-02-09
CN112348829B true CN112348829B (en) 2022-06-28

Family

ID=74355448

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011205299.6A Active CN112348829B (en) 2020-11-02 2020-11-02 Method for separating branches and leaves of ground LiDAR point cloud based on modal point evolution

Country Status (1)

Country Link
CN (1) CN112348829B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114022536B (en) * 2021-10-18 2023-03-10 电子科技大学 Leaf area solving method based on foundation laser radar point cloud data

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101783016A (en) * 2009-12-16 2010-07-21 中国科学院自动化研究所 Crown appearance extract method based on shape analysis
WO2010105712A1 (en) * 2009-03-16 2010-09-23 Tele Atlas B.V. System and method for verifying map update reports using probe data
CN103268729A (en) * 2013-05-22 2013-08-28 北京工业大学 Mobile robot cascading type map creating method based on mixed characteristics
CN106067039A (en) * 2016-05-30 2016-11-02 桂林电子科技大学 Method for mode matching based on decision tree beta pruning
CN107392222A (en) * 2017-06-07 2017-11-24 深圳市深网视界科技有限公司 A kind of face cluster method, apparatus and storage medium
CN108369737A (en) * 2015-12-11 2018-08-03 诺华股份有限公司 Using heuristic graph searching quickly and automatically to divide layered image
CN111507194A (en) * 2020-03-20 2020-08-07 东华理工大学 Foundation L iDAR branch and leaf point cloud separation method based on fractal dimension supervised learning
CN111652154A (en) * 2020-06-04 2020-09-11 河北工业大学 Underdetermined system mode identification method based on automatic frequency band segmentation
CN113487631A (en) * 2021-07-21 2021-10-08 智能移动机器人(中山)研究院 Adjustable large-angle detection sensing and control method based on LEGO-LOAM

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10712738B2 (en) * 2016-05-09 2020-07-14 Strong Force Iot Portfolio 2016, Llc Methods and systems for industrial internet of things data collection for vibration sensitive equipment
US11092443B2 (en) * 2018-11-20 2021-08-17 Here Global B.V. Method, apparatus, and system for categorizing a stay point based on probe data

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2010105712A1 (en) * 2009-03-16 2010-09-23 Tele Atlas B.V. System and method for verifying map update reports using probe data
CN101783016A (en) * 2009-12-16 2010-07-21 中国科学院自动化研究所 Crown appearance extract method based on shape analysis
CN103268729A (en) * 2013-05-22 2013-08-28 北京工业大学 Mobile robot cascading type map creating method based on mixed characteristics
CN108369737A (en) * 2015-12-11 2018-08-03 诺华股份有限公司 Using heuristic graph searching quickly and automatically to divide layered image
CN106067039A (en) * 2016-05-30 2016-11-02 桂林电子科技大学 Method for mode matching based on decision tree beta pruning
CN107392222A (en) * 2017-06-07 2017-11-24 深圳市深网视界科技有限公司 A kind of face cluster method, apparatus and storage medium
CN111507194A (en) * 2020-03-20 2020-08-07 东华理工大学 Foundation L iDAR branch and leaf point cloud separation method based on fractal dimension supervised learning
CN111652154A (en) * 2020-06-04 2020-09-11 河北工业大学 Underdetermined system mode identification method based on automatic frequency band segmentation
CN113487631A (en) * 2021-07-21 2021-10-08 智能移动机器人(中山)研究院 Adjustable large-angle detection sensing and control method based on LEGO-LOAM

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
An automated approach for wood-leaf separation from terrestrial lidar point clouds using the density based clustering algorithm DBSCAN;Ferrara, R;《Agr. Forest Meteorol》;20181231;434–444 *
Wood and leaf separation from terrestrial LiDAR point clouds based on mode points evolution;Hui, Zhenyang等;《ISPRS Journal of Photogrammetry and Remote Sensing》;20210831;219-239 *

Also Published As

Publication number Publication date
CN112348829A (en) 2021-02-09

Similar Documents

Publication Publication Date Title
Lucchi et al. Learning for structured prediction using approximate subgradient descent with working sets
CN109146948B (en) Crop growth phenotype parameter quantification and yield correlation analysis method based on vision
Zhang et al. A fuzzy classification of sub-urban land cover from remotely sensed imagery
KR20190082062A (en) Method and apparatus for determining a matching relationship between point cloud data
CN107341813B (en) SAR image segmentation method based on Structure learning and sketch characteristic inference network
Guan et al. DEM generation from lidar data in wooded mountain areas by cross-section-plane analysis
Ko et al. Tree genera classification with geometric features from high-density airborne LiDAR
Hui et al. Wood and leaf separation from terrestrial LiDAR point clouds based on mode points evolution
CN104036497A (en) Graph cut interactive image segmentation algorithm based on local coefficient of variation
Chen et al. Leaf segmentation by functional modeling
Khan et al. A modified adaptive differential evolution algorithm for color image segmentation
CN111950589B (en) Point cloud region growing optimization segmentation method combined with K-means clustering
CN109919955A (en) The tunnel axis of ground formula laser radar point cloud extracts and dividing method
CN112348829B (en) Method for separating branches and leaves of ground LiDAR point cloud based on modal point evolution
Zeybek et al. An automated approach for extracting forest inventory data from individual trees using a handheld mobile laser scanner
Harandi et al. How to make sense of 3D representations for plant phenotyping: a compendium of processing and analysis techniques
KR101708042B1 (en) Cylinder estimation device and method by ransac algorithm in point cloud
CN106683105A (en) Image segmentation method and image segmentation device
Belém et al. The importance of object-based seed sampling for superpixel segmentation
Tabb et al. Fast and robust curve skeletonization for real-world elongated objects
Marinelli et al. An approach to tree detection based on the fusion of multitemporal LiDAR data
CN111860359A (en) Point cloud classification method based on improved random forest algorithm
CN112102178A (en) Point cloud feature-preserving denoising method and device, electronic equipment and storage medium
Wang et al. Research on vehicle detection based on faster R-CNN for UAV images
Lin et al. In defense of iterated conditional mode for hyperspectral image classification

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant