A Multi-Scale Adaptive Representation Method and System Combining Dynamic and Static Elements for Point Clouds

By preprocessing and semantic segmenting point cloud data, and dynamically adjusting loading and rendering, a multi-scale adaptive representation combining static and dynamic elements is achieved. This solves the problem of low efficiency in point cloud representation and interaction in existing technologies and meets the needs of users in different application scenarios.

CN120852764BActive Publication Date: 2026-11-17WUHAN HANNING TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202510705138.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-05-29
Publication Date
2026-11-17
Estimated Expiration
2045-05-29

AI Technical Summary

Technical Problem

Existing technologies fail to effectively consider the properties of point clouds themselves during point cloud representation and interaction, resulting in an inability to meet diverse user needs in different application scenarios. Furthermore, the lack of a representation method that combines static and dynamic elements reduces interaction efficiency.

Method used

By preprocessing, semantic segmentation, and representation level classification of point cloud data, the loading and rendering of point clouds are dynamically adjusted to achieve multi-scale adaptive representation that combines static and dynamic elements, and to optimize the processing of point clouds for different categories.

Benefits of technology

It improves the application efficiency and user experience of point cloud data, enabling the acquisition of key information in dynamic interactions and the clear presentation of important details in static displays, thus adapting to the user needs of different application scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120852764B_ABST
    Figure CN120852764B_ABST
Patent Text Reader

Abstract

This invention discloses a dynamic and static combined multi-scale point cloud adaptive representation method and system. The method includes solving the original point cloud data, preprocessing the solved point cloud data to remove noise and irrelevant information, and performing semantic segmentation on the preprocessed scene point cloud data; determining the representation level of the scene point cloud corresponding to each attribute label based on the semantic segmentation results of the scene point cloud, and classifying dynamic and static point clouds of different scales during the interaction process; calculating the representation level parameters of each attribute target point cloud in the current field of view based on the representation level of the point cloud; loading scene point clouds of different representation levels in the field of view based on the representation level parameters of each attribute target point cloud in the current field of view; dynamically adjusting the loading and rendering of point clouds during user interaction to obtain the adaptively represented point cloud scene; and significantly improving the user experience by optimizing the representation and interaction efficiency of point cloud data based on application needs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of computer vision technology, and more specifically, relates to a dynamic and static combined multi-scale point cloud adaptive representation method and system. Background Technology

[0002] With the continuous development of 3D point cloud technology, its application in numerous fields such as architecture, transportation, and power is becoming increasingly widespread. How to efficiently represent and interact with point cloud data has become a research hotspot. However, existing technologies primarily aim to improve the efficiency of point cloud display and interaction, employing octrees / quadtrees as data organization methods. They utilize multi-threading technology to load octree node files in real-time, selecting different forms or nodes for display based on the height or density of the viewpoint. In the representation and interaction process, existing technologies neglect the inherent attributes of the point cloud itself, failing to reduce or even delete unimportant targets. For example, during user navigation of tunnel point clouds, rail inspectors are more concerned with the track and its ancillary facilities, while the tunnel wall structure can be reduced or even deleted; power engineers are more concerned with power poles and other power facilities, but also require the display of the terrain along the route. In this case, point clouds with the aforementioned attribute requirements can be retained, while other point clouds can be reduced or deleted. Existing technologies cannot meet this optimization need for specific targets, resulting in users having to process a large amount of irrelevant information in practical applications, reducing interaction efficiency. Furthermore, existing technologies struggle to meet diverse user needs when facing different application scenarios. Most existing technologies focus only on dynamic interaction or static display, lacking a representation method that combines both. In practical applications, users often need to obtain key information during dynamic interaction while simultaneously hoping for clear presentation of important details in static displays. Existing technologies cannot meet both needs at the same time, limiting the application value of point cloud data.

[0003] In summary, existing technologies have many shortcomings in point cloud representation and interaction, failing to meet the diverse needs of users in different scenarios. Therefore, there is an urgent need for a point cloud representation method that can focus on the inherent properties of point clouds, is applicable to multiple scenarios, and combines static and dynamic representations, in order to improve the application efficiency of point cloud data and the user experience. Summary of the Invention

[0004] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a dynamic and static combined multi-scale adaptive point cloud representation method and system. Through steps such as point cloud data preprocessing and semantic segmentation, determining representation levels, calculating representation parameters at different representation levels, and adaptive scene point cloud representation, it achieves efficient representation and interaction of point cloud data. This invention focuses on the inherent properties of point clouds, is applicable to various scenarios, and combines dynamic and static representation methods, thereby improving the application efficiency and user experience of point cloud data. It solves the problem of low efficiency caused by the homogeneous representation and interaction of massive point cloud data in existing technologies that do not consider the semantics of massive point cloud data.

[0005] To achieve the above objectives, one aspect of the present invention provides a dynamic and static combined multi-scale point cloud adaptive representation method, comprising the following steps: S1. Solve the original point cloud data, preprocess the solved point cloud data to remove noise and irrelevant information, and perform semantic segmentation on the preprocessed scene point cloud data. S2. Based on the semantic segmentation results of the scene point cloud, determine the representation level of the scene point cloud corresponding to each attribute label, and classify the dynamic and static point clouds at different scales during the interaction process. S3. Based on the representation level of the point cloud, calculate the representation level parameters of the target point cloud of each attribute under the current field of view; the representation level parameters include the field of view coordinates, target scale, target type, indexing method, and loading level; S4. Based on the representation level parameters of the target point clouds of each attribute under the current field of view calculated in step S3, load scene point clouds of different representation levels into the field of view. During user interaction, dynamically adjust the loading and rendering of point clouds to obtain the point cloud scene after adaptive representation.

[0006] Furthermore, step S1 involves processing the original point cloud data, including reading the original point cloud data file, processing the point cloud data, and extracting the three-dimensional coordinates, intensity, and color information of the point cloud. Step S1 involves preprocessing the solved point cloud data to remove noise and irrelevant information, and performing semantic segmentation on the preprocessed scene point cloud data, including: S11. Use a pass-through filter to remove points that are out of range; S12. Use neighborhood-based statistical filtering to remove noise points; S13. Depending on the application scenario requirements, use other filtering methods to remove other non-target points; S14. Based on artificial intelligence algorithms and skilled workers, perform semantic segmentation on the preprocessed scene point cloud, assign different attribute labels to different categories of scene point clouds, and classify and organize the attribute labels. Step S11: Use pass-through filtering to remove points outside the bounding box; this includes setting appropriate bounding box parameters based on the original point cloud range, retaining all points within the bounding box, and deleting all points outside the bounding box. Step S12, using neighborhood-based statistical filtering to remove noise points; includes: Set the search radius to determine the neighborhood range of each point in the original scenic spot cloud; Within the neighborhood of a given point, calculate statistical parameters such as the number of points in the neighborhood, average distance, and curvature of each point to help identify outliers. Based on these statistical parameters, thresholds are set to classify the original point cloud and remove noise points; Step S14 includes: The preprocessed scene point cloud data is semantically segmented using an AI-based semantic segmentation algorithm to obtain a preliminarily classified scene point cloud. If the scene point cloud after preliminary classification cannot meet the application requirements, then skilled workers will perform fine classification on the scene point cloud after preliminary classification and check whether the semantic attribute labels of the current classification are correct. After semantic segmentation, attribute labels are assigned to each point in the scene point cloud; The output includes preprocessed point cloud data and semantically segmented point cloud data, with each point having a semantic label.

[0007] Further, step S2 includes: S21. Statistically analyze the attribute information of the scene point cloud after semantic segmentation and extract key statistical information; S22. Based on the semantic segmentation results of the scene point cloud, determine the representation level of the scene point cloud corresponding to the attribute label; S23. Based on the representation level, classify the dynamic point cloud and static point cloud at different scales during the interaction process; S24. Output the representation level (reserved points, downsampled points, ignored points) of each semantic category, as well as the classification results of dynamic and static point clouds.

[0008] Furthermore, the characterization level assessment method described in step S22 involves both manual rating and automatic rating; Manual rating involves skilled workers rating the segmented scene point clouds based on the user's specific needs and concerns, and classifying the attribute labels. According to the task requirements of the scene point clouds, they are divided into three categories: reserved points. Sampling points and ignored points ; Automatic rating is based on the density of point cloud features to rate the segmented scene point cloud and classify the attribute labels. The segmented scene point cloud is rated based on the density of point cloud features, and the attribute labels are classified; including: S221. Extract the average distance density, average intensity, curvature, and SHOT features of the point cloud; S222. Construct a complexity criterion using the extracted features and calculate the complexity score for each category; S223. Construct similarity criteria and calculate the similarity with known categories; S224. Automatically classify representation levels based on complexity score and similarity score; S225. Classify the attribute tags and determine the representation method for each category.

[0009] Furthermore, the average distance density of the point cloud in step S221 is calculated using equation (1): (1) in, For the first Average distance density of point clouds for each semantic segmentation category; Indicates the first Total number of point clouds in each semantic segmentation category; Point The minimum distance to other points is expressed by equation (2): (2) in, Represents any point in the current category of the scene point cloud. With any point The distance between them; The average intensity characteristic of the point cloud is calculated using equation (3): (3) in, Represents any point The intensity value; The curvature of the point cloud is expressed by equation (4): (4) in, The mean curvature; Gaussian curvature; and For two different principal curvatures; It is the maximum curvature of a point on a quadratic surface in all directions. It is the minimum curvature; , , , , , ; , , , , It is the partial differential of the curved surface; It is the unit vector of the normal vector.

[0010] Furthermore, the complexity criterion in step S222 is expressed by equation (5): (5) in, As a complexity evaluation metric; yes Normalized score; yes Normalized score; yes Normalized score; Normalized value of average distance density feature The weights; Normalized value of average intensity characteristics The weights; Normalized value of Gaussian curvature characteristic The weights; The complexity criterion in step S222 is constructed as follows: The point cloud density features, intensity features, and curvature features are normalized to obtain normalized scores; Weights are assigned to each feature based on the different application scenarios. Calculate the complexity score for each category; Step S223 The similarity criterion is expressed by equation (6): (6) in, Similarity index; The 352-dimensional descriptor obtained for calculating SHOT features; SHOT descriptors for known categories; Further, step S3 includes: S31. Mapping scene point cloud attribute labels to representation levels: Mapping the attribute labels of the semantically segmented scene point cloud to the target representation levels after classification. To establish a connection; S32. Based on the number of point clouds displayed on the given screen, calculate the representation level parameters of each attribute target point cloud under the current field of view. S33. During user interaction, dynamically adjust the representation parameters according to the changes in the loading level to ensure that the display effect of the point cloud meets the current field of view requirements. S34. Output the calculated representation level parameters for use by the subsequent point cloud loading and rendering modules.

[0011] Further, step S32 includes: S321: Determine the position and field of view of the current viewpoint; wherein, the field of view includes the horizontal angle and the vertical angle; calculate the horizontal angle and the vertical angle based on the position of the current viewpoint; Specifically, let the current viewpoint be... The screen width within the field of view is The height is The equivalent focal length is Then the horizontal perspective for: (7) Vertical perspective for: (8); S322: Calculate the visible range under the current field of view based on the viewpoint position and field of view range; wherein, the visible range under the current field of view includes the visible range diameter and the maximum visible range; The diameter of the visible range is calculated based on the distance from the viewpoint to the target and the viewing angle; the relationship between the diameter of the visible range and the field of view angle is: (9) in, For the field of view; The object distance is the straight-line distance from the viewpoint to the target plane. The maximum visible range is calculated based on the maximum visible range at the maximum distance from the viewpoint to the target: the maximum visible range at different distances is expressed by equation (10): (10) in, Maximum visible width; Maximum visible height; From a horizontal perspective With vertical perspective Substitute the proportion and By referring to equation (10), the horizontal and vertical coordinates of the maximum visible range can be obtained: If the target point and the viewpoint fall within this plane, then the layer in which the point is located needs to be loaded. S323. Determine the attributes and priorities of the target point cloud; based on the semantic segmentation results and mapping table, obtain the attribute labels and representation levels of each target point cloud within the current field of view; S324. Based on the attribute labels and representation levels of each target point cloud in the current field of view, calculate the representation level parameters of each attribute target point cloud in the current field of view; the representation level parameters of the target point cloud include the field of view coordinates, target scale, target type, indexing method and loading level; The calculation for each target point cloud loading level includes: Based on screen resolution and user needs, determine the maximum number of point clouds that can be displayed within the current field of view. ; Calculate the number of target point clouds to be displayed in the current field of view. This includes retained points, downsampled points, and ignored points; the number of target point clouds to be displayed in the current field of view. This can be expressed by equation (11): (11) in, It is the number of reserved points. It is the number of downsampling points. It is the downsampling ratio; The loading level is calculated based on the maximum number of point clouds that can be displayed in the current field of view and the number of points to be displayed in all target point clouds in the current field of view; expressed by equation (12): (12) in, This represents the highest level of detail in the point cloud data loading process, i.e., the level at which the point cloud is not downsampled and retains all details completely. This level is determined by the octree hierarchical organization method; ceil() is the floor function. If the value is less than 0, then display the level 1 index; Further, step S4 includes: S41. Obtain the calculated characterization level parameters from step S3; S42. Based on the representation level parameters, locate the point cloud data in the corresponding level folder; S43. Use multi-threading technology to load point cloud data to improve loading efficiency: S44. During the loading process, the point cloud data is subjected to data decoding, coordinate transformation, and index optimization. S45. Use the Qt interface and OpenGL to render the loaded point cloud data; S46. During user interaction, dynamically adjust the loading and rendering of the point cloud: S47. Output the rendered point cloud data to the display device to ensure that the user can see the adaptively represented point cloud scene.

[0012] A second aspect of the present invention provides a dynamic-static combined multi-scale point cloud adaptive representation system for implementing the aforementioned dynamic-static combined multi-scale point cloud adaptive representation method, comprising: The data input module is used to read and import raw point cloud data; The data processing module is used to preprocess the raw point cloud data, remove noise and irrelevant information, and optimize data quality. The semantic segmentation module is used to perform semantic segmentation on the preprocessed point cloud data and assign semantic labels to each point. The representation level classification module is used to determine the representation level of each semantic category based on the semantic segmentation results; The characterization parameter calculation module is used to calculate the characterization parameters of the target point cloud of each attribute under the current field of view according to the characterization level. The point cloud loading and rendering module is used to load and render point cloud data of different representation levels in the field of view according to the representation parameters. The user interaction module provides a user interface to support user interaction with point cloud data. System management module; responsible for the overall management and coordination of the system, ensuring the collaborative work of all modules.

[0013] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) The dynamic and static combined multi-scale point cloud adaptive representation method and system of the present invention has significant advantages over the prior art. The prior art mainly starts from the specific technology in the computer field, and improves the display and interaction efficiency of point clouds through data organization methods such as octree / quadtree and multi-threading technology. However, these methods ignore the inherent properties of point clouds and the user needs in different application scenarios, resulting in users needing to process a large amount of irrelevant information in actual applications, which reduces the interaction efficiency. The present invention starts from the perspective of application needs, and introduces semantic segmentation technology to perform detailed semantic classification of point cloud data, and assigns clear semantic labels to each point, thereby realizing the refined management of point cloud data. Through this semantic processing method and representation level division, the system can perform targeted optimization processing on different categories of point cloud data according to the actual needs of users and application scenarios. For example, it retains the high-density details of key target point clouds, performs downsampling processing on secondary targets, and even ignores irrelevant point clouds, thereby effectively reducing the amount of irrelevant information that users need to process during the interaction process and significantly improving the representation interaction efficiency. In the power industry, the system can prioritize loading data from power poles, power lines, and ground-level points along the lines; in the construction industry, the system can prioritize loading data from buildings and their surrounding ancillary facilities. This targeted optimization allows the system to better meet the needs of users in different application scenarios.

[0014] (2) Most existing technologies focus only on dynamic interaction or static display, lacking a representation method that combines dynamic and static elements. In practical applications, users often need to obtain key information during dynamic interaction, while simultaneously hoping to clearly present important details in static display. The dynamic-static combined multi-scale point cloud adaptive representation method and system of the present invention satisfies both needs simultaneously through a dynamic-static combined representation method. This dynamic-static combined representation method overcomes the limitations of the separation of dynamic interaction and static display in existing technologies, providing users with a more coherent and natural interactive experience. For example, when a user quickly browses a scene, the system can dynamically load and render key target point clouds, while when the user focuses on a specific area, the system can provide a high-resolution static display of that area, ensuring that the user can clearly observe detailed information.

[0015] (3) The dynamic and static combined multi-scale point cloud adaptive representation method and system of the present invention significantly improves the loading and rendering efficiency of point cloud data through multi-threading technology and efficient index optimization strategy. During user interaction, the system can respond to user operations in real time, dynamically adjust the loading level and indexing method of the point cloud, and ensure a smooth interactive experience under different field of view and interactive scenarios. This efficient processing method not only improves the user experience, but also enables the system to adapt to larger-scale point cloud data processing and more complex interactive scenarios. For example, in the power inspection scenario, a 1km long three-lane spliced ​​point cloud contains 932,568,571 points, while the target tower and facilities along the line only contain 78,530,297 points, accounting for 8.42%. Loading all point clouds with the same priority not only wastes a lot of resources, but may also cause lag during interaction. The present invention greatly improves display efficiency by prioritizing the loading of the target point cloud and dynamically downsampling or even ignoring secondary or unimportant points.

[0016] (4) The dynamic and static combined multi-scale point cloud adaptive representation method and system of the present invention has a high degree of flexibility and scalability. Through modular design, each functional module can run and be optimized independently, and it is also easy to customize and expand according to different application scenarios and user needs. For example, users can adjust the semantic segmentation algorithm, representation level division rules and loading and rendering strategies according to their own needs, so that the system can better adapt to various complex application scenarios.

[0017] (5) The dynamic and static combined multi-scale point cloud adaptive representation method and system of the present invention, although semantic segmentation or manual segmentation are relatively time-consuming steps, are mainly applied to point cloud files that need to be displayed repeatedly or point cloud files that have already been pre-classified. For these files, one-time preprocessing can bring an efficient experience in multiple interactions, and therefore has high practical value. For example, in scenarios such as power inspection and building modeling, point cloud data usually needs to be loaded and displayed multiple times, and the optimization strategy of the present invention can significantly improve the user experience. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a dynamic and static combined multi-scale point cloud adaptive representation method according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the internal logic of a dynamic-static combined multi-scale point cloud adaptive representation method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a dynamic and static combined multi-scale point cloud adaptive representation system according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention.

[0019] The meanings of the terms used in this invention are uniformly expressed as follows: Point cloud representation: Displaying point clouds on software or web platforms; Point cloud interaction: The interaction between the user and the scene when the user operates the platform to translate, scale, and rotate the point cloud; Original point cloud: Original .las point cloud; Scene point cloud: The original point cloud after preprocessing; Characterization levels are mainly divided into three categories: retained points, downsampling points, and ignored points in the characterization process. Preservation points: Fully preserved during representation, and displayed at the highest level during representation interactions; Downsampling point: Downsampling is performed during representation, and the downsampled hierarchy is displayed during representation interaction; Ignore points: Not displayed during representation; Current field of view: The area that can be displayed on the screen from the viewpoint; Dynamic point cloud: A point cloud that undergoes hierarchical changes during interaction; generally, points are retained. Static point cloud: generally consists of downsampled points and ignored points; Hierarchy: When performing point cloud representation, the corresponding octree hierarchy is the octree hierarchy when loading scene point clouds on the screen; Multi-scale: The point cloud undergoes scale changes during interaction. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0021] like Figure 1 and Figure 2 As shown, one aspect of the present invention provides a dynamic and static combined multi-scale point cloud adaptive representation method, comprising the following steps: S1. Point cloud data preprocessing and semantic segmentation: Solve the original point cloud data, preprocess the solved point cloud data to remove noise and irrelevant information, and perform semantic segmentation on the preprocessed scene point cloud data to provide a basis for subsequent representation level classification. S2. Determine the representation level: Based on the semantic segmentation results of the scene point cloud, determine the representation level of the scene point cloud corresponding to each attribute label, and classify the dynamic and static point clouds at different scales during the interaction process. S3. Calculation of representation parameters under different representation levels: Based on the representation level of the point cloud, calculate the representation level parameters of the target point cloud of each attribute under the current field of view; the representation level parameters include the field of view coordinates, target scale, target type, indexing method, and loading level; S4. Adaptive Scene Point Cloud Representation: Based on the representation level parameters of the target point clouds of each attribute under the current field of view calculated in step S3, load scene point clouds of different representation levels into the field of view.

[0022] Furthermore, step S1 involves processing the original point cloud data, including reading the original point cloud data file, processing the point cloud data, and extracting information such as the three-dimensional coordinates, intensity, and color of the point cloud. Step S1 involves preprocessing the solved point cloud data to remove noise and irrelevant information, and performing semantic segmentation on the preprocessed scene point cloud data, including: S11. Use a pass-through filter to remove points that are out of range; S12. Use neighborhood-based statistical filtering to remove noise points; S13. Depending on the application scenario requirements, use other filtering methods to remove other non-target points; S14. Based on artificial intelligence algorithms and skilled workers, perform semantic segmentation on the preprocessed scene point cloud, assign different attribute labels to different categories of scene point clouds, and classify and organize the attribute labels. Further, in step S11, a pass-through filter is used to remove points outside the bounding box; this includes setting appropriate bounding box parameters based on the original point cloud range, retaining all points within the bounding box, and deleting all points outside the bounding box. Step S12, using neighborhood-based statistical filtering to remove noise points; includes: Set the search radius to determine the neighborhood range of each point in the original scenic spot cloud; Within the neighborhood of a given point, statistical parameters such as the number of points in the neighborhood, average distance, and curvature are calculated to help identify outliers; for example, if the number of points in the neighborhood of a point is much less than the average, then this point may be a noise point. Based on these statistical parameters, thresholds are set to classify the original point cloud and remove noise points. For example, a density threshold is set to mark points with fewer points in their neighborhood as noise points and remove them. In addition, curvature information can be used to distinguish between planes and edges to further optimize the filtering effect. Further, in step S14, semantic segmentation is performed on the preprocessed scene point cloud based on artificial intelligence algorithms and skilled workers, attribute labels are assigned to each point in the scene point cloud, and the attribute labels are categorized and organized; including: The preprocessed scene point cloud data is semantically segmented using an AI-based semantic segmentation algorithm (such as the point cloud semantic segmentation algorithm of RandLA-Net network) to obtain a preliminarily classified scene point cloud. Since current mainstream semantic segmentation algorithms perform poorly in real-world scenarios, if the initial classified scene point cloud cannot meet application requirements, skilled workers will correct and optimize the initial classified scene point cloud to ensure that each semantic category has a clear definition and consistent label, thereby ensuring the accuracy of semantic segmentation. After semantic segmentation, each point in the scene point cloud is assigned an attribute label. This label is unique within the same scene, meaning that all points belonging to that label category can be found by the label number. Specifically, each point in the scene point cloud is provided with an attribute label. This tag The size is 8 bits, corresponding to the category number from 0 to 255, where 0 represents an unclassified point; Output preprocessed point cloud data and semantically segmented point cloud data, with each point having a semantic label; Specifically, further, step S2 includes S21. Statistically analyze the attribute information of the scene point cloud after semantic segmentation and extract key statistical information; S22. Based on the semantic segmentation results of the scene point cloud, determine the representation level of the scene point cloud corresponding to the attribute label; S23. Based on the representation level, classify the dynamic point cloud and static point cloud at different scales during the interaction process; S24. Output the representation level (reserved points, downsampled points, ignored points) of each semantic category, as well as the classification results of dynamic and static point clouds.

[0023] Furthermore, step S21 involves collecting and analyzing the semantically segmented point cloud data to extract key statistical information, specifically including: Category statistics: Count the number of point clouds for each semantic category, such as how many points are in categories such as buildings, roads, and vegetation; Spatial distribution statistics: Analyze the spatial distribution of point clouds, such as calculating the center position, range, density, etc. of point clouds for each category; Attribute statistics: For point clouds with additional attributes (such as intensity, color, normal direction, etc.), the distribution of these attributes is statistically analyzed; for example, the average intensity value of building point clouds, the color distribution of vegetation point clouds, etc.

[0024] Dynamic characteristic statistics: For dynamic point clouds (such as moving vehicles, pedestrians, etc.), statistics are collected on their speed, direction of movement, and other information. Furthermore, the representation level assessment method described in step S22 involves both manual and automatic rating. Manual rating involves manually setting different categories of representation levels based on the user's specific needs and concerns. Specifically, skilled workers rate the segmented scene point clouds and classify the attribute labels, dividing them into three categories based on the task requirements of the scene point clouds: retained points... Sampling points and ignored points Automatic rating is achieved by automatically analyzing point cloud data through algorithms and automatically determining the representation level based on statistical information and predefined rules; that is, rating the segmented scene point cloud based on the density of point cloud features and classifying the attribute labels. The segmented scene point cloud is rated based on the density of point cloud features, and the attribute labels are classified; including: S221. Extract the average distance density, average intensity, curvature, and SHOT features of the point cloud; S222. Construct a complexity criterion using the extracted features and calculate the complexity score for each category; S223. Construct similarity criteria and calculate the similarity with known categories; S224. Automatically classify representation levels based on complexity score and similarity score; S225. Classify the attribute tags and determine the representation method for each category; Further, step S221 includes: For each semantic segmentation category, select the top few objects with the most points, calculate their point cloud average distance density feature, average intensity feature, curvature feature, and SHOT feature; extract key features. Point cloud average distance density feature: Calculate the average distance density of the point cloud in each category, that is, the average distance between each point and other points in its neighborhood; The average distance density of the point cloud is calculated using equation (1): (1) in, For the first Average distance density of point clouds for each semantic segmentation category; Indicates the first Total number of point clouds in each semantic segmentation category; Point The minimum distance to other points is expressed by equation (2): (2) in, Represents any point in the current category of the scene point cloud. With any point The distance between them; Point cloud average intensity characteristics: Calculate the average intensity value of the point cloud, as intensity reflects the reflection characteristics of the point cloud; the average intensity characteristics of the point cloud are calculated using equation (3): (3) in, Represents any point The intensity value; Point cloud curvature features: Calculate local differential geometric properties, including mean curvature. Gaussian curvature and two principal curvatures and etc., used to distinguish between planar and edge parts; among which It is the maximum curvature of a point on a quadratic surface in all directions. It is the minimum curvature; the curvature of the point cloud is expressed by equation (4): (4) in, , , , , , ; , , , , It is the partial differential of the curved surface; It is the unit vector of the normal vector; Point cloud SHOT features: Calculate a 352-dimensional SHOT descriptor for the point cloud, which is used for comparison with known categories; Furthermore, the complexity criterion in step S222 is expressed by equation (5): (5) in, As a complexity evaluation metric; yes Normalized score; yes Normalized score; yes Normalized score; Normalized value of average distance density feature The weights; Normalized value of average intensity characteristics The weights; Normalized value of Gaussian curvature characteristic The weights; The complexity criterion in step S222 is constructed as follows: The point cloud density features, intensity features, and curvature features are normalized to obtain normalized scores; Depending on the application scenario, weights are assigned to each feature; for airborne point cloud data, where density, intensity, and curvature are stable, the weights can be set as follows: For ground-based cloud data, density and intensity are significantly greater at closer locations than at farther locations, but curvature remains stable across all locations, allowing for the selection of weights: It should be noted that, ; Calculate the complexity score for each category; Step S223 The similarity criterion is expressed by equation (6): (6) in, Similarity index; The 352-dimensional descriptor obtained for calculating SHOT features; SHOT descriptors for known categories; Further, step 224 includes: when the SHOT descriptor of the target category is known, simultaneously calculating the complexity criterion and the similarity criterion; if the category of the current scene point cloud is consistent with the target category, then classifying according to the representation level of the target category; When the SHOT descriptor for the unknown target category is not known, only the complexity criterion is calculated; targets with higher complexity scores in the current scene point cloud are assigned higher priority, and vice versa. Further, step 225 includes: classifying targets into retained points, downsampled points, and ignored points based on complexity scores and similarity scores; for high-priority points, they are classified as retained points. Points with lower priority are classified as downsampling points. Points with lower priority are classified as ignored points. ; Furthermore, in step 23: Based on the representation level and the current field of view, dynamic point clouds can be classified as follows: High-priority dynamic point cloud: Points are retained to maintain a high-density effect during display; Medium-priority dynamic point cloud: downsampled points, which need to be downsampled during display; Low-priority dynamic point cloud: Ignore points, which can be ignored during display; Static point clouds are typically point clouds that users do not pay attention to, and can be divided into: Background static point clouds: For example, distant mountains or buildings can be ignored or displayed at a low resolution; Auxiliary static point cloud: For example, some auxiliary facilities or decorative structures can be displayed at low resolution; Dynamic point clouds adjust their loading levels and display effects in real time based on user actions; The static point cloud remains unchanged during user interaction and is only loaded once during the initial loading based on the representation level; Furthermore, in step S3, the calculation of representation parameters under different representation levels is a crucial step in achieving adaptive representation of point clouds. The goal of this step is to calculate the representation level parameters of the target point cloud for each attribute under the current field of view, based on the priority of the point cloud (retained points, downsampled points, ignored points) and the number of point clouds displayed within a given screen. These parameters include the field of view coordinates, target scale, target type, indexing method, and loading level. The specific implementation process of this step is as follows: S31. Mapping scene point cloud attribute labels to representation levels: Mapping the attribute labels of the semantically segmented scene point cloud to the target representation levels after classification. To establish a connection; S32. Based on the number of point clouds displayed on the given screen, calculate the representation level parameters of each attribute target point cloud under the current field of view. S33. During user interaction, dynamically adjust the representation parameters according to the changes in the loading level to ensure that the display effect of the point cloud meets the current field of view requirements. S34. Output the calculated representation level parameters for use by the subsequent point cloud loading and rendering modules.

[0025] Further, step S31 includes: S311. Obtain semantic segmentation results: Obtain the attribute labels for each point cloud from the semantic segmentation module. These labels represent the semantic category to which the point cloud belongs (such as buildings, roads, vegetation, etc.). S312. Determine the representation level: Based on the semantic segmentation results, assign different representation levels to the point clouds of each category: For higher priority targets, i.e., retain points... The point cloud with high density is preserved during display; for targets of lower priority, i.e., downsampling points... During display, the data is downsampled and retained; for lower priority targets, i.e., points are ignored. Ignore during display; S313. Establish mapping relationship: Create a mapping table to associate each attribute label with the corresponding representation level; S314. Output the mapping table for use in subsequent steps; Further, step S32 includes: S321: Determine the position and field of view of the current viewpoint; wherein, the field of view includes the horizontal angle and the vertical angle; calculate the horizontal angle and the vertical angle based on the position of the current viewpoint; Specifically, let the current viewpoint be... The screen width within the field of view is The height is The equivalent focal length is Then the horizontal perspective for: (7) Vertical perspective for: (8); S322: Calculate the visible range under the current field of view based on the viewpoint position and field of view range; wherein, the visible range under the current field of view includes the visible range diameter and the maximum visible range; The diameter of the visible range is calculated based on the distance from the viewpoint to the target plane and the field of view angle; the relationship between the diameter of the visible range and the field of view angle is: (9) in, The total field of view from the viewpoint; The object distance is the straight-line distance from the viewpoint to the target plane. The maximum visible range is calculated based on the maximum visible range at the maximum distance from the viewpoint to the target: the maximum visible range at different distances is expressed by equation (10): (10) in, Maximum visible width; Maximum visible height; From a horizontal perspective With vertical perspective Substitute the proportion and By referring to equation (10), the horizontal and vertical coordinates of the maximum visible range can be obtained: If the target point and the viewpoint fall within this plane, then the layer in which the point is located needs to be loaded. S323. Determine the attributes and priorities of the target point cloud; based on the semantic segmentation results and mapping table, obtain the attribute labels and representation levels of each target point cloud within the current field of view; S324. Based on the attribute labels and representation levels of each target point cloud in the current field of view, calculate the representation level parameters of each attribute target point cloud in the current field of view; the representation level parameters of the target point cloud include the field of view coordinates, target scale, target type, indexing method and loading level; Step S324 includes: S3241. Determine the coordinate range of each target point cloud in the current field of view; calculate its projected coordinate range on the screen based on the bounding box and viewpoint position of the target point cloud. S3242. Based on the distribution and density of the target point cloud, calculate the scale parameters of each target point cloud, including the length, width, and height of its bounding box; S3243. Determine the type of each target point cloud based on the semantic segmentation results; S3244. Select an appropriate indexing method based on the distribution and representation level of the target point cloud; S3245. Determine the loading level of each target point cloud based on the field of view and target scale parameters; Furthermore, in step S3242, the scale parameters of each target are calculated based on the distribution and density of the target point cloud; this includes: calculating the bounding box of each target point cloud, determining its length, width, and height, thereby obtaining the scale parameters of the target; calculating the point density of each target point cloud, i.e., the number of points in a unit volume; targets with high point density require higher resolution display. In step S3244, an appropriate indexing method is selected based on the distribution and representation level of the point cloud; including: for retained points, a high-resolution index is used to ensure that the details of the point cloud are preserved; for downsampled points, a medium-resolution index is used to reduce the number of points in the point cloud through downsampling while retaining key information; for ignored points, no indexing is performed and these point clouds are directly ignored. Step S3245 determines the loading level of each target point cloud based on the field of view and target scale; including: Based on screen resolution and user needs, determine the maximum number of point clouds that can be displayed within the current field of view. ; Calculate the number of target point clouds to be displayed in the current field of view. This includes retained points, downsampled points, and ignored points; the number of target point clouds to be displayed in the current field of view. This can be expressed by equation (11): (11) in, It is the number of reserved points. It is the number of downsampling points. It is the downsampling ratio; The loading level is calculated based on the maximum number of point clouds that can be displayed in the current field of view and the number of points to be displayed in all target point clouds in the current field of view; the loading level is expressed by equation (12): (12) in, This represents the actual loading level of the point cloud in the current view; ceil() is the floor function, if... If the value is less than 0, then display the level 1 index; The highest level of fine-grained loading of point cloud data, that is, the level at which the point cloud is not downsampled and all details are fully preserved, is determined by the octree hierarchical organization method; In step S33, during user interaction, the representation parameters are dynamically adjusted according to changes in the loading level, including: Real-time updates of field of view parameters: Based on user operations (such as panning, zooming, and rotating), the viewpoint coordinates, field of view range, and visible range diameter are updated in real time. Recalculate loading hierarchy: Based on the new view parameters, recalculate the loading hierarchy of each target point cloud to ensure that the display effect of the point cloud meets the current view requirements; Dynamically adjust indexing method: Based on changes in the loading level, dynamically adjust the indexing method of retained points, downsampling points, and ignored points to ensure that the point cloud display effect is both efficient and accurate; Traverse all scene point clouds and calculate the highest level of all edges that need to be loaded. ; For a set of edge points, when its scale is proportional to the original scale... When, its display parameters are That is, edge points have the highest priority; For non-edge point sets, the loading level is adjusted according to the target representation level; like The displayed parameters are ; like The displayed parameters are floor() is the floor function; like The displayed parameters are .

[0026] Further, step S4 includes: S41. Obtain the calculated characterization level parameters from step S3; S42. Based on the representation level parameters, locate the point cloud data in the corresponding level folder; S43. Use multi-threading technology to load point cloud data to improve loading efficiency: S44. During the loading process, perform necessary processing on the point cloud data: Data decoding: Decodes the loaded point cloud file into a point cloud data structure in memory; Coordinate transformation: Based on the current viewpoint and field of view, the point cloud data is transformed to ensure that it is correctly displayed in the field of view; Index optimization: Based on the indexing method, the point cloud data is indexed and optimized for faster rendering; S45. Use the Qt interface and OpenGL (or other rendering engines) to render the loaded point cloud data; S46. During user interaction, dynamically adjust the loading and rendering of the point cloud: Real-time update of loading level: Based on user operations, the loading level is updated in real time, and point cloud data at different levels of detail is dynamically loaded; Dynamically switch indexing mode: The indexing mode is dynamically switched according to changes in the loading level to optimize rendering performance; Performance monitoring: Monitor rendering performance to ensure a smooth interactive experience; if performance degrades, you can appropriately reduce the loading level or optimize the indexing method. S47. Output the rendered point cloud data to the display device to ensure that the user can see the adaptively represented point cloud scene.

[0027] The dynamic and static combined multi-scale adaptive point cloud representation method of this invention overcomes the shortcomings of existing technologies by optimizing the representation and interaction efficiency of point cloud data based on application requirements. Its dynamic and static combined representation method, efficient loading and rendering strategies, and flexible system design enable the system to better meet user needs in different application scenarios, significantly improving the application value of point cloud data and user experience.

[0028] like Figure 3 As shown, a second aspect of the present invention provides a dynamic and static combined multi-scale point cloud adaptive representation system for implementing the above-mentioned representation method, including a data input module, a data processing module, a semantic segmentation module, a representation level division module, a representation parameter calculation module, a point cloud loading and rendering module, a user interaction module, and a system management module. The data input module is used to read and import raw point cloud data, supporting multiple formats (such as LAS, PLY, XYZ, etc.); specifically, it reads point cloud data from files or databases; converts point cloud data of different formats into a unified internal format; and performs preliminary checks on the data to ensure data integrity and consistency. The data processing module is used to preprocess the raw point cloud data, remove noise and irrelevant information, and optimize data quality. Specifically, it removes point clouds that exceed the region of interest through pass-through filtering; removes noisy points through statistical filtering; removes ground points or other non-target points as needed through other filtering; reduces the amount of point cloud data and improves processing efficiency through downsampling; and removes duplicate points, outliers, etc. through data cleaning. The semantic segmentation module is used to perform semantic segmentation on the preprocessed point cloud data and assign semantic labels to each point; specifically, it includes: algorithm selection, selecting a suitable deep learning algorithm (such as RandLA-Net, PointCNN, etc.). Model training: Train the model using a labeled training dataset; Segmentation execution: Apply the trained model to the point cloud data and assign a semantic label to each point; Manual correction: Skilled workers check and correct the preliminary segmentation results; The representation level classification module is used to determine the representation level of each semantic category based on the semantic segmentation results. Specifically, it includes: manual rating, where skilled workers rate the semantic categories according to user needs; automatic rating, where algorithms automatically analyze point cloud features to determine the representation level; and dynamic and static classification, which divides the point cloud into dynamic point cloud and static point cloud. The representation parameter calculation module is used to calculate the representation parameters of the target point cloud for each attribute under the current field of view based on the representation level; specifically, it includes: Determine viewpoint and field of view: Obtain the current viewpoint position and field of view; Calculate the visible range: Calculate the visible range based on the viewpoint and field of view; Characterization parameter calculation, including: Field of view coordinates: Determines the coordinate range of the target point cloud within the field of view; Target Scale: Calculates the scale parameters of the target point cloud; Target type: Determines the semantic category of the target point cloud; Indexing method: Select the appropriate indexing method based on the representation level; Loading hierarchy: The loading hierarchy is determined based on the field of view and target scale; The point cloud loading and rendering module is used to load and render point cloud data of different representation levels in the field of view based on representation parameters; specifically, it includes: Locate the corresponding folder level: Find the corresponding point cloud file based on the representation parameters; Multi-threaded loading: Using multi-threading technology to load point cloud data; Point cloud indexing and organization: Indexing and organizing the loaded point cloud data; Point cloud rendering: Rendering point clouds using OpenGL or other rendering engines; Dynamic adjustment: Dynamically adjust loading and rendering strategies based on user interaction; The user interaction module provides a user interface to support user interaction with point cloud data. Specifically, it includes: view control (supporting panning, zooming, rotation, and other view operations); interactive response (responding to user operations in real time and updating point cloud loading and rendering); and parameter adjustment (allowing users to adjust representation parameters, such as loading level and indexing method). The system management module is responsible for the overall management and coordination of the system, ensuring the collaborative work of all modules; Through the collaborative work of the above modules, the dynamic and static combined multi-scale point cloud adaptive representation system can efficiently process and display point cloud data, while meeting the needs of users in different interactive scenarios.

[0029] It should be noted that the dynamic-static combined multi-scale point cloud adaptive representation system provided in this embodiment can be a computer program (including program code) running on a computer device. For example, the dynamic-static combined multi-scale point cloud adaptive representation system is an application software; the dynamic-static combined multi-scale point cloud adaptive representation system can be used to execute the corresponding steps in the above-described method provided in the embodiments of this application.

[0030] In some feasible implementations, the dynamic-static combined multi-scale point cloud adaptive representation system provided in this embodiment can be implemented in a hardware-software combined manner. As an example, the dynamic-static combined multi-scale point cloud adaptive representation system provided in this application embodiment can be a processor in the form of a hardware decoding processor, which is programmed to execute the dynamic-static combined multi-scale point cloud adaptive representation method provided in this application embodiment. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0031] In some feasible implementations, the dynamic-static combined multi-scale point cloud adaptive representation system provided in this embodiment can be implemented in software. It can be software in the form of programs and plug-ins, and includes a series of modules to implement the dynamic-static combined multi-scale point cloud adaptive representation method provided in this embodiment of the invention.

[0032] The dynamic and static combined multi-scale point cloud adaptive representation system provided in this embodiment significantly improves the representation efficiency and user experience of point cloud data through semantic segmentation, dynamic and static combined representation methods, and efficient loading and rendering strategies. It overcomes many shortcomings of existing technologies and provides strong technical support for the widespread application of 3D point cloud technology in fields such as architecture, transportation, and power.

[0033] A third aspect of the present invention also provides an electronic device, Figure 4 This is a schematic diagram of the electronic device in this embodiment, as shown below. Figure 4As shown, the electronic device 1000 in this embodiment may include: a processor 1001, a network interface 1004, and a memory 1005. Furthermore, the electronic device 1000 may also include: a user interface 1003, and at least one communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen and a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as at least one disk storage device. Optionally, the memory 1005 may also be at least one storage device located remotely from the aforementioned processor 1001. Figure 4 As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a device control application.

[0034] like Figure 4 In the electronic device 1000 shown, the network interface 1004 provides network communication functions; the user interface 1003 is mainly used to provide an input interface for users; and the processor 1001 can be used to call the device control application stored in the memory 1005 to implement the various steps of the above decision-making method.

[0035] It should be understood that in some feasible implementations, the processor 1001 described above may be a central processing unit (CPU), which may also be other general-purpose processors, DSPs, ASICs, FPGAs, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory. For example, the memory may also store device type information.

[0036] In specific implementation, the aforementioned electronic device 1000 can perform the above-described actions through its built-in functional modules. Figure 1 The implementation methods provided for each step are detailed in the above-mentioned implementation methods and will not be repeated here.

[0037] The electronic device provided in this embodiment significantly improves the representation efficiency and user experience of point cloud data through semantic segmentation, dynamic and static combined representation methods, and efficient loading and rendering strategies. It overcomes many shortcomings of existing technologies and provides strong technical support for the widespread application of 3D point cloud technology in fields such as architecture, transportation, and power.

[0038] This application also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement... Figure 1 The methods provided in each step are detailed in the implementation methods provided in the above steps, and will not be repeated here.

[0039] The computer-readable storage medium provided in this embodiment significantly improves the representation efficiency and user experience of point cloud data through semantic segmentation, dynamic and static combined representation methods, and efficient loading and rendering strategies. It overcomes many shortcomings of existing technologies and provides strong technical support for the widespread application of 3D point cloud technology in fields such as architecture, transportation, and power.

[0040] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0041] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic and static combined multi-scale adaptive representation method for point clouds, characterized in that, Includes the following steps: S1. Solve the original point cloud data, preprocess the solved point cloud data to remove noise and irrelevant information, and perform semantic segmentation on the preprocessed scene point cloud data. S2. Based on the semantic segmentation results of the scene point cloud, determine the representation level of the scene point cloud corresponding to each attribute label. The representation level includes retained points, downsampled points, and ignored points. The dynamic and static point clouds at different scales during the interaction process are classified. The retained points are classified as dynamic point clouds, and the downsampled points and ignored points are classified as static point clouds. During user interaction, the loading level of the downsampling points is as follows: The function's behavior depends on the loading level of the dynamic point cloud. Determined; among them, This is a floor function; the ignored points are never loaded. S3. Based on the representation level of the point cloud, calculate the representation level parameters of the target point cloud of each attribute under the current field of view; the representation level parameters include the field of view coordinates, target scale, target type, indexing method and loading level; S4. Based on the representation level parameters of the target point clouds of each attribute under the current field of view calculated in step S3, load scene point clouds of different representation levels into the field of view. During user interaction, dynamically adjust the loading and rendering of point clouds to obtain the point cloud scene after adaptive representation.

2. The dynamic-static combined multi-scale point cloud adaptive representation method according to claim 1, characterized in that, Step S1 involves processing the raw point cloud data, including reading the raw point cloud data file, processing the point cloud data, and extracting the three-dimensional coordinates, intensity, and color information of the point cloud. Step S1 involves preprocessing the solved point cloud data to remove noise and irrelevant information, and performing semantic segmentation on the preprocessed scene point cloud data, including: S11. Use a pass-through filter to remove points that are out of range; S12. Use neighborhood-based statistical filtering to remove noise points; S13. Depending on the application scenario requirements, use other filtering methods to remove other non-target points; S14. Based on artificial intelligence algorithms and skilled workers, perform semantic segmentation on the preprocessed scene point cloud, assign different attribute labels to different categories of scene point clouds, and classify and organize the attribute labels. Step S11: Use pass-through filtering to remove points outside the bounding box; this includes setting appropriate bounding box parameters based on the original point cloud range, retaining all points within the bounding box, and deleting all points outside the bounding box. Step S12, using neighborhood-based statistical filtering to remove noise points; includes: Set the search radius to determine the neighborhood range of each point in the original scenic spot cloud; Within the neighborhood of a given point, calculate the number of points in the neighborhood of each point, the average distance, and the curvature to help identify outliers; Based on these statistical parameters, thresholds are set to classify the original point cloud and remove noise points; Step S14 includes: The preprocessed scene point cloud data is semantically segmented using an AI-based semantic segmentation algorithm to obtain a preliminarily classified scene point cloud. If the scene point cloud after preliminary classification cannot meet the application requirements, then skilled workers will perform fine classification on the scene point cloud after preliminary classification and check whether the semantic attribute labels of the current classification are correct. After semantic segmentation, attribute labels are assigned to each point in the scene point cloud; The output includes preprocessed point cloud data and semantically segmented point cloud data, with each point having a semantic label.

3. The dynamic-static combined multi-scale point cloud adaptive representation method according to claim 2, characterized in that, Step S2 includes: S21. Statistically analyze the attribute information of the scene point cloud after semantic segmentation and extract key statistical information; S22. Based on the semantic segmentation results of the scene point cloud, determine the representation level of the scene point cloud corresponding to the attribute label; S23. Based on the representation level, classify the dynamic point cloud and static point cloud at different scales during the interaction process; S24. Output the representation level, dynamic point cloud, and static point cloud classification results for each semantic category.

4. The dynamic-static combined multi-scale point cloud adaptive representation method according to claim 3, characterized in that, The characterization rating method described in step S22 involves both manual rating and automatic rating; Manual rating involves skilled workers rating the segmented scene point clouds based on the user's specific needs and concerns, and classifying the attribute labels. According to the task requirements of the scene point clouds, they are divided into three categories: reserved points. Sampling points and ignored points ; Automatic rating is based on the density of point cloud features to rate the segmented scene point cloud and classify the attribute labels. The segmented scene point cloud is rated based on the density of point cloud features, and the attribute labels are classified; including: S221. Extract the average distance density, average intensity, curvature, and SHOT features of the point cloud; S222. Construct a complexity criterion using the extracted features and calculate the complexity score for each category; S223. Construct similarity criteria and calculate the similarity with known categories; S224. Automatically classify representation levels based on complexity score and similarity score; S225. Classify the attribute tags and determine the representation method for each category.

5. The dynamic-static combined multi-scale point cloud adaptive representation method according to claim 4, characterized in that, In step S221, the average distance density of the point cloud is calculated using equation (1): (1) in, For the first Average distance density of point clouds for each semantic segmentation category; Indicates the first Total number of point clouds in each semantic segmentation category; Point The minimum distance to other points is expressed by equation (2): (2) in, Represents any point in the current category of the scene point cloud. With any point The distance between them; The average intensity characteristic of the point cloud is calculated using equation (3): (3) in, Represents any point The intensity value; The curvature of the point cloud is expressed by equation (4): (4) in, The mean curvature; Gaussian curvature; and For two different principal curvatures; It is the maximum curvature of a point on a quadratic surface in all directions. It is the minimum curvature; It is the partial differential of the curved surface; It is the unit vector of the normal vector.

6. The dynamic-static combined multi-scale point cloud adaptive representation method according to claim 4, characterized in that, The complexity criterion in step S222 is expressed by equation (5): (5) in, As a complexity evaluation metric; yes Normalized score; yes Normalized score; yes Normalized score; Normalized value of average distance density feature The weights; Normalized value of average intensity characteristics The weights; Normalized value of Gaussian curvature characteristic The weights; The complexity criterion in step S222 is constructed as follows: The point cloud density features, intensity features, and curvature features are normalized to obtain normalized scores; Weights are assigned to each feature based on the different application scenarios. Calculate the complexity score for each category; Step S223 The similarity criterion is expressed by equation (6): (6) in, Similarity index; The 352-dimensional descriptor obtained for calculating SHOT features; SHOT descriptors for known categories.

7. A dynamic-static combined multi-scale adaptive representation method for point clouds according to any one of claims 1-6, characterized in that, Step S3 includes: S31. Mapping scene point cloud attribute labels to representation levels: Mapping the attribute labels of the semantically segmented scene point cloud to the target representation levels after classification. To establish a connection; S32. Based on the number of point clouds displayed on the given screen, calculate the representation level parameters of each attribute target point cloud under the current field of view. S33. During user interaction, dynamically adjust the representation parameters according to the changes in the loading level to ensure that the display effect of the point cloud meets the current field of view requirements. S34. Output the calculated representation level parameters for use by the subsequent point cloud loading and rendering modules.

8. The dynamic-static combined multi-scale point cloud adaptive representation method according to claim 7, characterized in that, Step S32 includes: S321: Determine the position and field of view of the current viewpoint; wherein, the field of view includes the horizontal angle and the vertical angle; calculate the horizontal angle and the vertical angle based on the position of the current viewpoint; Specifically, let the current viewpoint be... The screen width within the field of view is The height is The equivalent focal length is Then the horizontal perspective for: (7) Vertical perspective for: (8); S322: Calculate the visible range under the current field of view based on the viewpoint position and field of view range; wherein, the visible range under the current field of view includes the visible range diameter and the maximum visible range; The diameter of the visible range is calculated based on the distance from the viewpoint to the target and the viewing angle; the relationship between the diameter of the visible range and the field of view angle is: (9) in, For the field of view; The object distance is the straight-line distance from the viewpoint to the target plane. The maximum visible range is calculated based on the maximum visible range at the maximum distance from the viewpoint to the target: the maximum visible range at different distances is expressed by equation (10): (10) in, Maximum visible width; Maximum visible height; From a horizontal perspective and vertical perspective Substitute the proportion and By referring to equation (10), the horizontal and vertical coordinates of the maximum visible range can be obtained; S323. Determine the attributes and priorities of the target point cloud; based on the semantic segmentation results and mapping table, obtain the attribute labels and representation levels of each target point cloud within the current field of view; S324. Based on the attribute labels and representation levels of each target point cloud in the current field of view, calculate the representation level parameters of each attribute target point cloud in the current field of view; the representation level parameters of the target point cloud include the field of view coordinates, target scale, target type, indexing method, and loading level; The calculation for each target point cloud loading level includes: Based on screen resolution and user needs, determine the maximum number of point clouds that can be displayed within the current field of view. ; Calculate the number of target point clouds to be displayed in the current field of view. This includes retained points, downsampled points, and ignored points; the number of target point clouds to be displayed in the current field of view. This can be expressed by equation (11): (11) in, It is the number of reserved points. It is the number of downsampling points. It is the downsampling ratio; The loading level is calculated based on the maximum number of point clouds that can be displayed in the current field of view and the number of points to be displayed in all target point clouds in the current field of view; expressed by equation (12): (12) in, This represents the highest level of detail in the point cloud data loading process, i.e., the level at which the point cloud is not downsampled and retains all details completely. This level is determined by the octree hierarchical organization method; ceil() is the floor function. If the value is less than 0, then a level 1 index is displayed.

9. A dynamic-static combined multi-scale adaptive representation method for point clouds according to any one of claims 1-6 and 8, characterized in that, Step S4 includes: S41. Obtain the calculated characterization level parameters from step S3; S42. Based on the representation level parameters, locate the point cloud data in the corresponding level folder; S43. Use multi-threading technology to load point cloud data to improve loading efficiency: S44. During the loading process, the point cloud data is subjected to data decoding, coordinate transformation, and index optimization. S45. Use the Qt interface and OpenGL to render the loaded point cloud data; S46. During user interaction, dynamically adjust the loading and rendering of the point cloud: S47. Output the rendered point cloud data to the display device to ensure that the user can see the adaptively represented point cloud scene.

10. A dynamic and static combined multi-scale point cloud adaptive representation system, characterized in that, The method for implementing the dynamic-static combined multi-scale point cloud adaptive representation method as described in any one of claims 1-9 includes: The data input module is used to read and import raw point cloud data; The data processing module is used to preprocess the raw point cloud data, remove noise and irrelevant information, and optimize data quality; The semantic segmentation module is used to perform semantic segmentation on the preprocessed point cloud data and assign semantic labels to each point. The representation level classification module is used to determine the representation level of each semantic category based on the semantic segmentation results; The characterization parameter calculation module is used to calculate the characterization parameters of the target point cloud of each attribute under the current field of view according to the characterization level. The point cloud loading and rendering module is used to load and render point cloud data of different representation levels in the field of view according to the representation parameters. The user interaction module provides a user interface to support user interaction with point cloud data. The system management module is responsible for the overall management and coordination of the system, ensuring the collaborative work of all modules.

Citation Information

Patent Citations

  • Laser point cloud model rendering method and system

    CN118135080A

  • Bit-by-bit depth octree coding based on sparse tensor

    CN120035842A