Point cloud fusion quality detection method, device, storage medium and program product

Abstract

This specification provides a point cloud fusion quality inspection method, device, storage medium, and program product. The point cloud fusion quality inspection method includes: extracting local geometric feature values ​​of 3D points from a global point cloud obtained by multi-view point cloud fusion; performing initial clustering in the local geometric feature space based on the local geometric feature values ​​of the 3D points in the global point cloud to obtain at least one first cluster; identifying the first clusters that meet the overlapping region feature conditions as candidate overlapping clusters, where the overlapping region feature conditions are used to filter out 3D points located within the overlapping regions of the multi-view point cloud; performing secondary clustering in the 3D space based on the 3D coordinates of the 3D points contained in all candidate overlapping clusters to obtain at least one second cluster, where each second cluster corresponds to a continuously distributed overlapping region in the 3D space; and determining whether there are fusion anomalies in the overlapping region corresponding to each second cluster based on the fusion quality index value of each second cluster.

Description

Technical Field

[0001] This specification relates to the field of point cloud processing technology, and in particular to a point cloud fusion quality detection method, electronic device, computer-readable storage medium, and computer program product. Background Technology

[0002] In fields such as intelligent driving, robot navigation, digital twins, and engineering surveying, high-precision and complete 3D environment models are the foundation for perception, decision-making, and interaction. 3D sensors such as LiDAR actively detect and generate discrete point cloud data, becoming the core data source for constructing such models. To eliminate occlusion and field-of-view limitations inherent in single-view acquisition, multi-sensor arrays or single-sensor multi-station scanning are commonly used in practice to observe the same scene from different spatial locations. Subsequently, point cloud registration and fusion techniques are employed to generate a comprehensive, single global point cloud.

[0003] However, in the process of multi-source data fusion, the common observation area that inevitably exists between adjacent viewpoints or sensors, namely the overlapping area, becomes a challenge for quality control. In actual operations, due to factors such as inherent sensor noise, system calibration residuals, timing synchronization errors, and limitations of registration algorithms, point cloud data from different sources often cannot achieve complete spatial alignment and consistent representation in the overlapping area.

[0004] Currently, the quality assessment of the fused global point cloud, especially in overlapping areas, relies heavily on the experience and visual inspection of operators. The traditional process involves manually browsing and comparing in 3D rendering software to judge whether there are obvious misalignments or anomalies. This method is inefficient, highly subjective, and difficult to fully cover all potential problem areas in a large-scale point cloud. Summary of the Invention

[0005] In view of the above, one or more embodiments of this specification provide the following technical solutions: According to a first aspect of one or more embodiments of this specification, a point cloud fusion quality detection method is proposed, comprising: Obtain a global point cloud obtained by fusing point clouds from multiple perspectives, and extract the local geometric feature values ​​of each three-dimensional point in the global point cloud; Based on the local geometric feature values ​​of each three-dimensional point in the global point cloud, the global point cloud is first clustered in the local geometric feature space to obtain at least one first cluster. For the at least one first cluster, the first cluster that meets the overlapping region feature condition is determined as a candidate overlapping cluster, and the overlapping region feature condition is used to filter out three-dimensional points in the overlapping region of the multi-view point cloud. Based on the three-dimensional coordinates of the three-dimensional points contained in all the candidate overlapping clusters, a secondary clustering is performed in the three-dimensional space to obtain at least one second cluster, and any second cluster corresponds to an overlapping region that is continuously distributed in the three-dimensional space. Based on the fusion quality index value of each second cluster, it is determined whether there is a fusion anomaly in the overlapping region corresponding to the second cluster.

[0006] According to a second aspect of the embodiments of this specification, an electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; Wherein, when the processor executes the executable instructions, it is used to implement the method described in the first aspect.

[0007] According to a third aspect of the embodiments of this specification, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps of the method described in the first aspect.

[0008] According to a fourth aspect of the embodiments of this specification, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0009] As can be seen from the above embodiments, this specification extracts the local geometric feature values ​​of 3D points in the global point cloud obtained by multi-view point cloud fusion. It obtains at least one first cluster by performing initial clustering in the local geometric feature space. Based on the overlapping region feature conditions, it selects candidate overlapping clusters that may contain 3D points in the overlapping region of the multi-view point cloud from at least one first cluster. Then, it performs secondary clustering of the 3D points in all candidate overlapping clusters in 3D space to separate independent overlapping regions (second clusters) that are continuously distributed in 3D space. Finally, it calculates the fusion quality index value for each independent overlapping region to objectively determine whether there is a fusion anomaly in the overlapping region.

[0010] By adopting a two-level clustering strategy of "first clustering in local geometric feature space + second clustering in three-dimensional space", the first clustering accurately screens potential three-dimensional points in the overlapping areas of multi-view point clouds to eliminate invalid interference. The second clustering separates independent overlapping areas, realizing accurate positioning and independent evaluation of overlapping areas. The two-level clustering works together to achieve both the accuracy of overlapping point screening and the accuracy of overlapping area positioning, significantly improving the accuracy of fusion anomaly detection.

[0011] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this specification. Attached Figure Description

[0012] Figure 1 This is a flowchart of a point cloud fusion quality detection method provided in an exemplary embodiment.

[0013] Figure 2 This is a schematic diagram of a multi-angle scanning rectangular box acquisition scenario provided in an exemplary embodiment.

[0014] Figure 3 This is a schematic diagram of a global point cloud obtained by point cloud fusion based on multi-view scanning data, provided as an exemplary embodiment.

[0015] Figure 4 This is a schematic diagram illustrating an anomalous fusion regions in a global point cloud, provided as an exemplary embodiment.

[0016] Figure 5 This is a schematic diagram of the structure of a device provided in an exemplary embodiment. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0018] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this manual are all information and data authorized by the user or fully authorized by all parties. The collection, use and processing of related data shall comply with relevant laws, regulations and standards, and corresponding operation entry points shall be provided for users to choose to authorize or refuse.

[0019] The point cloud fusion quality inspection method provided in the embodiments of this specification is mainly applied to scenarios where, after fusing 3D point clouds acquired by multiple sensors or from multiple perspectives, it is necessary to perform automated quality assessment and problem localization on the fused global point cloud. Such scenarios are widely found in fields such as intelligent driving, robot environmental perception, digital twin city construction, 3D inspection of large structural components, and the digitization of cultural heritage.

[0020] For example, in the real-vehicle testing and data acquisition phase of intelligent driving systems, to construct a 360-degree environmental model around the vehicle without blind spots, multiple LiDAR sensors are typically installed on the roof or around the vehicle. These sensors simultaneously scan the surrounding environment from different perspectives, each generating a point cloud. Subsequently, calibration and registration algorithms are used to unify these point clouds in different coordinate systems to the vehicle coordinate system, fusing them into a complete global point cloud. However, due to the time synchronization errors of the sensors themselves, the slight displacements caused by mechanical vibrations, the residuals of calibration parameters, and the accuracy limitations of the registration algorithm, point cloud data belonging to the same physical surface from different sensors may not be perfectly aligned in the fused global point cloud, but rather exhibit spatial offsets, ghosting, or redundant superposition. These fusion defects directly interfere with the accuracy of subsequent advanced perception algorithms such as point cloud segmentation, target recognition, and tracking. Traditionally, these problems rely on engineers performing manual visual inspection in 3D display software, which is inefficient, highly subjective, and difficult to guarantee comprehensive quality control. Furthermore, in actual point cloud annotation scenarios, only the fused point cloud data can often be obtained, and the original acquisition perspective information (such as the source sensor or orientation label of each 3D point) is often not retained, which further increases the difficulty of identifying overlap problems.

[0021] Based on this, please refer to Figure 1 This specification provides a point cloud fusion quality inspection method, which can be executed by an electronic device, including but not limited to servers, personal computers, workstations, smartphones, tablets, or embedded devices; or dedicated devices for specific application scenarios, such as servers for intelligent driving systems, industrial control computers and edge servers for industrial quality inspection, and cloud computing resources and data centers. The point cloud fusion quality inspection method includes: In S100, for the global point cloud obtained by fusing multi-view point clouds, local geometric feature values ​​of 3D points in the global point cloud are extracted.

[0022] Multi-view point cloud fusion refers to the process of unifying point cloud data from different spatial locations, times, or sensors into the same coordinate system through coordinate transformation and merging the data, resulting in a global point cloud. Due to factors such as sensor calibration errors, registration algorithm accuracy limitations, and environmental interference, point clouds from different sources may not achieve ideal geometric consistency in overlapping areas.

[0023] For example, before extracting the local geometric feature values ​​of each 3D point in the global point cloud, the electronic device can preprocess the global point cloud to improve detection accuracy.

[0024] In the first possible preprocessing method, the electronic device counts the number of neighboring points within a preset radius for each 3D point in the global point cloud. If the number of neighboring points is lower than a set threshold, it is identified as an outlier and removed. The radius can be set according to the actual application scenario. For example, it can be determined based on the sensor type. For instance, for vehicle-mounted LiDAR used to collect road scene point clouds, the radius is set to 0.1~0.5m; for industrial structured light scanners used for quality inspection of small workpieces, the radius is set to 0.001~0.01m; and for LiDAR used to collect building / terrain point clouds, the radius is set to 0.5~2m.

[0025] In the second possible preprocessing method, it is assumed that the average neighborhood distance of effective 3D points follows a normal distribution. This assumption is consistent with the distribution law of point cloud neighborhood distance in most engineering scenarios. Under the normal distribution, the proportion of samples corresponding to the preset confidence level will fall within the range of "mean ± quantile coefficient Z × standard deviation". The quantile coefficient Z is determined by the preset confidence level. For example, Z is 1.645 when the confidence level is 90%, Z is 1.96 when the confidence level is 95%, and Z is 2.58 when the confidence level is 99%. The above confidence levels and corresponding quantile coefficients are only illustrative examples and do not constitute specific limitations. For each 3D point in the global point cloud, the electronic device can calculate the average neighborhood distance of its k nearest neighbors (k is a preset positive integer); then, based on the average neighborhood distance of all 3D points, calculate the mean (μ) and standard deviation (σ) of the distance set; determine the corresponding quantile coefficient Z according to the preset confidence level, and then set the screening threshold: lower limit = μ - Z × σ, upper limit = μ + Z × σ; finally, retain the 3D points whose average neighborhood distance is within the range of [lower limit, upper limit], which is the confidence interval corresponding to the preset confidence level, and the points within the interval are determined to be valid points.

[0026] In the process of implementing the embodiments of this specification, the inventors discovered that the fusion defects in the overlapping areas include, but are not limited to: ghosting caused by slight misalignment of point clouds on the same physical surface, abnormally high point density, and contradictions in the fitting of local curved surfaces. In other words, the essence of the fusion defects in the overlapping areas is the inconsistency of local spatial geometric properties. That is, the point clouds in normal non-overlapping areas follow a unified geometric law in local space, such as surface continuity, consistent normal vectors, and uniform density. However, the overlapping areas, due to registration deviations and resampling issues of multi-view point clouds, destroy this local geometric consistency, which manifests as specific abnormal local geometric features.

[0027] Therefore, this step extracts the local geometric feature values ​​of each 3D point in the global point cloud. By capturing the local spatial structure information around each 3D point, the geometric consistency is quantified into a computable feature value, thus providing a data basis for distinguishing normal regions from overlapping abnormal regions.

[0028] For example, for any 3D point in a global point cloud, an electronic device can determine a set of k nearest neighbors, where k is a positive integer, consisting of the k nearest 3D points centered on that point. Then, based on the local neighborhood information determined by the 3D point and the set of k nearest neighbors, the local geometric feature values ​​of the 3D point are calculated. The physical surface represented by the 3D point cloud has locally continuous geometric properties (such as flatness / curvature, smoothness / roughness)—that is, the geometric features (such as normal vector, curvature) of a 3D point are determined by its immediate neighbors and are independent of distant 3D points. The set of k nearest neighbors defines the scope of the local neighborhood, ensuring that the extracted features accurately reflect the geometric properties of the local region where the target 3D point is located.

[0029] The types of local geometric eigenvalues ​​include at least one of the following: point density distribution characteristics calculated based on the distance distribution between a 3D point and each point in the k-nearest neighbor set; normal direction consistency characteristics calculated based on the consistency relationship of the normal vector directions between a 3D point and each point in the k-nearest neighbor set; surface fitting residual characteristics calculated based on the degree of deviation of each point from the fitted surface after fitting the surface to the k-nearest neighbor set; and local surface curvature characteristics calculated based on the degree of anisotropic curvature of the local surface represented by the k-nearest neighbor set.

[0030] (1) Point density distribution characteristics are used to measure the density of point clouds in the region where a certain three-dimensional point is located. The correlation between point density distribution characteristics and fusion defects includes: due to repeated sampling from multiple perspectives, the local point density in the overlapping area is significantly higher than that in the normal area, which is manifested as small average spacing and high density in the neighborhood.

[0031] For example, the point density distribution characteristic value of any three-dimensional point includes: the reciprocal of the average Euclidean distance between the three-dimensional point and the set of k nearest neighbors, where the average Euclidean distance is the average of the sum of the Euclidean distances between the three-dimensional point and each three-dimensional point in the set of k nearest neighbors.

[0032] For any three-dimensional point Its point density distribution characteristic value Calculate using the following formula: ;in, Point The arithmetic mean of the Euclidean distances between each point in its k-nearest neighbor set is calculated using the following formula: In this formula, Point of Nearest neighbor set, for Three-dimensional points in a nearest neighbor set, It is a positive integer.

[0033] (2) Normal direction consistency feature, used to evaluate the direction consistency of local surfaces. The correlation between normal direction consistency feature and fusion defects includes: points in normal areas are coplanar or cocurved surfaces, the normal vector direction is consistent, and the dispersion is small; in overlapping areas, due to point cloud misalignment, the normal vector direction is disordered and the dispersion is large.

[0034] For example, the eigenvalue of normal direction consistency for any 3D point includes: the average angle between the unit normal vector of the 3D point and the unit normal vectors of the 3D points in the k-nearest neighbor set. The unit normal vector is obtained by eigenvalue decomposition of the covariance matrix, which is constructed based on the k-nearest neighbor set. The larger the eigenvalue of normal direction consistency, the more inconsistent the local normals are.

[0035] For any three-dimensional point Its normal direction consistency eigenvalue Calculate using the following formula: In this formula, and Points and its neighboring points The unit normal vector. The unit normal vector is obtained in the following way: based on the point. of Constructing the covariance matrix from the nearest neighbor set : ;in, Let be the centroid of this nearest neighbor set. For the covariance matrix... Perform eigenvalue decomposition and take the eigenvector corresponding to the smallest eigenvalue as the unit normal vector of that point.

[0036] (3) Local surface curvature features are used to describe the degree of curvature of local surfaces. The correlation between local surface curvature features and fusion defects includes: the curvature change of normal continuous surfaces is smooth; the overlapping area has local surface wrinkles or unnatural curvature due to point cloud superposition, and the curvature value changes abruptly, showing an abnormally high curvature response.

[0037] For example, the local surface curvature eigenvalue of any three-dimensional point includes: the ratio of the minimum eigenvalue of the covariance matrix to the sum of N eigenvalues ​​of the covariance matrix, wherein the N eigenvalues ​​include at least the minimum eigenvalue, the maximum eigenvalue, and any intermediate eigenvalue between the minimum eigenvalue and the maximum eigenvalue.

[0038] For any three-dimensional point Its local surface curvature characteristic value Calculate using the following formula: In this formula, , , For point-based of Covariance matrix constructed from nearest neighbor point sets The three eigenvalues ​​satisfy the following conditions: When the curvature characteristic value of a local surface is close to 0, it indicates that the three-dimensional point is located in a plane or flat region, and when it is close to 1, it indicates that the three-dimensional point is located in a highly curved or noisy region.

[0039] (4) Surface fitting residual characteristics are used to measure whether local point sets are coplanar or cosurface. The correlation between surface fitting residual characteristics and fusion defects includes: points in normal regions fit the real physical surface, and the fitting residual is small; in overlapping regions, due to point cloud misalignment, some points deviate from the real surface, forming "floating points" or "misaligned structures", and the fitting residual increases significantly.

[0040] For example, the surface fitting residual feature of any three-dimensional point includes: the average distance from the three-dimensional points in the k nearest neighbor set of the three-dimensional point to the first fitted surface, the first fitted surface being obtained by surface fitting based on the k nearest neighbor set.

[0041] For any three-dimensional point Its surface fitting residual eigenvalues Calculate using the following formula: In the formula, Indicates a point of The first fitted surface is obtained by fitting a local surface to the nearest neighbor set; Point To the first fitted surface The vertical distance. Local surface fitting can be achieved using least-squares plane fitting or quadratic surface fitting methods.

[0042] In the above formulas, the parameters The value can be adjusted according to the point cloud density and scene scale, for example, the value range is... The calculation results of each feature value will be used for subsequent cluster analysis and anomaly detection.

[0043] In S102, based on the local geometric feature values ​​of the three-dimensional points in the global point cloud, the global point cloud is first clustered in the local geometric feature space to obtain at least one first cluster.

[0044] For example, after calculating the above four features, a four-dimensional local geometric feature vector is constructed for each three-dimensional point pi: All items have been normalized to ensure comparability. This local geometric feature vector is a collection of local geometric feature values ​​used to transform the multi-dimensional local geometric attributes of 3D points into computable structured data. This provides a foundation for subsequent operations such as clustering and similarity comparison in the local geometric feature space. It comprehensively reflects the following characteristics of 3D points: whether they are dense; whether their orientation is disordered (discrete normals); whether they exhibit geometrical abrupt changes (curvature); and whether they are difficult to fit (residuals).

[0045] In addition to the above-mentioned full feature combination method, any one or more of the above four local geometric features can be selected and combined according to the needs of the actual detection scenario. For example, point density distribution features and normal direction consistency features can be selected to construct a two-dimensional feature vector, or point density distribution features, surface fitting residual features and local surface curvature features can be selected to construct a three-dimensional feature vector. The embodiments in this specification do not limit the number and combination method of local geometric features, and it is not necessary to use the full combination of the four features.

[0046] This step performs preliminary grouping of the global point cloud based on the local geometric similarity of 3D points, thereby quickly filtering out regions that may generate outlier point sets due to multi-view overlap. After representing each 3D point as its local geometric feature vector, all points constitute a high-dimensional feature space. In this space, 3D points with similar local geometric features will cluster together.

[0047] For example, the initial clustering can employ a density-based clustering algorithm, such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise). This clustering algorithm can autonomously discover clusters of arbitrary shapes and effectively identify and eliminate outlier noise points in the feature space. It does not require pre-specifying the number of clusters and is suitable for processing point cloud data with unknown feature distributions.

[0048] In density-based clustering algorithms (such as DBSCAN), the neighborhood radius parameter (denoted as ε) and the minimum number of points parameter (denoted as MinPts) are the core parameters defining the density determination criteria. These two parameters work together to determine the formation rules of clusters, directly affecting the accuracy and completeness of identifying candidate overlapping clusters in the local geometric feature space of the point cloud. The neighborhood radius parameter represents the radius threshold of a spherical neighborhood defined within the local geometric feature space, centered on any feature vector (corresponding to a single 3D point in the global point cloud). The minimum number of points parameter represents the minimum number of feature vectors required to form a cluster within the ε-neighborhood centered on any feature vector (i.e., the number of points in the neighborhood threshold).

[0049] To improve the robustness of the algorithm under different resolutions and scenarios, the neighborhood radius parameter and minimum number of points parameter in the density-based clustering algorithm can be adaptively determined to avoid relying on manual parameter tuning.

[0050] For example, the neighborhood radius parameter (ε) in a density-based clustering algorithm can be dynamically determined as follows: Calculate the m-th nearest neighbor distance for each 3D point in the global point cloud, where m is a preset positive integer, such as a value ranging from 3 to 10. Calculate the distance set formed by the m-th nearest neighbor distances of all 3D points, and use the median or a specified quantile of this distance set, such as the 25th quantile or 75th quantile, as the neighborhood radius parameter value. The m-th nearest neighbor distance is the distance between feature vectors calculated in the local geometric feature space based on the local geometric feature vectors corresponding to the 3D points, such as Euclidean distance or Manhattan distance.

[0051] The minimum number of points parameter in the clustering algorithm is positively correlated with the total number of 3D points in the global point cloud. When the total number of points in the point cloud is large, the minimum number of points parameter increases accordingly, which can filter out invalid small clusters formed by a large number of noise points and prevent candidate overlapping clusters in real overlapping areas from being overwhelmed by noise. When the total number of points in the point cloud is small, the minimum number of points parameter decreases appropriately to ensure that sparse but real overlapping area feature points can form effective clusters and avoid missed detections.

[0052] In clustering algorithms, the minimum number of points parameter is positively correlated with the dimensionality of local geometric features. When the dimensionality of local geometric features is high, the sparsity of the feature space increases, and the minimum number of points parameter increases with the dimensionality. This requires more similar feature points in the neighborhood to form a core point, avoiding clustering distortion caused by "pseudo-core points" in high-dimensional space. When the feature dimensionality is low, the minimum number of points parameter decreases accordingly, adapting to the dense characteristics of low-dimensional space and ensuring that no effective feature clusters are missed.

[0053] The embodiments in this specification do not require manual adjustment of the minimum number of points parameter value for different point cloud sizes and different feature combinations. They can autonomously adapt to the diverse scenarios of multi-view fused point clouds, reducing the cost of engineering implementation. At the same time, they ensure that the clustering results meet the requirements of "retaining dense and effective clusters and filtering sparse and noisy clusters", providing a reliable foundation for the accurate identification of overlapping areas in the future.

[0054] During the initial clustering process, the electronic device can determine the neighborhood relationships in the feature space. If the distance between two local geometric feature vectors is less than the neighborhood radius parameter value ε, they are considered to be similar in features. Only when there are more points in a certain region than the minimum number of points parameter value is it considered to be a valid first cluster. Traversing all points, densely connected points in the local geometric feature space are divided into the same cluster, and sparse regions are marked as noise. After the clustering is completed, the connected and dense point group in the local geometric feature space is divided into a first cluster.

[0055] In S104, for at least one first cluster, the first cluster that meets the overlapping region feature condition is determined as a candidate overlapping cluster. The overlapping region feature condition is used to filter out three-dimensional points in the overlapping region of the multi-view point cloud.

[0056] In this embodiment, not all first clusters correspond to the fusion problem; only those clusters that simultaneously exhibit typical anomalous features of multi-view overlapping regions need to be marked. The overlapping region feature condition is a pre-defined combination of logical judgments based on the geometric features of multi-view fusion errors. The overlapping region feature condition is used to filter out 3D points located within the overlapping regions of the multi-view point cloud.

[0057] In one possible implementation, the overlapping region feature conditions include at least one of the following: the point density distribution feature value is greater than the density threshold, the consistency feature value with the normal direction is greater than the normal dispersion threshold, the local surface curvature feature value is greater than the curvature threshold, and the surface fitting residual feature value is greater than the residual threshold.

[0058] In another possible implementation, the overlapping region feature conditions include at least one of the following two combinations: (1) First combination: at least two of the following combinations: point density distribution feature value is greater than density threshold, normal direction consistency feature value is greater than normal dispersion threshold, local surface curvature feature value is greater than curvature threshold, and surface fitting residual feature value is greater than residual threshold; such as point density distribution feature value + normal direction consistency feature value, local surface curvature feature value + surface fitting residual feature value, etc.; (2) Second combination: any one of the following: point density distribution feature value is greater than density threshold, or normal direction consistency feature value is greater than normal dispersion threshold, or local surface curvature feature value is greater than curvature threshold, or surface fitting residual feature value is greater than residual threshold, and the number of three-dimensional points in the corresponding first cluster is greater than the preset cluster point number threshold. The above two combinations ensure comprehensive coverage of all fusion anomalies and reduce the risk of misjudgment of non-fusion anomaly points through "multi-feature combination" or "preset number threshold constraint", thereby improving the accuracy of candidate overlapping cluster identification.

[0059] By performing statistical analysis on the local geometric feature values ​​of points within each first cluster, such as determining the local geometric feature value corresponding to the center of the first cluster, or calculating the mean of the local geometric feature values ​​of most points in the first cluster, and comparing them with the corresponding preset threshold, it can be automatically determined whether the cluster is a candidate overlapping cluster that needs further investigation.

[0060] For example, if the center or most of the 3D points of a certain first cluster have a combination of "high density + high normal discretization + high curvature + large residual", it is determined to be a candidate overlapping cluster. The points in the cluster are "suspected overlapping points", that is, 3D points that may be located in the overlapping area of ​​the multi-view point cloud.

[0061] In S106, based on the three-dimensional coordinates of the three-dimensional points contained in all candidate overlapping clusters, a secondary clustering is performed in the three-dimensional space to obtain at least one second cluster. Each second cluster corresponds to an overlapping region that is continuously distributed in the three-dimensional space.

[0062] The initial clustering in the local geometric feature space is based on the similarity of local surface properties (such as normal direction and curvature) of points. The formation of a candidate overlapping cluster means that the points within the cluster have highly consistent local geometric features, which meets the necessary condition that "the points within the cluster are located within overlapping regions." However, in scenarios with multi-view fusion problems, 3D points with the same or similar geometric features may not all originate from the same continuous surface in 3D space, but may come from multiple overlapping regions with similar local geometric features but different spatial locations, such as multiple facades facing different directions in a building. Therefore, a candidate overlapping cluster aggregated in the local geometric feature space may implicitly contain multiple potentially overlapping region fragments that are discretely distributed in 3D space.

[0063] Performing 3D spatial clustering on each candidate overlapping cluster separately may result in unreasonable segmentation. For example, point clouds that belong to the same large plane (such as a complete wall) but are assigned to two different candidate overlapping clusters due to subtle differences in local geometric features will be forcibly processed separately, which may incorrectly classify the same real physical overlapping area as two independent regions.

[0064] Therefore, after completing the initial clustering and selecting candidate overlapping clusters that meet the characteristics of overlapping regions, the electronic device can merge the three-dimensional points contained in all candidate overlapping clusters and perform a unified secondary clustering in three-dimensional space based on the merged point set. In this three-dimensional space, distance-based clustering algorithms, such as DBSCAN based on Euclidean distance or hierarchical clustering, can naturally aggregate continuously distributed points in three-dimensional space into a cluster, thereby accurately reconstructing each independent and complete physical overlapping region, i.e., the second cluster, without being affected by the inherent segmentation bias that may be caused by feature space clustering, and realizing an accurate mapping from "screening of suspected overlapping points based on local geometric features" to "locating the real overlapping region based on spatial location".

[0065] In S108, based on the fusion quality index value of each second cluster, it is determined whether there is a fusion anomaly in the overlapping region corresponding to the second cluster.

[0066] This step involves quantitative evaluation and diagnosis of each identified overlapping region (i.e., the second cluster). One or more fusion quality index values ​​are calculated for each second cluster, which directly reflect the severity of fusion defects in that local area.

[0067] For example, the types of fusion quality index values ​​include at least one of the following: positional deviation index, density anomaly index, and surface continuity index.

[0068] (1) Position deviation index is used to measure whether the spatial distribution of point clouds in the overlapping area is consistent and reflects the registration accuracy.

[0069] For example, the positional deviation index value of any second cluster includes the average Euclidean distance from each 3D point within the second cluster to the centroid of the second cluster. The larger this value, the more dispersed the point cloud distribution is in that region, the worse the spatial alignment, and the more likely there is a registration misalignment.

[0070] Let C be the set of three-dimensional points contained in any second cluster, and let N be the number of points. Let the positional deviation index be denoted as _____. The calculation formula is: ;in, This represents the centroid of the second cluster, which is the arithmetic mean of the three-dimensional coordinates of all points within the cluster: .

[0071] (2) Density anomaly index, used to determine whether the point cloud is too dense due to the superposition of multiple views in the area.

[0072] The density anomaly index value for any second cluster includes the ratio of the average point density distribution characteristic value of all 3D points within the second cluster to the average point density distribution characteristic value of all 3D points within the reference point set; wherein, the reference point set includes 3D points in the global point cloud that do not belong to any second cluster. A value significantly greater than 1 (e.g., exceeding the threshold of 2.0) indicates severe data redundancy caused by repeated overlay of multiple views.

[0073] Density anomaly index value The calculation formula is: ;in: This represents the average local point density of the second cluster, calculated by considering each point within the cluster. Calculate its point density distribution characteristic value And obtain by taking the arithmetic mean: ; This represents the average local point density of the reference region. The reference normal region is selected from 3D points in the global point cloud that do not belong to any second cluster. The arithmetic mean of the point density distribution characteristic values ​​of all points within this reference point set is calculated, thus yielding the value. .

[0074] (3) Surface continuity index is used to analyze whether the surface of the overlapping area is smooth and whether the normal changes abruptly, reflecting the geometric fidelity after fusion.

[0075] The surface continuity index value of any second cluster includes any one of the following, or a weighted fusion value of the following two: ① The statistical values ​​of the normal vectors of all 3D points within the second cluster, including but not limited to: the variance or mean angle of the normal vectors of all 3D points within the second cluster. If the difference in normal vectors is too large, such as the mean angle > 30°, it indicates that the surface orientation is inconsistent. ② The average distance from the three-dimensional points within the second cluster to the second fitted surface represents the degree to which the points within the cluster deviate from the ideal fitted surface, i.e., the overall fitting residual. The larger the residual, the more discontinuous the surface. The second fitted surface is obtained by fitting the surface based on the second cluster.

[0076] If either of the above two values ​​or the weighted value is too high, it indicates that there is distortion or fracture on the surface and poor geometric consistency.

[0077] For example, a normalized fusion quality score Q∈[0,1] can be constructed by combining the above three indicators to quantify the overall fusion level, for example, expressed as: ;in, D represents the positional consistency score. pos The smaller the value, the higher the score; R represents the density rationality score. den The closer to 1, the higher the score; The score represents the continuity of the surface; a higher score is given for consistent normals and small residuals. w 1+ w 2+ w 3=1, and the weight can be adjusted according to the application scenario.

[0078] In addition to the above-mentioned full feature combination method, any one or more of the above three fusion quality indicators can be selected and combined according to the needs of the actual detection scenario. For example, only the position deviation indicator can be selected for evaluation, or the density anomaly indicator and the surface continuity indicator can be selected for combined evaluation. The embodiments in this specification do not limit the number and combination of fusion quality indicators, and it is not necessary to use the full combination of the above three fusion quality indicators.

[0079] For example, a preset threshold for the fusion quality index value can be determined in advance through experiments or statistics. If the fusion quality index value of each second cluster is greater than or equal to the preset threshold, the electronic device determines that there is no fusion anomaly in the overlapping area corresponding to the second cluster and can proceed to the next processing stage. If the fusion quality index value of each second cluster is less than the preset threshold, the electronic device determines that there is a fusion anomaly in the overlapping area corresponding to the second cluster and requires manual review or re-registration.

[0080] For example, a preset threshold for the fusion quality score can be determined in advance through experiments or statistics. If the fusion quality score of each second cluster is greater than or equal to the preset threshold, the electronic device determines that there is no fusion anomaly in the overlapping area corresponding to the second cluster and can proceed to the next processing stage. If the fusion quality score of each second cluster is less than the preset threshold, the electronic device determines that there is a fusion anomaly in the overlapping area corresponding to the second cluster and requires manual review or re-registration.

[0081] If a fusion anomaly is identified in the overlapping region corresponding to the second cluster, and only the positional deviation index exceeds the corresponding threshold, the anomaly type can be determined as spatial misalignment. If only the density anomaly index exceeds its preset threshold, the anomaly type is data redundancy. If only the surface continuity index is greater than the corresponding threshold, the anomaly type is geometric distortion.

[0082] In real-world scenarios, multiple anomalies may occur simultaneously, which can be judged by combining indicators: (1) Misalignment and distortion: When both the positional deviation indicator and the surface continuity indicator exceed the limit, it indicates that the point cloud has both spatial offset and that the surface geometry after fusion is destroyed. (2) Redundancy and misalignment: When both the density anomaly indicator and the positional deviation indicator exceed the limit, it indicates that there are multiple superpositions and that the superimposed layers are not aligned. (3) Redundancy and distortion: When both the density anomaly indicator and the surface continuity indicator exceed the limit, it indicates that the high-density point cloud has not formed a smooth surface, and the superposition may have introduced complex geometric noise.

[0083] If all three indicators exceed the limits, there is a serious misalignment, redundancy, and geometric distortion, which is a serious fusion failure.

[0084] For example, please refer to Figure 2 , Figure 3 as well as Figure 4 , Figure 2 An exemplary diagram illustrates a data acquisition scenario of scanning a rectangular box from multiple angles. Specifically, it shows the process of multiple sensors (such as LiDAR and 3D scanner) scanning the target rectangular box from different directions. The scanning ranges of each perspective overlap to form an overlapping area, providing basic data for subsequent point cloud fusion. Figure 3 An example is shown in the schematic diagram of the global point cloud obtained after point cloud fusion based on the above multi-view scanning data; Figure 4 An exemplary embodiment illustrates the point cloud fusion quality detection method provided in the embodiments of this specification, for... Figure 3The diagram shows the result obtained after detecting a global point cloud with fusion defects. Through steps such as cluster analysis based on local geometric features and candidate overlapping cluster determination, the fusion anomaly region in the global point cloud is accurately identified and clearly marked (as shown by the dashed box in the figure). This anomaly region embodies typical defects such as redundant stacking generated during the fusion of multi-view point clouds, and intuitively verifies the accurate identification capability of the detection method for fusion anomalies.

[0085] The various technical features in the above embodiments can be combined arbitrarily, as long as there is no conflict or contradiction between the combinations of features. However, due to space limitations, they are not described one by one. Therefore, the arbitrary combination of various technical features in the above embodiments is also within the scope of this specification.

[0086] Figure 5 This is a schematic structural diagram of a device provided in an exemplary embodiment. For example... Figure 5 As shown, device 500 mainly consists of a communication interface 502, a user interface 504, a processor 506, and a data storage 508. These components are interconnected and communicate with each other via a system bus, network, or other connection mechanism 510. The communication interface 502 enables device 500 to communicate with other devices, access networks, and transmission networks via analog or digital modulation. For example, the communication interface 502 may include a processing module and an antenna for wireless communication with a radio access network or access point. Furthermore, the communication interface 502 can be a wired interface such as Ethernet, Token Ring, or a USB port, or a wireless interface such as Wi-Fi, Bluetooth, Global Positioning System (GPS), or a wide-area wireless interface (e.g., WiMAX or LTE). Of course, the communication interface 502 can also support other forms of physical layer interfaces and standard or proprietary communication protocols. The communication interface 502 may also include multiple physical communication interfaces, such as Wi-Fi interfaces, Bluetooth interfaces, and wide-area wireless interfaces.

[0087] User interface 504 includes receiving user input and providing output to the user. Therefore, user interface 504 may include input components such as a keypad, keyboard, touch-sensitive or presence-sensitive panel, computer mouse, trackball, joystick, microphone, still camera, and video camera, and output components such as a display screen (which may be combined with a touch-sensitive panel), CRT, LCD, LED, display using DLP technology, printer, and other similar devices known or developed in the future. User interface 504 may also generate auditory output via speakers, speaker jacks, audio output ports, audio output devices, headphones, and other similar devices known or developed in the future. In some embodiments, user interface 504 may include software, circuitry, or other forms of logic capable of transmitting and receiving data from external user input / output devices. Additionally or alternatively, device 500 may support remote access from other devices via communication interface 502 or another physical interface (not shown). User interface 504 may be configured to receive user input, the position and movement of which may be indicated by indicators or cursors described herein. User interface 504 may also be configured as a display device for rendering or displaying text fragments.

[0088] Processor 506 may contain one or more general-purpose processors and / or special-purpose processors.

[0089] Data storage 508 may include one or more volatile and / or non-volatile storage components and may be integrated wholly or partially with processor 506. Data storage 508 may include removable and non-removable components.

[0090] Processor 506 is capable of executing program instructions 518 (e.g., compiled or uncompiled program logic and / or machine code) stored in data storage 508 to perform the various functions described herein. Data storage 508 may contain a non-transitory computer-readable medium on which program instructions are stored, which, when executed by device 500, enable device 500 to perform any methods, processes, or functions disclosed in this specification and / or the accompanying drawings. Execution of program instructions 518 by processor 506 may result in processor 506 using data 512.

[0091] For example, program instructions 518 may include an operating system 522 (e.g., an operating system kernel, device drivers, and / or other modules) installed on device 500 and one or more applications 520 (e.g., a browser, social application, or game application). Similarly, data 512 may include operating system data 516 and application data 514. Operating system data 516 is primarily accessible to the operating system 522, while application data 514 is primarily accessible to one or more applications 520. Application data 514 may reside in a file system visible or hidden from the user of device 500.

[0092] Application 520 can communicate with operating system 522 through one or more application programming interfaces (APIs). These APIs help application 520 read and / or write application data 514, transmit or receive information via communication interface 502, receive or display information on user interface 504, etc.

[0093] In some terminology, application 520 may be simply referred to as "app". Furthermore, application 520 can be downloaded to device 500 through one or more online app stores or app markets. However, applications can also be installed on device 500 in other ways, such as through a web browser or a physical interface on device 500 (e.g., a USB port).

[0094] In some embodiments, the point cloud fusion quality inspection device can be applied to, for example... Figure 5 The device shown implements the technical solution described in this specification. The point cloud fusion quality inspection device may include: The feature extraction module is used to extract local geometric feature values ​​of 3D points in the global point cloud obtained by fusing multi-view point clouds.

[0095] The initial clustering module is used to perform initial clustering of the global point cloud in the local geometric feature space based on the local geometric feature values ​​of the 3D points in the global point cloud, so as to obtain at least one first cluster.

[0096] The candidate overlapping cluster determination module is used to determine the first cluster that meets the overlapping region feature conditions as a candidate overlapping cluster for at least one first cluster. The overlapping region feature conditions are used to filter out three-dimensional points in the overlapping region of the multi-view point cloud.

[0097] The secondary clustering module is used to perform secondary clustering in three-dimensional space based on the three-dimensional coordinates of the three-dimensional points contained in all candidate overlapping clusters, to obtain at least one second cluster. Each second cluster corresponds to an overlapping region that is continuously distributed in three-dimensional space.

[0098] The fusion anomaly detection module is used to determine whether there is a fusion anomaly in the overlapping region corresponding to each second cluster based on the fusion quality index value of each second cluster.

[0099] In one implementation, the feature extraction module is specifically used to determine, for any three-dimensional point in the global point cloud, a set of k nearest neighbors consisting of the k three-dimensional points centered on the three-dimensional point and closest to the three-dimensional point, where k is a positive integer; and to calculate the local geometric feature value of the three-dimensional point based on the local neighborhood information determined by the three-dimensional point and the set of k nearest neighbors.

[0100] In one implementation, the types of local geometric feature values ​​include at least one of the following: point density distribution features, normal direction consistency features, local surface curvature features, and surface fitting residual features.

[0101] The point density distribution characteristic value of any three-dimensional point includes: the reciprocal of the average Euclidean distance between the three-dimensional point and the set of k nearest neighbors, where the average Euclidean distance is the average of the sum of the Euclidean distances between the three-dimensional point and each three-dimensional point in the set of k nearest neighbors.

[0102] The uniformity eigenvalue of the normal direction of any 3D point includes: the average angle between the unit normal vector of the 3D point and the unit normal vectors of the 3D points in the k nearest neighbor set. The unit normal vector is obtained by eigenvalue decomposition of the covariance matrix, which is constructed based on the k nearest neighbor set.

[0103] The local surface curvature eigenvalues ​​of any three-dimensional point include: the ratio of the minimum eigenvalue of the covariance matrix to the sum of the N eigenvalues ​​of the covariance matrix, wherein the N eigenvalues ​​include at least the minimum eigenvalue, the maximum eigenvalue, and any intermediate eigenvalue between the minimum eigenvalue and the maximum eigenvalue.

[0104] The surface fitting residual characteristics of any three-dimensional point include: the average distance from each three-dimensional point in the k nearest neighbor set to the first fitted surface, which is obtained by surface fitting based on the k nearest neighbor set.

[0105] In one implementation, the preset overlapping region feature conditions include at least one of the following two combinations: The first type of combination is a combination of at least two of the following: point density distribution characteristic value is greater than density threshold, normal direction consistency characteristic value is greater than normal dispersion threshold, local surface curvature characteristic value is greater than curvature threshold, and surface fitting residual characteristic value is greater than residual threshold. The second type of combination is any one of the following: the point density distribution characteristic value is greater than the density threshold, the normal direction consistency characteristic value is greater than the normal dispersion threshold, the local surface curvature characteristic value is greater than the curvature threshold, and the surface fitting residual characteristic value is greater than the residual threshold, and the number of three-dimensional points in the corresponding first cluster is greater than the preset number threshold.

[0106] In one implementation, the type of fusion quality index value includes at least one of the following: positional deviation index, density anomaly index, and surface continuity index.

[0107] The positional deviation index of any second cluster includes the average Euclidean distance from each three-dimensional point within the second cluster to the centroid of the second cluster.

[0108] The density anomaly index value of any second cluster includes: the ratio of the average point density distribution characteristic value of all three-dimensional points in the second cluster to the average point density distribution characteristic value of all three-dimensional points in the reference point set; wherein, the reference point set includes three-dimensional points in the global point cloud that do not belong to any second cluster.

[0109] The surface continuity index value of any second cluster includes: the statistical value of the normal vectors of all three-dimensional points in the second cluster, and / or the average distance from the three-dimensional points in the second cluster to the second fitted surface, which is obtained by surface fitting based on the second cluster.

[0110] In one implementation, the fusion anomaly judgment module is used to determine that there is no fusion anomaly in the overlapping area corresponding to the second cluster if the fusion quality index value of each second cluster is greater than or equal to a preset threshold; and to determine that there is a fusion anomaly in the overlapping area corresponding to the second cluster if the fusion quality index value of each second cluster is less than the preset threshold.

[0111] In one implementation, the initial clustering uses a density-based clustering algorithm.

[0112] The neighborhood radius parameter value in density-based clustering algorithms includes: the median or a specified quantile of the m-th nearest neighbor distance among all 3D points in the global point cloud, where m is a positive integer, and the m-th nearest neighbor distance is calculated based on the local geometric feature values ​​of the 3D points.

[0113] The minimum number of points parameter in the clustering algorithm is positively correlated with the total number of 3D points in the global point cloud, and / or the minimum number of points parameter in the clustering algorithm is positively correlated with the dimension of local geometric features.

[0114] For ease of description, the above devices are described by dividing them into various modules or units based on their functions. Of course, when implementing one or more of these specifications, the functions of each module or unit can be implemented in the same or different software and / or hardware, or a module that performs the same function can be implemented by a combination of multiple sub-modules or sub-units, etc. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0115] Based on the same concept as the methods described above, this specification also provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor performs the steps of the method as described in any of the above embodiments by executing the executable instructions.

[0116] Based on the same concept as the methods described above, this specification also provides a computer-readable storage medium having computer instructions stored thereon that, when executed by a processor, implement the steps of the methods as described in any of the above embodiments.

[0117] Based on the same concept as the methods described above, this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the methods as described in any of the above embodiments.

[0118] What those skilled in the art will understand is: In this specification, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitation, the presence of additional identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded.

[0119] In this specification, “a,” “an,” and “the” do not specifically refer to the singular, but may also include the plural.

[0120] In this specification, ordinal numbers such as "first," "second," etc., do not necessarily indicate order; they are often used to distinguish between objects. For example, "first server" and "second server" usually refer to two servers. To differentiate between these two servers, they are described as "first server" and "second server." Of course, sometimes these two servers may be the same server.

[0121] In this specification, unless explicitly stated otherwise, "receiving and sending data" does not necessarily mean direct receiving and sending; it can also mean indirect receiving and sending. For example, A receiving data sent by B can be understood as A directly receiving the data sent by B, or it can be understood as A indirectly receiving the data sent by B through other entities such as C. Similarly, B sending data to A can be understood as B sending the data directly to A, or it can be understood as B indirectly sending the data to A through other entities such as C. Here, C can be one entity, or it can be two or more entities.

[0122] In this specification, unless explicitly stated otherwise, the relationships between structures can be direct or indirect. For example, when describing "A is connected to B," unless it is explicitly stated that A and B are directly connected, it should be understood that A can be directly connected to B or indirectly connected to B. Similarly, when describing "A is on top of B," unless it is explicitly stated that A is directly above B (AB is adjacent and A is above B), it should be understood that A can be directly above B or indirectly above B (AB is separated by other elements, and A is above B). And so on.

[0123] This specification uses specific terms to describe embodiments thereof. Terms such as "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described herein, as well as the features of those different embodiments or examples, without contradiction.

[0124] Although one or more embodiments of this specification provide method steps as described in the embodiments or flowcharts, it is understood that the order of steps listed in the embodiments or flowcharts is only one of many possible execution orders and does not represent the only execution order. Therefore, when the claims involve method steps, any changes or adjustments to the order of such steps, or the parallelism between steps, are also within the scope of protection of the claims.

Claims

1. A method for detecting the quality of point cloud fusion, comprising: For the global point cloud obtained by fusing multi-view point clouds, local geometric feature values ​​of three-dimensional points in the global point cloud are extracted; Based on the local geometric feature values ​​of the three-dimensional points in the global point cloud, the global point cloud is first clustered in the local geometric feature space to obtain at least one first cluster. For the at least one first cluster, the first cluster that meets the overlapping region feature condition is determined as a candidate overlapping cluster, and the overlapping region feature condition is used to filter out three-dimensional points in the overlapping region of the multi-view point cloud. Based on the three-dimensional coordinates of the three-dimensional points contained in all the candidate overlapping clusters, a secondary clustering is performed in the three-dimensional space to obtain at least one second cluster, and any second cluster corresponds to an overlapping region that is continuously distributed in the three-dimensional space. Based on the fusion quality index value of each second cluster, it is determined whether there is a fusion anomaly in the overlapping region corresponding to the second cluster.

2. The method according to claim 1, wherein extracting the local geometric feature values ​​of three-dimensional points in the global point cloud comprises: For any three-dimensional point in the global point cloud, determine the k nearest neighbor set of the k three-dimensional points centered at the point and closest to it, where k is a positive integer; Based on the local neighborhood information determined by the 3D point and the set of k nearest neighbors, the local geometric feature value of the 3D point is calculated.

3. The method according to claim 2, wherein the type of local geometric feature value includes at least one of the following: point density distribution feature, normal direction consistency feature, local surface curvature feature, and surface fitting residual feature; The point density distribution characteristic value of any three-dimensional point includes: the reciprocal of the average Euclidean distance between the three-dimensional point and the set of k nearest neighbors, wherein the average Euclidean distance is the average of the sum of the Euclidean distances between the three-dimensional point and each three-dimensional point in the set of k nearest neighbors; The consistency feature value of the normal direction of any three-dimensional point includes: the average angle between the unit normal vector of the three-dimensional point and the unit normal vector of the three-dimensional points in the k nearest neighbor set, wherein the unit normal vector is obtained by eigenvalue decomposition of the covariance matrix, and the covariance matrix is ​​constructed based on the k nearest neighbor set; The local surface curvature eigenvalue of any three-dimensional point includes: the ratio of the minimum eigenvalue of the covariance matrix to the sum of the N eigenvalues ​​of the covariance matrix, wherein the N eigenvalues ​​include at least the minimum eigenvalue, the maximum eigenvalue, and any intermediate eigenvalue between the minimum eigenvalue and the maximum eigenvalue; The surface fitting residual characteristics of any three-dimensional point include: the average distance from the three-dimensional points in the set of the k nearest neighbors of the three-dimensional point to the first fitted surface, wherein the first fitted surface is obtained by surface fitting based on the set of the k nearest neighbors.

4. The method according to claim 3, wherein the overlapping region feature conditions include at least one of the following two combinations: The first type of combination is a combination of at least two of the following: point density distribution characteristic value is greater than density threshold, normal direction consistency characteristic value is greater than normal dispersion threshold, local surface curvature characteristic value is greater than curvature threshold, and surface fitting residual characteristic value is greater than residual threshold. The second type of combination is any one of the following: the point density distribution characteristic value is greater than the density threshold, the normal direction consistency characteristic value is greater than the normal dispersion threshold, the local surface curvature characteristic value is greater than the curvature threshold, and the surface fitting residual characteristic value is greater than the residual threshold, and the number of three-dimensional points in the corresponding first cluster is greater than the preset number threshold.

5. The method according to claim 1, wherein the type of the fusion quality index value includes at least one of the following: positional deviation index, density anomaly index, and surface continuity index; The positional deviation index value for any second cluster includes: The average Euclidean distance from each 3D point within the second cluster to the centroid of the second cluster; The density anomaly index value of any second cluster includes: the ratio of the average point density distribution characteristic value of all three-dimensional points in the second cluster to the average point density distribution characteristic value of all three-dimensional points in the reference point set; wherein, the reference point set includes three-dimensional points in the global point cloud that do not belong to any second cluster; The surface continuity index value of any second cluster includes: the statistical value of the normal vectors of all three-dimensional points in the second cluster, and / or, the average distance from the three-dimensional points in the second cluster to the second fitted surface, which is obtained by surface fitting based on the second cluster.

6. The method according to claim 1 or 5, wherein determining whether there is a fusion anomaly in the overlapping region corresponding to the second cluster based on the fusion quality index value of each second cluster includes: If the fusion quality index value of each of the second clusters is greater than or equal to a preset threshold, it is determined that there is no fusion anomaly in the overlapping area corresponding to the second cluster; If the fusion quality index value of each of the second clusters is less than the preset threshold, it is determined that there is a fusion anomaly in the overlapping region corresponding to the second cluster.

7. The method according to claim 1, wherein the initial clustering employs a density-based clustering algorithm; The neighborhood radius parameter values ​​in the density-based clustering algorithm include: The median or a specified quantile of the m-th nearest neighbor distances of all 3D points in the global point cloud, where m is a positive integer, and the m-th nearest neighbor distance is calculated based on the local geometric feature values ​​of the 3D points; The minimum number of points parameter value in the clustering algorithm is positively correlated with the total number of 3D points in the global point cloud, and / or the minimum number of points parameter value in the clustering algorithm is positively correlated with the dimension of local geometric features.

8. An electronic device, comprising: processor; A memory for storing processor-executable instructions; wherein the processor implements the steps of the method as described in any one of claims 1-7 by executing the executable instructions.

9. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the steps of the method as claimed in any one of claims 1-7.

10. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method as claimed in any one of claims 1-7.

Images