A multi-period point cloud spatio-temporal correlation adaptive registration and dynamic updating method and device

Abstract

This invention belongs to the field of 3D laser scanning and point cloud data processing, and relates to a method and apparatus for adaptive registration and dynamic updating of multi-period point cloud spatiotemporal correlation. The method includes: performing time normalization and incremental encoding on multi-period point clouds to construct a spatiotemporal joint hash index; extracting multi-scale local geometric features of each point, and generating a change mask through inter-period residual analysis and density clustering; calculating the confidence of primitives by comprehensively considering land cover category, incident angle, temporal stability, and change mask penalty term, and globally normalizing and truncating; constructing a joint objective function containing adjacent period constraints, global closure error, and local smoothing regularization, and optimizing and solving the rigid body transformation of all periods in one go; performing local point cloud extraction and mesh reconstruction only in the change region, and embedding it into the global model; and adaptively adjusting hyperparameters until the accuracy and consistency requirements are met. It breaks through static assumptions, adapts to dynamic scenarios, eliminates cumulative drift, ensures spatiotemporal logic, reduces computational overhead, and constructs a closed-loop quality control.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional laser scanning and point cloud data processing technology, and in particular to a method and apparatus for adaptive registration and dynamic updating of multi-period point cloud spatiotemporal correlation. Background Technology

[0002] In existing technologies, multi-phase point cloud registration mainly relies on fixed feature points (such as target spheres and corner points) or the Iterative Closest Point (ICP) algorithm, which has the following drawbacks: it is difficult to cope with non-rigid scene changes such as vegetation growth, building demolition, and terrain erosion; it lacks consistency constraints in the time dimension, resulting in cumulative drift and spatiotemporal logic contradictions (such as buildings disappearing and then reappearing); and each update requires global reprocessing, which consumes huge computational resources. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method, employing the following technical solution, including the following steps:

[0004] Temporal normalization and incremental encoding are performed on multi-period point clouds, and a spatiotemporal joint hash index is constructed using an adaptive spatial grid partitioning strategy;

[0005] Calculate the local geometric feature tensor of each point at different scales, evaluate the stability of each point in the time dimension through intertemporal residual analysis, and extract the change region based on density clustering to generate a change mask;

[0006] Based on the geometric category of the ground feature, the scanning incident angle, the temporal stability, and the change mask penalty term, the comprehensive confidence of each registration primitive is calculated, and global normalization and dynamic truncation are performed.

[0007] A joint objective function is constructed, which includes relative transformation constraints between adjacent periods, global closure error constraints, and local smoothing regularization constraints in the change region. The rigid body transformation parameters for all periods are solved in one go through nonlinear least squares optimization.

[0008] Local point cloud extraction and triangulation reconstruction are performed only within the area marked by the changing mask. The local mesh is then embedded into the global model through boundary topology stitching, achieving local replacement rather than global repetition.

[0009] The root mean square error of registration is calculated using an independent validation set, and spatiotemporal logical inconsistencies are detected. Based on the evaluation results, the hyperparameters of the objective function are adaptively adjusted and optimized, triggering iterative re-registration until the accuracy and logical consistency requirements are met.

[0010] Preferably, the steps of performing temporal normalization and incremental encoding on multi-period point clouds, and constructing a spatiotemporal joint hash index using an adaptive spatial grid partitioning strategy, specifically include:

[0011] Using the initial data collection time as the baseline zero point, the original timestamps of each period are normalized using a preset standard time interval to generate a composite key value containing the period number and intraday time information for rapid filtering of time neighborhoods.

[0012] A dynamic octree recursive segmentation strategy is adopted, and a threshold for the number of point clouds in each leaf node is set. The spatial size of the leaf node is adaptively adjusted according to the local point density, so that the dense region is finely divided while the sparse region maintains a large grid.

[0013] A two-dimensional hash mapping is established using the spatial Morton code of the leaf nodes as the primary key and the time period sequence number as the secondary key. The point clouds of each period are organized and stored according to their respective spatial grids, and a cross-period doubly linked list is maintained to support constant-level neighborhood retrieval.

[0014] Preferably, the steps of calculating the local geometric feature tensor of each point at different scales, evaluating the stability of each point in the time dimension through intertemporal residual analysis, and extracting the change region based on density clustering to generate a change mask specifically include:

[0015] For each point, a neighborhood covariance matrix is ​​constructed with multiple different search radii. The eigenvalues ​​are solved and the planarity, linearity and randomness features are calculated. The geometric category of each point is determined by multi-scale voting.

[0016] Spatiotemporal indexing is used to coarsely align point clouds between two adjacent periods, and the three-dimensional residual vector between pairs of points with the same name is calculated. Combined with the system noise standard deviation and geometric category indicator function, the stability probability of each point in the time dimension is quantified.

[0017] Points with residuals exceeding a preset multiple of the system noise standard deviation are marked as change points. Spatial clustering algorithm is used to merge the change points, and morphological dilation processing is performed on the convex hull of each cluster to generate a binary change mask.

[0018] Preferably, the step of calculating the comprehensive confidence of each registration primitive based on the geometric category of the ground feature, the scanning incident angle, the temporal stability, and the change mask penalty term, and then performing global normalization and dynamic truncation specifically includes:

[0019] Different prior weights are assigned to each point according to its geometric category. At the same time, the incident angle between the laser beam and the local surface normal vector is calculated. Points with excessive incident angles or unfavorable geometric categories are weighted or eliminated.

[0020] Multiply the stability probability by the change mask penalty factor to make the overall stability factor of the points located inside the change mask zero, and the stable region maintains the original stability probability.

[0021] The product of the base confidence level and the stability weighting factor is normalized to the maximum value of all points in this period, and unreliable points with confidence levels below a preset threshold are removed, leaving only high-confidence primitives to participate in subsequent registration optimization.

[0022] Preferably, the step of constructing a joint objective function comprising relative transformation constraints between adjacent periods, global closure error constraints, and local smoothing regularization constraints in the change region, and solving for the rigid body transformation parameters of all periods in one go through nonlinear least squares optimization, specifically includes:

[0023] For each pair of adjacent periods, extract high-confidence matching point pairs, and minimize the sum of squared spatial distances between corresponding points after the transformation of adjacent periods, using the confidence of each point as the weight.

[0024] High-confidence corresponding feature points between the first and last periods are selected to construct a closed-loop error term. The corresponding points between the first and last periods are forced to coincide in the global coordinate system after their respective transformations in order to eliminate long-term cumulative drift.

[0025] Within the boundary zone of the changing mask, spatially adjacent point pairs are selected, and the change in local relative geometric relationship before and after the transformation is penalized to ensure a smooth transition at the edge of the changed region without tearing.

[0026] Preferably, the step of extracting local point clouds and reconstructing triangular meshes only within the area marked by the changing mask, and embedding the local meshes into the global model through boundary topology stitching to achieve local replacement rather than global repetition specifically includes:

[0027] Based on the change mask, the change point clusters are extracted from the current registered point cloud, transformed to the local coordinate system, and then a surface reconstruction algorithm is used to generate a local triangular mesh.

[0028] Extract the vertices of the boundary zone of the change region, use the boundary points of the new mesh as control points, and use radial basis function interpolation to smoothly transition the position of the vertices in the stitching area, so that the local mesh can be seamlessly embedded into the global model;

[0029] The reconstructed local mesh is subjected to triangle quality checks and optimizations, generating multiple versions at different levels of detail. The pointer index of the global model is updated, and dynamic model updates are completed through local replacement.

[0030] Preferably, the steps of calculating the root mean square error of registration using an independent validation set, detecting spatiotemporal logical inconsistencies, adaptively adjusting and optimizing the hyperparameters of the objective function based on the evaluation results, and triggering iterative re-registration until the accuracy and logical consistency requirements are met specifically include:

[0031] Extract the reserved verification point set that did not participate in the registration optimization in each period, project it to the global coordinate system through the transformation matrix, and calculate the root mean square error of the nearest distance to the reference point cloud as the unbiased estimation index of the registration accuracy.

[0032] A temporal logic test is performed on the change mask of multiple consecutive periods to identify contradictory regions where the previous and next periods are both changing but the intermediate period is unchanged, and the spatial range of the contradictory regions is statistically analyzed.

[0033] When the root mean square error exceeds the preset threshold or there is a logical contradiction, the global closed-loop constraint weight and / or smoothing regularization weight are increased by a preset step size, and the optimization solution is re-executed until the accuracy and consistency requirements are met or the preset iteration limit is reached.

[0034] To address the aforementioned technical problems, this invention also provides a multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update device, which employs the following technical solution, including:

[0035] The module is used to perform temporal normalization and incremental encoding on multi-period point clouds, and to construct a spatiotemporal joint hash index using an adaptive spatial grid partitioning strategy;

[0036] The identification module is used to calculate the local geometric feature tensor of each point at different scales, evaluate the stability of each point in the time dimension through intertemporal residual analysis, and extract the change region based on density clustering to generate a change mask.

[0037] The allocation module is used to calculate the overall confidence of each registration primitive based on the geometric category of the ground feature, the scanning incident angle, the temporal stability, and the change mask penalty term, and to perform global normalization and dynamic truncation.

[0038] The solver module is used to construct a joint objective function that includes relative transformation constraints between adjacent periods, global closure error constraints, and local smoothing regularization constraints in the change region. It solves the rigid body transformation parameters for all periods in one go through nonlinear least squares optimization.

[0039] The update module is used to extract local point clouds and reconstruct triangulation only within the area marked by the changing mask, and to embed the local mesh into the global model through boundary topology stitching, so as to achieve local replacement rather than global repetition.

[0040] The iterative module is used to calculate the root mean square error of registration using an independent validation set, detect spatiotemporal logical inconsistencies, adaptively adjust and optimize the hyperparameters of the objective function based on the evaluation results, and trigger iterative re-registration until the accuracy and logical consistency requirements are met.

[0041] To address the aforementioned technical problems, the present invention also provides a computer device that employs the technical solution described below, comprising a memory and a processor. The memory stores computer-readable instructions, and the processor, when executing the computer-readable instructions, implements the steps of the aforementioned multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method.

[0042] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium, which employs the technical solution described below. The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the aforementioned multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method.

[0043] Compared with the prior art, the present invention has the following main advantages:

[0044] (1) It can break through static assumptions and adapt to complex dynamic scenarios: By generating change masks through intertemporal residual analysis and density clustering, stable regions and change regions are explicitly distinguished. Local smoothing regularization constraints of change regions are introduced into the registration objective function to avoid misjudging change information as registration error, thus significantly improving robustness in complex scenarios.

[0045] (2) It can eliminate cumulative drift and ensure spatiotemporal logic consistency: By constructing a joint objective function that includes relative transformation between adjacent periods and global closure error constraints, all period transformation parameters are solved simultaneously at one time, fundamentally eliminating drift accumulation; at the same time, the validation set is used to detect spatiotemporal logic contradictions (such as buildings reappearing after disappearing), triggering iterative re-registration to ensure that the evolution sequence of ground features is reasonable.

[0046] (3) It can significantly reduce computation and storage overhead: By using spatiotemporal joint hash index to achieve fast association retrieval, and only performing local point cloud extraction and mesh reconstruction in the changing mask area, the global model is embedded through boundary topology stitching to achieve "on-demand update", avoiding repeated processing of stable areas, and reducing computational resource consumption by more than an order of magnitude.

[0047] (4) An adaptive closed-loop quality control was constructed: the registration RMSE was calculated based on the independent validation set, and the hyperparameters were dynamically adjusted and optimized to form a registration-evaluation-iteration closed loop, ensuring that the results not only meet the geometric accuracy threshold, but also pass the logical consistency test, which significantly improves the reliability and automation level of multi-period data fusion. Attached Figure Description

[0048] To more clearly illustrate the solutions in this invention, the accompanying drawings used in the description of the embodiments of this invention will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0049] Figure 1 This is a flowchart of an embodiment of the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method of the present invention;

[0050] Figure 2This is a schematic diagram of the structure of an embodiment of the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update device of the present invention;

[0051] Figure 3 This is a schematic diagram of the structure of an embodiment of the computer device of the present invention. Detailed Implementation

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the specification is for the purpose of describing particular embodiments only and is not intended to limit the invention; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings are used to distinguish different objects and not to describe a particular order.

[0053] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0054] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0055] It should be noted that the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method provided in the embodiments of the present invention is generally executed by a server / terminal device, and correspondingly, the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update device is generally set in the server / terminal device.

[0056] It should be understood that the number of terminal devices, networks, and servers is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be used.

[0057] Example 1

[0058] Please refer to Figure 1 The flowchart illustrates an embodiment of the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method of the present invention. The multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method includes the following steps:

[0059] Step S1: Perform time normalization and incremental encoding on the multi-period point cloud, and construct a spatiotemporal joint hash index using an adaptive spatial grid partitioning strategy.

[0060] In this embodiment, the electronic device (e.g., a server / terminal device) on which the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method runs can receive multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update requests via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra wideband) connections, and other currently known or future-developed wireless connection methods.

[0061] In this embodiment, step S1 may specifically include the following steps:

[0062] S11 uses the initial collection time as the baseline zero point and normalizes the original timestamps of each period using a preset standard time interval to generate a composite key value containing the period number and intraday time information for rapid filtering of time neighborhoods.

[0063] When collecting point cloud data for each period, the original timestamp of the collection time is recorded synchronously. (Unit: seconds, calculated from a certain epoch). Based on the time of the first data collection phase. As the global time base, the first time is calculated using the following formula. Normalized time values ​​of point clouds: .

[0064] in: — No. The normalized time value of the point cloud is a dimensionless real number. Its integer part represents the period number relative to the first period, and the decimal part represents the time of collection within the day.

[0065] — No. The original timestamp during point cloud acquisition, in seconds;

[0066] — The original timestamp during the initial point cloud collection, in seconds, serves as the global time zero point;

[0067] — Preset standard time interval (e.g., if the data is collected monthly, then...) (seconds), used to scale the time difference to an easily manageable order of magnitude.

[0068] Furthermore, regarding Perform binary incremental encoding to construct a composite primary key. ,in This represents the period number obtained by rounding down, used to quickly locate adjacent periods (e.g., the first period). Period and the The fractional part (of the key) is used to distinguish data blocks collected at different times within the same period. This encoding allows for the determination of temporal proximity in subsequent steps simply by comparing the key values, eliminating the need for repeated floating-point operations.

[0069] The purpose of step S11 is to unify the original time system of point clouds in each period (which may come from GPS, system clock or manual input) and convert it into dimensionless relative time sequence numbers, so as to facilitate the construction of time adjacency graph and quick filtering of adjacent periods.

[0070] S12 adopts a dynamic octree recursive segmentation strategy, sets a threshold for the number of point clouds in each leaf node, and adaptively adjusts the spatial size of the leaf nodes according to the local point density, so that the dense region is finely divided while the sparse region maintains a large grid.

[0071] A dynamic octree recursive segmentation strategy is adopted. An ideal threshold for the number of point clouds within each leaf node is set. (This value can be adjusted during implementation based on the overall point cloud density). From the root node of the entire scene (side length is...) Begin by evaluating each current node:

[0072] If the number of points contained in the node Then the node is evenly divided into 8 child nodes;

[0073] like If so, then merge the node with its sibling nodes back into the parent node;

[0074] Otherwise, keep the current node as a leaf node.

[0075] After recursive partitioning, the actual spatial side length of each leaf node It is no longer fixed, but dynamically determined by the local point density, as follows: .

[0076] in: — Adaptively adjusted side length of the leaf node space, in meters;

[0077] — The initial side length of the root node, in meters, is usually taken as 1.5 times the maximum span of the scene;

[0078] — Preset ideal point threshold, dimensionless integer (e.g., 2000).

[0079] — The estimated point cloud density of the local area where the leaf node is located, in points per cubic meter, is calculated by averaging the number of points of the node and its neighboring nodes;

[0080] — Root node volume, in cubic meters.

[0081] When local density At higher levels, Smaller size allows for finer partitioning; when density is low, The spatial resolution is relatively large, avoiding the generation of a large number of empty nodes. This achieves adaptive matching of spatial resolution to data density.

[0082] The purpose of step S12 is to overcome the defects of fixed-size grids in generating too many empty nodes in dense point cloud areas and losing details in sparse areas, so as to keep the number of point clouds in each grid cell relatively balanced, thereby improving index uniformity and retrieval efficiency.

[0083] S13. A two-dimensional hash mapping is established with the spatial Morton code of the leaf node as the primary key and the time period sequence number as the secondary key. The point clouds of each period are organized and stored according to their respective spatial grids, and a cross-period doubly linked list is maintained to support constant-level neighborhood retrieval.

[0084] Morton code is calculated for the spatial location of each leaf node. Morton code is derived by using three-dimensional coordinates. The binary bits are interleaved to generate a 64-bit integer, which can reduce the dimensionality of a three-dimensional space to one dimension while maintaining spatial proximity. Specifically, for the center coordinates of the leaf nodes... Normalize it to integer grid coordinates within the root node range. Then, the binary bits are interleaved to obtain the Morton code. .

[0085] Establish a spacetime hash mapping table: .

[0086] in: — The Morton code of the leaf node is a 64-bit unsigned integer, which serves as the spatial primary key;

[0087] — Issue number (i.e.) ), as a time subkey;

[0088] — Stored in the Period, Morton code is A dynamic array of the coordinates of all point clouds and their attributes (intensity, normal vector, etc.) within the spatial grid.

[0089] In addition, a doubly linked list spanning different periods is maintained for each spatial grid, linking the point cloud lists of the same grid across different periods. When inserting a new period's point cloud, it is only necessary to calculate the Morton code for each point. Through hash function Locate the bucket position and append the point to it. Simultaneously, the average distance between the current point and existing points within the bucket is calculated. If the average distance is less than the system resolution threshold... (Typically set to twice the scanner's nominal accuracy), then the point is marked as a potential stable candidate point for use in step S2. This hash table greatly reduces the time required for subsequent cross-period corresponding point searches.

[0090] The purpose of step S13 is to associate the three-dimensional spatial code with the time period and construct a two-dimensional lookup table, so that for any given spatial location and time period, it is possible to... It can directly locate the corresponding point cloud list within a constant time, providing a fast channel for cross-period neighborhood search.

[0091] The purpose of step S1 is to establish a unified temporal and spatial organization framework for massive unordered point clouds from different periods and different scanning sites, to enable rapid retrieval and neighborhood query of point clouds across periods, and to provide an efficient data access interface for all subsequent calculations.

[0092] Step S2: Calculate the local geometric feature tensor of each point at different scales, evaluate the stability of each point in the time dimension through intertemporal residual analysis, and extract the change region based on density clustering to generate a change mask.

[0093] In this embodiment, step S2 may specifically include the following steps:

[0094] S21: For each point, construct the neighborhood covariance matrix with multiple different search radii, solve for the eigenvalues ​​and calculate the planarity, linearity and randomness features, and determine the geometric category of each point through multi-scale voting.

[0095] For any point in the point cloud Define its three-dimensional coordinate vector as Choose three search radii. , , This covers local details, intermediate structure, and macroscopic form, respectively. For each radius... ( ), collect to Centered on All neighboring points inside a sphere with radius . ,in This represents the number of points within that radius. Calculate the mean vector of this neighborhood: .

[0096] Then construct the covariance matrix: .

[0097] in: — in radius Lower point The 3×3 covariance matrix reflects the degree of dispersion of neighborhood points in each direction;

[0098] — Within the neighborhood Three-dimensional coordinate vectors of points;

[0099] — The three-dimensional mean coordinate vector of all points in this neighborhood;

[0100] superscript — Represents the transpose of a vector or matrix.

[0101] right Perform eigenvalue decomposition to obtain eigenvalues. (All are real, non-negative values). Define the following three geometric characteristic quantities:

[0102] Planarity The closer this value is to 1, the more likely the point cloud distribution is on an ideal plane (such as a road or wall).

[0103] linear The closer this value is to 1, the more likely the point cloud distribution is an ideal straight line (such as a utility pole or tree trunk).

[0104] Scattered The closer this value is to 1, the more it indicates that the point cloud is spherically scattered (such as tree canopy or shrubbery).

[0105] The above three features satisfy For three scales After separate calculations, the principal geometric category of each point is determined using either majority voting or a weighted average method. The specific judgment rule is: if If the value is greater than 0.6, it is determined to be a road surface or facade (further distinguished as horizontal or vertical based on the direction of the normal vector); if If the maximum value is greater than 0.5, it is classified as a rod-shaped object; if If the value is greater than 0.5, it is classified as "vegetation".

[0106] The purpose of step S21 is to calculate the covariance matrix and its eigenvalues ​​of each point under different neighborhood radii, thereby quantifying the planarity, linearity and randomness of the point, and thus determining whether it belongs to an artificial rigid structure (such as a road or wall) or a non-rigid natural object (such as a tree canopy or shrub).

[0107] S22 uses spatiotemporal indexing to coarsely align point clouds between adjacent periods, calculates the three-dimensional residual vector between pairs of points with the same name, and combines the system noise standard deviation and geometric category indicator function to quantify the stability probability of each point in the time dimension.

[0108] Using the spatiotemporal hash index constructed in step S1, for two adjacent periods and In the same Morton code Within the corresponding spatial grid, a fast coarse alignment (containing only translation, no rotation) is first performed to eliminate the overall offset caused by different scan positions. Coarse translation amount It is obtained by calculating the difference between the centroids of the point clouds in the two-phase grid.

[0109] Then, for the first Each point in the period In its corresponding number Search for nearest neighbors in the same grid and adjacent grids of the same period. Calculate the residual vector: .

[0110] in: — No. The three-dimensional displacement residual vector of a point between two periods, in meters;

[0111] — No. Midterm Three-dimensional coordinate vectors of points;

[0112] — in the Midterm and The three-dimensional coordinate vector of the corresponding nearest neighbor.

[0113] Calculate the modulus of the residual Based on this, the stability probability of this point is defined as follows:

[0114] .

[0115] in: - point The stability probability takes a value between [0, 1], with the value closer to 1 indicating greater stability.

[0116] — The square of the residual modulus, i.e., the square of the displacement;

[0117] — System noise standard deviation, in meters, is given by the laser scanner's factory calibration document (e.g., for ground station scanners, a typical value is 0.003 m to 0.005 m).

[0118] — Natural exponential function;

[0119] — An indicator function that takes the value 1 when the condition inside the parentheses is true, and 0 otherwise;

[0120] - point The geometric category determined in step S21.

[0121] If the displacement of a point is much smaller than the system noise standard deviation (i.e.) If the exponent term is close to 1, it indicates that the point is highly stable in time. At the same time, only points belonging to artificial rigid structures such as road surfaces or facades are given a non-zero stability probability, because vegetation and poles may not have registration reference value even if the displacement is small due to growth or shaking.

[0122] The purpose of step S22 is to quantitatively assess the stability of each point in the time dimension by comparing the geometric differences between two adjacent point clouds near the same spatial location, thereby screening out rigid points that have not undergone displacement or deformation for many years.

[0123] S23, mark the points where the residual exceeds a preset multiple of the system noise standard deviation as change points, use a spatial clustering algorithm to merge the change points, perform morphological dilation processing on the convex hull of each cluster, and generate a binary change mask.

[0124] All satisfied The points marked as change points. Threshold This corresponds to a confidence level of approximately 99.7%, meaning that only points where the displacement exceeds three standard deviations of the system noise are considered to have actually undergone a physical change.

[0125] Spatial clustering was performed on all points of change using the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm. Two key parameters of DBSCAN were set as follows:

[0126] Neighborhood radius ,in The average side length of the leaf nodes in the adaptive partitioning in step S1;

[0127] Minimum points This indicates that a cluster must contain at least 10 points.

[0128] After clustering, several clusters of change points are obtained. For each cluster, its convex hull is calculated, and a binary morphological dilation operation is performed on the convex hull with a dilation radius of 0.5 meters to encompass the edge transition zone of the changing region. The final result is a binary mask function for the changing region.

[0129] .

[0130] in Let be an arbitrary position vector in three-dimensional space. This mask will be directly used for weight suppression in step S3: Any vector marked as In the region, the registration confidence of all points within it will be set to zero or significantly reduced.

[0131] The purpose of step S23 is to aggregate points with significant residuals into spatially continuous regions and generate a binary transformation mask, which is used in subsequent steps to forcibly reduce or mask the registration weight of this region and prevent non-rigid deformation from contaminating the registration parameter estimation.

[0132] The purpose of step S2 is to quantitatively extract the geometric type (road surface, facade, vegetation, etc.) of each point from the original point cloud, and through temporal comparison, identify rigid structures that remain unchanged over multiple periods (as high-quality registration primitives) and regions that undergo significant changes (which are suppressed or excluded in subsequent registration).

[0133] Step S3: Calculate the overall confidence level of each registration primitive based on the geometric category of the ground feature, the scanning incident angle, the temporal stability, and the change mask penalty term, and perform global normalization and dynamic truncation.

[0134] In this embodiment, step S3 may specifically include the following steps:

[0135] S31 assigns different prior weights to each point based on its geometric category, and calculates the incident angle between the laser beam and the local surface normal vector. Points with excessively large incident angles or unfavorable geometric categories are weighted or eliminated.

[0136] For each point First, calculate its local normal vector. (This can be obtained through the eigenvector corresponding to the smallest eigenvalue of the covariance matrix). Simultaneously, the scanner position corresponding to this point is obtained. (Provided by the device log file). Calculate the scan incident angle. : .

[0137] in: — The angle between the laser beam and the local surface normal vector, in radians, with a range of [0, 1]. ];

[0138] — The local normal vector of this point (3×1 column vector);

[0139] — The spatial position vector of the scanner at the moment of acquisition at that point;

[0140] — The vector of the line of sight from the scanner to that point;

[0141] — Vector dot product operation;

[0142] — The Euclidean magnitude of a vector.

[0143] Angle of incidence The larger the value, the more obliquely the laser beam hits the surface, resulting in a weaker echo signal and a greater ranging error. Based on engineering experience, a basic weight is defined. as follows:

[0144] .

[0145] Artificial planar structures contribute the most when scanned close to the front; rod-shaped objects are second; vegetation has a lower weight due to its unstable morphology; points scanned at the edge (with too large an incident angle) are directly eliminated.

[0146] The purpose of step S31 is to eliminate unreliable points caused by poor scanning geometry (such as edge scanning) and assign different prior confidence levels according to the land cover category.

[0147] S32, multiply the stability probability by the change mask penalty factor to make the comprehensive stability factor of the points located inside the change mask zero, and the stable region maintains the original stability probability.

[0148] Define the intermediate stability weighting factor: .

[0149] in: — A weighting factor that integrates stability and variation masking, dimensionless;

[0150] — The stability probability from step S22, with a value of [0,1];

[0151] — Change mask at point The value at that location is either 0 or 1.

[0152] If the point is located inside the changing mask ( Then, regardless of its own stability, The value is directly set to 0, thus completely eliminating the region of change; if the point is in a stable region and If it is higher, then Retain the larger value.

[0153] The purpose of step S32 is to introduce the stability probability and change mask obtained in step S2 into the weight system, so that the weight of the changed region is set to zero, while the weight of the stable region is maintained.

[0154] S33 normalizes the product of the base confidence level and the stability weighting factor to the maximum value of all points in this period, and removes unreliable points with confidence levels below the preset threshold, retaining only high-confidence primitives to participate in subsequent registration optimization.

[0155] Calculate the numerator of the overall confidence score: .

[0156] Perform global normalization for the first... All points in the point cloud: .

[0157] in: - point The final normalized confidence score takes values ​​of [0, 1].

[0158] — Unnormalized composite weights;

[0159] — No. The set of all points in the point cloud;

[0160] — No. Some points during the period The maximum value is used to scale the maximum confidence level to 1;

[0161] — A very small positive number (such as ( ), to prevent the denominator from being zero.

[0162] After normalization, for all satisfying Points with a confidence level below 5% are truncated and removed, meaning these points are not involved in subsequent registration optimization. The engineering significance of this truncation threshold of 0.05 is that it filters out unreliable points, typically reducing the number of points involved in registration by 40% to 60%, without losing effective information.

[0163] The purpose of step S33 is to integrate the above multi-source weights into a final confidence value, and to normalize it to balance the weight scale between different periods, while eliminating extremely low confidence points to reduce the amount of computation.

[0164] The purpose of step S3 is to integrate multi-source information such as geometric category, scan quality, temporal stability, and changing mask to calculate a continuous confidence value between 0 and 1 for each candidate point, so that the subsequent registration optimization process can automatically focus on the most reliable and stable rigid primitives, while discarding unreliable regions.

[0165] Step S4: Construct a joint objective function that includes relative transformation constraints between adjacent periods, global closure error constraints, and local smoothing regularization constraints in the change region. Solve the rigid body transformation parameters for all periods in one go using nonlinear least squares optimization.

[0166] In this embodiment, step S4 may specifically include the following steps:

[0167] S41. For each pair of adjacent periods, extract high-confidence matching point pairs, and minimize the sum of squared spatial distances between corresponding points after the adjacent periods are transformed, using the confidence level of each point as the weight.

[0168] Define the variable to be optimized For the first The rigid body transformation matrix from the point cloud to the global coordinate system is a 4×4 homogeneous transformation matrix, which can be written as: .

[0169] in A 3×3 rotation matrix (satisfying) , ), It is a 3×1 translation vector. It is a 3×1 zero vector.

[0170] For each pair of adjacent periods and ( Using the confidence level calculated in step S3, select the set of high-confidence matching point pairs. These matching point pairs are obtained as follows: For the first... Midterm confidence level point In step S1, search for the first... Nearest neighbor points within the same grid It requires nearest neighbors (i.e., bidirectional matching) to ensure matching reliability.

[0171] Define the adjacent period error term:

[0172] .

[0173] in: — A scalar value for the sum of errors between adjacent periods, in square meters;

[0174] — Total number of point cloud issues;

[0175] — No. Period and the The set of indexes of high-confidence matching pairs between periods;

[0176] — No. midpoint The confidence level (from step S3);

[0177] — No. The 4×4 homogeneous transformation matrix of the point cloud;

[0178] — No. Midterm Homogeneous coordinate vector of points (4×1);

[0179] — No. The 4×4 homogeneous transformation matrix of the point cloud;

[0180] — No. The homogeneous coordinate vectors of the corresponding matching points in the middle period;

[0181] — The square of the three-dimensional Euclidean distance (calculated as the difference between the first three components in homogeneous coordinates).

[0182] High confidence pairs are given greater weight, driving the optimization algorithm to prioritize aligning these reliable rigid structures.

[0183] The purpose of step S41 is to ensure that the point clouds of two adjacent periods in the time series are spatially precisely overlapped, which is the most basic requirement for local consistency in registration.

[0184] S42, select high-confidence corresponding feature points between the first and last periods, construct a closed-loop error term, and force the corresponding points in the first and last periods to coincide in the global coordinate system after their respective transformations, so as to eliminate long-term cumulative drift.

[0185] Select the first phase ( ) and the final stage ( The set of high-confidence corresponding points between ) These points are typically salient features that remain unchanged over long periods, such as building corners or road marking intersections. Define the global closed-loop error term: .

[0186] in: — Closed-loop error scalar, in square meters;

[0187] — A set of high-confidence pairs of identical points between the first and last periods;

[0188] — Transformation matrix of the initial point cloud;

[0189] — Transformation matrix of the final point cloud;

[0190] — First issue Homogeneous coordinate vectors of points with the same name;

[0191] — The homogeneous coordinate vector corresponding to the same point in the final stage.

[0192] This constraint is not explicitly required. and The relative transformation between them is not a unit matrix, but requires that the corresponding points at the beginning and end of the period coincide in the global coordinate system after their respective transformations, thereby indirectly achieving closed loop.

[0193] The purpose of step S42 is to establish a global closed-loop constraint by utilizing the same-name feature between the first and last periods, to force the cumulative drift in long-term registration to zero, and to prevent logical errors caused by non-closed beginnings and ends.

[0194] S43 selects spatially adjacent point pairs within the boundary band of the changing mask and penalizes the change in local relative geometric relationships before and after the transformation to ensure a smooth transition without tearing at the edges of the changed region.

[0195] Extracting the change mask The boundary zone (i.e., points within 0.5 meters of the boundary). Within this boundary zone, construct a set of neighboring point pairs. Each pair of points consists of two spatially adjacent points that are less than [a certain distance]. point and Composition. Define the smoothing regularization term:

[0196] .

[0197] in: — Smoothed regularized scalar, in square meters;

[0198] — The set of spatially adjacent point pairs within the changing boundary zone;

[0199] — No. The transformation matrix of the period;

[0200] — Homogeneous coordinate vectors of two adjacent points within the boundary zone;

[0201] — The difference vector between two points after transformation represents the local relative position after transformation;

[0202] — The difference vector between the two points before the transformation represents the original local relative position.

[0203] The change in local relative geometric relationships before and after the penalty transformation, that is, the rigid body transformation of the boundary region is forced to be as close as possible to the isometric transformation, so as to ensure that even if there is a change, its edge transition is smooth and there will be no tearing or overlapping.

[0204] The overall objective function combination is: .

[0205] in: — Overall optimization objective function value;

[0206] — The weight coefficients of the adjacent constraints are dimensionless and serve as the baseline weights;

[0207] — The weighting coefficient of the closed-loop constraint is slightly lower than that of the adjacent constraints to balance the reliability at the beginning and end of the period;

[0208] — The weighting coefficients of the smoothing regularization term are relatively low to avoid over-smoothing of real changes.

[0209] The optimization goal is The Levenberg-Marquardt (LM) algorithm is used for iterative solution. The LM algorithm is a hybrid of the Gauss-Newton method and gradient descent, insensitive to initial values ​​and exhibiting stable convergence. Each transformation matrix... Using Lie algebras a six-dimensional vector The parameters are parameterized, with the first three components corresponding to rotation and the last three to translation, and updated in each iteration. Until the change in the objective function is less than the threshold .

[0210] The purpose of step S43 is to prevent abrupt changes in registration parameters near the boundary of the changing region, which could lead to geometric discontinuities, and to force the transition between the changing and non-changing regions to maintain physical smoothness.

[0211] Step S4 serves as the core of the algorithm in this invention. It models the registration problem of multi-period point clouds as a joint optimization problem that includes adjacent-period alignment constraints, global closed-loop constraints, and smoothing constraints of change regions. It solves the rigid body transformation parameters of all periods at once, thereby completely avoiding the cumulative error caused by pairwise registration.

[0212] Step S5 involves extracting local point clouds and reconstructing triangular meshes only within the area marked by the changing mask. The local meshes are then embedded into the global model through boundary topology stitching, achieving local replacement rather than global repetition.

[0213] In this embodiment, step S5 may specifically include the following steps:

[0214] S51: Extract the change point clusters from the current registered point cloud based on the change mask, transform them to the local coordinate system, and then use the surface reconstruction algorithm to generate a local triangular mesh.

[0215] Using a changing mask (From step S2), extract the current period (e.g., the period) All points in the registered point cloud that satisfy the following conditions The points form a cluster of changing points. For each spatially connected variational cluster, calculate the origin of its local coordinate system as the centroid of the cluster. Transform all points within the cluster to local coordinates: .

[0216] in The rotation matrix (eigenvector matrix) obtained from the principal component analysis (PCA) of this cluster is such that the principal axes of the local coordinate system are aligned with the principal directions of the cluster.

[0217] Then, a triangular mesh is generated from the point cloud in the local coordinate system using the Poisson Surface Reconstruction algorithm. This algorithm converts the point cloud and its normal vectors into an implicit surface by solving the Poisson equation, and then extracts the isosurfaces to obtain the mesh. The reconstruction depth is then set. (i.e., the maximum level of the octree), this value strikes a balance between detail and smoothness. The output is a local triangular mesh. It contains a list of vertices and a list of triangle indices.

[0218] The purpose of step S51 is to cut out the point cloud of the changed region from the global coordinate system and establish an independent local triangular mesh to prepare for subsequent replacement.

[0219] S52 extracts the vertices of the boundary zone of the change region, uses the new mesh boundary points as control points, and employs radial basis function interpolation to smoothly transition the position of the vertices in the stitching area, so that the local mesh is seamlessly embedded into the global model.

[0220] Extracting the boundary zone of the change region: Expand outward by one grid cell. The distance is used to obtain the set of vertices on the boundary band. These vertices belong to both the old global model and the boundaries of the new mesh. Radial Basis Function (RBF) interpolation is used to smoothly transition the vertex positions in the stitching region.

[0221] For any vertex on the boundary band that needs adjustment (Vertices in the old model), calculate their updated positions: .

[0222] in: — The final three-dimensional coordinate vector of the vertex after stitching;

[0223] — Vertex coordinate vectors in the old model before stitching;

[0224] — The number of control points is taken from the boundary vertices of the new mesh;

[0225] — No. The undetermined weight coefficients for each control point (3×1 vector, corresponding to the x, y, z directions);

[0226] — No. Three-dimensional coordinate vectors of each control point (taken from the new mesh boundary);

[0227] — Gaussian radial basis functions, where For distance, This is a bandwidth parameter that controls the range of influence.

[0228] — Euclidean distance.

[0229] Weighting coefficient By solving the system of linear equations The stitching method ensures that the old and new positions perfectly match at the control points, while achieving smooth interpolation in other areas. This method guarantees the topological continuity and geometric smoothness of the mesh.

[0230] The purpose of step S52 is to seamlessly embed the newly generated local mesh into the global static model (the unchanged part), eliminating holes, overlaps or misalignments at the joints.

[0231] S53 performs triangle quality checks and optimizations on the reconstructed local mesh, generates multiple versions at various levels of detail, and updates the pointer index of the global model, thus completing the dynamic model update through local replacement.

[0232] Perform quality checks: Traverse the local mesh For all triangles in the array, calculate the minimum angle for each triangle. Ratio of the longest side (The ratio of the longest side to the shortest side). If it exists or If the triangle and its neighborhood are folded, edge collapse or edge split is performed until the quality meets the standard.

[0233] Generating three-level LOD (Level of Detail, a method used in 3D graphics and geographic information systems, the core idea of ​​which is to prepare multiple versions of the same 3D model with different levels of detail, so as to flexibly switch according to actual needs, thereby achieving the best balance between display effect and computing performance):

[0234] LOD 0 (High Detail): Original mesh resolution, number of vertices ;

[0235] LOD 1 (Medium Details): Simplifies vertices by 50% using the Quadratic Error Metric (QEM) algorithm; vertex count... ;

[0236] LOD 2 (Low Detail): Simplifies 90% of vertices, vertex count .

[0237] The three-level mesh is stored in a separate file, named {Mesh Morton Code}_{Period Number}_LOD{0 / 1 / 2}.ply. Finally, the pointer index table of the global model is updated, replacing the pointers of the corresponding regions in the old version with the pointers of the new version LOD0, and the old version data is moved to the history archive directory. This replacement operation only affects the changed areas (usually less than 10% of the total scene area), so the update efficiency is extremely high.

[0238] The purpose of step S53 is to ensure that the updated local mesh meets the geometric quality requirements (no distorted triangles) and to generate a multi-level of detail (LOD) version to support visualization and computational needs at different resolutions.

[0239] The purpose of step S5 is to reduce storage and computational overhead by performing local resampling, mesh reconstruction, and replacement only on the changed areas identified in step S2 after registration is completed, instead of regenerating the entire point cloud.

[0240] Step S6: Calculate the root mean square error of registration using an independent validation set, detect spatiotemporal logical inconsistencies, adaptively adjust and optimize the hyperparameters of the objective function based on the evaluation results, and trigger iterative re-registration until the accuracy and logical consistency requirements are met.

[0241] In this embodiment, step S6 may specifically include the following steps:

[0242] S61. Extract the reserved verification point set that did not participate in the registration optimization in each period, project it to the global coordinate system through the transformation matrix, and calculate the root mean square error of the nearest distance to the reference point cloud as the unbiased estimation index of the registration accuracy.

[0243] In each point cloud phase, 5% of the points are randomly selected as the validation set. These points were excluded in steps S1 to S4 and did not participate in any feature matching or optimization calculations. After registration, all validation points are transformed using their corresponding phase transformation matrices. Transform to the global coordinate system. For the first... Verification points of the period Calculate its distance to the first reference point cloud. Find the nearest distance (i.e., all point clouds in the first phase), and then calculate the root mean square error (RMSE):

[0244] .

[0245] in: — Root mean square error after registration, in meters;

[0246] — The total number of verification points across all periods;

[0247] — No. Issue No. Homogeneous coordinate vectors of verification points;

[0248] — No. The transformation matrix of the period;

[0249] — The three-dimensional coordinate vector of any point in the initial reference point cloud;

[0250] — An operator for finding the nearest neighbor in a reference point cloud;

[0251] — The square of the Euclidean distance.

[0252] The RMSE value objectively reflects the overall accuracy of registration. If the value is less than 1 meter (adjustable according to project requirements), then the current registration accuracy is considered substandard.

[0253] The purpose of step S61 is to use the reserved point set that has not participated in the registration optimization to estimate the true accuracy of the registration without bias.

[0254] S62 performs a temporal logic test on the change mask for multiple consecutive periods, identifies contradictory regions where the previous and next periods are both changing but the intermediate period is unchanged, and calculates the spatial range of the contradictory regions.

[0255] For every three consecutive periods Change mask , , Calculate the logical contradiction region: .

[0256] in: — A collection of logically contradictory regions;

[0257] — Intersection operation of spatial regions;

[0258] — No. The complement of the periodically changing mask (i.e., the non-changing region).

[0259] like The total area (or total volume) is greater than the threshold. If a significant logical contradiction exists, it indicates that the registration or change detection results are unreliable.

[0260] The purpose of step S62 is to detect whether there is a change logic that violates the physical time sequence, such as an abnormal situation where a building exists in phase 1, is demolished in phase 2, and is rebuilt in phase 3. Such contradictions usually stem from misjudgment of the change mask due to registration errors.

[0261] S63. When the root mean square error exceeds the preset threshold or there is a logical contradiction, increase the global closed-loop constraint weight and / or smoothing regularization weight by the preset step size, and re-execute the optimization solution until the accuracy and consistency requirements are met or the preset iteration limit is reached.

[0262] Set the maximum number of reregistration iterations In the first In the next iteration ( Adjust hyperparameters based on feedback:

[0263] like If the value is increased by 1 meter, then the weight of the global closed-loop constraint is increased: To enhance long-range constraints and suppress drift;

[0264] If a logical contradiction exists If the area is greater than the threshold, increase the smoothing regularization weight: And slightly reduce the confidence around the change mask (by step S22) (Shrinking it to 0.9 times makes it more sensitive to displacement).

[0265] After adjustment, return to step S4 and solve again. If the requirement is still not met after three iterations, output the result of the current iteration. The smallest set of transformation parameters is selected, and a manual quality inspection warning is issued. If during the iteration process... If the convergence falls below the threshold and there is no logical contradiction, the iteration is terminated early, and the final registration result and the updated dynamic model are output.

[0266] The function of step S63 is to automatically adjust the weighting coefficients in step S4 based on the error feedback from steps S61 and S62. Then, re-perform registration optimization until the accuracy and logic requirements are met or the maximum number of iterations is reached.

[0267] The purpose of step S6 is to serve as the final quality control step, using an independent validation set to evaluate the registration accuracy and check the spatiotemporal logic consistency. If any non-compliance is found, the hyperparameters are automatically adjusted and re-registration is triggered, forming an adaptive closed-loop optimization.

[0268] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0269] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0270] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0271] Example 2

[0272] Further reference Figure 2 As a response to the above Figure 1 The present invention provides an embodiment of a multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update device, which, according to the implementation of the method shown, provides an embodiment of such a device as... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0273] like Figure 2 As shown, the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update device 70 described in this embodiment includes: a construction module 71, an identification module 72, an allocation module 73, a solution module 74, an update module 75, and an iteration module 76. Wherein:

[0274] Module 71 is used to perform temporal normalization and incremental encoding on multi-period point clouds, and to construct a spatiotemporal joint hash index using an adaptive spatial grid partitioning strategy.

[0275] The identification module 72 is used to calculate the local geometric feature tensor of each point at different scales, evaluate the stability of each point in the time dimension through intertemporal residual analysis, and extract the change region based on density clustering to generate a change mask.

[0276] The allocation module 73 is used to calculate the comprehensive confidence of each registration primitive based on the geometric category of the ground feature, the scanning incident angle, the temporal stability, and the change mask penalty term, and to perform global normalization and dynamic truncation.

[0277] Solver module 74 is used to construct a joint objective function that includes relative transformation constraints between adjacent periods, global closure error constraints, and local smoothing regularization constraints in the change region. It solves the rigid body transformation parameters for all periods in one go through nonlinear least squares optimization.

[0278] The update module 75 is used to extract local point clouds and reconstruct triangular meshes only within the area marked by the changing mask, and to embed the local meshes into the global model through boundary topology stitching, so as to achieve local replacement rather than global repetition processing.

[0279] The iterative module 76 is used to calculate the root mean square error of registration using an independent validation set, detect spatiotemporal logical inconsistencies, adaptively adjust and optimize the hyperparameters of the objective function based on the evaluation results, and trigger iterative re-registration until the accuracy and logical consistency requirements are met.

[0280] Example 3

[0281] To address the aforementioned technical problems, embodiments of the present invention also provide a computer device. Please refer to [link / reference needed]. Figure 3 , Figure 3 This is a basic structural block diagram of the computer device in this embodiment.

[0282] The aforementioned computer device 8 includes a memory 81, a processor 82, and a network interface 83 that are interconnected via a system bus. It should be noted that only the computer device 8 with components 81, 82, and 83 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0283] The aforementioned computer devices can be desktop computers, laptops, handheld computers, and cloud servers, among other computing devices. These devices can facilitate human-computer interaction with users through keyboards, mice, remote controls, touchpads, or voice-activated devices.

[0284] The aforementioned memory 81 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the aforementioned memory 81 may be an internal storage unit of the aforementioned computer device 8, such as the hard disk or memory of the computer device 8. In other embodiments, the aforementioned memory 81 may also be an external storage device of the aforementioned computer device 8, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 8. Of course, the aforementioned memory 81 may also include both the internal storage unit and its external storage device of the aforementioned computer device 8. In this embodiment, the aforementioned memory 81 is typically used to store the operating system and various application software installed on the aforementioned computer device 8, such as computer-readable instructions for a multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method. In addition, the aforementioned memory 81 can also be used to temporarily store various types of data that have been output or will be output.

[0285] In some embodiments, the processor 82 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 82 is typically used to control the overall operation of the computer device 8. In this embodiment, the processor 82 is used to execute computer-readable instructions stored in the memory 81 or to process data, such as executing computer-readable instructions for the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method.

[0286] The network interface 83 may include a wireless network interface or a wired network interface, which is typically used to establish a communication connection between the computer device 8 and other electronic devices.

[0287] Example 4

[0288] The present invention also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method described above.

[0289] The beneficial effects of implementing the above embodiments are as follows:

[0290] (1) It can break through static assumptions and adapt to complex dynamic scenarios: By generating change masks through intertemporal residual analysis and density clustering, stable regions and change regions are explicitly distinguished. Local smoothing regularization constraints of change regions are introduced into the registration objective function to avoid misjudging change information as registration error, thus significantly improving robustness in complex scenarios.

[0291] (2) It can eliminate cumulative drift and ensure spatiotemporal logic consistency: By constructing a joint objective function that includes relative transformation between adjacent periods and global closure error constraints, all period transformation parameters are solved simultaneously at one time, fundamentally eliminating drift accumulation; at the same time, the validation set is used to detect spatiotemporal logic contradictions (such as buildings reappearing after disappearing), triggering iterative re-registration to ensure that the evolution sequence of ground features is reasonable.

[0292] (3) It can significantly reduce computation and storage overhead: By using spatiotemporal joint hash index to achieve fast association retrieval, and only performing local point cloud extraction and mesh reconstruction in the changing mask area, the global model is embedded through boundary topology stitching to achieve "on-demand update", avoiding repeated processing of stable areas, and reducing computational resource consumption by more than an order of magnitude.

[0293] (4) An adaptive closed-loop quality control was constructed: the registration RMSE was calculated based on the independent validation set, and the hyperparameters were dynamically adjusted and optimized to form a registration-evaluation-iteration closed loop, ensuring that the results not only meet the geometric accuracy threshold, but also pass the logical consistency test, which significantly improves the reliability and automation level of multi-period data fusion.

[0294] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0295] Obviously, the embodiments described above are merely some embodiments of the present invention, not all embodiments. The accompanying drawings show preferred embodiments of the present invention, but do not limit the patent scope of the present invention. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of this invention.

Claims

1. A method for adaptive registration and dynamic updating of spatiotemporal correlation of multi-period point clouds, characterized in that, Includes the following steps: Temporal normalization and incremental encoding are performed on multi-phase point clouds, and a spatiotemporal joint hash index is constructed using an adaptive spatial grid partitioning strategy; Calculate the local geometric feature tensor of each point at different scales, evaluate the stability of each point in the time dimension through intertemporal residual analysis, and extract the change region based on density clustering to generate a change mask; Based on the geometric category of the ground feature, the scanning incident angle, the temporal stability, and the change mask penalty term, the comprehensive confidence of each registration primitive is calculated, and global normalization and dynamic truncation are performed. A joint objective function is constructed, which includes relative transformation constraints between adjacent periods, global closure error constraints, and local smoothing regularization constraints in the change region. The rigid body transformation parameters for all periods are solved in one go through nonlinear least squares optimization. Local point cloud extraction and triangulation reconstruction are performed only within the area marked by the changing mask. The local mesh is then embedded into the global model through boundary topology stitching, achieving local replacement rather than global repetition. The root mean square error of registration is calculated using an independent validation set, and spatiotemporal logical inconsistencies are detected. Based on the evaluation results, the hyperparameters of the objective function are adaptively adjusted and optimized, triggering iterative re-registration until the accuracy and logical consistency requirements are met.

2. The multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method according to claim 1, characterized in that, The steps of performing time normalization and incremental encoding on multi-period point clouds, and constructing a spatiotemporal joint hash index using an adaptive spatial grid partitioning strategy, specifically include: Using the initial data collection time as the baseline zero point, the original timestamps of each period are normalized using a preset standard time interval to generate a composite key value containing the period number and intraday time information for rapid filtering of time neighborhoods. A dynamic octree recursive segmentation strategy is adopted, and a threshold for the number of point clouds in each leaf node is set. The spatial size of the leaf node is adaptively adjusted according to the local point density, so that the dense region is finely divided while the sparse region maintains a large grid. A two-dimensional hash mapping is established using the spatial Morton code of the leaf nodes as the primary key and the time period sequence number as the secondary key. The point clouds of each period are organized and stored according to their respective spatial grids, and a cross-period doubly linked list is maintained to support constant-level neighborhood retrieval.

3. The multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method according to claim 1, characterized in that, The steps of calculating the local geometric feature tensor of each point at different scales, evaluating the stability of each point in the time dimension through intertemporal residual analysis, and extracting the change region based on density clustering to generate a change mask specifically include: For each point, a neighborhood covariance matrix is ​​constructed with multiple different search radii. The eigenvalues ​​are solved and the planarity, linearity and randomness features are calculated. The geometric category of each point is determined by multi-scale voting. Spatiotemporal indexing is used to coarsely align point clouds between two adjacent periods, and the three-dimensional residual vector between pairs of points with the same name is calculated. Combined with the system noise standard deviation and geometric category indicator function, the stability probability of each point in the time dimension is quantified. Points with residuals exceeding a preset multiple of the system noise standard deviation are marked as change points. Spatial clustering algorithm is used to merge the change points, and morphological dilation processing is performed on the convex hull of each cluster to generate a binary change mask.

4. The multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method according to claim 1, characterized in that, The steps of calculating the comprehensive confidence of each registration primitive based on the geometric category of the ground feature, the scanning incident angle, the temporal stability, and the changing mask penalty term, and then performing global normalization and dynamic truncation, specifically include: Different prior weights are assigned to each point according to its geometric category. At the same time, the incident angle between the laser beam and the local surface normal vector is calculated. Points with excessive incident angles or unfavorable geometric categories are weighted or eliminated. Multiply the stability probability by the change mask penalty factor to make the overall stability factor of the points located inside the change mask zero, and the stable region maintains the original stability probability. The product of the base confidence level and the stability weighting factor is normalized to the maximum value of all points in this period, and unreliable points with confidence levels below a preset threshold are removed, leaving only high-confidence primitives to participate in subsequent registration optimization.

5. The multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method according to claim 1, characterized in that, The steps of constructing a joint objective function that includes relative transformation constraints between adjacent periods, global closure error constraints, and local smoothing regularization constraints in the changed region, and solving for the rigid body transformation parameters of all periods in one go through nonlinear least squares optimization, specifically include: For each pair of adjacent periods, extract high-confidence matching point pairs, and minimize the sum of squared spatial distances between corresponding points after the transformation of adjacent periods, using the confidence of each point as the weight. High-confidence corresponding feature points between the first and last periods are selected to construct a closed-loop error term. The corresponding points between the first and last periods are forced to coincide in the global coordinate system after their respective transformations in order to eliminate long-term cumulative drift. Within the boundary zone of the changing mask, spatially adjacent point pairs are selected, and the change in local relative geometric relationship before and after the transformation is penalized to ensure a smooth transition at the edge of the changed region without tearing.

6. The multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method according to claim 1, characterized in that, The steps of extracting local point clouds and reconstructing triangulations only within the area marked by the changing mask, and embedding the local mesh into the global model through boundary topology stitching to achieve local replacement rather than global repetition, specifically include: Based on the change mask, the change point clusters are extracted from the current registered point cloud, transformed to the local coordinate system, and then a surface reconstruction algorithm is used to generate a local triangular mesh. Extract the vertices of the boundary zone of the change region, use the boundary points of the new mesh as control points, and use radial basis function interpolation to smoothly transition the position of the vertices in the stitching area, so that the local mesh can be seamlessly embedded into the global model; The reconstructed local mesh is subjected to triangle quality checks and optimizations, generating multiple versions at different levels of detail. The pointer index of the global model is updated, and dynamic model updates are completed through local replacement.

7. The multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method according to any one of claims 1 to 6, characterized in that, The steps of calculating the root mean square error of registration using an independent validation set, detecting spatiotemporal logical inconsistencies, adaptively adjusting and optimizing the hyperparameters of the objective function based on the evaluation results, and triggering iterative re-registration until the accuracy and logical consistency requirements are met specifically include: Extract the reserved verification point set that did not participate in the registration optimization in each period, project it to the global coordinate system through the transformation matrix, and calculate the root mean square error of the nearest distance to the reference point cloud as the unbiased estimation index of the registration accuracy. A temporal logic test is performed on the change mask of multiple consecutive periods to identify contradictory regions where the previous and next periods are both changing but the intermediate period is unchanged, and the spatial range of the contradictory regions is statistically analyzed. When the root mean square error exceeds the preset threshold or there is a logical contradiction, the global closed-loop constraint weight and / or smoothing regularization weight are increased by a preset step size, and the optimization solution is re-executed until the accuracy and consistency requirements are met or the preset iteration limit is reached.

8. A multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update device, characterized in that, include: The module is used to perform temporal normalization and incremental encoding on multi-period point clouds, and to construct a spatiotemporal joint hash index using an adaptive spatial grid partitioning strategy; The identification module is used to calculate the local geometric feature tensor of each point at different scales, evaluate the stability of each point in the time dimension through intertemporal residual analysis, and extract the change region based on density clustering to generate a change mask. The allocation module is used to calculate the overall confidence of each registration primitive based on the geometric category of the ground feature, the scanning incident angle, the temporal stability, and the change mask penalty term, and to perform global normalization and dynamic truncation. The solver module is used to construct a joint objective function that includes relative transformation constraints between adjacent periods, global closure error constraints, and local smoothing regularization constraints in the change region. It solves the rigid body transformation parameters for all periods in one go through nonlinear least squares optimization. The update module is used to extract local point clouds and reconstruct triangulation only within the area marked by the changing mask, and to embed the local mesh into the global model through boundary topology stitching, so as to achieve local replacement rather than global repetition. The iterative module is used to calculate the root mean square error of registration using an independent validation set, detect spatiotemporal logical inconsistencies, adaptively adjust and optimize the hyperparameters of the objective function based on the evaluation results, and trigger iterative re-registration until the accuracy and logical consistency requirements are met.

9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the multi-period point cloud spatiotemporal correlation adaptive registration and dynamic update method as described in any one of claims 1 to 7.

Images