Forest uls-tls point cloud automatic registration method and system based on ground context description

Abstract

The application belongs to the technical field of ecological monitoring, and discloses a forest ULS-TLS point cloud automatic registration method based on ground context description; the application uses ground points as registration primitives, constructs ground context, and can directly register TLS single scanning data to ULS point cloud without pre-registration among different scanning point clouds of TLS; the application has high reliability: the extraction process of ground points is relatively standardized, and stable and reliable results can be usually obtained, and is not affected by the distribution, density and complexity of shapes of trees in the forest; the application improves the calculation efficiency: compared with the extraction of tree attributes, the use of ground points as registration primitives reduces the calculation cost and avoids high calculation overhead caused by the extraction of tree attributes; the application is dedicated to providing a forest ULS-TLS point cloud automatic registration method based on ground context description, so as to quickly, accurately and automatically realize the registration of forest ULS and TLS point clouds, comprehensively acquire forest structure information, and promote the application of laser radar technology in forest investigation.

Description

Technical Field

[0001] The present invention belongs to the technical field of ecological monitoring, and in particular relates to a forest ULS-TLS point cloud automatic registration method and system based on ground context description. Background Art

[0002] With the development of remote sensing technology, three-dimensional point cloud technology has been widely used in fields such as forest ecological monitoring, forestry resource surveys, and environmental protection. By capturing the three-dimensional coordinates of objects, point cloud technology can accurately depict the structural characteristics of forests, providing an important data foundation for forest management and research. Currently, laser scanning technology (LiDAR) is the main means of obtaining three-dimensional point clouds, including terrestrial laser scanning (TLS) and unmanned aerial laser scanning (ULS). TLS can obtain high-precision ground point cloud data and is suitable for high-precision monitoring of small areas; while ULS, due to the flexibility and maneuverability of drones, can cover large forest areas and collect point cloud data over a wide range.

[0003] ULS is efficient and flexible for mapping forest canopies, capable of generating georeferenced point clouds using the Global Navigation Satellite System (GNSS). However, it is limited in capturing information beneath the canopy. Conversely, TLS can effectively supplement understory measurements through close-range observations. However, due to terrain and line-of-sight limitations, data coverage is limited, and it is difficult to obtain complete information about the canopy. Therefore, the precise registration of ULS and TLS point cloud data, integrating their strengths to overcome their respective limitations and provide a more comprehensive and accurate representation of forest structure, has become an important research direction for point cloud applications in forest ecosystems.

[0004] Point cloud registration is a critical step in many LiDAR applications and has long been extensively studied and optimized in the fields of photogrammetry and computer vision. However, existing research has mostly focused on point cloud registration within a single platform, with relatively little research on the integration of point clouds across multiple platforms. Currently, multi-source point cloud registration methods primarily rely on extracting geometric features and establishing correspondences based on these features. While these methods are particularly suitable for urban scenarios, they are often difficult to directly apply to the registration of forest point cloud data, particularly when there is low overlap and high geometric similarity between tree trunks.

[0005] Existing multi-source forest point cloud automatic registration methods rely primarily on tree attributes. However, accurately acquiring attributes such as diameter at breast height, tree height, and tree position is challenging in situations where tree density is high, terrain is complex, and tree growth is irregular. Trees are affected by natural conditions such as wind, which reduces the reliability and accuracy of individual tree attributes acquired at different times and angles. Existing ULS-TLS forest point cloud registration methods rely heavily on the distribution and quality of tree attributes, and typically require pre-registration of multiple TLS scans to obtain sufficient features. Even if accurate tree attributes are acquired, the extracted tree attributes cannot achieve accurate registration for forests with regular tree arrangements and similar tree attributes.

[0006] Through the above analysis, the problems and defects of the existing technology are as follows:

[0007] Existing ULS-TLS forest point cloud registration methods rely heavily on the distribution and quality of tree attributes, often requiring pre-registration of multiple TLS scans to obtain sufficient features. Even with accurate tree attributes, the extracted tree attributes cannot achieve accurate registration for forests with regular tree arrangements and similar tree attributes. Summary of the Invention

[0008] In view of the problems existing in the prior art, the present invention provides a forest ULS-TLS point cloud automatic registration method based on ground context description.

[0009] The present invention is implemented as follows: a forest ULS-TLS point cloud automatic registration method based on ground context description includes:

[0010] Step 1: Data preprocessing;

[0011] Step 2: Initial correspondence establishment;

[0012] Using ground points to implement ULS-TLS registration, the ground can generally be approximated as consisting of planes and curved surfaces; the dimensional characteristics of the points are basically unaffected by the point cloud pose;

[0013] Three features are introduced to characterize the pre-processed point cloud: plane dimension, surface dimension, and the angle between the normal vector and the Z axis. These three geometric features are used to establish a preliminary correspondence between ULS key points and TLS scan positions.

[0014] Step 3: Point cloud registration based on ground context;

[0015] TLS collects surrounding environment data by rotating the scan around its scanning position; based on the LiDAR scanning method, through ground context matching, it identifies the key points in the ULS that are most similar to the TLS scanning position, and then estimates the transformation matrix for alignment.

[0016] Furthermore, the data is preprocessed:

[0017] (1) Ground point extraction;

[0018] ULS and TLS ground point cloud data are extracted through morphological filtering; all points in the point cloud are converted to a two-dimensional grid space; the lowest point in each pixel is selected as the seed point for clustering to obtain ground points;

[0019] (2) Local point cloud acquisition;

[0020] The ULS ground point cloud is divided into two-dimensional grids based on the range of x- and y-coordinates, with a step size of 0.5 meters. The center points of these grids are extracted as key points. The z-coordinates of the key points are set to 0.0 because they are irrelevant to the registration algorithm proposed in this invention.

[0021] The points in the TLS ground point cloud that are within a certain threshold of the horizontal neighborhood of the scanning position are filtered to generate a TLS local point cloud, and the points in the ULS ground point cloud that are within a certain threshold of the horizontal neighborhood of each key point are extracted to generate multiple ULS local point clouds for subsequent alignment; the threshold is usually set to 5m.

[0022] Furthermore, the ground point extraction:

[0023] ULS and TLS ground point cloud data are extracted by morphological filtering; all points in the point cloud are converted into a two-dimensional grid space; the lowest point in each pixel is selected as the seed point for clustering to obtain ground points.

[0024] Furthermore, the local point cloud is obtained:

[0025] The ULS ground point cloud is divided into two-dimensional grids based on the range of x- and y-coordinates, with a step size of 0.5 meters. The center points of these grids are extracted as key points. The z-coordinates of the key points are set to 0.0 because they are irrelevant to the registration algorithm proposed in this invention.

[0026] The points in the TLS ground point cloud that are within a certain threshold of the horizontal neighborhood of the scanning position are filtered to generate a TLS local point cloud, and the points in the ULS ground point cloud that are within a certain threshold of the horizontal neighborhood of each key point are extracted to generate multiple ULS local point clouds for subsequent alignment; the threshold is usually set to 5m.

[0027] Furthermore, the initial corresponding relationship is established:

[0028] For a given point, principal component analysis (PCA) is applied to analyze the local point cloud of its neighborhood, and three eigenvalues ​​λ1, λ2 and λ3 (satisfying λ1≥λ2≥λ3≥0) and the normal vector can be obtained. Based on these results, the plane feature a2, the surface feature a3 and the angle f between the normal vector and the Z axis of this point can be derived;

[0029]

[0030] Given a TLS scan position p t and its related geometric features and f t , a key point p of ULS u and its related geometric features and f u ; If the following conditions are met, p u Considered as p t The initial corresponding points are added to the initial corresponding set C init middle:

[0031]

[0032] Among them, δ, θ and ε are three different thresholds, set to 0.1, 0.1 and 2.0° respectively;

[0033] By using the above method, the geometric features of all ULS key points and the local point cloud of the TLS scanning position are calculated and compared to generate the initial corresponding set C of the TLS scanning position. init ={p u |u=1,2…n init}.

[0034] Furthermore, the point cloud registration based on ground context:

[0035] Similarity score sim(H ULS ,H TLS ) is calculated as follows:

[0036]

[0037] In the similarity score calculation, the matrix element M ij The median value of NaN is excluded from the calculation; the variable l represents M ij The number of non-NaN values ​​in the matrix; For the rotation transformation of the TLS point cloud, it can be achieved by performing corresponding transformation operations on the columns of the matrix:

[0038] r m :[z1,z2,…z Na ]→[zNa-m+1 ,…z Na ,z1,…,z Na-m ]#(4)

[0039] Among them, H TLS =[z1,z2,…z Na ], zj represents H TLS A column of , and m∈[1,Na) is the rotation parameter, and the corresponding rotation angle is θ m =2πm / Na;

[0040] If a pair of ground contexts of ULS and TLS are correctly aligned, then the similarity score function represents the standard deviation of the height difference between the corresponding point pairs; since the Z axes of ULS and TLS are perpendicular, ideally, the similarity score should be close to zero;

[0041] After obtaining the initial correspondence, a ground context Ht is generated for the TLS scan position; at the same time, the initial correspondence set C init The ULS key points in the ground context Hu = {H u i |i=1,2…n init}; To complete the registration, it is necessary to compare each H u i With Ht at a rotation angle θ m The similarity score under the formula is score=sim(H u i ,r m (H t )); The matching item with the smallest score will be selected for registration;

[0042] Using the rotation angle θ in the best match best The horizontal coordinates (xu, yu) of the TLS scanning position and its corresponding point in the ULS can be used to complete the planar registration. At the same time, according to formula (3), when calculating the similarity score of the best match, the average height difference Zmean between the corresponding points can also be calculated to determine the Z-axis translation transformation in the vertical registration.

[0043] Another object of the present invention is to provide a forest ULS-TLS point cloud automatic registration system based on ground context description, comprising:

[0044] Preprocessing module, used for data preprocessing;

[0045] A module is built to establish initial correspondences. ULS-TLS registration is performed using ground points, which can generally be approximated as consisting of planes and curved surfaces. The dimensional characteristics of the points are largely unaffected by the point cloud pose. Three features are introduced to characterize the preprocessed point cloud: plane dimension, surface dimension, and the angle between the normal vector and the Z axis. These three geometric features are used to establish preliminary correspondences between ULS keypoints and TLS scan positions.

[0046] The point cloud registration module is used for ground context-based point cloud registration. The TLS collects surrounding environment data by rotating the scan around its scanning position. Based on the LiDAR scanning method, the key points in the ULS that are most similar to the TLS scanning position are identified through ground context matching, and the transformation matrix is ​​estimated for registration.

[0047] 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 performs the steps of the forest ULS-TLS point cloud automatic registration method based on ground context description.

[0048] 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 forest ULS-TLS point cloud automatic registration method based on ground context description.

[0049] Another object of the present invention is to provide an information data processing terminal, which is used to implement the forest ULS-TLS point cloud automatic registration system based on ground context description.

[0050] 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:

[0051] First, the present invention is committed to proposing a forest ULS-TLS point cloud automatic registration method based on ground context description, so as to quickly, accurately and automatically realize the registration of forest ULS and TLS point clouds, comprehensively obtain forest structure information, and promote the application of lidar technology in forest surveys.

[0052] This paper proposes a novel method for automatic ULS-TLS point cloud registration in forest environments. This method uses ground points as registration primitives to construct a ground context, enabling direct registration of single TLS scans to ULS point clouds without requiring pre-registration between different TLS scans. Compared to point cloud registration based on tree attributes, using ground points for registration offers the following significant advantages:

[0053] (1) High reliability: The extraction process of ground points is relatively standardized, and stable and reliable results can usually be obtained, unaffected by the complexity of tree distribution, density and shape in the forest.

[0054] (2) Improved computational efficiency: Compared to tree attribute extraction, using ground points as registration primitives reduces computational cost and avoids the high computational overhead caused by tree attribute extraction.

[0055] (3) Adapt to complex environment: Ground point registration method can better adapt to complex and variable forest environment, and is not limited by the accuracy and reliability of tree attribute acquisition.

[0056] (4) Strong scene applicability: Using ground points for registration can deal with forest scenes with regular tree arrangement and similar tree trunk attributes, helping to achieve more accurate registration results in these environments.

[0057] In addition, this method has the following advantages when building ground context for registration:

[0058] (1) When building TLS ground context, points within a certain horizontal distance from the scan center are selected to ensure the quality of TLS data and effectively reduce the impact of obstacles such as trees on subsequent registration results.

[0059] (2) In the process of building TLS ground context, point clouds are discretized into a certain number of sections by equidistant azimuth and radial distance, and converted into a matrix. As the distance from the scan position increases, these sections gradually expand to adapt to the decrease in point density caused by laser beam divergence.

[0060] (3) When building ground context, the discretization process is used to encode point clouds into a matrix based on height information, effectively addressing the impact of significant resolution differences between TLS and ULS data on registration accuracy.

[0061] Second, as a creative evidence of the invention's claims, it is also reflected in the following important aspects:

[0062] (1) The expected revenue and commercial value of the technical solution of the present invention after transformation is:

[0063] The present invention proposes a forest ULS-TLS point cloud automatic registration method and system based on ground context description, which can automatically and efficiently and accurately realize the registration between TLS and ULS point cloud data, thereby providing important technical support for ecological protection, forestry management and forest resource monitoring, promoting the development of forestry economy, helping to achieve the double carbon goal, and having wide commercial application prospects.

[0064] (2) The technical solution of the present invention fills the technical gap in the industry at home and abroad:

[0065] Currently, there are no examples of forest multi-source point cloud registration based on ground points in related technical fields, both domestically and internationally. The present invention innovatively proposes using ground points as registration primitives and designs a ground context to establish a correspondence between TLS and ULS point clouds, thereby achieving automatic point cloud registration. This invention fills this gap in the technology and provides an innovative and effective solution for the automatic registration of forest laser point clouds.

[0066] (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:

[0067] Point cloud registration is a key step in many LiDAR applications and has long been widely studied and developed in the fields of photogrammetry and computer vision. However, most research focuses on the registration of point clouds from the same platform, and the exploration of point cloud registration from different platforms is relatively limited. Currently, multi-source point cloud registration methods mainly focus on the extraction of geometric features and establishing correspondences based on these features. Commonly selected geometric features include points, lines, and planes, but these methods are generally more suitable for urban scenes. For forest point clouds, these methods are difficult to directly apply to multi-source forest point cloud registration due to the low overlap rate and the high similarity of tree trunk geometric features.

[0068] In complex forest environments, achieving fast and accurate automatic registration of ULS-TLS point clouds is an urgent problem to be solved. At present, automatic registration methods for multi-source point clouds in forests usually rely on tree attributes (such as tree position and diameter at breast height), but these methods are often not suitable for forests with diverse tree species, complex terrain, irregular tree layout, and lack of sufficient common trees. The present invention proposes a method and system for automatic registration of forest ULS-TLS point clouds based on ground context description, which innovatively uses ground points that are widely present in various forest environments as registration primitives, successfully solves the difficult problem of achieving fast and accurate registration of ULS and TLS point clouds in complex forest environments, and meets the long-standing demand of people to efficiently obtain forest structure information with the help of lidar technology.

[0069] (4) The technical solution of the present invention overcomes technical prejudice:

[0070] Traditional forest multi-source point cloud registration methods mostly rely on tree attributes. This invention abandons this approach and innovatively achieves registration through ground points and ground context, reflecting the breakthrough and creativity of the technical ideas.

[0071] Fourth, the innovation and technological progress of the present invention

[0072] 1. Innovation of the present invention

[0073] 1. Introducing ground context

[0074] This invention introduces ground context for the first time in point cloud registration between ULS (UAV laser scanning) and TLS (terrestrial laser scanning), enabling differentiated recognition of point cloud data from different scanning angles. Registration is achieved by using similarity scoring of ground context, significantly improving point cloud matching accuracy across different environments and scanning positions.

[0075] 2. Initial correspondence construction based on geometric features

[0076] This method extracts geometric features such as plane dimensions, surface dimensions, and the angle between the normal vector and the Z axis, and uses principal component analysis (PCA) to analyze the local point cloud. This method overcomes the challenges of high noise interference and low initial matching efficiency in traditional methods. This combined application of multiple features ensures a more accurate initial registration match and reduces subsequent computational effort.

[0077] 3. Optimized calculation model for similarity scoring

[0078] The similarity scoring model utilizes a difference matrix and calculates the standard deviation of non-NaN values. We also incorporate angular rotation matching to enhance the rotational invariance of the point cloud. By calculating the rotation transformation parameters through matrix column transformation, we ensure accurate initial point cloud matching while reducing matching time and achieving efficient registration between ULS and TLS.

[0079] 4. Innovative algorithms that prioritize both time efficiency and accuracy

[0080] By improving ground context and optimizing the similarity score calculation model, the time and computational resources required for ULS-TLS point cloud registration are significantly reduced. While retaining high-level information accuracy, this method balances registration speed and spatial complexity, making it highly practical for applications such as real-time drone inspections and ground-based 3D modeling.

[0081] 2. Existing technical problems solved

[0082] 1. Difficulty in constructing initial correspondences in point cloud registration

[0083] Traditional ULS-TLS point cloud registration relies heavily on manual annotation or local matching algorithms, but is significantly affected by noise and inconsistent point cloud features in complex environments. This invention automatically constructs the initial correspondence between ULS and TLS by introducing planar and curved surface features, as well as geometric features such as the angle between the normal vector and the Z axis, significantly improving the efficiency and accuracy of automated registration.

[0084] 2. Limitations of traditional point cloud registration methods in complex scenes

[0085] Conventional registration methods are inapplicable to height differences and vertical dimensions, and have difficulty handling ULS-TLS scan discrepancies. This paper optimizes the calculation of similarity scores by incorporating ground context and a rotational transformation model, achieving efficient and high-precision point cloud matching under different scanning conditions.

[0086] 3. It is difficult to strike a balance between registration speed and accuracy

[0087] The complex initial matching and subsequent optimization in traditional point cloud registration methods make it difficult to balance speed and accuracy. This invention uses a rapid ground-context-based screening registration method that accelerates the matching process through similarity scoring, improving speed while ensuring accuracy. This approach can meet the real-time needs of drone and terrestrial laser scanning in real-world projects.

[0088] 3. Significant Technological Advances

[0089] 1. Improved the automation level of point cloud registration

[0090] The initial correspondence is constructed using a variety of geometric features, and the point cloud is automatically registered through the ground context, eliminating the need for manual intervention and greatly improving the convenience and automation level of drone scanning applications.

[0091] 2. Improved registration accuracy at different angles

[0092] Through innovative similarity scoring and angle rotation calculation methods, this invention achieves high-precision registration of point cloud data from different angles, significantly improving the registration results of ULS and TLS. This technological advancement provides reliable support for applications such as 3D modeling, forestry management, and topographic mapping.

[0093] 3. Enhanced applicability in complex environments

[0094] The algorithm of the present invention improves the noise suppression capability and registration efficiency through filtering and optimized point cloud processing methods, so that the system still has strong robustness in complex terrain and harsh environments, can accurately adapt to different application scenarios and quickly complete data processing tasks.

[0095] Fifth, the technical solution of the present invention effectively solves the following key problems existing in the prior art in industrial applications, achieving significant technological progress:

[0096] 1. Technical Problems of Existing Technologies

[0097] 1. Point cloud registration accuracy is not high

[0098] Traditional point cloud registration technology usually relies on simple feature point matching or ICP algorithms when processing ULS (UAV laser scanning) and TLS (terrestrial laser scanning) data. However, it is difficult to achieve high-precision registration in environments with high noise or poor data quality. In particular, the accuracy will drop significantly in scenes with complex terrain or unclear target features.

[0099] 2. Low registration efficiency, difficult to meet real-time needs

[0100] Conventional registration methods are computationally intensive and, especially when dealing with large amounts of data, are slow, making them difficult to meet real-time requirements in practical applications. Applications like drone inspections demand high real-time data processing, so the speed bottleneck of existing technologies limits their application.

[0101] 3. Registration is easily affected by changes in scanning angle

[0102] There are inconsistencies in feature matching between point cloud data at different angles. Conventional feature point matching algorithms perform poorly when dealing with different viewing angles and angles, which can easily lead to a decrease in registration accuracy. Especially when the scanning angles of ULS and TLS differ greatly, the registration effect is difficult to guarantee.

[0103] 4. Inability to adequately handle noisy and unstructured data

[0104] Point cloud data is often noisy, especially in outdoor environments, due to interference from lighting, vegetation, and the scanning device. Traditional methods struggle to adequately filter out this noise during the registration process, making the registration results susceptible to interference, impacting accuracy and stability.

[0105] 2. Significant technological advancement of the present invention

[0106] 1. Improved the automation and accuracy of point cloud registration

[0107] This paper significantly improves the initial matching accuracy and automation of ULS-TLS point cloud data by introducing multiple geometric features (such as planar dimensions, curved surface dimensions, and the angle between the normal vector and the Z axis) and ground context. The initial correspondences established using these features reduce the need for human intervention, enabling the system to automatically and accurately identify and match key points, achieving highly accurate registration results.

[0108] 2. Significantly improved registration speed to meet real-time requirements

[0109] Through an innovative similarity scoring model and rotation transformation method, this paper achieves an efficient registration process. In particular, the combination of similarity scoring and ground context enables rapid identification and matching of point cloud data, reducing complex calculations and enabling the algorithm to meet real-time requirements in practical applications.

[0110] 3. Enhanced adaptability to different scanning angles and noise data

[0111] This paper designs a rotational matching model within the ground context, enabling the system to handle point cloud data from different scanning angles and being less susceptible to differences in viewing angles. Furthermore, through morphological filtering and matching methods based on local features, the system can maintain high registration accuracy even in noisy environments, adapting to complex scanning scenarios.

[0112] 4. Significantly improved data fusion processing effect and expanded the scope of application

[0113] This advancement in ULS-TLS registration technology addresses data fusion challenges in areas such as drone inspections, geographic mapping, and architectural modeling, laying the foundation for efficient data integration and subsequent data analysis. The algorithm can be widely applied in forest mapping, terrain modeling, and smart city management, significantly improving data availability and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0114] Figure 1 This is a flow chart of a forest ULS-TLS point cloud automatic registration method based on ground context description provided by an embodiment of the present invention.

[0115] Figure 2 This is a structural block diagram of a forest ULS-TLS point cloud automatic registration system based on ground context description provided by an embodiment of the present invention.

[0116] Figure 3 This is a ground context description construction diagram provided by an embodiment of the present invention.

[0117] Figure 4 This is a schematic diagram of local point cloud extraction provided by an embodiment of the present invention. (a, b, c) represent a local point cloud from a TLS scan, and (d, e, f) represent a local point cloud from a key point in a ULS scan. (a, d) represent the top view, (b, e) the front view, and (c, f) the left view.

[0118] Figure 5 Schematic diagram of the registration result provided by an embodiment of the present invention, (a) the sample plot after registration, (b) the cross-section, and (c) a partially enlarged view of the cross-section. DETAILED DESCRIPTION

[0119] In order to make the purpose, technical solutions 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 intended to limit the present invention.

[0120] like Figure 1As shown, an embodiment of the present invention provides a forest ULS-TLS point cloud automatic registration method based on ground context description, which includes the following steps:

[0121] S101: data preprocessing;

[0122] S102: Initial correspondence relationship is established;

[0123] Using ground points to implement ULS-TLS registration, the ground can generally be approximated as consisting of planes and curved surfaces; the dimensional characteristics of the points are basically unaffected by the point cloud pose;

[0124] Three features are introduced to characterize the pre-processed point cloud: plane dimension, surface dimension, and the angle between the normal vector and the Z axis. These three geometric features are used to establish a preliminary correspondence between ULS key points and TLS scan positions.

[0125] S103: Point cloud registration based on ground context;

[0126] TLS collects surrounding environment data by rotating the scan around its scanning position; based on the LiDAR scanning method, through ground context matching, it identifies the key points in the ULS that are most similar to the TLS scanning position, and then estimates the transformation matrix for alignment.

[0127] The proposed method for automatic forest ULS-TLS point cloud registration based on ground context descriptions achieves precise registration of point cloud data from unmanned laser scanning (ULS) and terrestrial laser scanning (TLS) through steps such as data preprocessing, establishing initial correspondences, and context-based point cloud registration. It is particularly suitable for complex forest environments. The detailed working principle is as follows:

[0128] First, in the data preprocessing stage (S101), the system processes the raw point cloud data to remove noise and unify the data format, providing high-quality input for subsequent registration. Because forest point cloud data contains trees, ground, and other debris, a filtering algorithm is used in the preprocessing stage to extract ground points from the entire point cloud data to improve registration accuracy. This step ensures that subsequent steps are performed only on meaningful ground points, improving registration efficiency.

[0129] During the initial correspondence establishment phase (S102), the system leverages the stable features of ground points to achieve preliminary alignment of the ULS-TLS point cloud. Forest floors can typically be approximated as planes or curved surfaces, and these dimensional features are unaffected by pose in ULS and TLS point cloud data. Therefore, the system introduces three features: plane dimension, surface dimension, and the angle between the normal vector and the Z axis. The stability of these geometric features enables the establishment of preliminary correspondences between ULS and TLS data, forming preliminary point pairs even in complex forest environments, thus providing a reliable foundation for subsequent precise alignment.

[0130] During the ground context-based point cloud registration phase (S103), the system uses the scanning characteristics of the TLS to further optimize the registration process. The TLS collects 3D data of the surrounding environment by rotating the scan around a fixed position, while the ULS data comes from laser scanning by drones, which has a larger range but less detail. To achieve registration between the two, the system constructs a ground context descriptor to match the data context of the TLS scan position with the key point data of the ULS. This context descriptor can identify the key points that best match the TLS position and the ULS data, making the registration process more targeted.

[0131] After ground context matching, the system further estimates the transformation matrix between the ULS and TLS. This matrix, which contains translation and rotation parameters, accurately reflects the relative position and pose of the ULS and TLS point cloud data in 3D space. By estimating the transformation matrix, the system can effectively perform a preliminary spatial alignment of the ULS and TLS data, representing both point clouds in the same coordinate system.

[0132] The system then uses an iterative optimization algorithm to refine the registration results. By gradually adjusting the transformation matrix, the error between the ULS and TLS point clouds is reduced, thereby improving registration accuracy. After each iteration, the system re-evaluates the point-pair relationships between the ULS and TLS and adjusts the matching relationship based on the error results. This step ensures the accuracy of the registration results, ultimately achieving highly accurate point cloud alignment.

[0133] Finally, the proposed registration method has been repeatedly validated and is particularly suitable for ULS-TLS point cloud registration in forest environments. In complex forest scenes, thanks to the inclusion of ground context, this method can quickly and accurately identify the correspondence between ULS and TLS, significantly improving the automation and accuracy of point cloud registration, providing effective data support for fields such as forest monitoring and ecological research.

[0134] Data preprocessing provided by the embodiment of the present invention:

[0135] (1) Ground point extraction;

[0136] ULS and TLS ground point cloud data are extracted through morphological filtering; all points in the point cloud are converted to a two-dimensional grid space; the lowest point in each pixel is selected as the seed point for clustering to obtain ground points;

[0137] (2) Local point cloud acquisition;

[0138] The ULS ground point cloud is divided into two-dimensional grids based on the range of x- and y-coordinates, with a step size of 0.5 meters. The center points of these grids are extracted as key points. The z-coordinates of the key points are set to 0.0 because they are irrelevant to the registration algorithm proposed in this invention.

[0139] The points in the TLS ground point cloud that are within a certain threshold of the horizontal neighborhood of the scanning position are filtered to generate a TLS local point cloud, and the points in the ULS ground point cloud that are within a certain threshold of the horizontal neighborhood of each key point are extracted to generate multiple ULS local point clouds for subsequent alignment; the threshold is usually set to 5m.

[0140] The ground point extraction provided by the embodiment of the present invention:

[0141] ULS and TLS ground point cloud data are extracted by morphological filtering; all points in the point cloud are converted into a two-dimensional grid space; the lowest point in each pixel is selected as the seed point for clustering to obtain ground points.

[0142] The local point cloud acquisition provided by the embodiment of the present invention:

[0143] The ULS ground point cloud is divided into two-dimensional grids based on the range of x- and y-coordinates, with a step size of 0.5 meters. The center points of these grids are extracted as key points. The z-coordinates of the key points are set to 0.0 because they are irrelevant to the registration algorithm proposed in this invention.

[0144] The points in the TLS ground point cloud that are within a certain threshold of the horizontal neighborhood of the scanning position are filtered to generate a TLS local point cloud, and the points in the ULS ground point cloud that are within a certain threshold of the horizontal neighborhood of each key point are extracted to generate multiple ULS local point clouds for subsequent alignment; the threshold is usually set to 5m.

[0145] The initial correspondence relationship provided by the embodiment of the present invention is established:

[0146] For a given point, principal component analysis (PCA) is applied to analyze the local point cloud of its neighborhood, and three eigenvalues ​​λ1, λ2 and λ3 (satisfying λ1≥λ2≥λ3≥0) and the normal vector can be obtained. Based on these results, the plane feature a2, the surface feature a3 and the angle f between the normal vector and the Z axis of this point can be derived;

[0147]

[0148] Given a TLS scan position p t and its related geometric features and f t , a key point p of ULS u and its related geometric features and f u ; If the following conditions are met, p u Considered as p t The initial corresponding points are added to the initial corresponding set C init middle:

[0149]

[0150] Among them, δ, θ and ε are three different thresholds, set to 0.1, 0.1 and 2.0° respectively;

[0151] By using the above method, the geometric features of all ULS key points and the local point cloud of the TLS scanning position are calculated and compared to generate the initial corresponding set C of the TLS scanning position. init ={p u |u=1,2…n init}.

[0152] The ground context-based point cloud registration provided by the embodiment of the present invention:

[0153] Similarity score sim(H ULS ,H TLS ) is calculated as follows:

[0154]

[0155] In the similarity score calculation, the matrix element M ij The median value of NaN is excluded from the calculation; the variable l represents M ij The number of non-NaN values ​​in the matrix; For the rotation transformation of the TLS point cloud, it can be achieved by performing corresponding transformation operations on the columns of the matrix:

[0156] r m :[z1,z2,…z Na ]→[z Na-m+1 ,…z Na ,z1,…,z Na-m ]#(4)

[0157] Among them, H TLS =[z1,z2,…z Na ], zj represents H TLSA column of , and m∈[1,Na) is the rotation parameter, and the corresponding rotation angle is θ m =2πm / Na;

[0158] If a pair of ground contexts of ULS and TLS are correctly aligned, then the similarity score function represents the standard deviation of the height difference between the corresponding point pairs; since the Z axes of ULS and TLS are perpendicular, ideally, the similarity score should be close to zero;

[0159] After obtaining the initial correspondence, a ground context Ht is generated for the TLS scan position; at the same time, the initial correspondence set C init The ULS key points in the ground context Hu = {H u i |i=1,2…n init}; To complete the registration, it is necessary to compare each H u i With Ht at a rotation angle θ m The similarity score under the formula is score=sim(H u i ,r m (H t )); The matching item with the smallest score will be selected for registration;

[0160] Using the rotation angle θ in the best match best The horizontal coordinates (xu, yu) of the TLS scanning position and its corresponding point in the ULS can be used to complete the planar registration. At the same time, according to formula (3), when calculating the similarity score of the best match, the average height difference Zmean between the corresponding points can also be calculated to determine the Z-axis translation transformation in the vertical registration.

[0161] like Figure 2 As shown, an embodiment of the present invention provides a forest ULS-TLS point cloud automatic registration system based on ground context description, including:

[0162] Preprocessing module, used for data preprocessing;

[0163] A module is built to establish initial correspondences. ULS-TLS registration is performed using ground points, which can generally be approximated as consisting of planes and curved surfaces. The dimensional characteristics of the points are largely unaffected by the point cloud pose. Three features are introduced to characterize the preprocessed point cloud: plane dimension, surface dimension, and the angle between the normal vector and the Z axis. These three geometric features are used to establish preliminary correspondences between ULS keypoints and TLS scan positions.

[0164] The point cloud registration module is used for ground context-based point cloud registration. The TLS collects surrounding environment data by rotating the scan around its scanning position. Based on the LiDAR scanning method, the key points in the ULS that are most similar to the TLS scanning position are identified through ground context matching, and the transformation matrix is ​​estimated for registration.

[0165] 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 performs the steps of the forest ULS-TLS point cloud automatic registration method based on ground context description.

[0166] 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 forest ULS-TLS point cloud automatic registration method based on ground context description.

[0167] Another object of the present invention is to provide an information data processing terminal, which is used to implement the forest ULS-TLS point cloud automatic registration system based on ground context description.

[0168] The mathematical model of the present invention solves the following existing technical problems:

[0169] 1. Low accuracy of ULS-TLS point cloud registration

[0170] In the existing ULS (Unmanned Aerial Vehicle Laser Scanning) and TLS (Terrestrial Laser Scanning) point cloud data registration methods, due to the differences in resolution, viewing angle and point cloud density of the data sources, the registration process is prone to feature loss and matching errors.

[0171] 2. Registration failure in complex terrain environments

[0172] The complex terrain in forest environments makes it difficult for traditional methods to accurately extract ground points and further use ground features for registration. Especially when the ground height varies greatly, the registration algorithm is prone to failure.

[0173] 3. Low efficiency of initial corresponding point calculation

[0174] Traditional initial corresponding point generation algorithms based on geometric features require a lot of calculations in high-dimensional point clouds, resulting in low efficiency and making it difficult to meet the real-time processing requirements of large-scale point clouds in forest scenes.

[0175] 4. Problems with inconsistent multi-scale context

[0176] Existing methods find it difficult to construct a unified contextual descriptor to combine the different resolutions and point cloud characteristics of ULS and TLS, resulting in large rotation angle estimation errors in the initial registration stage.

[0177] Significant technological advancements:

[0178] 1. Mathematical model for ground point extraction improves accuracy

[0179] Morphological filtering and rasterization: Morphological filtering is used to convert the point cloud into a two-dimensional grid space. The lowest point of each pixel is selected as the seed point and clustered to ensure accurate ground point extraction. This method solves the problem of accurate ground point extraction in complex forest environments.

[0180] Local point cloud generation model: Generate local point clouds of TLS and ULS through two-dimensional grid division and neighborhood radius setting, unify the representation of key points, and reduce the error between different point cloud resolutions.

[0181] 2. Initial correspondence relationship modeling improves efficiency and robustness

[0182] The geometric feature extraction formula (1) based on principal component analysis (PCA) effectively obtains the plane features (a2), surface features (a3), and normal vector angle (f) of the point. These features remain unchanged under posture changes, significantly improving the reliability of feature matching.

[0183] Conditional formula (2) reduces the computational complexity by constraining the feature difference threshold and quickly generates the initial corresponding point set C.

[0184] 3. Innovative design of highly contextual descriptors improves matching robustness

[0185] The height context descriptor is introduced. Based on LiDAR scanning data, the similarity score of ULS and TLS point clouds is calculated using the matrix difference and standard deviation formula (3), which overcomes the complexity of height changes in forest environments.

[0186] Through the rotation transformation operation of formula (4), the similarity scores under different rotation angles are quickly calculated, the best matching rotation angle is selected, and the joint registration of rotation and horizontal coordinates is achieved.

[0187] 4. Multi-dimensional feature fusion improves overall registration accuracy

[0188] The present invention combines the optimal rotation angle (θ m ) and mean height difference (Z), completing both planar and vertical registration, thus achieving full consistency between ULS and TLS point clouds.

[0189] 5. Unification of Context Description and Multi-resolution Matching

[0190] The height context descriptor is used to uniformly represent the ground context features of ULS and TLS point clouds, overcoming the differences in data resolution and sampling density between the two.

[0191] In forest scenes, the algorithm can maintain high robustness even with complex terrain.

[0192] 6. Efficient computing models adapt to large-scale data processing

[0193] Through the design of matrix operations and contextual similarity scoring, the present invention can quickly process ULS and TLS point cloud data in large-scale forest scenes while significantly reducing the calculation time.

[0194] The innovation of this invention in mathematical model directly solves the common problems of insufficient accuracy, efficiency and adaptability in ULS-TLS point cloud registration in the existing technology. The main technical advances are reflected in the following aspects:

[0195] 1. Improved accuracy: By combining principal component analysis with contextual descriptors, the accuracy of point cloud registration is ensured, and stable matching results can be obtained even in complex forest environments.

[0196] 2. Efficiency improvement: Matrix processing and threshold constraints optimize the feature extraction and initial corresponding point generation process, significantly reducing the registration time.

[0197] 3. Enhanced adaptability: The introduction of highly contextual descriptors enables the algorithm to adapt to the matching requirements of point clouds of various resolutions, and it exhibits extremely high robustness, especially in forest environments.

[0198] Example 1: Precision Measurement and Ecological Assessment of Forest Resources

[0199] Forest resource surveys require accurate 3D data to estimate tree growth, terrain changes, and carbon storage. The limitations of traditional single data sources (such as TLS or ULS) limit measurement accuracy, but ULS-TLS point cloud fusion technology can address this shortcoming.

[0200] 1. Data Collection

[0201] Unmanned aerial vehicle laser scanning (ULS) was used to acquire high-altitude point clouds of large forest areas, covering topography and canopy structure.

[0202] Terrestrial laser scanning (TLS) stations were deployed in the selected plots to collect high-density ground point cloud data, especially the tree trunk and ground morphology.

[0203] 2. Data Processing

[0204] By the point cloud automatic registration method of the present invention:

[0205] Ground Point Extraction: Extract ground points from ULS and TLS point clouds using morphological filtering.

[0206] Highly contextual descriptor registration: Data fusion is achieved by matching highly contextual descriptors of ULS-TLS point clouds.

[0207] Multi-dimensional feature fusion: Calculate indicators such as trunk diameter, tree height, and forest topography changes through registered point cloud data.

[0208] 3. Results Analysis

[0209] Obtain a three-dimensional model of the entire forest resource to accurately assess the growth status, tree species distribution and carbon storage of forest vegetation.

[0210] Use point cloud data in forest management systems to provide a basis for long-term monitoring of ecological changes.

[0211] Application Value

[0212] - Accurately assess forest resources and provide a scientific basis for forestry decision-making.

[0213] - Improve the accuracy of forest carbon sink measurements and support the formulation of carbon emissions trading and ecological protection policies.

[0214] Example 2: Forest Disaster Assessment and Restoration Monitoring

[0215] After a forest fire, landslide, or insect infestation, rapid assessment of damage and subsequent vegetation recovery is necessary. ULS-TLS point cloud fusion technology can quickly generate high-precision 3D models to support post-disaster recovery planning.

[0216] 1. Data Collection

[0217] Deploy drone laser scanning in the disaster-stricken areas to quickly cover a large area of ​​disaster-affected point cloud data, focusing on recording the overall morphology of tree canopies and terrain.

[0218] TLS was used to conduct high-precision scans in the disaster-stricken areas, capturing details of fallen trees, exposed ground, and tree trunks.

[0219] 2. Data Processing

[0220] Ground point extraction: Morphological filtering is used to extract ground points in the disaster area and analyze the scope of landslides or burned surfaces.

[0221] Initial corresponding point establishment and registration: The method of the present invention is used to realize the high context descriptor registration of the ULS-TLS point cloud and generate a unified 3D model.

[0222] Recovery monitoring: multi-period point cloud data comparison and analysis (e.g. ULS and TLS data over a year), to evaluate the natural recovery speed and trend of forest vegetation.

[0223] 3. Result output

[0224] Output high-precision three-dimensional disaster model, including information such as the distribution of fallen trees and the extent of damaged ground.

[0225] Provide recovery monitoring report, analyze vegetation recovery rate and spatial distribution.

[0226] Application value

[0227] Quickly assess disaster damage range and provide support for post-disaster emergency management.

[0228] Track recovery progress, optimize vegetation recovery plan, and improve ecological restoration efficiency.

[0229] Summary

[0230] The present invention has important industrial application value in the fields of forest resource measurement and ecological assessment, disaster assessment and recovery monitoring, and can significantly improve the accuracy and processing efficiency of data acquisition, providing advanced technical support for forestry and ecological protection.

[0231] Specific implementation of the present invention:

[0232] Step 1: Data preprocessing

[0233] (1) Ground point extraction

[0234] Extract ULS and TLS ground point cloud data through morphological filtering. All points in the point cloud are converted to a two-dimensional grid space. The lowest point in each pixel is selected as the seed point for clustering to obtain the ground points.

[0235] (2) Local point cloud acquisition

[0236] In order to determine the potential corresponding position of the TLS scanning position within the ULS, a certain step size is set as the grid size, which is set to 0.5 meters. According to the range of x and y coordinates, the ULS ground point cloud is divided into a two-dimensional grid, and the center points of these grids are extracted as key points. The z coordinate of the key points is set to 0.0, as they are irrelevant to the registration algorithm of the present invention.

[0237] Filter the points in the TLS ground point cloud that are within a certain threshold horizontal neighborhood of the scanning position to generate a TLS local point cloud, and extract the points in the ULS ground point cloud that are within a certain threshold horizontal neighborhood of each key point to generate multiple ULS local point clouds, which are used for subsequent registration. The threshold is usually set to 5m, and users can set the threshold size according to different forest scenarios.

[0238] Step 2: Initial correspondence establishment

[0239] The proposed method uses ground points to achieve ULS-TLS registration. The ground can generally be approximated as consisting of planes and curved surfaces. The dimensional characteristics of the points are basically unaffected by the point cloud pose. In addition, since the Z axis of the TLS and ULS coordinate systems in this study is vertically downward, the angle between the normal vector of the corresponding point in the TLS and ULS point clouds and the Z axis is consistent. Therefore, this paper introduces three features to characterize the preprocessed point cloud: plane dimension, surface dimension, and the angle between the normal vector and the Z axis. To reduce the amount of computation, these three geometric features are used to establish a preliminary correspondence between the ULS key points and the TLS scan positions.

[0240] For a given point, principal component analysis (PCA) is applied to analyze the local point cloud of its neighborhood, and three eigenvalues ​​λ1, λ2 and λ3 (satisfying λ1≥λ2≥λ3≥0) and the normal vector can be obtained. Based on these results, the plane feature a2, the surface feature a3, and the angle f between the normal vector and the Z axis of the point can be deduced.

[0241]

[0242] Given a TLS scan position p t and its related geometric features and f t , a key point p of ULS u and its related geometric features and f u If the following conditions are met, p u Considered as p t The initial corresponding points are added to the initial corresponding set C init middle:

[0243]

[0244] Among them, δ, θ and ε are three different thresholds, set to 0.1, 0.1 and 2.0° respectively.

[0245] By using the above method, the geometric features of all ULS key points and the local point cloud of the TLS scanning position are calculated and compared to generate the initial corresponding set C of the TLS scanning position. init ={p u |u=1,2…n init}.

[0246] Step 3: Point cloud registration based on ground context

[0247] The TLS collects surrounding data by rotating the scan around its scanning position. Based on the LiDAR scanning method, this paper proposes ground context. By matching the ground context, key points in the ULS that are most similar to the TLS scanning position are identified, and the transformation matrix is ​​estimated for registration.

[0248] The cylindrical neighborhood around a point is defined as a cylinder radius r, which is uniformly discretized in radial distance and azimuth. Specifically, the radial direction is divided into Nr=ceil(r / dr) radial segments, and the azimuth direction is divided into Na angular segments. Among them, the cylinder radius r is set to 5.0 meters, the radial resolution dr is set to 0.5 meters, and the number of azimuth segments Na is set to 360. The discretization space can be represented as an Nr×Na matrix H defined as the ground context. Each element in the matrix corresponds to a segment. If the segment contains at least one point, it is assigned the average height of the points in the segment; if it is empty, it is assigned NaN (not a number). The generation process of the ground context is as follows Figure 3 shown.

[0249] In order to measure the similarity between two ground contexts, the present invention proposes a similarity scoring function. Similarity score sim(H ULS ,H TLS ) is calculated as follows:

[0250]

[0251] In the similarity score calculation, the matrix element M ij The median value of NaN is excluded from the calculation. ij The number of non-NaN values ​​in the . For the rotation transformation of the TLS point cloud, it can be achieved by performing the corresponding transformation operation on the columns of the matrix:

[0252] r m :[z1,z2,…z Na ]→[z Na-m+1 ,…z Na ,z1,…,z Na-m ]#(4)

[0253] Among them, H TLS =[z1,z2,…z Na ], zj represents H TLS A column of , and m∈[1,Na) is the rotation parameter, and the corresponding rotation angle is θ m =2πm / Na.

[0254] If a pair of ground contexts of ULS and TLS are correctly aligned, the similarity score function represents the standard deviation of the height difference between the corresponding point pairs. Since the Z axes of ULS and TLS are both perpendicular, ideally, the similarity score should be close to zero.

[0255] After obtaining the initial correspondence, a ground context Ht is generated for the TLS scan position. At the same time, the initial correspondence set C init The ULS key points in the ground context Hu = {H u i |i=1,2…n init}. To complete the registration, it is necessary to compare each H u i With Ht at a rotation angle θ m The similarity score under the formula is score=sim(H u i ,r m (H t )). The matches with the lowest scores will be selected for registration.

[0256] Using the rotation angle θ in the best match best The horizontal coordinates (xu, yu) of the TLS scan position and its corresponding point in the ULS can be used to complete the planar registration. At the same time, according to formula (3), when calculating the similarity score of the best match, the average height difference Zmean between the corresponding points can also be calculated to determine the Z-axis translation transformation in the vertical registration.

[0257] In the preprocessing stage, in order to reduce the computational burden, this paper sets the grid size to 0.5 meters. This results in a decrease in registration accuracy, so further fine matching is required. best Then, select the value at intervals of 0.05 meters within 1 meter around xu and yu, and at the same time best Values ​​are selected at intervals of 0.5° within a 5° radius. These refined values ​​are then freely combined and registered using the ground context. The similarity score is calculated for all combinations, and the one with the smallest score is selected as the final result of the coarse registration.

[0258] After coarse registration, fine registration is required to improve accuracy. Given that the ground surface is typically composed of multiple planar or curved surfaces, a point-to-plane iterative closest point (ICP) algorithm is used for fine registration. To ensure high overlap between datasets and avoid local optima during the registration process, two local point clouds, representing the TLS scan position and its corresponding point in the ULS ground point cloud, are used as input to the ICP algorithm.

[0259] Technical solution of the embodiment of the present invention

[0260] Taking a 60m×20m forest plot as an example, the technical solution of the forest ULS-TLS point cloud automatic registration method based on ground context description is as follows.

[0261] (1) Sample site TLS data collection

[0262] Within the forest plot, operators collected data using a terrestrial laser scanner at 11 scanning locations to ensure complete forest coverage. Separately, on the same day, after the TLS data collection was completed, operators collected data from the plot using a drone laser scanner, flying at an altitude of approximately 50 meters and scanning at a frequency of 1800 kHz. The ULS data provide greater coverage, and the TLS data are a subset of the ULS data.

[0263] (2) Data preprocessing

[0264] For the TLS data from different scanning locations at the site, morphological filtering was first used to extract ground points, and then local point clouds were extracted for each scanning location. For the ULS data from the site, morphological filtering was also used to extract ground points, followed by a 2D gridding method to extract key points, and finally a local point cloud was generated for each key point. Figure 4 The extraction results of the local point cloud of a certain TLS scanning position in the sample site and the local point cloud of a key point in the ULS point cloud are displayed.

[0265] (3) Point cloud registration

[0266] After extracting the TLS local point cloud and the local point cloud of the ULS key points, the geometric features are first used to establish preliminary correspondences. Then, the ground context is used to filter out the correct correspondences to complete the coarse registration. Finally, the point-to-surface ICP algorithm is used to optimize the coarse registration results.

[0267] 2. Technical Effects of the Embodiments of the Present Invention

[0268] The technical results are demonstrated using a forest plot as an example. Eleven scanned TLS point clouds of the plot are registered using the forest ULS-TLS point cloud automatic registration method based on ground context description. Scans S1-11 are all registered to the ULS point cloud.

[0269] The visual registration effect is as follows Figure 5 As shown in the figure, the different colored point clouds represent different TLS point clouds and ULS point clouds from the site. All TLS scans were successfully registered to the ULS point cloud, with sufficient registration accuracy to describe the structural features of the branches and trunks. The error statistics of the registration results are shown in Table 1. The average translation error and average rotation error of the registration results for this site are 7.89 cm and 1.01 mrad, respectively.

[0270] Table 1. Accuracy evaluation of registration results

[0271]

[0272]

[0273] The present invention is applicable to fields such as forest ecological monitoring, forestry management, and environmental protection, and is particularly applicable to aspects such as forest structure analysis, tree species identification, biomass estimation, and carbon stock monitoring, which require precise three-dimensional information. By utilizing a method combining terrestrial laser scanning (TLS) and unmanned aerial vehicle laser scanning (ULS), the present invention can obtain high-precision tree structure data in forest plots, thereby improving the comprehensiveness and accuracy of forest structure information. In addition, the technology can be applied to environmental monitoring systems, geographic information systems (GIS), forest resource survey equipment, and software tools related to ecology and remote sensing, as an important component of data registration and information fusion modules.

[0274] This paper proposes a ground context-based method for automatic registration of single-scan TLS and ULS data in a forest environment. First, by comparing geometric features, potential ULS points corresponding to TLS scan positions are identified and a preliminary correspondence is established. Subsequently, the ground context is constructed using the ground points of ULS and TLS for coarse registration, and then fine registration is achieved using the point-to-plane ICP algorithm. In the experiment, 11 TLS scans and ULS data of a forest plot were used for verification. Figure 5 As shown in Table 1, all TLS scans were successfully registered to the ULS point cloud, demonstrating excellent registration results and enabling more comprehensive and accurate information about the forest structure. As shown in Table 1, the average translation error and rotation error of the ULS-TLS registration results for this site were 7.89 cm and 1.01 mrad, respectively, validating the effectiveness of this method.

[0275] 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 portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, 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. The devices and 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, 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, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0276] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A forest ULS-TLS point cloud automatic registration method based on ground context description, characterized by: The following steps are involved: Step 1: Data preprocessing; Step 2: Initial correspondence establishment; Use ground points to implement ULS-TLS registration. The ground is approximated as consisting of planes and curved surfaces. The dimensional characteristics of the points are not affected by the point cloud pose. Three features are introduced to characterize the pre-processed point cloud: plane dimension, surface dimension, and the angle between the normal vector and the Z axis. These three geometric features are used to establish a preliminary correspondence between ULS key points and TLS scan positions. Step 3: Point cloud registration based on ground context; The TLS collects surrounding environment data by rotating the scan around its scanning position. Based on the LiDAR scanning method, the key points in the ULS that are most similar to the TLS scanning position are identified through ground context matching, and the transformation matrix is ​​estimated for registration. The point cloud registration based on ground context: Similarity score is calculated as follows: （3） A cylindrical neighborhood around a point is defined with a cylindrical radius of r, which is uniformly discretized in radial distance and azimuth. The radial direction is divided into Nr = ceil (r / dr) radial segments, and the azimuth direction is divided into Na angular segments, where dr is the radial resolution and Na is the number of azimuth segments. The discretization space is represented by an Nr × Na matrix H, which is defined as the ground context. Each element in the matrix corresponds to a segment. If the segment contains at least one point, it is assigned the average height of the points in the segment. If it is empty, it is assigned NaN. In the similarity score calculation, the matrix elements The median value of NaN is excluded from the calculation; the variable l represents The number of non-NaN values ​​in the matrix; For the rotation transformation of the TLS point cloud, it is achieved by performing corresponding transformation operations on the columns of the matrix: （4） in, , zj represents A column of is the rotation parameter, and the corresponding rotation angle is ; If a pair of ground contexts of ULS and TLS are correctly corresponding, then the similarity score function represents the standard deviation of the height difference between the corresponding point pairs; since the Z axes of ULS and TLS are both vertical, the similarity score is close to zero; After obtaining the initial correspondence, a ground context Ht is generated for the TLS scan position; at the same time, Generate a set of ground context from the ULS keypoints in ; To complete the registration, it is necessary to compare each With Ht in rotation angle The similarity score under , the formula is ;The matches with the smallest scores will be selected for registration; Using the rotation angle in the best match The horizontal coordinates (xu, yu) of the TLS scanning position and its corresponding point in the ULS can be used to complete the planar registration; at the same time, according to formula (3), when calculating the similarity score of the best match, the average height difference Zmean between the corresponding points is also calculated to determine the Z-axis translation transformation in the vertical registration.

2. The forest ULS-TLS point cloud automatic registration method based on ground context description according to claim 1, characterized in that: The data preprocessing: (1) Ground point extraction; ULS and TLS ground point cloud data are extracted through morphological filtering; all points in the point cloud are converted to a two-dimensional grid space; the lowest point in each pixel is selected as the seed point for clustering to obtain ground points; (2) Local point cloud acquisition; The ULS ground point cloud is divided into two-dimensional grids according to the range of x- and y-coordinates, with a step size of 0.5 meters. The center points of these grids are extracted as key points. The z-coordinates of the key points are set to 0.0 because they are not relevant to the registration algorithm. Points in the TLS ground point cloud that are within a certain threshold of the horizontal neighborhood of the scanning position are filtered to generate a TLS local point cloud, and points in the ULS ground point cloud that are within a certain threshold of the horizontal neighborhood of each key point are extracted to generate multiple ULS local point clouds for subsequent alignment; the threshold is set to 5m.

3. The method for automatic forest ULS-TLS point cloud registration based on ground context description according to claim 1, characterized in that: The initial correspondence relationship is established: For a given point, principal component analysis (PCA) is applied to analyze its neighborhood local point cloud to obtain three eigenvalues ​​λ1, λ2 and λ3, satisfying λ1≥λ2≥λ3≥0, and the normal vector Based on these results, the plane feature a2, the surface feature a3 and the angle f between the normal vector and the Z axis of the point are derived; （1） Given a TLS scan location and its related geometric features , and , a key point of ULS and its related geometric features , and ; If the following conditions are met, regarded as The initial corresponding points are added to the initial corresponding set C init middle: （2） in, , and There are three different thresholds, set at 0.1, 0.1, and 2.0°; By using the above method, the geometric features of all ULS key points and the local point cloud of the TLS scan position are calculated and compared to generate the initial corresponding set of TLS scan positions. .

4. A forest ULS-TLS point cloud automatic registration system based on ground context description, which implements the forest ULS-TLS point cloud automatic registration method based on ground context description as described in any one of claims 1 to 3, characterized in that: The forest ULS-TLS point cloud automatic registration system based on ground context description includes: Preprocessing module, used for data preprocessing; A module is built for establishing initial correspondences. ULS-TLS registration is performed using ground points, where the ground is approximated as consisting of planes and curved surfaces. The dimensional characteristics of the points are not affected by the point cloud pose. Three features are introduced to characterize the preprocessed point cloud: the plane dimension, the surface dimension, and the angle between the normal vector and the Z axis. These three geometric features are used to establish preliminary correspondences between ULS keypoints and TLS scan locations. The point cloud registration module is used for ground context-based point cloud registration. The TLS collects surrounding environment data by rotating the scan around its scanning position. Based on the LiDAR scanning method, the key points in the ULS that are most similar to the TLS scanning position are identified through ground context matching, and the transformation matrix is ​​estimated for registration.

5. 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 performs the steps of the forest ULS-TLS point cloud automatic registration method based on ground context description as described in any one of claims 1 to 3.

6. 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 forest ULS-TLS point cloud automatic registration method based on ground context description as described in any one of claims 1 to 3.

7. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the forest ULS-TLS point cloud automatic registration system based on ground context description as claimed in claim 4.

Images