Multi-station foundation laser scanning point cloud registration method and system suitable for forest environment

Abstract

The invention belongs to the technical field of laser scanning, and discloses a multi-station foundation laser scanning point cloud registration method and system suitable for a forest environment. Firstly, an original three-dimensional point cloud is converted into a two-dimensional distance image, a trunk is extracted based on continuity constraint, and the position of a trunk root point is searched; secondly, constructing a triangular descriptor, describing the spatial relationship among the trunks in scanning, and storing the spatial relationship in a hash table; and searching the corresponding triangular pairs by using the search operation of the hash table, calculating the similarity between scans, and extracting three-dimensional transformation parameters. And taking a to-be-solved scanning station pose as a node, taking relative transformation as a factor constraint, and obtaining a scanning station pose parameter with optimal overall consistency through factor graph optimization. And finally, by taking the conversion parameter after the factor graph optimization as an initial value, optimizing and adjusting the conversion parameter by using an ICP method, deriving the pose of each scanning station in the unified coordinate system through a depth-first strategy, converting the local point cloud into the unified coordinate system, and realizing global point cloud registration.

Description

Technical Field

[0001] The invention belongs to the field of laser scanning technology, and in particular relates to a multi-station ground-based laser scanning point cloud registration method and system suitable for forest environments. Background Art

[0002] LiDAR (Light Detection and Ranging, LiDAR) has natural advantages such as speed, high precision, and penetration. It can accurately depict the three-dimensional spatial structure around the sensor and has been widely used in forestry surveys. The three-dimensional point cloud it obtains can accurately digitize trees and calculate a series of attribute parameters such as biomass, stem curve, canopy structure, etc., which is difficult to achieve in traditional artificial forestry surveys.

[0003] At present, the terrestrial laser scanning system (TLS) includes two general scanning modes: single-station scanning and multi-station scanning. Among them, multi-station scanning sets up stations at different locations inside the forest plot and obtains scanning point clouds from different perspectives, which can reduce the occlusion effect of dense branches and shrubs and cover the entire measurement area. However, the point cloud data in the multi-station scanning mode are in different local coordinate systems, and the multi-station data need to be spliced ​​in a unified coordinate system for subsequent practical applications. The most commonly used point cloud splicing method is to arrange multiple artificial targets inside the plot, and then ensure that three or more targets are scanned during the operation process, manually extract the targets with the same name during the post-processing process, and calculate the three-dimensional transformation parameters. This method is accurate and reliable enough, but it is time-consuming and labor-intensive in field operations and internal processing, and the operation efficiency is low.

[0004] According to the different registration primitives, the existing TLS point cloud registration methods can be roughly divided into three categories: point-based, feature-based, and object-based.

[0005] Point-based registration methods directly establish corresponding associated point pairs in the original data, minimize the distance between all point pairs and obtain the optimal transformation parameters. This type of method includes typical iterative nearest neighbor points (ICP), generalized ICP, global optimal ICP, etc. These methods can perform precise optimization to obtain accurate transformation parameters under the condition of good initial values, but the method is sensitive to the initial values, and inappropriate initial transformation can easily cause the point-based method to fall into a local optimal solution. Forest scenes have more noise points and isolated points than urban buildings. This poses new challenges to point-based registration methods, making it difficult to establish correct point pair relationships. Although advanced methods such as TEASER, GORE, and GROR expect to remove the influence of erroneous point pairs by adding a series of robust gross error elimination methods, they are still difficult to apply to forest scenes. In addition, the massive point clouds scanned by TLS in forest scenes also bring huge computational burdens to such methods.

[0006] The performance of feature-based registration methods depends on two key steps, namely feature extraction and feature description. First, representative key points are extracted from the original point cloud. Subsequently, feature descriptors are used to describe the local neighborhood of the key points and hypothetical associations are established for the key points based on the similarity between the descriptors. Various feature extraction and description algorithms have been proposed, including hand-designed feature points such as LSP, ISS, Harris 3D, and popular feature descriptors such as FPFH and SHOT. These hand-designed methods require expert knowledge with a deep understanding of the scene, and have poor robustness and are difficult to generalize to ordinary users. Another type is deep learning-based methods, such as PPFNet, D3Feat, SpinNet, etc., which use deep learning networks to mine knowledge of the scene and effectively extract feature points. Although these feature extraction methods can better cope with the influence of noise points, the changes in scanning angle, point density, and occlusion effects in forest point clouds will greatly cause the failure of manual features. At the same time, deep learning methods have the problems of requiring a large amount of labeled point cloud data for training, high computing power requirements, and poor applicability to different scenarios.

[0007] Object-based registration methods use fragments in point clouds as registration primitives to establish corresponding associations. This type of method can effectively deal with the influence of noise points, avoid complex feature design, and retain the original geometric information. Among them, the trunk part of the forest is a stable and highly representative part, which is widely used in various TLS point cloud registration methods. Large-scale and high-trunk density plots will greatly increase the computational burden of existing methods.

[0008] Through the above analysis, the problems and defects of the prior art are as follows:

[0009] The most commonly used point cloud stitching method is to place multiple artificial targets inside the sample site, and then ensure that three or more targets are scanned during the operation. In the post-processing process, targets with the same name are manually extracted and three-dimensional transformation parameters are calculated. This method is time-consuming and labor-intensive in field operations and office processing, and has low operating efficiency. Summary of the invention

[0010] In view of the problems existing in the prior art, the present invention provides a multi-station ground-based laser scanning point cloud registration method and system suitable for forest environments.

[0011] The present invention is implemented as follows: a multi-station ground-based laser scanning point cloud registration method and system suitable for a forest environment comprises:

[0012] Step 1: Generate range image;

[0013] Step 2: Continuity constraints of trunk points;

[0014] Perform distance continuity constraints in the column direction to extract the stable part of the point cloud;

[0015] Step 3: Trunk point direction constraint;

[0016] The tree trunk points extracted by the continuity constraint still contain some dense branches and scattered ground fragments. Clustering is used to divide the point cloud into clusters, and then the geometric characteristics of each cluster are used to remove branches and ground points, and retain the tree trunk points;

[0017] Step 4: Construct a triangle descriptor;

[0018] Triangles are used to describe the relative relationship between discrete tree trunk positions; multiple triangles are combined into a triangulated network to describe the overall distribution of tree trunks in the scan; the number of triangles depends on the size of the scanned sample and the density of tree trunks;

[0019] Step 5: Rough registration of 3D point cloud;

[0020] Through efficient search operations in hash tables, triangle matching pairs are found and the similarity between matching pairs of scanning stations is calculated. If the similarity meets the threshold, the coarse registration parameters of the scanning point cloud pairs are output;

[0021] Step 6: Factor graph optimization;

[0022] The factor graph optimization method is introduced to model the inconsistency between relative transformations of multiple stations, and globally consistent transformation parameters are obtained after overall optimization.

[0023] Step 7: Point cloud precise registration and pose estimation;

[0024] The relative pose transformation parameters after factor graph optimization are used as initial values ​​and optimized through the point-to-point ICP method. The pose of each scanning station in the unified coordinate system is derived through the optimized transformation parameters, and all scanning station clouds are converted to the unified coordinate system to form an overall consistent global point cloud.

[0025] Further, the distance image is generated:

[0026] (1) Ground-based LiDAR data acquisition;

[0027] TLS scanning stations are arranged according to the size of the forest plot. For a plot of 20m×20m, a scanning method of one station in the middle and one station in each of the four corners can be adopted to obtain complete plot scanning data to the greatest extent possible;

[0028] (2) Converting 3D point cloud into range image;

[0029] Process the single-site cloud data and set the three-dimensional coordinates of any point i in the point cloud to be (x i ,y i ,z i ), can also be expressed as polar coordinates (α i ,β i ,γ i ) in the form of i , β i are the horizontal angle and the vertical angle respectively, γ i is the distance from the scanning point i to the scanning center; then, the index of the scanning point i in the distance image can be defined by formula (1);

[0030]

[0031] In the formula, row i ,col i are the row index and column index of the scanning point i in the range image, respectively, and θ is the angular resolution of the range image;

[0032] Each pixel in the distance image uses γ i Filling: If multiple scan points correspond to the same index position after conversion, the pixel retains the scan point information closest to the center.

[0033] Furthermore, the trunk point continuity constraint is:

[0034] (1) Define the search step (Step), pixel difference threshold (R diff ), group pixel number threshold (Seg th ) The three parameters are set to 15 pixels, 0.03m, and 100 pixels respectively;

[0035] (2) For each column in the distance image, starting from the first point, compare the pixel value difference between the search point and the current point within the search step Step range. If it is less than the preset threshold R diff , the search point and the current point are grouped into the same group; otherwise, a new group is created;

[0036] (3) Repeat the above operation until all pixels in the range image are visited;

[0037] (4) Count the number of pixels in each group. If it exceeds the preset threshold Seg th If the pixel is not positive, it is retained and the pixels in the group are marked as 1; otherwise, it is discarded and the pixels in the group are marked as 0;

[0038] (5) Traverse all pixels. If the pixel is marked as 1, the corresponding scan point is retained. Otherwise, if the pixel is marked as 0, the scan point is discarded.

[0039] Furthermore, the trunk point direction constraint is:

[0040] (1) Using Euclidean distance clustering, the point cloud data extracted in step 2 is processed and divided into different point cloud clusters;

[0041] (2) Perform principal component analysis (PCA) on each point cloud cluster and calculate the linear L λ , Planarity P λ , Dispersion S λ Three parameters, describing the geometric distribution of point cloud clusters in space, are defined by formula (2);

[0042]

[0043] In the formula, λ1>λ2>λ3 are the eigenvalues ​​of the point cloud cluster, and are arranged in descending order, corresponding to the degree of change in the direction of each eigenvector;

[0044] (3) The trunk points are linearly distributed in space, that is, they have a large linear L λ , while the ground point has a larger planarity P λ , the branch and leaf points have a large dispersion S λ Therefore, this feature is used to judge the type of the extracted point cloud cluster. If a point cloud cluster meets the condition L λ >0.95, the point cloud cluster is considered to be a tree trunk; distribution of point cloud clusters before and after using geometric constraints;

[0045] (4) Using the extracted tree trunk point cloud cluster, extract the corresponding tree trunk position; its plane position (x, y) is the center of the point cloud cluster, and the z component is selected as the root point of the trunk, that is, the intersection of the trunk and the ground. The black dot at the bottom is the extracted trunk position.

[0046] Further, the triangle descriptor is constructed:

[0047] (1) Screening triangles;

[0048] 1) For any tree trunk position, search for 10 neighboring points with it as the center;

[0049] 2) Pick any two neighboring points and combine them with the current point to form a triangle;

[0050] 3) If the constructed triangle is an isosceles / equilateral triangle, remove it; otherwise, if the three sides are different, keep it;

[0051] 4) Repeat the above steps to traverse all tree trunk positions;

[0052] (2) Hash table storage

[0053] 1) Use the “division remainder method” to construct a hash function, defined by formula (3);

[0054] h(l1,l2,l3)=mod[mod(l1*m,N)+l2*m,N]+l3 (3)

[0055] Where (l1, l2, l3) is the side length of any triangle and is arranged in descending order (l1>l2>l3); m is a large prime number, and N represents the size of the hash table;

[0056] 2) Perform the above operation for each triangle in the triangulated network so that each triangle can be projected to a unique index in the hash table;

[0057] 3) All TLS scans in the sample plot are projected to the corresponding positions in the hash table; if a hash conflict occurs, the chaining method is used to handle it.

[0058] Further, the three-dimensional point cloud is roughly registered:

[0059] (1) Search for triangle matching pairs

[0060] 1) For any triangle in the scan station to be searched, use the hash function to directly map the triangle descriptor to the corresponding index in the hash table;

[0061] 2) Find the triangle descriptor contained in the current index position, and calculate the direction vector difference of the corresponding vertex using formula (4). If it is less than the threshold value 0.1, the tree trunk matching pair is retained, otherwise it is discarded;

[0062]

[0063] In the formula, and Represent the direction vectors of the query trunk and the candidate trunk respectively;

[0064] (2) Calculation of three-dimensional transformation coarse parameters

[0065] 1) Count the scanning station IDs corresponding to each trunk match, vote, and calculate the similarity, which is defined by formula (5);

[0066]

[0067] In the formula, S q With S c Represent query scan and candidate scan respectively, V c is the number of votes scanned by the candidate, N q is the number of trunks that are candidates for scanning;

[0068] 2) If the similarity between the query scan and the candidate scan is greater than a preset threshold, it is considered that there is an adjacency relationship between the two stations, and the paired trunk positions are used to solve the 3D transformation parameters;

[0069] The factor graph optimization:

[0070] (1) Factor graph modeling

[0071] For a certain forest plot, a multi-station TLS scan was performed. Then, the pose to be determined in the unified coordinate system of each station scan was used as a node, and the coarse transformation between adjacent stations was used as a factor constraint to establish a factor graph model.

[0072] (2) Global Optimization

[0073] Comprehensively consider all nodes and constraints in the factor graph and convert it into a least squares problem under the Bayesian estimation framework; reduce the inconsistency between relative transformations and solve the optimal scanning station posture and relative transformation;

[0074] The point cloud precise registration and pose estimation:

[0075] (1) Point cloud precise registration;

[0076] The relative pose transformation parameters after factor graph optimization are used as initial values, and the transformation parameters between each valid paired scanning station are optimized and adjusted using the point-to-point ICP precise registration method.

[0077] (2) Global pose derivation;

[0078] Based on the optimized relative transformation parameters, the depth-first strategy is used to deduce the position and posture of each scanning station in the unified coordinate system;

[0079] (3) Unified point cloud coordinate system;

[0080] After the above steps, the position and posture of each scanning station in the unified coordinate system are known and applied to each scanning point cloud to convert it from an independent local coordinate system to an overall point cloud in the unified coordinate system.

[0081] Another object of the present invention is to provide a multi-station ground-based laser scanning point cloud registration system suitable for forest environments, comprising:

[0082] A distance image generation module, used for generating a distance image;

[0083] The continuity constraint module is used for the continuity constraint of the tree trunk points. It performs distance continuity constraint in the column direction and extracts the stable part of the point cloud.

[0084] Direction constraint module, used for direction constraint of tree trunk points; removes the influence of dense branches and leaves and scattered ground fragments that still exist;

[0085] Descriptor construction module, used to construct triangle descriptors; triangles are used to describe the relative relationship between discrete tree trunk positions; multiple triangles are combined into a triangulated network to describe the overall distribution of tree trunks in the scan; the number of triangles depends on the size of the scanned sample and the density of tree trunks;

[0086] The registration module is used for the rough registration of 3D point clouds. It searches for triangle matching pairs through efficient search operations in hash tables and calculates the similarity between matching pairs of scanning stations. If the similarity meets the threshold, it outputs the rough registration parameters of the scanning point cloud pairs.

[0087] Optimization module, used for factor graph optimization; introduces factor graph optimization method to model the inconsistency between relative transformations of multiple stations, and obtains globally consistent transformation parameters after overall optimization;

[0088] The pose estimation module is used for precise point cloud registration and pose estimation. The relative pose transformation parameters after factor graph optimization are used as initial values ​​and optimized through the point-to-point ICP method. The pose of each scanning station in the unified coordinate system is derived through the optimized transformation parameters, and all scanning station clouds are converted to the unified coordinate system to form an overall consistent global point cloud.

[0089] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the multi-station ground-based laser scanning point cloud registration method applicable to a forest environment.

[0090] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the multi-station ground-based laser scanning point cloud registration method applicable to a forest environment.

[0091] Another object of the present invention is to provide an information data processing terminal, which is used to implement the multi-station ground-based laser scanning point cloud registration system suitable for forest environments.

[0092] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0093] First, the present invention proposes a multi-station ground-based laser scanning point cloud registration method and system suitable for forest environment. The method does not require any prior information and realizes the automatic conversion of multi-station cloud to a unified coordinate system. It solves the time-consuming and labor-intensive problem of laying out artificial targets during multi-station TLS scanning, and greatly improves the work efficiency. The method fills the gap between the original TLS scanning data and the actual application, and provides a new solution for forest surveys. The present invention proposes a multi-station ground-based laser scanning point cloud registration method and system suitable for forest environment, which solves the problems of large noise point influence, low calculation efficiency and poor applicability. The method can convert multi-station local scans in forest plots into a unified coordinate system without any prior information, such as initial transformation, scanning station adjacency, etc., to provide data support for subsequent applications. First, the present invention converts the original three-dimensional point cloud into a two-dimensional range image, extracts the trunk based on the continuity constraint and finds the position of the trunk root point. Secondly, a triangle descriptor is constructed to describe the spatial relationship between the trunks in the scan and store it in a hash table. The corresponding triangle pairs are found by the search operation of the hash table, the similarity between the scans is calculated, and the three-dimensional transformation parameters are extracted at the same time. Since the three-dimensional transformation calculated in the above steps is relatively rough and there are inconsistencies, it is difficult to accurately derive the pose of each TLS scanning station. The present invention introduces factor graph modeling to solve the problem of inconsistent rough transformations between multi-site clouds. The pose of the scanning station to be determined is used as a node, and the relative transformation is used as a factor constraint to obtain the pose parameters of the scanning station with the best overall consistency. Finally, the conversion parameters optimized by the factor graph are used as the initial values, and the ICP method is used to optimize and adjust them. The pose of each scanning station in the unified coordinate system is derived through a depth-first strategy, and the local point cloud is converted to the unified coordinate system to achieve global point cloud alignment without any prior information.

[0094] Second, as auxiliary evidence of the inventiveness of the claims of the present invention, it is also reflected in the following important aspects:

[0095] (1) The expected benefits and commercial value of the technical solution of the present invention after transformation are:

[0096] The present invention proposes a multi-station ground-based laser scanning point cloud registration method and system suitable for forest environments. The trunk position is used as the registration unit, and multi-station cloud registration can be completed without the need to deploy artificial targets in the field. It builds a bridge from field data collection to actual production applications, and can provide data support for the automated extraction of single tree attributes and forest surveys. Therefore, the present invention can provide guarantees for the realization of "carbon peak" and "carbon neutrality", and provide a new evaluation scheme for the carbon trading market, which has important commercial and economic value.

[0097] (2) The technical solution of the present invention fills the technical gap in the industry at home and abroad:

[0098] The technical solution of the present invention innovatively proposes three contents: 1) Use the trunk position as the registration primitive to construct a triangle descriptor (the simplest polygon) to represent the relative position between each trunk. 2) Use a hash table to efficiently store the triangle descriptors of multi-station scans, use the hash key value and the direction constraint of the trunk point cloud cluster to find paired triangles, calculate the similarity between the two-station scan point clouds and solve the rough transformation parameters. 3) Use the pose to be determined of each scanning station in the unified coordinate system as a node and the relative transformation as a factor constraint to construct a factor graph to obtain the pose parameters with global consistency and optimality. Finally, use the optimized relative pose as the initial transformation, and fine-tune it through the point-to-point iterative nearest neighbor registration algorithm to obtain high-precision pose parameters, and all point cloud data are converted to a unified coordinate system.

[0099] The above three innovations fill the gap in the unreliable multi-site cloud registration technology in forest environments at home and abroad, and realize the efficient, accurate and automatic registration of multi-site TLS scans in complex forest environments without any prior information, any manual operation or any special markings.

[0100] (3) The technical solution of the present invention solves the technical problems that people have been eager to solve but have never been able to solve successfully:

[0101] The traditional multi-station laser scanning method requires the deployment of targets on site, and the manual selection of points as homonymous points in the internal processing is time-consuming and labor-intensive, greatly reducing the work efficiency. In addition, there are no regular geometric features in the forest scene, and there are many noise points. The existing point cloud registration method is difficult to meet the needs of practical applications. Therefore, a robust, accurate and automated method is urgently needed. The method and system proposed in the present invention can solve the technical problems of multi-station TLS scanning point cloud registration in complex forest environments, improve the efficiency of forest survey operations, and promote the application of TLS in forestry.

[0102] Third, the multi-station ground-based laser scanning point cloud registration system and method of the present invention is aimed at the registration requirements of multi-station laser scanning data in forest environments, and significantly solves the problem that the existing technology is difficult to obtain high-precision global point clouds in forest environments. Its key parameter selection, algorithm design and mathematical model have achieved technical breakthroughs in accuracy and efficiency, which are specifically reflected in the following aspects:

[0103] 1. Efficient point cloud filtering and feature extraction

[0104] The present invention utilizes multiple filtering methods such as distance images and trunk point continuity constraints to effectively remove dense branches and ground interference fragments in the point cloud. By setting the search step size, pixel difference threshold, and grouping pixel number threshold, the point cloud data is efficiently processed and stable trunk data is retained, which is difficult to achieve in traditional forest environment laser scanning point cloud processing. Fast Euclidean distance clustering combined with principal component analysis (PCA) classifies point cloud clusters by geometric characteristics, ensuring that the extracted trunk point data is purer, thereby greatly improving the registration accuracy.

[0105] 2. Accurate triangle descriptor construction and matching

[0106] The triangle descriptor proposed in the present invention is constructed through the relative position relationship of the trunk positions, and can effectively describe the overall distribution characteristics of the trunks in the forest environment. By storing the constructed triangles in a hash table and using the "division and remainder method" to construct a hash function, fast and accurate triangle matching is achieved, reducing redundant data calculations in traditional point cloud registration. Using these descriptors, the system can efficiently find similar trunk matching pairs between each scanning station, laying the foundation for subsequent registration parameter calculations. Compared with the traditional direct point-to-point registration method, this method has higher matching efficiency.

[0107] 3. Factor graph optimization and global pose derivation

[0108] In order to solve the common problem of cumulative error in multi-site cloud registration, the present invention introduces factor graph optimization and models the rough transformation result as a least squares problem under the Bayesian estimation framework. The poses of multiple scanning stations are modeled by factor graphs, and the relative transformations between stations are comprehensively considered to reduce the inconsistency between the relative transformations, thereby optimizing the global pose. Factor graph optimization not only enhances the robustness of the system, but also significantly improves the global consistency, ensuring that the final output global point cloud is accurate and reliable.

[0109] 4. Point cloud precise registration and global pose estimation using depth-first strategy

[0110] Based on the factor graph optimization results, the present invention uses the ICP algorithm for precise registration. The relative pose transformation parameters after factor graph optimization are used as initial values, and the transformation parameters between each paired scanning station are gradually optimized to reduce the pose error. Subsequently, the pose of all scanning stations is derived in a unified coordinate system through a depth-first strategy. Compared with traditional methods, this method can reduce the calculation time while ensuring the registration accuracy, so that each scanning station cloud can be quickly converted to a unified coordinate system to form a complete global point cloud, which provides a solid data foundation for the subsequent processing of large-scale forest data.

[0111] In general, the present invention achieves efficient and accurate multi-station laser scanning point cloud registration in a forest environment. Compared with the prior art, the innovation of the present invention is mainly reflected in the refinement of data filtering, the efficiency of feature matching, and the consistency of global pose optimization. This technology can significantly improve the efficiency and accuracy of data acquisition in forest ecological research and resource management, and has broad application prospects, especially in the fields of environmental monitoring and forest resource census.

[0112] Fourth, the mathematical model of the present invention solves the following problems in the prior art:

[0113] 1. The problem of insufficient accuracy of existing point cloud registration

[0114] In complex forest environments, since point cloud data contains a large number of branch points and ground noise, it is difficult for existing technologies to accurately extract trunk points from them, resulting in large registration errors.

[0115] Point cloud matching relies on simple features (such as nearest neighbor search), which can easily lead to matching failures due to environmental complexity and occlusion.

[0116] 2. Poor global consistency

[0117] Among the multi-site cloud registration methods, common local feature matching (such as the ICP method) is difficult to ensure global consistency and easily causes cumulative errors, which ultimately leads to distortion of the overall point cloud.

[0118] There is a lack of systematic modeling approaches to describe the global relationships among multi-station scanning stations.

[0119] 3. Lack of effective optimization methods to deal with the inconsistency of postures between multiple stations

[0120] The existing technology cannot effectively eliminate the local errors of posture between multiple stations during multi-station cloud registration, causing the point cloud under the unified coordinate system to show obvious mismatch.

[0121] Significant technical advances of the mathematical model of the present invention:

[0122] 1. Distance image and trunk point extraction model improves point cloud processing accuracy

[0123] Distance image generation formula: Convert the point cloud from a three-dimensional coordinate system into a two-dimensional distance image, and map it through pixel values ​​(the distance from the point to the scanning center), thus solving the problem of high computational complexity in traditional three-dimensional point cloud processing.

[0124] Mathematical formula (1) ensures that the scan points are accurately mapped to two-dimensional pixel indices, which facilitates subsequent operations.

[0125] Continuity constraint model: Through the dual threshold constraints of pixel difference and grouped pixel number, the tree trunk points are accurately separated from the branches and ground points.

[0126] Geometric direction constraint formula (2): The geometric characteristic calculation based on principal component analysis (PCA) significantly improves the classification ability of tree trunk points, branch points and ground points.

[0127] 2. Triangle descriptor and hash storage model improve feature matching efficiency

[0128] Triangle descriptor model: The geometric distribution between tree trunk points is described by using the side length and angle relationship of the triangle. By constructing a sparse but stable triangulated network, the problem of unstable point cloud features is solved.

[0129] Hash table storage formula (3): The triangle descriptor is projected to a unique index position through a hash function, and the conflict is handled by combining the linking method to efficiently achieve fast matching between multi-site clouds.

[0130] 3. Factor graph optimization model achieves global consistency

[0131] Factor graph modeling: By taking the scanning station poses as nodes and the coarse transformations between adjacent stations as factor constraints, a global optimization model is established to make up for the global consistency defect of the traditional ICP method.

[0132] Least squares optimization under the Bayesian estimation framework: By minimizing the constraint inconsistency between nodes, the registered point cloud is ensured to have global consistency.

[0133] 4. Multi-site cloud precision registration and pose inference model improves global accuracy

[0134] ICP optimization model: The parameters after factor graph optimization are used as initial values, and the point-to-point ICP method is used to further accurately adjust the posture transformation between multiple stations.

[0135] Depth-first pose derivation: Based on the optimized transformation parameters, the global pose in the unified coordinate system is derived through depth-first search to achieve the unification and accuracy of the global point cloud.

[0136] 5. Multi-level modeling and distributed processing solve adaptability problems in complex scenarios

[0137] The present invention establishes a complete point cloud processing system from multiple levels (range image, trunk points, triangle network, global optimization), combined with a distributed computing framework, which can efficiently process large-scale, multi-station scanning data and adapt to forest scenes of different densities and complexities.

[0138] The present invention solves the problems of insufficient registration accuracy, poor global consistency, inconsistent postures among multiple stations, etc. in the prior art through the mathematical model of distance image generation, continuity constraint, geometric direction constraint, triangle descriptor, factor graph optimization, and ICP precise registration. The significant technical progress is reflected in the following aspects:

[0139] - Accuracy improvement: Through multi-level modeling and optimization, the registration accuracy of multi-site clouds in forest environments is significantly improved.

[0140] - Efficiency improvement: The use of hash tables and sparse triangle descriptors greatly reduces the computational cost of feature matching and pose optimization.

[0141] - Enhanced adaptability: The model shows good processing effects on point cloud data of different forest densities, occlusion conditions and complex terrains, providing strong technical support for industrial applications such as forest resource surveys and ecological monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0142] Figure 1 It is a flow chart of a multi-station ground-based laser scanning point cloud registration method suitable for a forest environment provided by an embodiment of the present invention.

[0143] Figure 2 It is a structural block diagram of a multi-station ground-based laser scanning point cloud registration system suitable for a forest environment provided by an embodiment of the present invention.

[0144] Figure 3 The three-dimensional point cloud is converted into a distance image according to an embodiment of the present invention. (a) TLS scanning principle. (b) Original three-dimensional point cloud. (c) Conversion distance image.

[0145] Figure 4 1 is a continuous constraint in a range image provided by an embodiment of the present invention. (a) Range image of point cloud projection. (b) Continuous segments obtained by continuity constraint.

[0146] Figure 5 1 is the geometric constraint of the tree trunk point cloud provided by the embodiment of the present invention. (a) All point cloud clusters generated after Euclidean clustering. (b) Tree trunk point cloud cluster after geometric constraint.

[0147] Figure 6 It is the extracted trunk position provided by the embodiment of the present invention.

[0148] Figure 7It is a triangle descriptor provided by an embodiment of the present invention. (a) Screening triangle shape. (b) Triangle rotation invariance.

[0149] Figure 8 The hash table storage triangle descriptor provided by the embodiment of the present invention. (a) Single triangle storage. (b) Multi-scanning station data storage.

[0150] Fig. 9 1 is a search triangle matching pair provided by an embodiment of the present invention. (a) Hash search. (b) Trunk direction constraint.

[0151] Fig.10 The algorithm flow chart of factor graph optimization provided by the embodiment of the present invention is as follows: (a) Multi-station TLS scanning of forest plots; (b) Factor graph modeling.

[0152] Fig.11 It is a flow chart of an algorithm provided by an embodiment of the present invention.

[0153] Fig.12 Schematic diagram of trunk point extraction provided by an embodiment of the present invention. Different trunks have different grayscales, and black dots represent the positions of the trunks. (a) Shennongjia data sample. (b) ETH data sample. (c) FGI data sample. (d) Tongji University data sample.

[0154] Fig.13 : This is a histogram of point cloud rough registration accuracy evaluation provided by an embodiment of the present invention. The shadow in the histogram represents whether the factor graph optimization has been performed. (a) Rotation error. (b) Translation error.

[0155] Fig.14 The point cloud precise registration error assessment provided by the embodiment of the present invention is as follows: (a) Rotation error. (b) Translation error.

[0156] Fig.15 Schematic diagram of the overall point cloud after registration provided by the embodiment of the present invention. (a) Shennongjia dataset. (b) ETH dataset. (c) FGI dataset. (d) Tongji University dataset. DETAILED DESCRIPTION

[0157] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0158] The technical solution of the present invention solves the following key problems in the prior art in industrial applications and has achieved significant technical progress:

[0159] Technical issues:

[0160] 1. Difficulty of point cloud registration in complex forest environments

[0161] When existing technologies align multi-station ground-based laser scanning point clouds in complex forest environments, it is difficult to effectively separate and match tree trunk points, branch points, and ground points due to the mixing of tree trunk points, branch points, and ground points. Traditional methods rely on manual screening or a single algorithm, which is inefficient and lacks accuracy.

[0162] 2. Insufficient global consistency in the registration process

[0163] Existing registration methods mostly rely on local feature matching, which makes it difficult to ensure the consistency and robustness of the global point cloud. Especially in multi-station scanning data, cumulative errors are prone to occur, resulting in deformation or distortion of the overall point cloud.

[0164] 3. Challenges of pose estimation in dynamic forest environments

[0165] In a dynamic forest environment, due to the uneven distribution of trees and severe local occlusion, existing methods find it difficult to accurately estimate the global pose of the scanning station and generate a global point cloud in a unified coordinate system.

[0166] Significant technological advancement

[0167] 1. Accurate extraction of tree trunk points improves registration efficiency

[0168] The present invention uses distance image generation, trunk point continuity constraints and geometric direction constraints, pixel difference thresholds, Euclidean distance clustering and geometric characteristics to determine trunk points, and accurately separates trunk points from branches and ground points. This improvement significantly improves the efficiency and accuracy of point cloud processing, laying a solid foundation for subsequent registration.

[0169] 2. Triangle descriptor improves feature matching accuracy

[0170] By constructing a triangle descriptor, the present invention describes the geometric relationship between discrete tree trunk points and uses a hash table for efficient storage and search. The method can quickly and accurately match tree trunk points in multi-station scanning point clouds, avoiding redundancy or mismatching problems that are prone to occur in traditional feature matching methods.

[0171] 3. Factor graph optimization to achieve global consistency

[0172] The factor graph optimization method is introduced to model the multi-station scanning station poses and their relative transformations as factor graphs. The problem of local error accumulation is solved through least squares optimization under the Bayesian estimation framework, which significantly improves the global consistency and robustness of point cloud registration.

[0173] 4. Organic combination of precise registration and pose derivation

[0174] Based on factor graph optimization, the point-to-point ICP method is used for precise registration, and the global pose of each scanning station is derived in combination with the depth-first strategy. This process not only improves the registration accuracy, but also successfully realizes the conversion of each station cloud from the local coordinate system to the global coordinate system.

[0175] 5. Application adaptability in dynamic forest environment

[0176] Through deep modeling and optimization of the distribution of tree trunk points in forest environments, the present invention can adapt to the scanning tasks of forest plots with different densities and structures, and can maintain high alignment accuracy and efficiency in dynamic environments with severe occlusion and complex structures.

[0177] 6. Industrial application value

[0178] This invention significantly improves the registration accuracy and efficiency of multi-station ground-based laser scanning technology in forest environments, and can be widely used in forest resource surveys, carbon sink measurements, vegetation growth analysis and other fields. At the same time, its universal design allows it to be extended to other complex environments (such as urban or industrial scenes), providing an innovative solution for the industrialization of lidar point cloud technology.

[0179] Example 1: Forest resource survey and carbon sink measurement

[0180] Forest resource survey is an important means to assess forest health, tree density, tree species distribution and carbon sequestration capacity. Traditional methods rely on manual measurement or drone remote sensing, which is difficult to achieve high-precision requirements at the detailed level.

[0181] 1. Data Collection

[0182] Multi-station ground-based laser scanning (TLS) equipment was deployed in designated forest plots (e.g., 20m×20m area), with center and four-corner station deployment adopted according to the plot shape and vegetation density.

[0183] Each station scan obtains high-precision point cloud data, covering tree trunks, branches, leaves and ground information.

[0184] 2. Data processing

[0185] The point cloud registration method of the present invention is used to perform tree trunk point extraction, geometric direction constraint and point cloud matching:

[0186] The tree trunk points are accurately extracted to remove branches, leaves and ground noise, and retain the tree trunk information;

[0187] Triangle descriptors and factor graph optimization for globally consistent registration of multi-site clouds.

[0188] 3. Result Output

[0189] Obtain high-precision global point cloud data, including the spatial position, diameter, density and distribution of tree trunks.

[0190] Extract the total biomass and carbon stock of forest plots to provide support for the assessment of forest carbon sequestration capacity.

[0191] Value

[0192] -It has achieved high-precision quantitative assessment of forest resources and provided a reliable data basis for forestry management departments.

[0193] - Improve the measurement accuracy of carbon sink capacity and facilitate the development of the carbon emissions trading market.

[0194] Example 2: Monitoring of forest ecological restoration after natural disasters

[0195] Forest recovery often needs to be assessed after natural disasters (such as fire, storms or insect infestations), but traditional methods are inefficient due to the instability and complexity of the damaged areas.

[0196] 1. Data Collection

[0197] Multi-station laser scanning is set up in the disaster-affected area, and multi-station TLS scanning is carried out on key sample sites (such as the core disaster area or key recovery area) to obtain point cloud data.

[0198] The station layout is flexible and can adapt to irregular terrain and distribution of remaining trees.

[0199] 2. Data processing

[0200] The method of the present invention is used to process the point cloud of the damaged and restored areas:

[0201] Extract the trunk points of surviving trees and fallen trees.

[0202] The distribution of tree trunk positions is reconstructed using triangle descriptors to monitor changes in the spatial relationship between trees.

[0203] Compare post-disaster data with historical point cloud data to analyze the progress of vegetation restoration.

[0204] 3. Result Output

[0205] Output the tree recovery rate, spatial density variation, and number distribution of fallen or broken trees in the area.

[0206] Generate a 3D visualization model to intuitively present the restoration process.

[0207] - Provide quantitative indicators for ecological protection and restoration, and guide subsequent vegetation reconstruction and artificial intervention.

[0208] -Reduces the risk of manual on-site surveys and significantly improves monitoring efficiency and accuracy.

[0209] like Figure 1 As shown, a multi-station ground-based laser scanning point cloud registration method applicable to a forest environment provided by an embodiment of the present invention comprises the following steps:

[0210] S101: Generate a distance image;

[0211] S102: trunk point continuity constraint;

[0212] Perform distance continuity constraints in the column direction to extract the stable part of the point cloud;

[0213] S103: Trunk point direction constraint;

[0214] The tree trunk points extracted by the continuity constraint still contain some dense branches and scattered ground fragments. Clustering is used to divide the point cloud into clusters, and then the geometric characteristics of each cluster are used to remove branches and ground points, and retain the tree trunk points;

[0215] S104: construct a triangle descriptor;

[0216] Triangles are used to describe the relative relationship between discrete tree trunk positions; multiple triangles are combined into a triangulated network to describe the overall distribution of tree trunks in the scan; the number of triangles depends on the size of the scanned sample and the density of tree trunks;

[0217] S105: 3D point cloud coarse registration;

[0218] Through efficient search operations in hash tables, triangle matching pairs are found and the similarity between matching pairs of scanning stations is calculated. If the similarity meets the threshold, the coarse registration parameters of the scanning point cloud pairs are output;

[0219] S106: Factor graph optimization;

[0220] The factor graph optimization method is introduced to model the inconsistency between relative transformations of multiple stations, and globally consistent transformation parameters are obtained after overall optimization.

[0221] S107: Point cloud precise registration and pose estimation;

[0222] The relative pose transformation parameters after factor graph optimization are used as initial values ​​and optimized through the point-to-point ICP method. The pose of each scanning station in the unified coordinate system is derived through the optimized transformation parameters, and all scanning station clouds are converted to the unified coordinate system to form an overall consistent global point cloud.

[0223] The generation of distance images provided by the embodiment of the present invention:

[0224] (1) Ground-based LiDAR data acquisition;

[0225] TLS scanning stations are arranged according to the size of the forest plot. For a plot of 20m×20m, a scanning method of one station in the middle and one station in each of the four corners can be adopted to obtain complete plot scanning data to the greatest extent possible;

[0226] (2) Converting 3D point cloud into range image;

[0227] Process the single-site cloud data and set the three-dimensional coordinates of any point i in the point cloud to be (x i ,y i ,z i ), can also be expressed as polar coordinates (α i ,β i ,γ i ) in the form of i , β i are the horizontal angle and the vertical angle respectively, γ i is the distance from the scanning point i to the scanning center; then, the index of the scanning point i in the distance image can be defined by formula (1);

[0228]

[0229] In the formula, row i ,col i are the row index and column index of the scanning point i in the range image, respectively, and θ is the angular resolution of the range image;

[0230] Each pixel in the distance image uses γ i Filling: If multiple scan points correspond to the same index position after conversion, the pixel retains the scan point information closest to the center.

[0231] The trunk point continuity constraint provided by the embodiment of the present invention is as follows:

[0232] (1) Define the search step (Step), pixel difference threshold (R diff ), group pixel number threshold (Seg th ) The three parameters are set to 15 pixels, 0.03m, and 100 pixels respectively;

[0233] (2) For each column in the distance image, starting from the first point, compare the pixel value difference between the search point and the current point within the search step Step range. If it is less than the preset threshold R diff , the search point and the current point are grouped into the same group; otherwise, a new group is created;

[0234] (3) Repeat the above operation until all pixels in the range image are visited;

[0235] (4) Count the number of pixels in each group. If it exceeds the preset threshold Seg thIf the pixel is not positive, it is retained and the pixels in the group are marked as 1; otherwise, it is discarded and the pixels in the group are marked as 0;

[0236] (5) Traverse all pixels. If the pixel is marked as 1, the corresponding scan point is retained. Otherwise, if the pixel is marked as 0, the scan point is discarded.

[0237] The trunk point direction constraint provided by the embodiment of the present invention is:

[0238] (1) Using Euclidean distance clustering, the point cloud data extracted in step 2 is processed and divided into different point cloud clusters;

[0239] (2) Perform principal component analysis (PCA) on each point cloud cluster and calculate the linear L λ , Planarity P λ , Dispersion S λ Three parameters, describing the geometric distribution of point cloud clusters in space, are defined by formula (2);

[0240]

[0241] In the formula, λ1>λ2>λ3 are the eigenvalues ​​in descending order, corresponding to the degree of change in the direction of the eigenvector;

[0242] (3) The trunk points are linearly distributed in space, that is, they have a large linear L λ , while the ground point has a larger planarity P λ , the branch and leaf points have a large dispersion S λ Therefore, this feature is used to judge the type of the extracted point cloud cluster. If a point cloud cluster meets the condition L λ >0.95, the point cloud cluster is considered to be a tree trunk; distribution of point cloud clusters before and after using geometric constraints;

[0243] (4) Using the extracted tree trunk point cloud cluster, extract the corresponding tree trunk position; its plane position (x, y) is the center of the point cloud cluster, and the z component is selected as the root point of the trunk, that is, the intersection of the trunk and the ground. The black dot at the bottom is the extracted trunk position.

[0244] The triangle descriptor constructed according to the embodiment of the present invention is as follows:

[0245] (1) Screening triangles;

[0246] 1) For any tree trunk position, search for 10 neighboring points with it as the center;

[0247] 2) Pick any two neighboring points and combine them with the current point to form a triangle;

[0248] 3) If the constructed triangle is an isosceles / equilateral triangle, remove it; otherwise, if the three sides are different, keep it;

[0249] 4) Repeat the above steps to traverse all tree trunk positions;

[0250] (2) Hash table storage

[0251] 1) Use the “division remainder method” to construct a hash function, defined by formula (3);

[0252] h(l1,l2,l3)=mod[mod(l1*m,N)+l2*m,N]+l3 (3)

[0253] Where (l1, l2, l3) is the side length of any triangle and is arranged in descending order (l1>l2>l3); m is a large prime number, and N represents the size of the hash table;

[0254] 2) Perform the above operation for each triangle in the triangulated network so that each triangle can be projected to a unique index in the hash table;

[0255] 3) All TLS scans in the sample plot are projected to the corresponding positions in the hash table; if a hash conflict occurs, the chaining method is used to handle it.

[0256] The three-dimensional point cloud coarse registration provided by the embodiment of the present invention:

[0257] (1) Search for triangle matching pairs

[0258] 1) For any triangle in the scan station to be searched, use the hash function to directly map the triangle descriptor to the corresponding index in the hash table;

[0259] 2) Find the triangle descriptor contained in the current index position, and calculate the direction vector difference of the corresponding vertex using formula (4). If it is less than the threshold value 0.1, the tree trunk matching pair is retained, otherwise it is discarded;

[0260]

[0261] In the formula, and Represent the direction vectors of the query trunk and the candidate trunk respectively;

[0262] (2) Calculation of three-dimensional transformation coarse parameters

[0263] 1) Count the scanning station IDs corresponding to each trunk match, vote, and calculate the similarity, which is defined by formula (5);

[0264]

[0265] In the formula, S q With S c Represent query scan and candidate scan respectively, Vc is the number of votes scanned by the candidate, N q is the number of trunks that are candidates for scanning;

[0266] 2) If the similarity between the query scan and the candidate scan is greater than a preset threshold, it is considered that there is an adjacency relationship between the two stations, and the paired trunk positions are used to solve the 3D transformation parameters;

[0267] The factor graph optimization:

[0268] (1) Factor graph modeling

[0269] For a certain forest plot, a multi-station TLS scan was performed. Then, the pose to be determined in the unified coordinate system of each station scan was used as a node, and the coarse transformation between adjacent stations was used as a factor constraint to establish a factor graph model.

[0270] (2) Global Optimization

[0271] Comprehensively consider all nodes and constraints in the factor graph and convert it into a least squares problem under the Bayesian estimation framework; reduce the inconsistency between relative transformations and solve the optimal scanning station posture and relative transformation;

[0272] The point cloud precise registration and pose estimation:

[0273] (1) Point cloud precise registration;

[0274] The relative pose transformation parameters after factor graph optimization are used as initial values, and the transformation parameters between each valid paired scanning station are optimized and adjusted using the point-to-point ICP precise registration method.

[0275] (2) Global pose derivation;

[0276] Based on the optimized relative transformation parameters, the depth-first strategy is used to deduce the position and posture of each scanning station in the unified coordinate system;

[0277] (3) Unified point cloud coordinate system;

[0278] After the above steps, the position and posture of each scanning station in the unified coordinate system are known and applied to each scanning point cloud to convert it from an independent local coordinate system to an overall point cloud in the unified coordinate system.

[0279] like Figure 2 As shown, a multi-station ground-based laser scanning point cloud registration system suitable for a forest environment provided by an embodiment of the present invention includes:

[0280] A distance image generation module, used for generating a distance image;

[0281] The continuity constraint module is used for the continuity constraint of the tree trunk points. It performs distance continuity constraint in the column direction and extracts the stable part of the point cloud.

[0282] Direction constraint module, used for direction constraint of tree trunk points; removes the influence of dense branches and leaves and scattered ground fragments that still exist;

[0283] Descriptor construction module, used to construct triangle descriptors; triangles are used to describe the relative relationship between discrete tree trunk positions; multiple triangles are combined into a triangulated network to describe the overall distribution of tree trunks in the scan; the number of triangles depends on the size of the scanned sample and the density of tree trunks;

[0284] The registration module is used for the rough registration of 3D point clouds. It searches for triangle matching pairs through efficient search operations in hash tables and calculates the similarity between matching pairs of scanning stations. If the similarity meets the threshold, it outputs the rough registration parameters of the scanning point cloud pairs.

[0285] Optimization module, used for factor graph optimization; introduces factor graph optimization method to model the inconsistency between relative transformations of multiple stations, and obtains globally consistent transformation parameters after overall optimization;

[0286] The pose estimation module is used for precise point cloud registration and pose estimation. The relative pose transformation parameters after factor graph optimization are used as initial values ​​and optimized through the point-to-point ICP method. The pose of each scanning station in the unified coordinate system is derived through the optimized transformation parameters, and all scanning station clouds are converted to the unified coordinate system to form an overall consistent global point cloud.

[0287] The multi-station ground-based laser scanning point cloud registration system of the present invention mainly realizes the accurate registration of multi-station clouds in a forest environment by means of a modular approach. The following is a detailed description of its working principle:

[0288] 1. Range image generation and tree trunk point extraction

[0289] The system first generates a range image through the range image generation module, which performs preliminary processing on the point cloud data collected by each scanning station. Then, the system uses the continuity constraint module to constrain the continuity of the trunk points, extracts points with continuous distances from the column direction, and retains a stable trunk point cloud. This step is further processed by the fast Euclidean distance clustering method to remove dense branches and ground fragments and retain only the trunk point cloud. This method takes advantage of the relatively stable continuity characteristics of tree trunks in forest environments and effectively reduces the interference of branches and cluttered ground on point cloud registration.

[0290] 2. Direction constraints and descriptor construction

[0291] The direction constraint module constrains the direction of the trunk points. Since the growth direction of the trunk is usually relatively vertical in the forest, the system uses the direction constraint as a criterion for removing residual branches and cluttered points on the ground. On this basis, the system constructs a triangle descriptor in the descriptor construction module, forms triangles by describing the relative positions between discrete trunk points, and combines multiple triangles into a triangulated network. This triangulated network can characterize the spatial distribution characteristics of the trunk and has strong geometric stability. The number of triangles is determined by the size of the scanning area and the density of the trunk. This descriptor construction method helps to improve the matching accuracy of the trunk point cloud during the registration process.

[0292] 3. Coarse Registration and Factor Graph Optimization

[0293] In the registration module, the system uses hash tables for efficient search, calculates the similarity between scanning stations by finding triangle matching pairs, and the matching pairs that meet the similarity threshold are considered to be valid pairs for coarse registration, and outputs the coarse registration parameters. This registration method based on triangle descriptors and hash search can effectively achieve coarse registration of different scanning stations. Then, the system enters the optimization module and introduces the factor graph optimization method to model the relative transformation of multi-site clouds to solve the inconsistency caused by errors between multiple stations. Through factor graph optimization, the system obtains globally consistent transformation parameters, providing accurate initial values ​​for the next step of fine registration.

[0294] 4. Precise registration and global pose estimation

[0295] In the pose estimation module, the system first uses the relative pose transformation parameters obtained by factor graph optimization as the initial value, and further performs precise registration based on the point-to-point ICP (iterative closest point) method. Finally, the system derives the pose of each scanning station in a unified coordinate system, and converts the cloud data of all scanning stations to the same coordinate system to form an overall consistent global point cloud. This method can effectively reduce errors and ensure the registration accuracy of point clouds in complex forest environments, providing high-precision three-dimensional point cloud data support for subsequent forest environment analysis and data processing.

[0296] The present invention is specifically implemented:

[0297] The multi-station ground-based laser scanning point cloud registration method and system suitable for forest environment, the specific implementation steps are as follows:

[0298] Step 1: Generate a range image

[0299] (1) Ground-based LiDAR data collection

[0300] TLS scanning stations are arranged according to the size of the forest sample plot. For example, for a sample plot of 20m×20m, a scanning method with one station in the middle and one station in each of the four corners can be adopted to obtain complete sample plot scanning data to the greatest extent possible.

[0301] (2) Conversion of 3D point cloud into distance image

[0302] Process the single-site cloud data and set the three-dimensional coordinates of any point i in the point cloud to be (x i ,y i ,z i ), can also be expressed as polar coordinates (α i ,β i ,γ i ) in the form of i , β i are the horizontal angle and the vertical angle respectively, γ i is the distance from scanning point i to the scanning center; then, the index of scanning point i in the distance image can be defined by formula (1).

[0303]

[0304] In the formula, row i ,col i are the row index and column index of the scanning point i in the range image, respectively, and θ is the angular resolution of the range image;

[0305] Each pixel in the distance image uses γ i Filling: If multiple scan points correspond to the same index position after conversion, the pixel retains the scan point information closest to the center. Figure 3 The working principle of ground-based laser scanners and the process of converting raw point cloud data into distance images are demonstrated in the paper.

[0306] Step 2: Trunk point continuity constraints

[0307] The converted distance image retains the geometric structure information in the original point cloud while reducing the data dimension. The present invention performs distance continuity constraints in the column direction to extract the stable part of the point cloud. Then, the fast Euclidean distance clustering method is used for further processing, and the dense branches and scattered ground fragments are removed through geometric constraints.

[0308] (1) Define the search step size (Step), pixel difference threshold (R diff ), group pixel number threshold (Seg th ) are set to 15 pixels, 0.03m, and 100 pixels respectively.

[0309] (2) For each column in the distance image, starting from the first point, compare the pixel value difference between the search point and the current point within the search step Step range. If it is less than the preset threshold R diff , the search point and the current point are grouped together. Otherwise, a new group is created.

[0310] (3) Repeat the above steps until all pixels in the range image are visited.

[0311] (4) Count the number of pixels in each group. If it exceeds the preset threshold Seg th If the pixel is not in the group, it is retained and the pixel in the group is marked as 1. Otherwise, it is discarded and the pixel in the group is marked as 0. Figure 4 as shown in .

[0312] (5) Traverse all pixels. If the pixel is marked as 1, the corresponding scan point is retained. Otherwise, if the pixel is marked as 0, the scan point is discarded.

[0313] Step 3: Trunk point orientation constraints

[0314] There are still some dense branches and leaves and sporadic ground fragments in the trunk points extracted by continuity constraints. These parts are continuously distributed in the distance image and are difficult to distinguish. However, their distribution in three-dimensional space is obviously different from that of the trunk points. For example, the trunk points are obviously distributed linearly, while the branch and leaf points are evenly distributed in space, and the distribution of the ground points is close to a plane.

[0315] (1) Using Euclidean distance clustering, the point cloud data extracted in step 2 is processed and divided into different point cloud clusters.

[0316] (2) Perform principal component analysis (PCA) on each point cloud cluster and calculate the linear L λ , Planarity P λ , Dispersion S λ Three parameters, describing the geometric distribution of point cloud clusters in space, are defined by formula (2).

[0317]

[0318] Where λ1>λ2>λ3 are the eigenvalues ​​in descending order, corresponding to the degree of change in the direction of the eigenvector.

[0319] (3) The trunk points are linearly distributed in space, that is, they have a large linear L λ , while the ground point has a larger planarity P λ , the branch and leaf points have a large dispersion S λ Therefore, the present invention uses this characteristic to judge the type of the extracted point cloud cluster. If a point cloud cluster meets the condition L λ >0.95, then the point cloud cluster is considered to be a tree trunk. The distribution of point cloud clusters before and after using geometric constraints is shown in the figure below. Figure 5 as shown in .

[0320] (4) Using the extracted tree trunk point cloud cluster, extract the corresponding tree trunk position. Its plane position (x, y) is the center of the point cloud cluster, and the z component is selected as the root point of the tree trunk, that is, the intersection of the tree trunk and the ground. Figure 6 As shown in the figure, the black dot at the bottom is the extracted tree trunk position.

[0321] Step 4: Construct triangle descriptor

[0322] Triangles are used to describe the relative relationship between discrete tree trunk positions. Subsequently, multiple triangles are combined into a triangulated network to describe the overall distribution of tree trunks in the scan. The number of triangles depends on the size of the scanned plot and the density of tree trunks, so how to store triangle information is one of the key steps. The present invention uses a hash table to efficiently store triangulated networks, and each triangle is directly mapped to the corresponding position through a hash function.

[0323] (1) Filter triangles

[0324] A triangle is the simplest geometric shape that can be described by only three sides and has good three-dimensional rotation / translation invariance. In addition, except for isosceles / equilateral triangles, if the lengths of the three sides of two triangles are equal, then their three vertices also correspond one to one, such as Figure 7 as shown in .

[0325] 1) For any tree trunk position, search for 10 neighboring points with it as the center.

[0326] 2) Pick any two neighboring points and combine them with the current point to form a triangle.

[0327] 3) If the constructed triangle is an isosceles / equilateral triangle, remove it. Otherwise, if the three sides are different, keep it, such as Figure 7 as shown in .

[0328] 4) Repeat the above steps to traverse all tree trunk positions.

[0329] (2) Hash table storage

[0330] 1) Use the “division remainder method” to construct a hash function, defined by formula (3).

[0331] h(l1,l2,l3)=mod[mod(l1*m,N)+l2*m,N]+l3 (3)

[0332] Where (l1, l2, l3) is the side length of any triangle and is arranged in descending order (l1>l2>l3); m is a large prime number and N represents the size of the hash table.

[0333] 2) Perform the above operation for each triangle in the triangulated network so that each triangle can be projected into a unique index in the hash table, such as Figure 8 as shown in .

[0334] 3) All TLS scans in the sample plot are projected to the corresponding positions in the hash table. If a hash conflict occurs, the chaining method is used to handle it, such as Figure 8 as shown in .

[0335] Step 5: Rough registration of 3D point cloud

[0336] The present invention relies on efficient search operations of hash tables to find triangle matching pairs, calculate the similarity between scanning station matching pairs, and output the coarse registration parameters of the scanning point cloud pair if the similarity meets a threshold.

[0337] (1) Search for triangle matching pairs

[0338] 1) For any triangle in the scan station to be searched, use the hash function to directly map the triangle descriptor to the corresponding index in the hash table, such as Fig. 9 As shown in (a).

[0339] 2) Find the triangle descriptor contained in the current index position, and calculate the direction vector difference of the corresponding vertex using formula (4). If it is less than the threshold value 0.1, the tree trunk matching pair is retained, otherwise it is discarded. Fig. 9 As shown in (b).

[0340]

[0341] In the formula, and Represent the direction vectors of the query trunk and the candidate trunk respectively.

[0342] (2) Calculation of three-dimensional transformation coarse parameters

[0343] 1) Count the scanning station IDs corresponding to each trunk match, vote, and calculate the similarity, which is defined by formula (5).

[0344]

[0345] In the formula, S q With S c Represent query scan and candidate scan respectively, V c is the number of votes scanned by the candidate, N q The number of trunks that are candidates for scanning.

[0346] 2) If the similarity between the query scan and the candidate scan is greater than a preset threshold (set to 30% in the present invention), it is considered that there is an adjacency relationship between the two stations, and the paired trunk positions are used to solve the three-dimensional transformation parameters.

[0347] Step 6: Factor Graph Optimization

[0348] The three-dimensional transformation parameters obtained by the rough registration of point clouds have large inconsistencies, and the errors will gradually accumulate after being transmitted through multiple stations, making it difficult to provide reliable initial values ​​for the subsequent fine registration steps. Therefore, the present invention introduces a factor graph optimization method to model the inconsistency between the relative transformations of multiple stations, and obtains globally consistent transformation parameters after overall optimization.

[0349] (1) Factor graph modeling

[0350] For a certain forest plot, a multi-station TLS scan is performed, such as Fig.10 Then, the pose to be determined in the unified coordinate system of each station scan is used as a node, and the coarse transformation between adjacent stations is used as a factor constraint to establish a factor graph model, as shown in (a). Fig.10 As shown in (b).

[0351] (2) Global Optimization

[0352] Considering all nodes and constraints in the factor graph, it is converted into a least squares problem under the Bayesian estimation framework. The inconsistency between relative transformations is reduced, and the optimal scanning station posture and relative transformation are solved.

[0353] Step 7: Point cloud registration and pose estimation

[0354] The present invention uses the relative pose transformation parameters optimized by the factor graph as the initial value, and optimizes it by the point-to-point ICP method. Then, the pose of each scanning station in the unified coordinate system is derived by the optimized transformation parameters, and all scanning station clouds are converted to the unified coordinate system to form an overall consistent global point cloud.

[0355] (1) Point cloud registration

[0356] The relative pose transformation parameters after factor graph optimization are used as initial values, and the transformation parameters between each valid paired scanning station are optimized and adjusted using the point-to-point ICP precise registration method.

[0357] (2) Global pose derivation

[0358] Based on the optimized relative transformation parameters, the depth-first strategy is used to deduce the position and posture of each scanning station in the unified coordinate system.

[0359] (3) Unified point cloud coordinate system

[0360] After the above steps, the position and posture of each scanning station in the unified coordinate system are known and applied to each scanning point cloud to convert it from an independent local coordinate system to an overall point cloud in the unified coordinate system.

[0361] Technical solution flow of multi-station ground-based laser scanning point cloud registration method and system suitable for forest environment Fig.11 .

[0362] Technical solution of the embodiment of the present invention

[0363] Taking the self-collected dataset of Shennongjia National Nature Reserve in Hubei Province and the public datasets released by Swiss Federal Institute of Technology (ETH), Finnish Geoscience Institute (FGI) and Tongji University, a total of 12 forest plots as examples, the technical solutions of the multi-station ground-based laser scanning point cloud registration method and system suitable for forest environment are as follows.

[0364] (1) Trunk point extraction

[0365] In the forest plot, the operator deploys multiple scanning stations from different angles to try to cover the plot range completely. For any of the TLS scanning data of the station, the trunk points are extracted. Then, the corresponding position of each trunk point is calculated. The present invention uses its root point, that is, the intersection of the trunk point and the ground, as a representative point to describe the distribution of the trunk. The following is an example of processing representative plots in four data sets of Shennongjia, ETH, FGI, and Tongji University. The extracted trunk points and their corresponding positions are as follows: Fig.12 as shown in .

[0366] Fig.12 The four data sets all have obvious characteristics. The Shennongjia data belongs to a typical subtropical natural forest, with a messy and dense distribution of tree trunks, which makes extraction difficult. The ETH data and the FGI data belong to the northern natural forest, with a high density and a few shrubs. The Tongji University data was collected on campus and is an artificially managed forest with low density and regular distribution. The present invention shows good trunk extraction effects in the examples of the four data sets, with complete structures and accurate tree root point locations.

[0367] (2) Point cloud coarse registration

[0368] Based on the extracted tree trunk points, triangle descriptors are constructed, and the hash table database is used to efficiently store triangle descriptors in all scans and search for triangle pairs with the same name. Direction constraints are added to the paired tree trunks, the similarity of the two-station scans is calculated, and it is determined whether the two scanning stations are adjacent and the three-dimensional transformation parameters of the rough registration are output.

[0369] Subsequently, the inconsistency problem in the coarse registration parameters is modeled, with the scan station poses to be determined as nodes and the adjacent transformations between scan stations as factor constraints to form a factor graph. After optimization, the overall consistent transformation parameters are obtained, providing reliable initial values ​​for the point cloud fine registration.

[0370] Fig.13The figure shows the distribution of the rough registration error of the point cloud in four data sets, including rotation and translation errors. The figure also shows the changing trends of the errors before and after factor graph optimization. The results show that factor graph optimization can effectively reduce the impact of the rough registration error. In most plots, a rotation error of 20 mrad and a translation error of less than 30 cm can be achieved.

[0371] (3) Point cloud registration

[0372] Based on the reliable transformation initial value provided by factor graph optimization, the present invention uses the point-to-point ICP registration method for precise optimization, and uses the depth-first strategy to derive the position and posture of each scanning station in the unified coordinate system. The error evaluation results are as follows: Fig.14 as shown in .

[0373] Fig.14 The experimental results of the precise registration of the present invention in four data sets and twelve sample plots are shown in Figure 2. After a large number of experimental tests, the method proposed in the present invention can achieve a rotation error of 0-3mrad and a translation error of 0-3cm in most sample plots. The high-precision registration results demonstrate the superiority of the method proposed in the present invention.

[0374] 2. Technical Effects of Embodiments of the Invention

[0375] The embodiments of the present invention have carried out a large number of experiments in 12 forest plots, including the self-collected dataset of the Shennongjia National Nature Reserve in Hubei Province, the public datasets released by the Swiss Federal Institute of Technology (ETH), the Finnish Geodesy Institute (FGI), and Tongji University. The multi-station ground-based laser scanning point cloud registration method and system proposed in the present invention for forest environments can achieve good experimental results and realize automatic registration of multi-station clouds without any prior information. The overall point cloud after registration is as follows: Fig.15 As shown in , four representative examples of the datasets are shown.

[0376] An embodiment of the present invention provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the multi-station ground-based laser scanning point cloud registration method applicable to a forest environment.

[0377] like Figure 2 As shown, an embodiment of the present invention provides a multi-station ground-based laser scanning point cloud registration system suitable for a forest environment that implements the robust real-time positioning and mapping method suitable for a complex understory environment, the system comprising:

[0378] A distance image generation module, used for generating a distance image;

[0379] The continuity constraint module is used for the continuity constraint of the tree trunk points. It performs distance continuity constraint in the column direction and extracts the stable part of the point cloud.

[0380] Direction constraint module, used for direction constraint of tree trunk points; removes the influence of dense branches and leaves and scattered ground fragments that still exist;

[0381] Descriptor construction module, used to construct triangle descriptors; triangles are used to describe the relative relationship between discrete tree trunk positions; multiple triangles are combined into a triangulated network to describe the overall distribution of tree trunks in the scan; the number of triangles depends on the size of the scanned sample and the density of tree trunks;

[0382] The registration module is used for the rough registration of 3D point clouds. It searches for triangle matching pairs through efficient search operations in hash tables and calculates the similarity between matching pairs of scanning stations. If the similarity meets the threshold, it outputs the rough registration parameters of the scanning point cloud pairs.

[0383] Optimization module, used for factor graph optimization; introduces factor graph optimization method to model the inconsistency between relative transformations of multiple stations, and obtains globally consistent transformation parameters after overall optimization;

[0384] The pose estimation module is used for precise point cloud registration and pose estimation. The relative pose transformation parameters after factor graph optimization are used as initial values ​​and optimized through the point-to-point ICP method. The pose of each scanning station in the unified coordinate system is derived through the optimized transformation parameters, and all scanning station clouds are converted to the unified coordinate system to form an overall consistent global point cloud.

[0385] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0386] The present invention processes multi-station ground-based laser scanning (TLS) point cloud data collected in complex forest (artificial / natural forest) environments and realizes automatic registration of multi-station clouds. In order to comprehensively evaluate the registration performance, the embodiments of the present invention were tested on 12 plots in four data sets (Shennongjia National Nature Reserve in Hubei Province, as well as Swiss Federal Institute of Technology (ETH), Finnish Geodesy Institute (FGI), and Tongji University). The accuracy evaluation results obtained are as follows: Fig.13 And Table 1, Fig.14 shown.

[0387] Fig.13 The rough registration errors of the present invention in various locations are shown. (a) shows the rotation component, and (b) shows the translation component. Table 1 summarizes the quantitative evaluation data of the rough registration errors in each data set. The different color columns in the figure represent different rotation / translation components, and whether there is a dotted line shadow inside the column represents the error value before and after the factor graph optimization. Fig.13 As can be concluded from Table 1, the error amount after factor graph optimization is significantly reduced, especially for the case of large coarse registration error. For example, the overall rotation error of plot 1 is reduced from 140mrad to less than 70mrad, and the translation error component is also reduced from 60cm to less than 30cm. In addition, the other plots also have good performance, showing the superiority of the factor graph optimization module in the system proposed in the present invention.

[0388] Table 1 Quantitative evaluation results of coarse registration error (including four public datasets of Shennongjia, ETH, FGI, and Tongji University)

[0389]

[0390] Fig.14 The precise registration errors of the present invention in various plots are shown, where (a) and (b) represent the rotation and translation error components, respectively. It can be seen from the figure that the method proposed in the present invention has achieved good results in all 12 test plots. For the rotation component, except for the slightly larger error in plot 4, the overall rotation errors of the other plots are all below 3mrad. For the translation component, the errors of a single coordinate axis of all plots are about 3cm. Among them, high-precision results within 1cm were even obtained in the two artificial forests of plots 11 and 12, verifying the accuracy and robustness of the method proposed in the present invention.

[0391] Finally, the precise position and posture obtained by the method of the present invention are used to transform the cloud data of each scanning station into a unified coordinate system to form a global point cloud of the sample site. Fig.15As shown in the figure (one plot is selected as a representative for each data set). It can be seen that the multi-site cloud after automatic registration can be properly restored to each scanning position and completely covers the entire plot. The goal of the method proposed in the present invention is well achieved, that is, to achieve efficient, accurate and automatic registration of multi-site TLS scans in complex forest environments without any prior information, any manual operation or any special marking.

[0392] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A multi-station ground-based laser scanning point cloud registration method suitable for forest environments, characterized in that: The following steps are involved: Step 1: Generate range image; Step 2: Continuity constraints of trunk points; Perform distance continuity constraints in the column direction to extract the stable part of the point cloud; Step 3: Trunk point direction constraint; The tree trunk points extracted by the continuity constraint still contain some dense branches and scattered ground fragments. Clustering is used to divide the point cloud into clusters, and then the geometric characteristics of each cluster are used to remove branches and ground points, and retain the tree trunk points; Step 4: Construct a triangle descriptor; Triangles are used to describe the relative relationship between discrete tree trunk positions; multiple triangles are combined into a triangulated network to describe the overall distribution of tree trunks in the scan; the number of triangles depends on the size of the scanned sample and the density of tree trunks; Step 5: Rough registration of 3D point cloud; Through efficient search operations in hash tables, triangle matching pairs are found and the similarity between matching pairs of scanning stations is calculated. If the similarity meets the threshold, the coarse registration parameters of the scanning point cloud pairs are output; Step 6: Factor graph optimization; The factor graph optimization method is introduced to model the inconsistency between relative transformations of multiple stations, and globally consistent transformation parameters are obtained after overall optimization. Step 7: Point cloud precise registration and pose estimation; The relative pose transformation parameters after factor graph optimization are used as initial values ​​and optimized through the point-to-point ICP method. The pose of each scanning station in the unified coordinate system is derived through the optimized transformation parameters, and all scanning station clouds are converted to the unified coordinate system to form an overall consistent global point cloud.

2. The multi-station ground-based laser scanning point cloud registration method suitable for forest environment as claimed in claim 1, characterized in that: The generated distance image is: (1) Ground-based LiDAR data acquisition; TLS scanning stations are arranged according to the size of the forest plot. For a plot of 20m×20m, a scanning method of one station in the middle and one station in each of the four corners can be adopted to obtain complete plot scanning data to the greatest extent possible; (2) Converting 3D point cloud into range image; Process the single-site cloud data and set the three-dimensional coordinates of any point i in the point cloud to be (x i ,y i ,z i ), can also be expressed as polar coordinates (α i ,β i ,γ i ) in the form of i , β i are the horizontal angle and the vertical angle, γ i is the distance from the scanning point i to the scanning center; then, the index of the scanning point i in the distance image can be defined by formula (1); In the formula, row i ,col i are the row index and column index of the scanning point i in the range image, respectively, and θ is the angular resolution of the range image; Each pixel in the distance image uses γ i Filling: If multiple scan points correspond to the same index position after conversion, the pixel retains the scan point information closest to the center.

3. The multi-station ground-based laser scanning point cloud registration method suitable for forest environment as claimed in claim 1, characterized in that: The trunk point continuity constraint: (1) Define the search step (Step), pixel difference threshold (R diff ), group pixel number threshold (Seg th ) The three parameters are set to 15 pixels, 0.03m, and 100 pixels respectively; (2) For each column in the distance image, starting from the first point, compare the pixel value difference between the search point and the current point within the search step Step range. If it is less than the preset threshold R diff , the search point and the current point are grouped into the same group; otherwise, a new group is created; (3) Repeat the above operation until all pixels in the range image are visited; (4) Count the number of pixels in each group. If it exceeds the preset threshold Seg th If the pixel is not positive, it is retained and the pixels in the group are marked as 1; otherwise, it is discarded and the pixels in the group are marked as 0; (5) Traverse all pixels. If the pixel is marked as 1, the corresponding scan point is retained. Otherwise, if the pixel is marked as 0, the scan point is discarded.

4. The multi-station ground-based laser scanning point cloud registration method suitable for forest environment as claimed in claim 1, characterized in that: The trunk point orientation constraint: (1) Using Euclidean distance clustering, the point cloud data extracted in step 2 is processed and divided into different point cloud clusters; (2) Perform principal component analysis (PCA) on each point cloud cluster and calculate the linear L λ , Planarity P λ , Dispersion S λ Three parameters, describing the geometric distribution of point cloud clusters in space, are defined by formula (2); In the formula, λ1>λ2>λ3 are the eigenvalues ​​in descending order, corresponding to the degree of change in the direction of the eigenvector; (3) The trunk points are linearly distributed in space, that is, they have a large linear L λ , while the ground point has a larger planarity P λ , the branch and leaf points have a large dispersion S λ Therefore, this feature is used to judge the type of the extracted point cloud cluster. If a point cloud cluster meets the condition L λ >0.95, the point cloud cluster is considered to be a tree trunk; distribution of point cloud clusters before and after using geometric constraints; (4) Using the extracted tree trunk point cloud cluster, extract the corresponding tree trunk position; its plane position (x, y) is the center of the point cloud cluster, and the z component is selected as the root point of the trunk, that is, the intersection of the trunk and the ground. The black dot at the bottom is the extracted trunk position.

5. The multi-station ground-based laser scanning point cloud registration method suitable for forest environment as claimed in claim 1, characterized in that: The triangle descriptor is constructed as follows: (1) Screening triangles; 1) For any tree trunk position, search for 10 neighboring points with it as the center; 2) Pick any two neighboring points and combine them with the current point to form a triangle; 3) If the constructed triangle is an isosceles / equilateral triangle, remove it; otherwise, if the three sides are different, keep it; 4) Repeat the above steps to traverse all tree trunk positions; (2) Hash table storage 1) Use the "division remainder method" to construct a hash function, defined by formula (3); h(l1,l2,l3)=mod[mod(l1*m,N)+l2*m,N]+l3 (3) Where (l1, l2, l3) is the side length of any triangle and is arranged in descending order (l1>l2>l3); m is a large prime number, and N represents the size of the hash table; 2) Perform the above operation for each triangle in the triangulated network so that each triangle can be projected to a unique index in the hash table; 3) All TLS scans in the sample plot are projected to the corresponding positions in the hash table; If a hash conflict occurs, the chaining method is used to resolve it.

6. The multi-station ground-based laser scanning point cloud registration method suitable for forest environment as claimed in claim 1, characterized in that: The three-dimensional point cloud coarse registration: (1) Search for triangle matching pairs 1) For any triangle in the scan station to be searched, use the hash function to directly map the triangle descriptor to the corresponding index in the hash table; 2) Find the triangle descriptor contained in the current index position, and calculate the direction vector difference of the corresponding vertex using formula (4). If it is less than the threshold value 0.1, the tree trunk matching pair is retained, otherwise it is discarded; In the formula, and Represent the direction vectors of the query trunk and the candidate trunk respectively; (2) Calculation of three-dimensional transformation coarse parameters 1) Count the scanning station IDs corresponding to each trunk match, vote, and calculate the similarity, which is defined by formula (5); In the formula, S q With S c Represent query scan and candidate scan respectively, V c is the number of votes scanned by the candidate, N q is the number of trunks that are candidates for scanning; 2) If the similarity between the query scan and the candidate scan is greater than a preset threshold, it is considered that there is an adjacency relationship between the two stations, and the paired trunk positions are used to solve the 3D transformation parameters; The factor graph optimization: (1) Factor graph modeling For a certain forest plot, a multi-station TLS scan was performed. Then, the pose to be determined in the unified coordinate system of each station scan was used as a node, and the coarse transformation between adjacent stations was used as a factor constraint to establish a factor graph model. (2) Global Optimization Comprehensively consider all nodes and constraints in the factor graph and convert it into a least squares problem under the Bayesian estimation framework; reduce the inconsistency between relative transformations and solve the optimal scanning station posture and relative transformation; The point cloud precise registration and pose estimation: (1) Point cloud precise registration; The relative pose transformation parameters after factor graph optimization are used as initial values, and the transformation parameters between each valid paired scanning station are optimized and adjusted using the point-to-point ICP precise registration method. (2) Global pose derivation; Based on the optimized relative transformation parameters, the depth-first strategy is used to deduce the position and posture of each scanning station in the unified coordinate system; (3) Unified point cloud coordinate system; After the above steps, the position and posture of each scanning station in the unified coordinate system are known and applied to each scanning point cloud to convert it from an independent local coordinate system to an overall point cloud in the unified coordinate system.

7. A multi-station ground-based laser scanning point cloud registration system suitable for a forest environment, which implements the multi-station ground-based laser scanning point cloud registration method suitable for a forest environment as claimed in any one of claims 1 to 6, characterized in that: The multi-station ground-based laser scanning point cloud registration system suitable for forest environment includes: A distance image generation module, used for generating a distance image; The continuity constraint module is used for the continuity constraint of the tree trunk points. It performs distance continuity constraint in the column direction and extracts the stable part of the point cloud. Direction constraint module, used for direction constraint of tree trunk points; removes the influence of dense branches and leaves and scattered ground fragments that still exist; Descriptor construction module, used to construct triangle descriptors; triangles are used to describe the relative relationship between discrete tree trunk positions; multiple triangles are combined into a triangulated network to describe the overall distribution of tree trunks in the scan; the number of triangles depends on the size of the scanned sample and the density of tree trunks; The registration module is used for the rough registration of 3D point clouds. It searches for triangle matching pairs through efficient search operations in hash tables and calculates the similarity between matching pairs of scanning stations. If the similarity meets the threshold, it outputs the rough registration parameters of the scanning point cloud pairs. Optimization module, used for factor graph optimization; introduces factor graph optimization method to model the inconsistency between relative transformations of multiple stations, and obtains globally consistent transformation parameters after overall optimization; The pose estimation module is used for precise point cloud registration and pose estimation. The relative pose transformation parameters after factor graph optimization are used as initial values ​​and optimized through the point-to-point ICP method. The pose of each scanning station in the unified coordinate system is derived through the optimized transformation parameters, and all scanning station clouds are converted to the unified coordinate system to form an overall consistent global point cloud.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the multi-station ground-based laser scanning point cloud registration method applicable to a forest environment as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the multi-station ground-based laser scanning point cloud registration method applicable to a forest environment as described in any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the multi-station ground-based laser scanning point cloud registration system suitable for forest environments as described in claim 7.

Images