A target change rapid detection method based on multi-temporal point cloud registration

Abstract

This application discloses a method, system, electronic device, and storage medium for rapid target change detection based on multi-temporal point cloud registration. By acquiring multi-temporal point cloud data, voxelization downsampling and local geometric feature construction are performed. A random consistency sampling strategy is used to achieve global coarse registration, and iterative optimization is combined to complete fine registration. Difference point clouds are extracted and optimized through bidirectional consistency detection. Finally, a component reference model is introduced to calculate the change ratio of each component, thereby achieving rapid and accurate assessment of complex target changes.

Description

Technical Field

[0001] This application relates to the field of 3D point cloud processing and spatial information analysis technology, specifically to a method, device, electronic device and storage medium for rapid detection of target changes based on multi-temporal point cloud registration. Background Technology

[0002] With the rapid development of 3D sensing and laser scanning technologies, target state analysis based on point cloud data has been widely applied in fields such as engineering inspection, equipment evaluation, facility inspection, and damage analysis. By acquiring 3D point cloud data of a target at different times or in different states, it is possible to perform detailed analysis of structural changes, morphological differences, and local damage, providing important information for decision-making. However, existing point cloud change detection methods still face many challenges in practical applications. On the one hand, multi-temporal point clouds usually originate from different acquisition conditions, exhibiting significant differences in spatial location, orientation, and scale. Insufficient registration accuracy will directly affect the accuracy of subsequent change detection. On the other hand, existing methods mostly focus on change detection at the overall point cloud level, making it difficult to achieve detailed change analysis of individual functional components within complex targets, and unable to effectively quantify the degree or proportion of change at the component level. Furthermore, when dealing with large-scale point cloud data, traditional point-by-point comparison methods have high computational costs, making it difficult to meet the practical needs of rapid analysis. Therefore, how to achieve an efficient, stable, and component-level analysis-supporting multi-temporal point cloud change detection method while ensuring registration accuracy has become an urgent technical problem to be solved. Summary of the Invention

[0003] This application aims to address the limitations of existing point cloud change detection technologies, such as insufficient registration accuracy, coarse granularity of change analysis, and difficulty in component-level damage assessment. It provides a rapid target change detection method based on multi-temporal point cloud registration to achieve rapid and accurate assessment of changes in complex targets under different states.

[0004] To achieve the above objectives, the first aspect of this application provides a method for rapid target change detection based on multi-temporal point cloud registration, comprising: acquiring multi-temporal point cloud data of the same target collected at different times or under different working conditions, wherein the point cloud collected at an earlier time or under a baseline state is a reference point cloud, and the point cloud collected at the time to be detected or under a changing state is a point cloud to be detected; performing voxelization downsampling processing on the reference point cloud and the point cloud to be detected respectively, and constructing local geometric feature descriptors on the downsampled point clouds to characterize the spatial structure characteristics of the point clouds; based on the local geometric feature descriptors, establishing a feature matching relationship between the reference point cloud and the point cloud to be detected using a random consistency sampling strategy, estimating initial spatial transformation parameters, and achieving global coarse registration of the two temporal point clouds; and, based on the global coarse registration result, minimizing the spatial distance between corresponding points... The spatial transformation parameters are iteratively optimized to improve the distance error, achieving high-precision registration of the two temporal point clouds and placing the reference point cloud and the point cloud to be detected in a unified spatial coordinate system. Based on a spatial neighborhood search mechanism, bidirectional consistency detection is performed on the registered two temporal point clouds. Points that cannot find a corresponding neighborhood within a preset spatial distance threshold are identified as change points, thus extracting a set of difference point clouds. Statistical characteristic analysis is performed on the set of difference point clouds. By calculating the distribution density of each change point in a spherical neighborhood and removing isolated noise points below a preset minimum neighborhood point number threshold, an optimized change point cloud is obtained. The original reference point cloud model of each component of the target is introduced, and the optimized change points are fused with the point clouds of each component for spatial correlation analysis. The number of change points corresponding to each component and its proportion in the total number of points in the component are statistically analyzed to obtain the change ratio of each component.

[0005] Optionally, the voxelization downsampling process performed on the reference point cloud and the point cloud to be detected includes: setting the voxel size and replacing multiple points located in the same voxel with their centroid points to obtain a downsampled point cloud with a reduced size.

[0006] Optionally, the construction of the local geometric feature descriptor includes: constructing a local neighborhood for each sampled point in the downsampled point cloud, and calculating a local feature vector representing the geometric attributes of the point based on the spatial distribution of points within the neighborhood.

[0007] Optionally, the step of estimating the initial spatial transformation parameters includes: randomly selecting the smallest matching subset from the feature matching point pairs, eliminating erroneous matches by introducing geometric consistency constraints and distance constraints, and filtering the optimal spatial transformation matrix and translation vector.

[0008] Optionally, the iterative optimization of the spatial transformation parameters includes: searching for the nearest neighbor point in the reference point cloud corresponding to the point in the point cloud to be detected, constructing an optimization function with the objective of minimizing the Euclidean distance error, and updating the transformation matrix through multiple iterations until convergence.

[0009] Optionally, the bidirectional consistency detection includes: for any point in the point cloud to be detected, searching for its nearest neighbor in the reference point cloud; if the distance between the two exceeds a preset spatial distance threshold, then marking the point as a change point; and performing the same operation on the reference point cloud, taking the intersection or union of the two as the set of difference point clouds.

[0010] Optionally, calculating the change ratio of each component includes: Statistics m Number of change points corresponding to each component and the total number of points in the reference point cloud of this component. ; Using formula Calculate the percentage change of this component.

[0011] A second aspect of this application also provides a rapid target change detection system based on multi-temporal point cloud registration, comprising: a point cloud acquisition module for acquiring multi-temporal point cloud data of the same target collected at different times or under different working conditions, wherein the point cloud collected at an earlier time or under a reference state is a reference point cloud, and the point cloud collected at the time to be detected or under a changing state is a point cloud to be detected; a processing module for performing voxel downsampling processing on the reference point cloud and the point cloud to be detected respectively, and constructing local geometric feature descriptors on the downsampled point clouds to characterize the spatial structure characteristics of the point clouds; based on the local geometric feature descriptors, using a random consistency sampling strategy to establish a feature matching relationship between the reference point cloud and the point cloud to be detected, estimating initial spatial transformation parameters, and achieving global coarse registration of the two temporal point clouds; and based on the global coarse registration result, minimizing the difference between corresponding points... The spatial distance error iteratively optimizes the spatial transformation parameters to achieve high-precision registration of the two temporal point clouds, ensuring that the reference point cloud and the point cloud to be detected are in a unified spatial coordinate system. Based on a spatial neighborhood search mechanism, bidirectional consistency detection is performed on the registered two temporal point clouds. Points that cannot find a corresponding neighborhood within a preset spatial distance threshold are identified as change points, thereby extracting a set of difference point clouds. Statistical characteristic analysis is performed on the set of difference point clouds. By calculating the distribution density of each change point in a spherical neighborhood and removing isolated noise points below a preset minimum neighborhood point number threshold, an optimized change point cloud is obtained. The output module is used to introduce the original reference point cloud model of each component of the target, perform spatial correlation analysis between the optimized change point cloud and the point clouds of each component, count the number of change points corresponding to each component and its proportion in the total number of points in the component, and calculate the change ratio of each component.

[0012] A third aspect of this application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the target change rapid detection method based on multi-temporal point cloud registration as described in any one of claims 1 to 7.

[0013] A fourth aspect of this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the target change rapid detection method based on multi-temporal point cloud registration as described in any one of claims 1 to 7.

[0014] This application discloses a rapid target change detection method based on multi-temporal point cloud registration. The method includes acquiring multi-temporal point cloud data of the same target collected at different times or under different working conditions. Point clouds collected at earlier times or under a baseline state are designated as reference point clouds, while point clouds collected at the time of detection or under changing conditions are designated as target point clouds. The reference point cloud and the target point cloud are subjected to voxel downsampling processing, and local geometric feature descriptors are constructed on the downsampled point clouds to characterize their spatial structure. Based on the local geometric feature descriptors, a random consistency sampling strategy is used to establish feature matching relationships between the reference point cloud and the target point cloud, and initial spatial transformation parameters are estimated to achieve global coarse registration of the two temporal point clouds. Based on the global coarse registration result, the method minimizes the spatial distance error iteration between corresponding points. The spatial transformation parameters are optimized to achieve high-precision registration of the two temporal point clouds, ensuring that the reference point cloud and the point cloud to be detected are in a unified spatial coordinate system. Based on a spatial neighborhood search mechanism, bidirectional consistency detection is performed on the registered two temporal point clouds. Points that cannot find a corresponding neighborhood within a preset spatial distance threshold are identified as change points, thus extracting a set of difference point clouds. Statistical characteristic analysis is performed on the set of difference point clouds. By calculating the distribution density of each change point in a spherical neighborhood and removing isolated noise points below a preset minimum neighborhood point number threshold, an optimized change point cloud is obtained. The original reference point cloud model of each component of the target is introduced, and the optimized change points are fused with the point clouds of each component for spatial correlation analysis. The number of change points corresponding to each component and its proportion in the total number of points in the component are statistically analyzed to obtain the change ratio of each component. This application significantly improves the spatial alignment accuracy of multi-temporal point clouds in complex scenes by constructing a multi-stage point cloud registration strategy that combines global coarse registration and local fine registration, providing a reliable foundation for subsequent change detection. Simultaneously, by employing a change detection method based on spatial neighborhood consistency analysis, it can accurately extract real change areas, reducing the impact of background noise and mismatches on the detection results. Building upon this, a component-level point cloud mapping and damage ratio calculation mechanism is introduced, enabling a refined extension of change analysis from the overall target to the component level, allowing for quantitative assessment of the degree of change in each component. While ensuring detection accuracy, this method effectively reduces computational complexity and improves overall processing efficiency through voxelization and local computation strategies, making it suitable for rapid change detection and damage analysis of complex structural targets. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an embodiment of this application; Figure 2 This is a schematic diagram of the multi-temporal point cloud registration results in an embodiment of this application; Figure 3 This is a schematic diagram of the point cloud change detection results in an embodiment of this application; Figure 4 This is a schematic diagram illustrating the component-level change analysis and damage ratio visualization in the embodiments of this application.

[0016] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0018] Reference point cloud: refers to the target's three-dimensional point cloud data acquired at an earlier time or under a reference state, which serves as the reference coordinate system for registration.

[0019] Point cloud to be detected: refers to the target 3D point cloud data collected during the detection time or under changing conditions, used to compare with the reference point cloud to detect changes.

[0020] Voxelization downsampling: A preprocessing method that aims to reduce data size by dividing the space into equal-sized voxels and replacing all points within the voxels with centroids.

[0021] Local geometric feature descriptor: A vector calculated based on the spatial distribution of a point and its neighborhood, used to characterize the structural properties of the point in three-dimensional space.

[0022] Global coarse registration: Initial transformation parameters are estimated by feature matching and random consistency sampling to achieve preliminary alignment in large-scale space.

[0023] Fine registration: Based on coarse registration, iterative optimization is used to minimize distance error and achieve high-precision spatial alignment.

[0024] Difference point cloud set: A set of points extracted through bidirectional consistency detection that cannot find corresponding points within a preset threshold.

[0025] Component-level variation ratio: refers to the ratio of the number of variation points corresponding to a specific component to the total number of points in the reference point cloud of that component.

[0026] Key Existing Technology Description Existing point cloud change detection methods typically employ point-by-point comparison or simple distance calculations. However, multi-temporal point clouds often reside in different local coordinate systems. Relying solely on a single registration algorithm can easily lead to local optima when dealing with complex structural targets, resulting in insufficient registration accuracy and a large number of false detections. Furthermore, most existing technologies output the overall change area, lacking the ability to map the change results to specific functional components and perform quantitative analysis, making it difficult to meet the needs of refined evaluation. To address the problems of insufficient multi-temporal point cloud registration accuracy, coarse change detection granularity, and difficulty in component-level change evaluation in existing technologies, this invention proposes a component-level fast change detection method based on multi-temporal point cloud registration.

[0027] This method constructs a multi-stage point cloud registration process combining global coarse registration and local fine registration to achieve high-precision spatial alignment of point cloud data at different time points. Based on this, it rapidly extracts changed regions of the point cloud through bidirectional spatial neighborhood consistency analysis. Furthermore, by combining a component-level reference point cloud, it achieves accurate mapping of the changed point cloud at the component level and calculation of the change ratio, thereby enabling rapid and quantitative analysis of changes at the component level of complex targets. This invention proposes a rapid target change detection method based on multi-temporal point cloud registration to address the problems of insufficient registration accuracy, unstable extraction of changed regions, and difficulty in achieving component-level quantitative analysis in existing point cloud change detection methods. This method achieves rapid and accurate assessment of changes in complex targets under different states through multi-stage point cloud registration, change detection, and component-level mapping analysis.

[0028] Example 1 refer to Figure 1 The first embodiment of this application provides a method for rapid detection of target changes based on multi-temporal point cloud registration, to solve the technical problems mentioned in the background art, such as insufficient registration accuracy of multi-temporal data, unstable extraction of change areas, and difficulty in achieving component-level quantitative analysis. This method can be executed by a processor, which can be set in a terminal or a server. The execution process of this method is as follows: Step s101: Obtain multi-temporal point cloud data of the same target collected at different times or under different working conditions. Among them, the point cloud collected at an earlier time or under the baseline condition is the reference point cloud, and the point cloud collected at the time to be detected or under the changing condition is the point cloud to be detected.

[0029] The acquisition of multi-temporal point cloud data of the same target collected at different times or under different operating conditions includes multi-temporal point cloud data acquisition: acquiring multi-temporal point cloud data of the same target collected at different times or under different operating conditions, wherein the point cloud collected at an earlier time or under a reference condition is recorded as the reference point cloud. The point cloud collected during the time period or under changing conditions to be detected is recorded as the point cloud to be detected. The point clouds mentioned above are all located in their own independent local coordinate systems, providing the original data basis for subsequent registration and change analysis.

[0030] Step s102: Voxelization downsampling is performed on the reference point cloud and the point cloud to be detected, and a local geometric feature descriptor is constructed on the downsampled point cloud to characterize the spatial structure characteristics of the point cloud.

[0031] In one embodiment of this application, the voxelization downsampling process performed on the reference point cloud and the point cloud to be detected includes: setting the voxel size and replacing multiple points located in the same voxel with their centroid points, thereby obtaining a downsampled point cloud with a reduced size.

[0032] In an optional embodiment of this application, the construction of the local geometric feature descriptor includes: constructing a local neighborhood for each sampled point in the downsampled point cloud, and calculating a local feature vector representing the geometric attributes of the point based on the spatial distribution of points within the neighborhood. Further, on the downsampled point cloud, a local neighborhood is constructed for each sampled point, and a local geometric feature descriptor is calculated based on the spatial distribution of points within the neighborhood, denoted as:

[0033] in, The local neighborhood search operator for a point. This represents a local feature extraction function.

[0034] Step s103: Based on the local geometric feature descriptor, a random consistency sampling strategy is used to establish a feature matching relationship between the reference point cloud and the point cloud to be detected, estimate the initial spatial transformation parameters, and realize global coarse registration of the two temporal point clouds.

[0035] In one embodiment of this application, the step of estimating the initial spatial transformation parameters includes: randomly selecting a minimum matching subset from feature matching point pairs, eliminating erroneous matches by introducing geometric consistency constraints and distance constraints, and filtering for the optimal spatial transformation matrix and translation vector. Further, a random consistency sampling strategy is adopted to randomly select a minimum matching subset from feature matching point pairs to estimate the initial rigid spatial transformation relationship between point clouds, including the rotation matrix and translation vector, such that:

[0036] The optimal spatial transformation parameters are selected by evaluating the number of interior points or the matching error. This enables the initial alignment of two temporal point clouds on a large-scale spatial scale.

[0037] Step s104: Based on the global registration result, the spatial transformation parameters are iteratively optimized by minimizing the spatial distance error between corresponding points to achieve high-precision registration of the two temporal point clouds, so that the reference point cloud and the point cloud to be detected are in a unified spatial coordinate system.

[0038] In one embodiment of this application, the iterative optimization of the spatial transformation parameters includes: searching for the nearest neighbor point in the reference point cloud corresponding to the point in the point cloud to be detected; constructing an optimization function with the objective of minimizing the Euclidean distance error; and updating the transformation matrix through multiple iterations until convergence. Further, the optimization objective function is constructed as follows:

[0039] By iteratively updating the spatial transformation parameters, high-precision point cloud registration results are obtained, ensuring accurate alignment of multi-temporal point clouds in a unified spatial coordinate system. .

[0040] By iteratively updating the spatial transformation parameters, high-precision point cloud registration results are obtained, ensuring accurate alignment of multi-temporal point clouds in a unified spatial coordinate system.

[0041] Step s105: Based on the spatial neighborhood search mechanism, perform bidirectional consistency detection on the two-phase point clouds after fine registration. Points that cannot find a corresponding neighborhood within the preset spatial distance threshold will be identified as change points, thereby extracting the set of difference point clouds.

[0042] In one embodiment of this application, the bidirectional consistency detection includes: for any point in the point cloud to be detected, searching for its nearest neighbor in the reference point cloud; if the distance between the two exceeds a preset spatial distance threshold, then marking the point as a changed point; simultaneously performing the same operation on the reference point cloud, taking the intersection or union of the two as the set of differing point clouds. Further, for any point in the point cloud to be detected, searching for its nearest neighbor in the reference point cloud; if the distance between the two satisfies:

[0043] in If a preset spatial distance threshold is used, the point is identified as a change point. A bidirectional detection strategy is employed to further enhance the stability of the change detection results, ultimately yielding a set of difference point clouds reflecting changes in the target structure.

[0044] Step s106: Perform statistical characteristic analysis on the difference point cloud set. Calculate the distribution density of each change point in the spherical neighborhood and remove isolated noise points below the preset minimum neighborhood point number threshold to obtain the optimized change point cloud.

[0045] In an optional embodiment of this application, for any changing point in the changing point cloud First, taking this point as the center, construct a spatial neighborhood with a radius of... r Determine a spherical neighborhood region and count the number of change points within that neighborhood, denoted as . This neighborhood is used to characterize the distribution density characteristics of the change points in the local space. Subsequently, the number of neighborhood points counted... , and the preset minimum neighborhood point threshold Compare the following conditions:

[0046] If the change point lacks sufficient spatial support information, it is considered an isolated noise point and is removed; otherwise, the change point is retained as a valid change point.

[0047] The above-described screening process based on local spatial density can effectively remove anomalous change points introduced by registration errors, point cloud sparsity, or sensor noise, thereby enhancing the spatial continuity and structural consistency of the changed point cloud and obtaining more stable and reliable change detection results. Step s107: Introduce the original reference point cloud model of each component of the target, perform spatial correlation analysis between the optimized change point cloud and the point cloud of each component, count the number of change points corresponding to each component and its proportion in the total number of points of the component, and calculate the change ratio of each component.

[0048] In one embodiment of this application, for any component Count the number of corresponding change points. Total number of points in the point cloud of components And calculate the percentage change of the component:

[0049] This enables a quantitative assessment of the degree of change in each component of the target. The registered overall point cloud, the changed point cloud, and the component-level change results are fused and displayed, and statistical information on the change ratio of each component is output, thereby achieving an intuitive display and analysis of the changes in the overall target and its components.

[0050] This invention proposes a rapid target change detection method based on multi-temporal point cloud registration, effectively addressing the limitations of existing point cloud change detection technologies, such as insufficient registration accuracy, coarse granularity of change analysis, and difficulty in component-level damage assessment. By constructing a multi-stage point cloud registration strategy combining global coarse registration and local fine registration, the spatial alignment accuracy of multi-temporal point clouds in complex scenes is significantly improved, providing a reliable foundation for subsequent change detection. Simultaneously, through a change detection method based on spatial neighborhood consistency analysis, the actual change areas can be accurately extracted, reducing the impact of background noise and mismatches on the detection results. Furthermore, by introducing a component-level point cloud mapping and damage ratio calculation mechanism, change analysis is refined from the overall target to the component level, enabling quantitative assessment of the degree of change in each component. While ensuring detection accuracy, this method effectively reduces computational complexity and improves overall processing efficiency through voxelization and local calculation strategies. It is suitable for rapid change detection and damage analysis of complex structural targets, possessing high engineering practical value and promotional significance.

[0051] In another embodiment, this application also provides simulation conditions and a simulation process, as follows: This invention is based on simulation using Python software and the Open3d framework on a CPU consisting of an Intel® i7-6800K 3.40GHz CPU, an NVIDIA GeForce GTX3090 GPU, and a Windows operating system.

[0052] refer to Figures 2-3 In the aforementioned simulation environment, multi-temporal 3D point cloud data is first imported into the Open3D platform for processing. Point clouds acquired at earlier times are used as reference point clouds, while those acquired at later times are used as the point clouds to be detected. The original point clouds are first subjected to voxel downsampling. By setting the voxel size, the point cloud is meshed, and the voxel centers replace the original points, thus reducing the data size. Then, the normal vector information of the points is calculated on the downsampled point clouds, and local geometric feature descriptors are constructed to characterize the spatial structure of the point clouds. Based on this, an initial correspondence between the two temporal point clouds is established through feature matching, and the initial rigidity transformation between the two sets of point clouds is estimated using a random consistency sampling algorithm to achieve global coarse registration. Subsequently, the iterative nearest-point method is used to optimize the coarse registration result. By continuously minimizing the Euclidean distance error between corresponding points, the rotation matrix and translation vector are iteratively updated to obtain accurate point cloud registration results.

[0053] After obtaining high-precision registration results, change detection analysis is performed on the two-phase point clouds. Specifically, a spatial neighborhood search structure based on a KD-tree is constructed. Under a given distance threshold, each point in the point cloud to be detected performs a nearest neighbor search in the reference point cloud. When the distance between the two exceeds a preset threshold, the point is marked as a change point. A bidirectional consistency detection strategy is also employed to reduce misjudgments caused by registration errors or local noise. Subsequently, the extracted change point cloud is subjected to statistical filtering based on neighborhood density to remove spatially isolated outliers, thus obtaining continuous and stable change regions. Finally, the change point cloud is spatially matched with the reference point cloud models of each component. By statistically analyzing the number of change points in each component's point cloud and calculating their proportion in the total number of points in the component, the component-level change ratio is calculated and analyzed. Finally, the registration results, change regions, and the degree of change in each component are displayed through 3D visualization.

[0054] Example 2 To address the aforementioned technical problems, the second embodiment of this application provides a rapid target change detection system based on multi-temporal point cloud registration to solve the same technical problems as the method embodiment. This system 1000 may include the following modules: The point cloud acquisition module 1001 is used to acquire multi-temporal point cloud data of the same target collected at different times or under different working conditions. Among them, the point cloud collected at an earlier time or under the baseline state is the reference point cloud, and the point cloud collected at the time to be detected or under the changing state is the point cloud to be detected. Processing module 1002 is used to perform voxelization downsampling on the reference point cloud and the point cloud to be detected, and to construct local geometric feature descriptors on the downsampled point clouds to characterize the spatial structure characteristics of the point clouds; based on the local geometric feature descriptors, a random consistency sampling strategy is used to establish feature matching relationships between the reference point cloud and the point cloud to be detected, estimate the initial spatial transformation parameters, and realize global coarse registration of the two temporal point clouds; based on the global coarse registration result, the spatial transformation parameters are iteratively optimized by minimizing the spatial distance error between corresponding points to realize high-precision fine registration of the two temporal point clouds, so that the reference point cloud and the point cloud to be detected are in a unified spatial coordinate system; based on the spatial neighborhood search mechanism, bidirectional consistency detection is performed on the finely registered two temporal point clouds, and points that cannot find corresponding neighborhoods within a preset spatial distance threshold are judged as change points, thereby extracting a set of difference point clouds; statistical characteristic analysis is performed on the set of difference point clouds, and the distribution density of each change point in the spherical neighborhood is calculated and isolated noise points below the preset minimum neighborhood point number threshold are removed to obtain the optimized change point cloud; The output module 1003 is used to import the original reference point cloud model of each component of the target, perform spatial correlation analysis between the optimized change point cloud and the point cloud of each component, count the number of change points corresponding to each component and its proportion in the total number of points of the component, and calculate the change ratio of each component.

[0055] It is understood that this embodiment is the system authority corresponding to the method authority, and therefore this embodiment has all the technical effects of the method embodiment, which will not be repeated here.

[0056] Example 3 To address the aforementioned technical problems, the third embodiment of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the target change rapid detection method based on multi-temporal point cloud registration provided in the previous embodiment.

[0057] Example 4 To address the aforementioned technical problems, the fourth embodiment of this application provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the target change rapid detection method based on multi-temporal point cloud registration provided in the preceding embodiments.

[0058] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for rapid target change detection based on multi-temporal point cloud registration, characterized in that, include: Acquire multi-temporal point cloud data of the same target collected at different times or under different working conditions. Among them, the point cloud collected at an earlier time or under the baseline condition is the reference point cloud, and the point cloud collected at the time to be detected or under the changing condition is the point cloud to be detected. The reference point cloud and the point cloud to be detected are subjected to voxelization downsampling processing, and local geometric feature descriptors are constructed on the downsampled point cloud to characterize the spatial structure characteristics of the point cloud. Based on the local geometric feature descriptor, a random consistency sampling strategy is used to establish a feature matching relationship between the reference point cloud and the point cloud to be detected, estimate the initial spatial transformation parameters, and achieve global coarse registration of the two temporal point clouds. Based on the global coarse registration result, the spatial transformation parameters are iteratively optimized by minimizing the spatial distance error between corresponding points to achieve high-precision fine registration of the two temporal point clouds, so that the reference point cloud and the point cloud to be detected are in a unified spatial coordinate system. Based on the spatial neighborhood search mechanism, bidirectional consistency detection is performed on the two-phase point clouds after fine registration. Points that cannot find a corresponding neighborhood within a preset spatial distance threshold are identified as change points, thereby extracting the set of difference point clouds. Statistical characteristic analysis is performed on the difference point cloud set. By calculating the distribution density of each change point in the spherical neighborhood and removing isolated noise points below the preset minimum neighborhood point number threshold, the optimized change point cloud is obtained. The original reference point cloud model of each component of the target is introduced, and the optimized change point cloud is spatially correlated with the point cloud of each component. The number of change points corresponding to each component and its proportion in the total number of points of the component are counted, and the change ratio of each component is calculated.

2. The method according to claim 1, characterized in that, The step of performing voxelization downsampling processing on the reference point cloud and the point cloud to be detected includes: By setting the voxel size, multiple points located within the same voxel are replaced by their centroids, thereby obtaining a downsampled point cloud with a reduced scale.

3. The method according to claim 1, characterized in that, The construction of the local geometric feature descriptor includes: For each sampled point in the downsampled point cloud, a local neighborhood is constructed, and a local feature vector representing the geometric properties of the point is calculated based on the spatial distribution of points within the neighborhood.

4. The method according to claim 1, characterized in that, The step of estimating the initial spatial transformation parameters includes: The minimum matching subset is randomly selected from the feature matching point pairs. By introducing geometric consistency constraints and distance constraints, erroneous matches are eliminated, and the optimal spatial transformation matrix and translation vector are selected.

5. The method according to claim 1, characterized in that, The iterative optimization of the spatial transformation parameters includes: Search for the nearest neighbor of the point in the reference point cloud that corresponds to the point in the point cloud to be detected, construct an optimization function with the objective of minimizing the Euclidean distance error, and update the transformation matrix through multiple iterations until convergence.

6. The method according to claim 1, characterized in that, The bidirectional consistency detection includes: For any point in the point cloud to be detected, search for its nearest neighbor in the reference point cloud. If the distance between the two exceeds a preset spatial distance threshold, then mark the point as a changed point. At the same time, the same operation is performed on the reference point cloud, and the intersection or union of the two is taken as the set of difference point clouds.

7. The method according to claim 1, characterized in that, The calculation of the change ratio of each component includes: The calculation of the change ratio of each component includes: statistically analyzing the percentage change of the first component. m Number of change points corresponding to each component and the total number of points in the reference point cloud of this component. ; Using formula Calculate the percentage change of this component.

8. A rapid target change detection system based on multi-temporal point cloud registration, characterized in that, include: The point cloud acquisition module is used to acquire multi-temporal point cloud data of the same target collected at different times or under different working conditions. Among them, the point cloud collected at an earlier time or under the baseline state is the reference point cloud, and the point cloud collected at the time to be detected or under the changing state is the point cloud to be detected. The processing module performs voxelization downsampling on the reference point cloud and the point cloud to be detected, and constructs local geometric feature descriptors on the downsampled point clouds to characterize the spatial structure characteristics of the point clouds. Based on the local geometric feature descriptors, a random consistency sampling strategy is used to establish feature matching relationships between the reference point cloud and the point cloud to be detected, estimate the initial spatial transformation parameters, and achieve global coarse registration of the two temporal point clouds. Based on the global coarse registration result, the spatial transformation parameters are iteratively optimized by minimizing the spatial distance error between corresponding points to achieve high-precision fine registration of the two temporal point clouds, so that the reference point cloud and the point cloud to be detected are in a unified spatial coordinate system. Based on the spatial neighborhood search mechanism, bidirectional consistency detection is performed on the finely registered two temporal point clouds. Points that cannot find corresponding neighborhoods within a preset spatial distance threshold are identified as change points, thereby extracting a set of difference point clouds. Statistical characteristic analysis is performed on the set of difference point clouds. By calculating the distribution density of each change point in a spherical neighborhood and removing isolated noise points below a preset minimum neighborhood point number threshold, an optimized change point cloud is obtained. The output module is used to import the original reference point cloud model of each component of the target, perform spatial correlation analysis between the optimized change point cloud and the point cloud of each component, count the number of change points corresponding to each component and its proportion in the total number of points of the component, and calculate the change ratio of each component.

9. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the target change rapid detection method based on multi-temporal point cloud registration as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the target change rapid detection method based on multi-temporal point cloud registration as described in any one of claims 1 to 7.

Images