A method and system for constructing 3D changing point cloud datasets for urban scenes

Abstract

This invention discloses a method and system for constructing a 3D change point cloud dataset for urban scenes, belonging to the field of 3D GIS technology. The method includes: collecting multi-temporal 3D change point cloud data for urban scenes and performing preprocessing; using an octree algorithm to coarsely screen the preprocessed multi-temporal 3D change point cloud data; and based on the coarsely screened multi-temporal 3D change point cloud data, obtaining the final labeled multi-temporal point cloud data through manual annotation and cross-validation, thus completing the construction of the 3D change point cloud dataset. This invention, by collecting real multi-temporal 3D change point cloud data for urban scenes, using an octree algorithm for coarse screening of change information, and employing a semi-automated method of manual annotation, helps improve the efficiency of constructing 3D change point cloud datasets, thereby providing efficient data support for the construction of 3D change detection methods.

Description

Technical Field

[0001] This invention belongs to the field of 3D GIS technology, specifically relating to a method and system for constructing a 3D change point cloud dataset for urban scenes. Background Technology

[0002] Cities are the primary venues for human activity, undertaking major social and productive activities. In today's technologically advanced world, cities are constantly in a state of dynamic change, including urban expansion, deforestation, changes in lake area, and alterations in landforms following natural disasters. To accurately grasp these changes and achieve precise urban management, change detection is necessary. Change detection compares data from the same area at different times to accurately locate areas of change and uncover key information such as the degree and type of change. Three-dimensional change detection, on the other hand, compares three-dimensional data, such as point clouds, DEMs, and DSMs, to obtain changes in geographic space over different time periods. Compared to two-dimensional change detection, three-dimensional change detection can detect changes in the vertical direction, providing a more comprehensive and accurate assessment of the changes.

[0003] Deep learning-based 3D change detection methods are currently the mainstream approach, offering better stability, robustness, and accuracy compared to algebraic interpolation and machine learning. However, deep learning-based 3D change detection requires a robust 3D change detection dataset. To achieve city-level 3D change detection, a dataset tailored to urban scenarios is needed. Since point cloud data is easy to collect, highly accurate, and has a simple data structure, most 3D change detection datasets are built upon point clouds. Currently, the main method for constructing a city-level 3D change detection dataset involves comparing differences between point clouds from different time periods and annotating change information point-by-point across multiple time periods. This method boasts extremely high accuracy, but the multi-period data and point-by-point comparison significantly increase the workload of data annotation, greatly raising the cost of dataset creation.

[0004] To improve the speed and reduce the cost of dataset production, a semi-automated approach is currently used. This involves first generating multi-temporal virtual point cloud data using a simulation engine, and then creating variations between the data based on defined change rules or manual intervention, resulting in rich 3D point cloud variation data. While this method quickly obtains a large amount of 3D variation data, it still has certain limitations because the virtual data has simple geometry and low texture realism, making it unable to accurately describe changes in the real urban landscape.

[0005] In summary, a 3D change detection dataset is a necessary condition for constructing a deep learning-based 3D change detection method. Although current methods for constructing 3D change detection datasets can provide rich and accurate 3D change data, the following problems still exist: 1) While manual annotation methods can achieve extremely high data accuracy, the cost and difficulty of data annotation are too high, greatly hindering the creation and expansion of 3D change detection datasets. 2) Semi-automatic methods that generate virtual data through simulation engines can effectively improve the speed of dataset creation and reduce the cost of dataset production, but the virtual data has poor realism and cannot accurately describe changes in urban scenes, which will greatly limit the implementation of large-scale urban 3D change detection. Therefore, it is urgent to study an efficient method for constructing 3D change detection datasets for urban scenes.

[0006] Therefore, it is necessary to design a method and system for constructing a 3D change point cloud dataset for urban scenarios to address the above problems. Summary of the Invention

[0007] The purpose of this invention is to address the problems existing in the prior art by providing a method and system for constructing a 3D change point cloud dataset for urban scenes. The method involves preprocessing the data through data cleaning and point cloud registration, then using an octree algorithm to coarsely screen the change information of multi-temporal point cloud data. The coarsely screened change information is manually labeled and corrected, overcoming the problems of high cost and difficulty in data labeling, obtaining a real dataset describing changes in urban scenes, and realizing the construction of a 3D change point cloud dataset.

[0008] According to one aspect of this specification, a method for constructing a 3D changing point cloud dataset for urban scenes is provided, including:

[0009] Step 1: Collect multi-temporal point cloud data for urban scenarios and perform preprocessing;

[0010] Step 2: Use the octree algorithm to perform coarse screening on the preprocessed multi-temporal point cloud data, specifically including:

[0011] First, based on the spatial extent of the point cloud data, an octree is constructed for each time series of point cloud data.

[0012] Secondly, features are extracted for each node in the octree, including the number of nodes and the node density.

[0013] Finally, based on the number of nodes and node density, the feature differences of each node in different temporal octrees are compared to determine the point cloud data that have changed and those that have not changed.

[0014] Step 3: Based on the coarsely screened multi-temporal 3D change point cloud data, perform annotation and cross-validation to obtain the final annotated multi-temporal point cloud data, thus obtaining the constructed 3D change point cloud dataset for urban scenes.

[0015] Furthermore, the construction of the octree includes:

[0016] Based on the spatial range of the point cloud data, determine the cubic space represented by the root node of the octree;

[0017] The cube space is divided into eight sub-cubes. Each sub-cube is determined to reach a predetermined maximum recursion depth or the number of point clouds in each sub-cube meets a certain threshold, thus obtaining the constructed octree.

[0018] Further, identify the point cloud data that has changed and that has not changed, including:

[0019] The nearest neighbor algorithm is used to determine the corresponding position of each node in the octree of different time series, and the threshold is determined based on the point cloud density.

[0020] Octrees of different time sequences are divided into voxels by partitioning the space. The node density is calculated based on the number of nodes and the volume of the voxels. The node density is then used as a feature difference for comparison.

[0021] When the feature difference exceeds the threshold, the point cloud data is determined to have changed; when the feature difference does not exceed the threshold, the point cloud data is determined to have changed.

[0022] Furthermore, annotation and cross-validation are performed, including:

[0023] The multi-time series point cloud data after coarse screening is labeled, and the unchanged data is corrected to change data;

[0024] Cross-validation of data with different manual corrections yields the final labeled multi-temporal point cloud data.

[0025] Furthermore, after preprocessing, point cloud registration is also included:

[0026] Multiple feature points are selected to perform preliminary alignment of multi-temporal point cloud data, and the iterative nearest point algorithm is used to perform fine registration of the preliminary aligned point cloud data.

[0027] According to one aspect of this specification, a system for constructing a 3D changing point cloud dataset for urban scenes is provided, comprising:

[0028] The data acquisition module is used to collect multi-temporal 3D changing point cloud data for urban scenarios and perform preprocessing.

[0029] The coarse screening module is used to perform coarse screening on preprocessed multi-temporal point cloud data using an octree algorithm. Specifically, it includes:

[0030] First, based on the spatial extent of the point cloud data, an octree is constructed for each time series of point cloud data.

[0031] Secondly, features are extracted for each node in the octree, including the number of nodes and the node density.

[0032] Finally, based on the number of nodes and node density, the feature differences of each node in different temporal octrees are compared to determine the point cloud data that have changed and those that have not changed.

[0033] The dataset construction module is used to label and cross-validate the multi-temporal 3D change point cloud data after coarse screening to obtain the final labeled multi-temporal point cloud data, thus creating a constructed 3D change point cloud dataset for urban scenarios.

[0034] According to one aspect of this specification, an electronic device is provided, including a memory and a processor, the memory storing a computer program, characterized in that the processor executes the computer program to implement the steps of the method for constructing a three-dimensional change point cloud dataset for urban scenes.

[0035] According to one aspect of this specification, a computer-readable storage medium is provided having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the method for constructing a three-dimensional change point cloud dataset for urban scenes.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] 1. This invention preprocesses data through methods such as data cleaning and point cloud registration. Then, it uses an octree algorithm to coarsely screen the change information of multi-time series point cloud data. The coarsely screened change information is manually labeled and corrected. The labeled data is cross-validated to ensure the accuracy of the data, thereby efficiently and accurately constructing a three-dimensional change point cloud dataset.

[0038] 2. This invention addresses the challenges of labeling and identifying multi-temporal data in 3D change datasets. By collecting real multi-temporal 3D change point cloud data for urban scenarios, employing an octree algorithm for coarse screening of change information, and using a semi-automated method of manual labeling, it helps improve the efficiency of constructing 3D change point cloud datasets, thereby providing efficient data support for the construction of 3D change detection methods. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a method for constructing a 3D changing point cloud dataset for urban scenes according to an embodiment of the present invention;

[0041] Figure 2 This is a flowchart of the data preprocessing process according to an embodiment of the present invention;

[0042] Figure 3 This is a flowchart illustrating the semi-automated annotation process according to an embodiment of the present invention;

[0043] Figure 4 This is a sample image of a three-dimensional change detection dataset constructed for an embodiment of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] This invention provides a method for constructing a 3D change point cloud dataset for urban scenes, such as... Figure 1 As shown, it includes:

[0046] Step 1: Collect multi-temporal point cloud data for urban scenarios and perform preprocessing;

[0047] Step 2: Use the octree algorithm to perform coarse screening on the preprocessed multi-temporal point cloud data, specifically including:

[0048] First, an octree is constructed for the point cloud data of each time series;

[0049] Secondly, features are extracted from the point cloud region corresponding to each node in the octree;

[0050] Finally, based on the extracted features, the feature differences of each node in the octree at different time series are compared to determine the point cloud data that have changed and those that have not changed.

[0051] Step 3: Based on the coarsely screened multi-temporal 3D change point cloud data, the final labeled multi-temporal point cloud data is obtained through manual annotation and cross-validation, thus completing the construction of a 3D change point cloud dataset for urban scenes.

[0052] Specifically, this embodiment of the invention also provides the acquisition of multi-temporal point clouds. The city-level 3D change detection dataset is designed for urban scenarios and requires airborne equipment for data acquisition. Furthermore, change detection requires multiple observations of the same area at different time intervals. When acquiring data at different times, it is necessary to ensure that the performance and parameters of the equipment remain consistent for each data acquisition, minimizing false change samples caused by the acquisition equipment and ensuring the accuracy of the change data. The specific implementation process is described below:

[0053] First, for UAV data acquisition, the specific area to be collected needs to be selected. Based on the terrain undulations of the area, relevant UAV flight parameters, such as flight altitude, flight speed, and aerial survey tilt angle, are determined. Following the principle of comprehensive coverage, UAV flight paths are delineated, and the complexity of the area determines whether intersecting flight paths are necessary. After the aerial survey mission is completed, the image data or LiDAR point clouds need to be cleaned to ensure the accuracy of the experimental data. Finally, based on the principle of single variable, the same area is repeatedly collected at different time intervals to obtain multi-time series data. Then, to address the time-consuming and labor-intensive nature of UAV data acquisition, a multi-UAV collaborative acquisition method is adopted to obtain experimental data. However, simultaneous data acquisition by multiple UAVs can lead to the problem of data "bubble" (or "data fusion"), so a multi-UAV data fusion method is proposed. This method uses ground control points as a reference to fuse data from multiple UAVs with different starting points, and filters the point cloud boundaries using statistical data to eliminate the influence of different data boundaries. Finally, based on the oblique photographic images of the relevant area, the image pose parameters and camera parameters calculated by aerial triangulation are used to obtain the 3D point cloud of the relevant area through dense image matching. Currently, 3D point cloud reconstruction is a relatively mature engineering technology. It can be achieved through algorithm design or through automated 3D modeling software, such as Smart3D and Photoscan. After reconstruction, multi-temporal real-world point cloud data for urban scenes can be obtained, providing a rich data foundation for subsequent data annotation.

[0054] Specifically, the implementation scheme for acquiring multi-temporal point clouds provided in this embodiment of the invention is as follows: taking the acquisition of 3D real-world point clouds from the School of Information Science at Wuhan University as an example, and using a DJI M300 RTK + CERNET 102S as an example, firstly, based on the average ground elevation of the School of Information Science at Wuhan University and the actual size requirement of each pixel, the ground resolution is determined to be 1.5cm, the flight altitude to be 96 meters, the flight speed to be 11.9m / s, and the operating area to be 0.35km². 2The daily workload is 5.3 km. 2 The oblique photogrammetry imagery acquisition for the School of Information Science at Wuhan University can be completed within a day. If multi-drone coordinated acquisition is required, it's important to note that during flight path design, the forward overlap should generally be 60%–80%, with a minimum of 53%; the lateral overlap should generally be 15%–60%, with a minimum of 8%. When dealing with sparsely built-up areas, the minimum forward and lateral overlap for the UAV's survey area should be no less than 70%; while when facing densely built-up areas, the image overlap can be designed to be at most 80%–90%. After acquisition, 3D point cloud reconstruction can be performed. Based on the Smart3D algorithm, a control interval can be provided from the final aerial triangulation feature point cloud level, i.e., one control point is placed every 20,000 to 40,000 pixels. If differential POS data exists (which represents a relatively accurate initial value), this interval can be widened to 40,000 pixels; if differential POS data is unavailable, at least one control point should be placed every 20,000 pixels. Furthermore, the rules should be applied flexibly according to the actual terrain and features of each task. For example, in cases where the terrain is particularly undulating, there is a large area of ​​vegetation, and the feature points of the surface water are extremely scarce, control points should be added as appropriate. The control points are measured by traverse surveying to obtain high-precision location information.

[0055] Specifically, this embodiment of the invention also provides preprocessing for 3D point clouds. Before data annotation, the multi-temporal point cloud data collected in the above steps needs to be preprocessed, mainly including data cleaning and data registration. Data cleaning is mainly to remove noise and outliers from the data, reduce the impact of noisy data on the main data, and improve the quality and accuracy of the data. Data registration unifies the multi-temporal point cloud data into the same coordinate system, clarifies the correspondence between point clouds of different temporal periods, and facilitates subsequent annotation of changes. The overall process is as follows: Figure 2 As shown. The specific implementation process is explained below:

[0056] First, data cleaning is performed, primarily using statistical filtering to clean the point cloud data, removing noise and outliers to reduce the impact of noise on the main data and improve data quality and accuracy. Statistical filtering reduces noise by analyzing the statistical characteristics of the point cloud, improving its quality. This requires determining the neighborhood of each point in the given input point cloud. For each point, a suitable neighborhood is defined, such as a spherical neighborhood with a certain radius centered on the point, or a cubic neighborhood based on spatial coordinates. The size of this neighborhood is determined comprehensively based on the characteristics of the point cloud data, the scale of the target object, and the noise level, preparing for subsequent statistical analysis of the relevant features of points within the neighborhood. Within the defined neighborhood, key features of each point are statistically analyzed, commonly including distances from the point to other points in the neighborhood. For example, the distances from a point in the point cloud to all other points in its spherical neighborhood are calculated, and then the mean, standard deviation, and other statistical measures of these distances are calculated. Finally, based on the statistically obtained features and the desired cleaning effect of the point cloud data, appropriate thresholds are set. For example, a distance threshold can be set as a multiple of the average distance. If the distance from a point to other points in its neighborhood exceeds this threshold, the point is initially identified as a potential noise point. Setting the threshold usually requires experimentation and adjustment, as different point cloud datasets may have different suitable threshold ranges. Finally, based on the set threshold, points marked as noise or outliers are removed from the original point cloud data, resulting in cleaned point cloud data with relatively less noise. Point cloud registration then proceeds, including manual coarse alignment and automatic fine registration. Since the coordinate systems of point clouds from different time periods may differ significantly, leading to large distances between point clouds from different time periods, it is necessary to manually select multiple feature points and perform preliminary alignment of the multi-time-series point clouds based on the correspondence between these feature points. On this basis, the Iterative Closest Point algorithm is used to achieve precise registration of the multi-time-series point clouds.

[0057] Specifically, the implementation scheme for preprocessing three-dimensional point clouds provided in this embodiment of the invention is as follows:

[0058] Taking the real-world point cloud obtained from the School of Information Science at Wuhan University as an example, the data is first cleaned. This method determines whether a point is noise by calculating the distance from each point to the underlying surface. If a point is too far from the fitted plane, it is considered noise, which can be seen as a low-pass filter. When estimating the distance from the fitted plane, the neighborhood points of each point need to be determined. This method uses K-nearest neighbors to determine the neighborhood set of any point in the point cloud. Based on the density of the point cloud, the number of nearest neighbors is determined to be 6 for K-nearest neighbor judgment. Furthermore, a maximum input error needs to be determined during the judgment to determine whether the input point is judged as noise. The error can be relative or absolute. This embodiment uses relative error, and the maximum input error is set to 1.0. Then, data registration is performed. During manual coarse alignment, this embodiment selects 3 feature points, preferably those with obvious features such as building inflection points, for coarse alignment. Then, the Iterative Closest Point algorithm is used, with the RMS difference set to 1.0 × 10⁻⁶. -5 The final coverage rate reached 95%. The average spacing between the multi-temporal point clouds after registration was small, reaching 0.2m, which effectively achieved the correspondence of multi-temporal data.

[0059] Specifically, this embodiment of the invention also provides a method for coarsely screening change information using an octree algorithm. Change information is labeled based on the output multi-temporal point cloud pairs. To minimize the difficulty and cost of change information labeling, a semi-automated labeling method is proposed. For point clouds of different temporal sequences, the resolution of the octree detection needs to be set according to their distribution characteristics to balance the false changes caused by differences in detection accuracy and density, thereby screening out as many true changes as possible. The semi-automated labeling process is as follows: Figure 3 As shown. The specific implementation process is explained below:

[0060] The octree algorithm effectively filters out potential changes in multi-temporal point clouds, providing information guidance for subsequent data annotation and reducing its difficulty. First, an octree is constructed for each temporal point cloud based on parameters such as the maximum recursion depth. During construction, the cube space represented by the root node of the octree is determined according to the spatial range covered by the point cloud data. This cube space is then recursively divided into eight sub-cubes until a predetermined maximum depth is reached or the number of points in each sub-cube meets a certain threshold, thus constructing a complete octree structure. Next, relevant features, such as point cloud density, normal vector, and curvature, are extracted for each node or the corresponding point cloud region. These features serve as the basis for judging changes. Finally, the octree is constructed by dividing the space into voxels. The density of points within each voxel is calculated based on the number of points and the volume of the voxel (distribution location is ignored since it's a voxel), serving as a comparative feature for octree nodes. Comparisons are made at corresponding locations to determine the area and degree of change in 3D space. A certain threshold can be set, and when the feature difference exceeds the threshold, it is determined that a change has occurred.

[0061] Specifically, the implementation scheme for coarsely screening change information using the octree algorithm provided in this embodiment of the invention is as follows:

[0062] The Point Cloud Library (PCL) open-source library is used to perform octree-based 3D change detection. The PointCloud() data structure is used to store the read multi-temporal point cloud data. The make_octreeChangeDetector() function is used to construct the octree structure for the first phase of point cloud data. The resolution of the octree is set to 3.0 based on the density of the input point cloud. Then, the second phase of point cloud data is read in the same way, and the second phase of point cloud data is input into the octree structure based on the first phase of point cloud data. The get_PointIndicesFromNewVoxels() function can be used to obtain the changes in the second phase of point cloud data, and the changes in the first phase of point cloud data can be obtained similarly. When outputting, it is important to output changed points and unchanged points separately to facilitate subsequent correction of changes in the data.

[0063] Specifically, this embodiment of the invention also provides manual correction of change information. The coarse-screened change information obtained in the previous step is not accurate, and it is necessary to manually check this change information, correct false change samples, and supplement true change samples. Then, the data processed by different people are cross-validated to ensure data consistency and accuracy. In addition, during the data annotation process, it is also necessary to clarify and unify the annotation principles to ensure that the annotated changes can provide effective, clear, and consistent information support. An example of an annotated 3D change dataset is shown below. Figure 4As shown. The specific implementation process is explained below:

[0064] Manual correction of changes primarily involves checking and correcting changes identified in the initial screening. To ensure data usability and validity, data correction must adhere to the following principles: 1) Point clouds from different time periods must exhibit clear changes at corresponding locations, and these changes must align with the needs of social and economic activities, such as building project management and tree monitoring. It's important to note that inconsistencies in point cloud density caused by sampling and occlusion cannot be classified as changes. 2) Focus should be placed on three types of changes: additions, reductions, and replacements. Additions refer to changes from non-existent to existing objects, such as the addition of a building; reductions refer to changes from existing to non-existent objects, such as the withering of a tree; and replacements refer to changes in the category of an object at the same location, such as changing from a tree to a building. Following a process of first correcting misclassified points and then supplementing changed points during the correction process will lead to better data correction.

[0065] Specifically, the implementation scheme for manually correcting change information provided in this embodiment of the invention is as follows:

[0066] Manually correcting changes is a data annotation task, which is implemented using Cloud Compare software in this embodiment of the invention. For multi-time-series point cloud data from the School of Information Science at Wuhan University, corrections are performed sequentially across different time series. When processing each time-series point cloud, firstly, point clouds identified as changed by the octree algorithm are checked. By comparing different time-series point clouds, false changes caused by density and occlusion are corrected, and these are reclassified as unchanged data. This type of data is mostly found in objects such as buildings and vegetation. Then, point clouds identified as unchanged by the octree algorithm are checked. Similarly, by comparing different time-series point clouds, false unchanged samples caused by improper octree resolution settings are corrected, and these are reclassified as changed data. This type of data is mostly found in small-volume objects in cities, such as vehicles and street facilities. Finally, change data annotated by different individuals are cross-validated to further ensure the accuracy of the change data annotation.

[0067] The implementation of the various embodiments of the present invention is based on programmed processing through a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a rapid construction system for a 3D change detection dataset for urban scenes. This system is used to execute the method for constructing a 3D change point cloud dataset for urban scenes in the above method embodiments.

[0068] The system includes: a data acquisition module for acquiring multi-temporal 3D change point cloud data for urban scenarios and performing preprocessing; a coarse screening module for using an octree algorithm to coarsely screen the preprocessed multi-temporal 3D change point cloud data; and a dataset construction module for obtaining the final labeled multi-temporal point cloud data based on the coarsely screened multi-temporal 3D change point cloud data through manual annotation and cross-validation, thus completing the construction of the 3D change point cloud dataset.

[0069] The present invention provides a rapid construction system for a 3D change detection dataset for urban scenarios. Addressing the issues of high cost and difficulty in data annotation and data inaccuracies, the system employs several modules. Data preprocessing is completed through methods such as data cleaning and point cloud registration. Then, an octree algorithm is used to coarsely screen change information from multi-temporal point cloud data. The coarsely screened change information is manually annotated and corrected. The annotated data is cross-validated to ensure accuracy, thereby efficiently and accurately constructing a 3D change point cloud dataset.

[0070] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention also provides an electronic device, including a memory and a processor. The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize a method for constructing a three-dimensional change point cloud dataset for urban scenes as proposed in the above embodiments.

[0071] This invention also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, this program overcomes the problems of high cost and difficulty in data annotation, acquires a real dataset describing changes in urban scenes, and constructs a three-dimensional change point cloud dataset. The storage medium can be any non-volatile storage device such as a hard disk, solid-state drive, flash drive, or optical disk, used to store computer program code and necessary data files. The stored computer program includes: a data acquisition module, a coarse screening module, and a dataset construction module.

[0072] Finally, it should be noted that the above specific embodiments are merely representative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments and many variations are possible. Any simple modifications, equivalent changes, and alterations made to the above specific embodiments based on the technical essence of the present invention should be considered within the protection scope of the present invention.

Claims

1. A method for constructing a three-dimensional varying point cloud dataset for an urban scene, characterized in that, The method comprises the following steps: Step 1, collecting multi-time series three-dimensional change point cloud data for urban scenes, and preprocessing; Step 2, using an octree algorithm to coarsely screen the multi-time series three-dimensional change point cloud data after preprocessing, specifically including: Firstly, according to the spatial range of the point cloud data, an octree is constructed for the point cloud data of each time series; Secondly, related features are extracted for each node of the octree or the point cloud region corresponding to the node; Finally, the octree divides the space to form voxels, and then, for the points located in the voxels, the density is calculated according to the number of points and the volume of the voxels, and the corresponding positions are compared to determine the changed region and degree in the three-dimensional space; by setting a certain threshold, when the feature difference exceeds the threshold, the changed point cloud data is obtained; when the feature difference does not exceed the threshold, the unchanged point cloud data is obtained; Step 3, based on the multi-time series three-dimensional change point cloud data after coarse screening, labeling and cross-validation are performed to obtain the final labeled multi-time series three-dimensional change point cloud data, and a three-dimensional change point cloud data set for urban scenes is constructed; wherein, when processing each time series point cloud, firstly, the point cloud data determined as changed by the octree algorithm is compared with the point cloud data of different time series, and the false change data caused by density and occlusion is corrected and classified as unchanged data, then, the point cloud data determined as unchanged by the octree algorithm is also compared with the point cloud data of different time series, and the false unchanged sample caused by improper setting of the resolution of the octree is corrected and classified as changed data, and finally, the changed data labeled by different persons is cross-verified. 2.The method of claim 1, wherein, The construction of the octree comprises: According to the spatial range of the point cloud data, a cube space represented by the root node of the octree is determined; Based on the cube space, eight sub-cubes are obtained, and each sub-cube reaches a predetermined maximum recursion depth or the number of point clouds in each sub-cube meets a certain threshold, thereby obtaining the constructed octree. 3.The method of claim 1, wherein, After preprocessing, point cloud registration is further performed: A plurality of feature points are selected, the multi-time series three-dimensional change point cloud data is preliminarily aligned, and the point cloud data after preliminary alignment is precisely registered by using an iterative closest point algorithm.

4. A city oriented three dimensional varying point cloud dataset construction system, characterized in that, The method comprises: A data acquisition module is configured to collect multi-time series three-dimensional change point cloud data for urban scenes, and perform preprocessing; A coarse screening module is configured to use an octree algorithm to coarsely screen the multi-time series three-dimensional change point cloud data after preprocessing, specifically including: Firstly, according to the spatial range of the point cloud data, an octree is constructed for the point cloud data of each time series; Secondly, related features are extracted for each node of the octree or the point cloud region corresponding to the node; Finally, the octree divides the space to form voxels, and then, for the points located in the voxels, the density is calculated according to the number of points and the volume of the voxels, and the corresponding positions are compared to determine the changed region and degree in the three-dimensional space; by setting a certain threshold, when the feature difference exceeds the threshold, the changed point cloud data is obtained; when the feature difference does not exceed the threshold, the unchanged point cloud data is obtained; A data set construction module is configured to perform labeling and cross verification based on the multi-temporal three-dimensional change point cloud data after coarse screening, obtain the final labeled multi-temporal three-dimensional change point cloud data, and obtain the constructed three-dimensional change point cloud data set for urban scenes; wherein, when processing each time sequence point cloud, firstly, the point cloud data determined as changed by the octree algorithm is checked, the different time sequence point clouds are compared, the false change data caused by density and occlusion is corrected and classified as unchanged data, then the point cloud data determined as unchanged by the octree algorithm is checked, the different time sequence point clouds are compared, the false unchanged sample caused by improper setting of the octree resolution is corrected and classified as change data, and finally, the change data labeled by different persons is cross verified. 5.An electronic device comprising a memory and a processor, the memory storing a computer program, wherein, The computer program is executed by the processor to implement the steps of the three-dimensional change point cloud data set construction method for urban scenes in any one of claims 1-3.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the three-dimensional change point cloud data set construction method for urban scenes in any one of claims 1-3.

Images