A vertical space modeling method based on virtual point cloud enhancement and related equipment

Abstract

The application provides a vertical space modeling method based on virtual point cloud enhancement and related equipment, and relates to the technical field of space modeling. The method comprises the following steps: obtaining environment point cloud data and vertical distance information; generating a virtual point cloud set based on the geometric characteristics of the target space cross section, and adjusting the sampling interval distance between adjacent virtual points according to the size of the vertical distance, so that the distribution density of the virtual point cloud set decreases with the increase of the vertical distance; fusing the environment point cloud data and the virtual point cloud set to obtain enhanced point cloud data; performing inter-frame registration and global optimization on the enhanced point cloud data, and outputting a three-dimensional structure model of the target vertical space. The method aims to solve the problems of pose drift and incomplete modeling caused by insufficient elevation direction constraint and sensor measurement uncertainty in the prior art in the modeling of vertical narrow spaces, and can accurately model typical square structures such as elevator shafts.

Description

Technical Field

[0001] This invention relates to the field of spatial modeling technology, and more specifically, to a vertical spatial modeling method and related equipment based on virtual point cloud enhancement. Background Technology

[0002] As a critical vertical passageway in high-rise buildings, the geometric flatness and structural integrity of elevator shafts are crucial for elevator installation and safe operation. After building construction, accurate 3D structural data of the elevator shaft is essential for subsequent elevator installation design and construction verification, as well as safety assessment after commissioning. Failure to detect discrepancies between the actual shaft structure and the design in a timely manner can not only hinder elevator installation but also create serious safety hazards for long-term elevator operation. Traditional shaft surveying relies on manual climbing of measuring tools floor by floor or using static 3D laser scanners to collect data segment by segment. These methods are not only inefficient and time-consuming but also require personnel to enter unfinished high-rise buildings for high-altitude work, posing significant safety risks. Furthermore, limitations in human error and equipment setup make it difficult to guarantee overall modeling accuracy, especially in complex environments such as deep or narrow shafts with large depths and small cross-sectional dimensions. This further increases the difficulty of measurement work, and the final modeling results often fail to meet the accuracy requirements of subsequent engineering applications.

[0003] In recent years, mobile mapping technology using drones equipped with LiDAR has been increasingly applied to 3D modeling of various indoor enclosed spaces. Through simultaneous localization and mapping (SLAM) algorithms, it enables automatic 3D reconstruction of scenes, eliminating the need for manual point-by-point measurements within the work area, significantly improving the efficiency of surveying operations and reducing the safety risks associated with manual work. However, vertical spaces, including elevator shafts, are typical examples of vertically elongated and degenerate scenarios. These scenarios have highly repetitive internal structures, extending vertically with insufficient differentiated textures in the horizontal direction. This results in insufficient constraints in the elevation direction for SLAM algorithms based on LiDAR point clouds, leading to pose estimation drift and even mapping failure. Furthermore, since the LiDAR on mobile platforms typically scans horizontally or at small angles, the scanning range is insufficient to cover the bottom and top areas of vertical spaces, making it difficult to effectively acquire point cloud data for these areas. This further weakens the modeling accuracy in the elevation direction, resulting in incomplete 3D structural models that fail to meet application requirements.

[0004] To improve positioning robustness, some existing solutions attempt to integrate inertial measurement units (IMUs) with barometers or ultrasonic sensors to assist in altitude estimation, compensating for insufficient elevation constraints. However, these sensors have inherent limitations, are highly susceptible to environmental interference (e.g., airflow fluctuations significantly affect barometer measurements), or have limited accuracy (e.g., ordinary ultrasonic ranging has significant errors). None of these solutions fundamentally address the elevation drift problem in vertical spatial modeling. While altitude-fixed radar can output highly accurate vertical distance measurements, its output is only a single-point altitude value, representing one-dimensional measurement data. This makes it difficult to directly integrate into point cloud-based simultaneous localization and mapping (SMR) frameworks, failing to provide effective constraints for point cloud registration and pose optimization. Furthermore, the measurement uncertainty of altitude-fixed radar increases with the measurement distance; the farther from the reference plane (i.e., the greater the vertical distance), the greater the measurement error. Directly using altitude-fixed radar measurements at different altitudes interchangeably introduces additional errors, affecting the final modeling accuracy.

[0005] Therefore, how to effectively utilize the high-precision vertical information of fixed-altitude radar, overcome its characteristic that measurement uncertainty increases with distance, solve the problem of insufficient elevation direction constraints and easy pose drift in point cloud synchronous positioning and mapping in vertical narrow spaces, and supplement the structural data of the top and bottom of the vertical space to improve the accuracy and completeness of vertical space 3D modeling, has become a key technical bottleneck for improving the accuracy of automatic modeling of vertical spaces such as elevator shafts.

[0006] There is currently no effective technical solution to the above problems. Summary of the Invention

[0007] The purpose of this invention is to provide a vertical space modeling method and related equipment based on virtual point cloud enhancement, which aims to solve the problems of pose drift and incomplete modeling caused by insufficient elevation direction constraints and sensor measurement uncertainties in the existing technology for vertical narrow space modeling. It can accurately model typical square structures such as elevator shafts.

[0008] In a first aspect, the present invention provides a vertical space modeling method based on virtual point cloud enhancement, comprising the following steps: S1. Acquire environmental point cloud data collected by the mobile platform in the vertical space of the target, and simultaneously acquire the vertical distance information of the mobile platform relative to the reference plane; S2. Based on the preset geometric features of the target space cross section, combined with the vertical distance information, a virtual point cloud set located on the height plane corresponding to the vertical distance is generated, and the sampling interval distance between adjacent virtual points in the virtual point cloud set is dynamically adjusted according to the size of the vertical distance, so that the distribution density of the virtual point cloud set decreases as the vertical distance increases; S3. Enhanced point cloud data is obtained by fusing the environmental point cloud data with the adjusted virtual point cloud set; S4. By performing inter-frame registration and global optimization on the enhanced point cloud data, the three-dimensional structural model of the target vertical space is output.

[0009] The vertical space modeling method based on virtual point cloud enhancement provided by this invention can effectively solve the problems of insufficient synchronous positioning and mapping elevation direction constraints of point clouds in narrow vertical spaces, which can easily cause pose drift. At the same time, it supplements the structural data of the top and bottom of the vertical space, overcomes the defect that the uncertainty of fixed-altitude radar measurement increases with distance, and has the advantages of improving the accuracy and completeness of vertical space three-dimensional modeling and meeting the needs of engineering applications.

[0010] Secondly, the present invention provides a vertical space modeling device based on virtual point cloud enhancement, comprising: The acquisition module is used to acquire environmental point cloud data collected by the mobile platform in the vertical space of the target, and simultaneously acquire the vertical distance information of the mobile platform relative to the reference plane. The generation module is used to generate a set of virtual point clouds located on the height plane corresponding to the vertical distance, based on the preset geometric features of the target space cross section and the vertical distance information, and to dynamically adjust the sampling interval between adjacent virtual points in the set of virtual point clouds according to the size of the vertical distance, so that the distribution density of the set of virtual point clouds decreases as the vertical distance increases. The fusion module is used to obtain enhanced point cloud data by fusing the environmental point cloud data with the adjusted virtual point cloud set; The output module is used to output a three-dimensional structural model of the target vertical space by performing inter-frame registration and global optimization on the enhanced point cloud data.

[0011] Thirdly, the present invention provides an electronic device including a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, perform the steps of the vertical spatial modeling method based on virtual point cloud enhancement provided in the first aspect above.

[0012] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the vertical spatial modeling method based on virtual point cloud enhancement provided in the first aspect above.

[0013] As can be seen from the above, the vertical space modeling method based on virtual point cloud enhancement provided by this invention generates a virtual point cloud by combining the vertical distance information obtained by a fixed-altitude radar. The distribution density of the virtual point cloud is dynamically adjusted so that it decreases as the vertical distance increases. Then, the virtual point cloud and the environmental point cloud collected in reality are fused to complete the mapping optimization. It can adapt to the characteristic that the uncertainty of fixed-altitude radar measurement increases with distance, effectively utilize the high-precision vertical information of fixed-altitude radar, and solve the problem of insufficient synchronous positioning of point cloud and easy pose drift in the elevation direction of mapping in a narrow vertical space. At the same time, it supplements the structural data of the top and bottom of the vertical space. It has the advantages of suppressing the accumulation of errors in the elevation direction, improving the accuracy and completeness of vertical space 3D modeling, improving the efficiency of modeling operations, and reducing the safety risks of manual operations.

[0014] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0015] Figure 1 This is a flowchart of a vertical space modeling method based on virtual point cloud enhancement provided in an embodiment of the present invention.

[0016] Figure 2 The overall process architecture diagram of the vertical space modeling method based on virtual point cloud enhancement provided in the embodiments of the present invention is shown.

[0017] Figure 3 This is a schematic diagram of a vertical space modeling device based on virtual point cloud enhancement provided in an embodiment of the present invention.

[0018] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0019] Label Explanation: 100. Acquisition module; 200. Generation module; 300. Fusion module; 400. Output module; 13. Electronic device; 1301. Processor; 1302. Memory; 1303. Communication bus. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0021] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] In the automatic modeling process of vertical and narrow spaces such as elevator shafts, the laser synchronous positioning and mapping algorithm has insufficient constraints in the elevation direction due to the high repetition of the internal structure and the lack of horizontal feature textures, resulting in positioning drift. At the same time, the single-point height information output by the fixed-elevation radar is difficult to directly integrate into the point cloud-based mapping framework, and its measurement uncertainty increases with the vertical distance, further weakening the modeling accuracy and robustness. The insufficient elevation constraint mainly stems from the lack of horizontal features in the degraded vertical space scenario, and the difference in format between single-point height information and point cloud data makes it impossible for the two to be effectively integrated.

[0023] For example, in the mapping of elevator shafts in high-rise buildings, the shafts are vertical and narrow with a fixed cross-sectional geometry and smooth inner wall surfaces lacking significant texture features. When UAVs equipped with lidar and altitude-fixing radar collect data, the lidar, due to its horizontal scanning characteristics, struggles to acquire sufficient point cloud data for the bottom and top areas of the shaft, resulting in sparse point clouds in the elevation direction. This makes it easy for the synchronous positioning and mapping algorithms to accumulate errors during inter-frame registration. At the same time, when the altitude-fixing radar is far from the reference plane, the measurement uncertainty increases, and its single-point height information cannot effectively supplement the elevation constraints of the point cloud data, thus leading to a decrease in elevation direction positioning accuracy during the modeling process.

[0024] If the above problems are not solved, the accuracy of 3D modeling in vertical space will not meet the needs of engineering applications, which may lead to elevator installation positioning deviations and increase operational safety hazards. In addition, the robustness of the modeling process will be reduced, making it susceptible to interference from environmental factors and failure, affecting the efficiency and reliability of surveying operations. The cumulative effect of positioning drift will cause the global optimization results to deviate from the actual structure, further triggering adaptation problems in the subsequent installation stage.

[0025] For reference, see the appendix. Figure 1 and attached Figure 2 This invention proposes a vertical space modeling method based on virtual point cloud enhancement, comprising the following steps: S1. Acquire environmental point cloud data collected by the mobile platform in the vertical space of the target, and simultaneously acquire the vertical distance information of the mobile platform relative to the reference plane; S2. Based on the preset geometric features of the target space cross section and combined with the vertical distance information, a set of virtual point clouds is generated on the height plane corresponding to the vertical distance. The sampling interval between adjacent virtual points in the set of virtual point clouds is dynamically adjusted according to the size of the vertical distance so that the distribution density of the set of virtual point clouds decreases as the vertical distance increases. S3. Enhanced point cloud data is obtained by fusing environmental point cloud data with the adjusted virtual point cloud set; S4. Employing a simultaneous localization and mapping algorithm, the system performs inter-frame registration and global optimization on the enhanced point cloud data to output a three-dimensional structural model of the target in the vertical space.

[0026] For ease of understanding, the following explains some key terms in this embodiment: Mobile platform: refers to a carrier capable of moving within the vertical space of a target, such as a drone. This mobile platform typically carries multiple sensors to collect environmental data and its own status information.

[0027] Target vertical space: refers to spaces with significant vertical extension, such as elevator shafts, mine shafts, and the interior of chimneys. These spaces typically have relatively fixed cross-sectional geometry, but significant vertical depth, and their internal environment may be complex.

[0028] Environmental point cloud data: A set of three-dimensional points representing the vertical interior environment of a target, collected by sensors such as LiDAR. These points contain geometric information about the surfaces of objects in space, used to construct a three-dimensional model of the space.

[0029] Vertical distance information: refers to the vertical height or depth of the mobile platform relative to a reference plane (such as the bottom or top of the shaft). This information is typically obtained through high-precision ranging sensors such as altitude-fixing radar.

[0030] Virtual point cloud set: Point cloud data generated on the corresponding height plane based on the prior geometric features of the target's vertical space and the vertical distance information of the moving platform. These virtual point clouds are used to compensate for the deficiencies of the actual acquired point clouds in certain areas (such as the bottom or top of the shaft or areas with sparse features) and to enhance the constraints in the elevation direction.

[0031] Sampling interval distance: The spatial distance between adjacent virtual points in a virtual point cloud set. By dynamically adjusting this distance, the distribution density of the virtual point cloud can be controlled.

[0032] Enhanced point cloud data: This is data obtained by fusing actual collected environmental point cloud data with a generated virtual point cloud set. This data combines real-world environmental information with prior structural information, providing richer and more reliable input for subsequent 3D modeling.

[0033] Simultaneous Localization and Mapping (SLAM) algorithm: a technique that enables a mobile platform to simultaneously perform self-localization and environmental map construction in an unknown environment. In this embodiment, the algorithm is used to process augmented point cloud data to output a three-dimensional structural model of the target in vertical space.

[0034] Inter-frame registration: a key step in the SLAM algorithm, which estimates the pose changes of the mobile platform at different times by matching point cloud data from consecutive frames.

[0035] Global optimization: Another key step in the SLAM algorithm is to jointly optimize the poses and map points of all frames by constructing an optimization problem (e.g., factor graph) to eliminate accumulated errors and improve the global consistency of the map.

[0036] Three-dimensional structural model: A three-dimensional model that reflects the vertical spatial geometry and structural features of the target after processing enhanced point cloud data through the SLAM algorithm. It is usually represented as a closed point cloud model or a mesh model.

[0037] This application provides a vertical space modeling method based on virtual point cloud enhancement, aiming to solve the problems of low accuracy, poor efficiency, and insufficient elevation constraints in traditional methods for modeling vertically elongated spaces. The method first acquires environmental point cloud data collected by a mobile platform within the target vertical space, and simultaneously acquires the vertical distance information of the mobile platform relative to a reference plane. The mobile platform can be equipped with a lidar and an altitude-fixing radar, where the lidar is used to acquire environmental point cloud data, and the altitude-fixing radar is used to acquire vertical distance information. For example, a Livox MID360 lidar can be used, which primarily scans horizontally and can effectively acquire point clouds of the shaft sidewalls. The altitude-fixing radar can be installed at the center of the bottom of the mobile platform, measuring the distance vertically downwards to the bottom of the shaft. Simultaneous acquisition of environmental point cloud data and vertical distance information ensures temporal alignment of the two types of data, providing an accurate foundation for subsequent data fusion and modeling.

[0038] Furthermore, based on the preset geometric features of the target space cross-section and combined with vertical distance information, a virtual point cloud set located on the height plane corresponding to the vertical distance is generated. The geometric features of the target space cross-section can be preset according to the actual application scenario. For example, for an elevator shaft, its cross-sectional geometric features are usually square or rectangular. After obtaining the current vertical distance of the mobile platform, a virtual point cloud can be generated on the height plane corresponding to that vertical distance. For example, it can be generated based on the preset cross-sectional dimensions (such as...). * ) and reference sampling interval (e.g. and This generates a uniform grid set of points as the initial virtual point cloud. This approach utilizes prior information about the target's vertical spatial structure, avoiding the difficulty of extracting cross-sectional features from a sparse original point cloud, and providing structural constraints in the elevation direction.

[0039] Based on this, the sampling interval between adjacent virtual points in the virtual point cloud is dynamically adjusted according to the vertical distance, so that the distribution density of the virtual point cloud decreases as the vertical distance increases. This dynamic adjustment strategy takes into account the characteristic that the measurement uncertainty of the fixed-altitude radar increases with distance. When the vertical distance is small, the measurement accuracy of the fixed-altitude radar is relatively high, and a smaller sampling interval can be used to generate a higher density virtual point cloud to provide more refined constraints. However, when the vertical distance is large, the measurement uncertainty of the fixed-altitude radar increases, and the sampling interval can be increased to reduce the density of the virtual point cloud, thereby reducing the interference of high uncertainty information on the subsequent SLAM optimization process. For example, a linear increasing relationship can be defined between the sampling interval and the vertical distance, that is, the sampling interval increases with the vertical distance.

[0040] Subsequently, enhanced point cloud data is obtained by fusing the environmental point cloud data with the adjusted virtual point cloud set. Data fusion combines the real environmental point cloud actually collected by LiDAR with the virtual point cloud generated based on prior information. This fusion method retains the detailed information of the real environment while supplementing the structural constraints in the elevation direction through the virtual point cloud, effectively compensating for the lack of features in the original point cloud in vertically elongated scenes. For example, the two parts of the point cloud can be directly merged at the set level to form a more complete and constrained point cloud dataset.

[0041] Finally, a simultaneous localization and mapping (SLAM) algorithm is employed to output a 3D structural model of the target's vertical space through inter-frame registration and global optimization of the enhanced point cloud data. The enhanced point cloud data serves as input to the SLAM algorithm, and its virtual point cloud provides additional elevation direction constraints for inter-frame registration and global optimization. During inter-frame registration, the Normal Distribution Transform (NDT) algorithm or the Iterative Closest Point (ICP) algorithm can be used to perform feature matching on the enhanced point cloud data of consecutive frames to obtain the pose transformation information of the mobile platform. In the global optimization stage, a backend factor map can be constructed, incorporating the observations corresponding to the virtual point clouds. Explicit uncertainty modeling is achieved by assigning a covariance matrix to the observations corresponding to the virtual point clouds, and their weights in the optimization process are adaptively adjusted based on the covariance matrix to suppress the accumulation of elevation direction errors. Ultimately, the SLAM algorithm outputs a high-precision, well-closed 3D point cloud model of the target's vertical space, which includes complete sidewall, bottom, and top structures.

[0042] The following example will provide a more detailed explanation of the above technical solution: Suppose we need to create a 3D model of an elevator shaft with a depth of 50 meters. The shaft has a cross-section of 1.5 meters x 1.5 meters. Traditional modeling methods might require manual climbing of each floor with measuring tools, or segmented data acquisition using a static 3D laser scanner, which is inefficient and poses safety risks. Furthermore, due to the highly repetitive internal structure of the elevator shaft and the lack of horizontal texture features, laser SLAM suffers from insufficient vertical constraints, making it prone to drift.

[0043] To address this, this application proposes a vertical space modeling method based on virtual point cloud enhancement. First, a UAV equipped with a Livox MID360 lidar and a Surertech WT50A altitude-fixing radar is deployed. The UAV descends at a constant speed along the central axis of the shaft, starting from the top of the shaft, maintaining a stable attitude. During the descent, the lidar primarily scans horizontally, acquiring environmental point cloud data from the shaft sidewalls in real time, while the altitude-fixing radar simultaneously measures the vertical distance from the bottom of the UAV to the bottom of the shaft. For example, at a certain time t, the lidar acquires the current environmental point cloud data... The vertical distance measured by the fixed-altitude radar is .

[0044] Next, using the preset geometric features of the shaft cross-section (e.g., square, with dimensions of...) =1.5m, =1.5m) and current vertical distance This generates a virtual point cloud set. Specifically, based on the vertical distance... The size of the virtual point cloud can be dynamically adjusted to change the sampling interval. For example, the reference sampling interval distances in the x-axis and y-axis directions can be set. and Both are 0.02 meters, with a density attenuation coefficient. It is 0.002m -1 So, at the current altitude At this location, the sampling interval distance along the x-axis. Sampling interval distance in the y-axis direction .when When it increases, and This also increases, causing a decrease in the distribution density of the virtual point cloud. Then, at z= Within the height plane, according to the calculated and A uniform grid is created, and the grid intersections are designated as virtual points, thus obtaining a virtual point cloud set. This dynamic adjustment strategy results in a lower density of the virtual point cloud when the UAV is far from the bottom of the well and the uncertainty of the altitude-fixed radar measurement is high, thus reducing the impact of high uncertainty information on subsequent optimization.

[0045] Subsequently, the environmental point cloud collected by the lidar was... With the generated virtual point cloud set Data fusion is performed to obtain enhanced point cloud data. This enhanced point cloud data not only contains the true geometric information of the shaft sidewalls, but also supplements the prior structural information of the shaft cross-section through virtual point clouds. Especially in areas where lidar is difficult to observe effectively, such as the bottom of the shaft, the virtual point cloud provides crucial elevation direction constraints.

[0046] Finally, The data is input into an improved laser SLAM algorithm. The front end of the SLAM algorithm uses the Normal Distribution Transform (NDT) algorithm to perform inter-frame registration of the enhanced point cloud data across consecutive frames, obtaining the UAV's pose transformation information. The back-end optimization module constructs a factor graph, incorporating the observations corresponding to the virtual point cloud and assigning it a covariance matrix for explicit uncertainty modeling. Based on the covariance matrix, the weights of the virtual point cloud in the factor graph optimization process are adaptively adjusted according to its distribution density to suppress the accumulation of errors in the elevation direction. In this way, the SLAM algorithm can more accurately estimate the UAV's pose and construct a high-precision, well-closed 3D point cloud model of the wellbore, which includes complete sidewall, bottom, and top structures.

[0047] As can be seen from the above examples, the technical solution of this application effectively solves the problems encountered by traditional methods in modeling vertically elongated spaces. Compared with existing technologies that rely solely on lidar for SLAM modeling, this application introduces high-precision vertical distance information from a fixed-altitude radar and transforms it into a virtual point cloud with a dynamic density adjustment mechanism, providing a strong elevation direction constraint for the SLAM algorithm. In traditional solutions, the single-point height value of a fixed-altitude radar is difficult to directly integrate into the point cloud mapping framework. However, this application successfully transforms single-point height information into a point cloud form that can be utilized by the SLAM algorithm by generating a virtual point cloud set. Furthermore, considering the characteristic that the measurement uncertainty of fixed-altitude radar increases with distance, this application designs a dynamic density adjustment strategy for the virtual point cloud, which reduces the weight of the virtual point cloud when the measurement uncertainty is high, avoiding the negative impact of high uncertainty information on the overall modeling accuracy. This method, which combines multi-sensor fusion with structural prior knowledge, significantly improves the accuracy and robustness of 3D modeling in vertically elongated and degraded scenarios such as elevator shafts. The final output 3D structural model can be applied to building information modeling, installation verification, or safety assessment, and has significant practical application value.

[0048] In some embodiments, the specific steps in step S2 include: S21. Based on the magnitude of the vertical distance, determine the corresponding sampling interval distance using a preset linear increasing relationship, so that the sampling interval distance increases as the vertical distance increases; S22. Based on the contour dimensions of the geometric features of the target space cross section, uniform grids are divided in the height plane corresponding to the vertical distance according to the determined sampling interval distance, and the grid intersections are determined as virtual points to obtain a virtual point cloud set. Thus, the distribution density of the virtual point cloud set is inversely proportional to the vertical distance.

[0049] The "predefined linear increasing relationship" refers to a predefined mathematical function or lookup table whose output value (sampling interval distance) increases linearly with the input value (vertical distance). For example, it could be a simple linear equation such as "sampling interval = base interval + scaling factor × vertical distance," where the base interval and scaling factor are pre-defined constants. Alternatively, it can be defined using a series of discrete vertical distance points and their corresponding sampling intervals, and in practical applications, the sampling interval at any vertical distance can be obtained through interpolation. "Determining the corresponding sampling interval distance" refers to the system calculating, based on the current vertical distance information of the moving platform relative to the reference plane, the sampling interval distance that virtual points should maintain on the height plane corresponding to that vertical distance, using the aforementioned predefined linear increasing relationship. This process can be real-time calculation or obtained through a lookup table. "The outline dimensions of the target space cross-section geometric features" represents the geometric boundary dimensions of the target vertical space in the horizontal direction. For example, for a square cross-section, its outline dimensions can be the side lengths in the x-axis direction and the y-axis direction. These dimensions are known in advance or obtained through other means and are used to limit the range of virtual point cloud generation. The "uniform mesh generation" operation generates a regular two-dimensional mesh within the contour of the target space's cross-sectional geometry on the height plane corresponding to the vertical distance, based on a determined sampling interval. The mesh lines are evenly spaced along the x and y axes at determined sampling intervals. "Mesh intersections are designated as virtual points" means that after uniform mesh generation, all mesh line intersections are treated as virtual points. These virtual points collectively constitute a virtual point cloud set located on that height plane. "The distribution density of the virtual point cloud set is inversely proportional to the vertical distance" is achieved by dynamically adjusting the sampling interval. As the vertical distance increases, the sampling interval also increases, resulting in fewer virtual points generated within the same area, thus reducing the virtual point cloud's distribution density; conversely, as the vertical distance decreases, the sampling interval decreases, and the virtual point cloud's distribution density increases.

[0050] This solution is implemented as follows: First, based on the current vertical distance of the mobile platform, the system accurately calculates the sampling interval distance that should be maintained between virtual points on the height plane using a preset linear increasing relationship. This linear increasing relationship design ensures that the sampling interval distance increases with the vertical distance, directly and intuitively linking the vertical distance information of the altitude-fixed radar with the sampling parameters of the virtual point cloud. This association method perfectly suits the characteristic that the uncertainty of altitude-fixed radar measurement increases with the vertical distance. That is, when the distance is far and the uncertainty is high, the sampling interval of the virtual point cloud is larger, avoiding the introduction of too many virtual points that may have large errors; while when the distance is close and the accuracy is high, the sampling interval is smaller, ensuring a sufficient density of virtual points to provide strong constraints. Second, the system uses the known contour dimensions of the target's spatial cross-sectional geometry to uniformly divide the height plane corresponding to the current vertical distance according to the determined sampling interval distance. Subsequently, the intersections of these grids are determined as virtual points, thereby constructing a virtual point cloud set on the height plane. This meshing method based on cross-sectional contour dimensions ensures that the generated virtual point cloud accurately matches the shape and size of the target's vertical space, avoiding the generation of invalid points beyond the actual spatial range and thus reducing unnecessary computational burden. Simultaneously, using mesh intersections directly as virtual points is simple and efficient, quickly generating a uniformly distributed virtual point cloud set with the required density. Through the synergistic effect of the above steps, the design goal of the virtual point cloud set's distribution density being inversely proportional to the vertical distance is ultimately achieved. This method of dynamically adjusting the virtual point cloud density provides specific and operable implementation details compared to the basic scheme's approach of only dynamically adjusting the sampling interval. It makes the virtual point cloud generation process more precise and controllable, stably providing reliable elevation direction constraints for subsequent synchronous positioning and mapping algorithms. When the vertical distance is small, the accuracy of the altitude-fixed radar measurement is high, resulting in a high virtual point cloud density, providing rich and high-confidence elevation constraints for synchronous positioning and mapping. When the vertical distance is large, the uncertainty of the altitude-fixed radar measurement increases, leading to a lower virtual point cloud density, which effectively suppresses the negative impact of high uncertainty information on pose estimation and avoids error accumulation. Therefore, this solution significantly improves the accuracy and robustness of 3D modeling in a vertical, narrow space by refining the virtual point cloud generation mechanism.

[0051] The following is a concrete example to illustrate this. Assume the target vertical space is an elevator shaft with a square cross-section, and its x-axis dimensions are known. and y-axis profile dimensions As a specific implementation method, when the mobile platform descends or ascends within the elevator shaft, the height-fixing radar will acquire its vertical distance relative to the bottom of the shaft in real time. The system can pre-determine the sampling interval distance using a linearly increasing relationship. For example, the sampling interval distances along the x-axis and y-axis can be calculated using a linear function, with the vertical distance as the metric. As input, output a random number. A linearly increasing sampling interval value. For example, the function could include a baseline sampling interval (e.g., 0.02 meters) and a density attenuation coefficient (e.g., 0.002 m). -1 This causes the sampling interval to increase proportionally with the vertical distance. Subsequently, the system uses the calculated sampling interval distance to measure the current vertical distance... On the corresponding height plane, based on the outline dimensions of the square cross-section and Perform uniform mesh generation. Specifically, this can be done along the x-axis from - Starting from / 2, a series of x-coordinate points are generated at intervals defined by a fixed x-axis sampling interval; in the y-axis direction, from - Starting at / 2, a series of y-coordinate points are generated at intervals defined by a fixed y-axis sampling interval. All combinations of these y-coordinate points and y-coordinate points are represented at a height of [missing information]. This constitutes a virtual point cloud set on that height plane. In this way, when the vertical distance... When the sampling interval increases, the same * Within the cross-sectional area, the number of generated virtual points will decrease, thereby reducing the distribution density of the virtual point cloud and achieving the effect that the distribution density is inversely proportional to the vertical distance.

[0052] Through the above technical solution, this application provides a stable and adaptive virtual point cloud generation mechanism, effectively solving the technical problem of insufficient elevation direction constraints and difficulty in guaranteeing modeling accuracy in vertically narrow spaces due to the increasing uncertainty of elevation radar measurements with increasing vertical distance. Specifically, by dynamically determining the sampling interval distance using a preset linear increasing relationship, the distribution density of the virtual point cloud can accurately match the measurement uncertainty of the elevation radar. When the vertical distance is small and the elevation radar accuracy is high, the virtual point cloud density is high, providing rich and high-confidence elevation constraints for the synchronous positioning and mapping algorithm, significantly improving the accuracy of short-range modeling. When the vertical distance is large and the elevation radar uncertainty is high, the virtual point cloud density is correspondingly reduced, effectively avoiding the introduction of virtual points with large errors into the synchronous positioning and mapping optimization process, thereby suppressing the accumulation of elevation direction errors and enhancing the robustness of long-range modeling. In addition, uniform mesh division based on the contour dimensions of the target space cross-sectional geometry ensures that the generated virtual point cloud accurately fits the actual space, avoiding the generation of invalid points and improving computational efficiency. Overall, this solution provides a refined, adaptive, and efficient virtual point cloud generation strategy for vertical space modeling based on virtual point cloud enhancement, significantly improving the accuracy and stability of 3D modeling in complex vertical scenarios such as elevator shafts.

[0053] In some embodiments, the specific steps in step S21 include: S211. Determine the sampling interval distance according to the following formula: ; ; ; ; in, Let be the sampling interval distance along the x-axis on the height plane corresponding to the vertical distance at time t. Let be the perpendicular distance at time t. The preset x-axis reference sampling interval distance. The preset density attenuation coefficient, Let be the sampling interval distance along the y-axis on the height plane corresponding to the vertical distance at time t. The preset y-axis reference sampling interval distance. The total vertical length (i.e., total height or total depth) of the target vertical space.

[0054] This application's solution, by introducing the aforementioned quantization calculation formula, provides a clear dynamic adjustment mechanism for the sampling interval distance between adjacent virtual points in the virtual point cloud set. When the mobile platform moves in the target's vertical space, the vertical distance acquired in real-time by the altitude-fixed radar... Substituting these values ​​into the formula, the sampling intervals of the x and y axes on the current height plane are calculated. and .because Greater than 0, as The increase, and A linear increase in density means that the sampling points of the virtual point cloud become sparser, resulting in a decrease in the distribution density of the virtual point cloud set. Conversely, when... When the sampling interval decreases, the virtual point cloud density increases. This dynamic adjustment strategy matches the measurement uncertainty characteristics of altitude-fixed radar, which typically has higher measurement accuracy at close range and lower accuracy at long range. Therefore, when the measurement accuracy of altitude-fixed radar is relatively high (…), the virtual point cloud density increases. In areas with relatively low (smaller) virtual point cloud density, the virtual point cloud density is higher, providing stronger constraints for subsequent synchronous localization and mapping algorithms; while in areas with lower (lower) altitude-fixed radar measurement accuracy... In larger areas, the virtual point cloud density is lower, which avoids the negative impact of low-precision information on the overall modeling results. Furthermore, by setting the reference sampling intervals for the x-axis and y-axis separately... and This solution can flexibly adapt to vertical spaces with different cross-sectional sizes and shapes, such as square or rectangular cross-sections, thereby improving its applicability and robustness. This dynamic and adaptive sampling interval adjustment mechanism enables the virtual point cloud to more effectively enhance the constraints in the elevation direction and suppress error accumulation, thus improving the accuracy and reliability of vertical space modeling.

[0055] In some embodiments, the target space cross-sectional geometry is square; The specific steps in step S22 include: S221. Obtain the virtual point cloud set according to the following formula: ; in, Let be the set of virtual point clouds located on the vertical distance-corresponding height plane at time t. Let be the x-axis coordinate of the i-th virtual point along the x-axis direction on the height plane corresponding to the vertical distance at time t. Let be the y-axis coordinate of the j-th virtual point along the y-axis direction on the height plane corresponding to the vertical distance at time t. The x-axis profile dimension is a square (i.e., the side length in the x-axis direction). Let be the index of the i-th virtual point along the x-axis on the height plane corresponding to the vertical distance at time t. The y-axis profile dimension is a square (i.e., the side length in the y-axis direction). It is the index of the j-th virtual point along the y-axis on the vertical distance plane at time t.

[0056] This application's solution addresses the problem of accurately generating virtual point clouds that meet position and density requirements in a square vertical space by explicitly defining the target space's cross-sectional geometry as square and providing a precise formula for generating virtual point cloud sets. Specifically, the solution first utilizes the aforementioned square geometric features to set clear boundary conditions for virtual point cloud generation. At each time t, when the mobile platform moves within the target vertical space and acquires the vertical distance... At that time, the system will determine the preset x-axis profile dimensions. and y-axis profile dimensions Combined with previously determined vertical distance Dynamically adjusted sampling interval distance and This is used to calculate the precise three-dimensional coordinates of each virtual point. The calculation process uses the center of the square cross-section as the origin, and... / 2 and - / 2 is used as the starting point, combined with the virtual point number. , and dynamic sampling interval , Accumulate, thus achieving the desired result at the current height plane. A series of uniformly distributed virtual points are generated on top. These virtual points together constitute... Set. Due to and It is based on vertical distance The density of the generated virtual point cloud set increases linearly, thus decreasing with increasing vertical distance. This matches the characteristic of uncertainty in altitude-fixed radar measurements increasing with distance, allowing distant virtual points to be assigned relatively low weights in subsequent synchronous positioning and mapping algorithms, effectively suppressing the accumulation of errors in the elevation direction. In this way, this scheme transforms the high-precision single-point vertical distance information acquired by altitude-fixed radar into virtual point cloud data with well-defined geometric shapes and dynamic density characteristics, enabling seamless fusion with environmental point cloud data acquired by lidar. This fusion not only compensates for the insufficient height constraint of lidar in vertically elongated "degraded scenarios," but also ensures the quality and effectiveness of the virtual point cloud through precise coordinate generation and adaptive density adjustment. This provides reliable augmentation data for subsequent inter-frame registration and global optimization, thereby improving the overall accuracy and closure of the target's vertical three-dimensional structural model.

[0057] In some embodiments, the specific steps in step S3 include: S31. Calculate the enhanced point cloud data according to the following formula: ; in, For the augmented point cloud data at time t, The data represents the environmental point cloud data collected at time t.

[0058] This technology aims to integrate point cloud data from different sources to form a more comprehensive and information-rich point cloud dataset. This can be achieved by having software programs call specific data processing functions or algorithm modules to perform calculations. For example, in a point cloud processing library (such as PCL or Open3D), code can be written to read, process, and merge point cloud objects. Alternatively, customized hardware acceleration units, such as FPGAs or GPUs, can be used to process point cloud data in parallel to improve computational efficiency. The above formula clarifies the specific composition of the enhanced point cloud data, namely, by using environmental point cloud data collected at time t. and located at vertical distance Virtual point cloud set on the corresponding height plane Perform a union operation. The union operation combines all points from two point cloud sets into a new set, ensuring that all original and virtual points are included in the final augmented point cloud data. In practice, this typically means simply concatenating or merging two point cloud data structures (e.g., two arrays or lists of point coordinates) to form a new point cloud data structure. Augmented point cloud data This refers to the comprehensive point cloud dataset obtained after fusion processing at time t. This dataset contains environmental information actually collected by LiDAR and virtual structure information generated based on prior knowledge, providing richer and more constrained data input for subsequent 3D reconstruction algorithms. Environmental point cloud data This refers to the 3D point cloud data of the target's vertical space environment, actually scanned and collected by the lidar on the mobile platform at time t. This point cloud data reflects the real geometric features of the target's vertical space (e.g., the sidewalls of an elevator shaft). Virtual point cloud set. This refers to the vertical distance of the moving platform relative to the reference plane at time t. In addition to the preset geometric features of the target space cross section, a set of virtual three-dimensional points with a specific distribution pattern are generated on the corresponding height plane. These virtual points are designed to compensate for the problem of sparse or missing data from lidar in certain areas (such as the bottom or top of the well, or areas lacking features), and to provide additional structural constraints in the elevation direction.

[0059] The operational logic of this solution is as follows: First, the mobile platform acquires environmental point cloud data in the vertical space of the target using LiDAR. Simultaneously, the vertical distance information of the mobile platform relative to the reference plane is obtained using a fixed-altitude radar. Subsequently, based on the preset geometric features of the target space cross-section (e.g., square), and combined with the real-time acquired vertical distance information, The system generates a virtual point cloud set on the height plane corresponding to the vertical distance. During this generation process, the sampling interval between adjacent virtual points in the virtual point cloud set will be adjusted according to the vertical distance. The size of the virtual point cloud set is dynamically adjusted so that the distribution density of the virtual point cloud set decreases as the vertical distance increases. This dynamic adjustment mechanism, for example, determines the sampling interval distance through a linear increasing relationship and calculates it using the formula in the above embodiment. This ensures that the generation of the virtual point cloud can reflect the characteristic that the uncertainty of the constant-altitude radar ranging increases with distance, thereby adaptively adjusting its weights in subsequent optimization. Based on this, this scheme uses a union operation to process the dynamically density-adjusted virtual point cloud set. Simultaneously collected environmental point cloud data The data is then fused. This fusion method ensures precise alignment in the time dimension, meaning that each frame of LiDAR data is merged with the virtual point cloud generated at the corresponding time. The characteristics of union operation guarantee that all environmental point information observed by the original LiDAR is completely preserved, avoiding the loss of any effective structural features. Simultaneously, pre-generated virtual point cloud information with highly adaptive density is seamlessly integrated, effectively supplementing the structural constraints in the vertical space along the height direction, especially in areas where LiDAR data is sparse or lacks features. Through this explicit and unified fusion rule, the final enhanced point cloud data... Not only is the data format unified, but it also includes real-world environmental information and intelligently generated structural prior information, providing a high-quality and highly reliable data foundation for subsequent synchronous localization and mapping algorithms (e.g., inter-frame registration and back-end optimization through an improved laser SLAM front-end). This significantly improves the modeling accuracy and robustness of the target's vertical three-dimensional structural model and effectively avoids inter-frame registration errors and global optimization interference caused by improper data fusion.

[0060] In some embodiments, the specific steps in step S4 include: S41. Perform inter-frame registration on the enhanced point cloud data according to the following steps S411: S411. Using the normal distribution transformation algorithm or the iterative nearest point algorithm, feature matching is performed on the enhanced point cloud data of consecutive frames to obtain the pose transformation information of the mobile platform in order to complete the inter-frame registration. S42. Perform global optimization of the enhanced point cloud data according to the following steps S421-S423: S421. Construct a backend factor graph containing observations corresponding to the virtual point cloud set; S422. Explicit uncertainty modeling is performed by assigning a covariance matrix to the observations corresponding to the virtual point cloud set; S423. Based on the covariance matrix, the weight of the virtual point cloud set in the back-end factor graph optimization process is adaptively adjusted according to the distribution density of the virtual point cloud set to suppress the accumulation of errors in the elevation direction; S43. Based on the enhanced point cloud data that has completed inter-frame registration and global optimization, output a closed 3D point cloud model containing the complete side wall structure, bottom structure and top structure, and use it as the 3D structural model of the target vertical space; the 3D structural model is used for building information modeling, installation verification or safety assessment.

[0061] In some of the solutions mentioned above in this application, a simultaneous localization and mapping algorithm is proposed. By performing inter-frame registration and global optimization on the enhanced point cloud data, a three-dimensional structural model of the target vertical space is output to complete the three-dimensional modeling of the vertical space. However, in this process, the conventional registration and optimization process does not design an optimization strategy adapted to the fused virtual point cloud, which cannot effectively utilize the elevation constraints brought by the virtual point cloud and is difficult to suppress the accumulation of errors in the elevation direction. At the same time, the native laser point cloud itself is difficult to obtain complete data of the top and bottom of the vertical space. The conventionally output model is often incomplete and not closed, which cannot meet the needs of subsequent practical applications.

[0062] To address this, this application further proposes a method for performing inter-frame registration and global optimization on enhanced point cloud data, and outputting a closed 3D point cloud model containing complete sidewall, bottom, and top structures. Specifically, the method includes the following steps: First, according to step S411, inter-frame registration is performed on the enhanced point cloud data. This involves using either the normal distribution transform algorithm or the iterative nearest point algorithm to obtain the pose transformation information of the mobile platform by performing feature matching on the enhanced point cloud data of consecutive frames, thus completing the inter-frame registration. The normal distribution transform algorithm is a point cloud registration method based on probability density function matching. It represents the reference point cloud as a combination of normal distributions and optimizes the transform parameters to maximize the probability that points in the point cloud to be registered have a normal distribution in the reference point cloud. This method has good robustness to point cloud density changes and noise. The iterative nearest point algorithm iteratively finds the nearest point correspondence between two sets of point clouds and calculates the optimal rigid body transformation based on these correspondences to minimize the distance between the two sets of point clouds. Variations of this algorithm include point-to-point ICP and point-to-surface ICP. Feature matching refers to identifying and associating points or regions with similar geometric or intensity characteristics in consecutive frames of enhanced point cloud data. This can be done by extracting keypoints and descriptors for matching, or by implicitly or explicitly establishing point correspondences in NDT or ICP algorithms. Pose transformation information refers to the relative position and attitude changes of the mobile platform between two consecutive moments, usually represented as a rigid body transformation matrix. The purpose of inter-frame registration is to align two or more consecutively acquired point cloud data frames in the same coordinate system, eliminating relative displacement and rotation caused by the motion of the mobile platform, and providing local consistency for subsequent global optimization.

[0063] Secondly, the enhanced point cloud data is globally optimized according to steps S421-S423. Step S421 involves constructing a backend factor graph containing observations corresponding to the virtual point cloud set. The backend factor graph is commonly used for backend optimization in Simultaneous Localization and Mapping (SLAM) systems. It is a bipartite graph composed of variable nodes (representing poses, map points, etc. to be estimated) and factor nodes (representing constraints such as sensor observations and motion models). The optimal estimate of the variable nodes is solved by minimizing the error function defined by all factor nodes. The virtual point cloud set corresponding to observations refers to introducing the virtual point cloud set as a special type of sensor observation into the factor graph, providing information about the vertical position of the mobile platform. Step S422 performs explicit uncertainty modeling by assigning a covariance matrix to the observations corresponding to the virtual point cloud set. The covariance matrix is ​​used to quantify the correlation and uncertainty between random variables, and in SLAM, it is typically used to represent the noise characteristics of sensor observations or the uncertainty of state estimation. Explicit uncertainty modeling refers to explicitly considering and representing the uncertainty of sensor observations or state estimates during the optimization process. By assigning a covariance matrix, the optimization algorithm can adjust its weight in the objective function based on the reliability of the observations. Step S423 adaptively adjusts the weight of the virtual point cloud set in the back-end factor graph optimization process based on the distribution density of the covariance matrix to suppress the accumulation of errors in the elevation direction. Adaptive weight adjustment refers to dynamically changing the influence of the virtual point cloud set in the factor graph optimization based on its characteristics (such as distribution density, which is related to vertical distance).

[0064] Finally, based on the enhanced point cloud data with completed inter-frame registration and global optimization, a closed 3D point cloud model containing complete sidewall, bottom, and top structures is output as the 3D structural model of the target vertical space. A closed 3D point cloud model refers to a complete 3D point cloud representation without obvious holes or missing areas, accurately reflecting the geometry and topology of the target vertical space. The 3D structural model can be applied to Building Information Modeling (BIM) modeling, installation verification, or safety assessment. Building Information Modeling (BIM) is an information-based building design and management method based on 3D models; high-precision 3D point cloud models can serve as the foundational data for BIM modeling. Installation verification refers to checking whether the position, dimensions, and orientation of equipment or components meet requirements by comparing them with the design model or standards before or after installation. Safety assessment refers to evaluating the structural integrity, potential hazardous areas, and escape routes of the target vertical space (such as elevator shafts).

[0065] This application's solution achieves high-precision modeling of vertical space by inputting enhanced point cloud data incorporating virtual point clouds into improved laser SLAM front-end and back-end optimization modules. During inter-frame registration, feature matching of the enhanced point cloud data is performed using a normal distribution transformation algorithm or an iterative nearest-point algorithm. This enables the acquisition of accurate mobile platform pose transformation information based on the enhanced point cloud incorporating virtual point clouds, providing a reliable initial pose foundation for subsequent global optimization. In the global optimization phase, a back-end factor graph containing observations corresponding to the virtual point cloud set is first constructed. This transforms the single-point altitude information output by the original fixed-altitude radar into observation constraints acceptable to the point cloud SLAM framework, solving the problem of high-precision fixed-altitude information being difficult to directly integrate into the point cloud SLAM framework and providing additional constraints in the elevation direction. Subsequently, explicit uncertainty modeling is performed by assigning a covariance matrix to the observations corresponding to the virtual point cloud set, quantifying the uncertainty of the virtual point cloud itself as a function of altitude, providing a basis for subsequent weight adjustment. Finally, based on the obtained covariance matrix, the weights of virtual point cloud observations in the back-end factor graph optimization process are adaptively adjusted according to the distribution density of the virtual point cloud. Because the distribution density of virtual point clouds decreases with increasing vertical distance, while their inherent uncertainty increases with distance, adaptive adjustment allows for higher weight constraints on closer virtual point clouds (which have higher accuracy and lower uncertainty), and lower weights on farther virtual point clouds (which have lower accuracy and higher uncertainty). This perfectly aligns with the uncertainty variation pattern of virtual point clouds, effectively suppressing error accumulation in the elevation direction and mitigating the drift problem in vertically degraded SLAM scenarios. Finally, based on the enhanced point cloud with completed registration optimization, a closed 3D point cloud model containing complete sidewalls, bottom, and top structures is output. This supplements the missing top and bottom region data from the original laser point cloud, resulting in a complete closed 3D structural model. Furthermore, the application direction of the output model is clearly defined, directly supporting downstream building information modeling, installation verification, and safety assessment, thus enhancing the overall practicality of the solution.

[0066] As a specific implementation method, in the inter-frame registration step S411, the NDT (Normal Distributions Transform) class or the NDT (Iterative Closest Point) class in PCL (Point Cloud Library) can be used. For example, when using the normal distribution transformation algorithm, the target point cloud can be set to the enhanced point cloud data of the previous frame, and the source point cloud can be set to the enhanced point cloud data of the current frame. By calling the registration function, the pose transformation matrix of the current frame relative to the previous frame can be calculated. In the global optimization step S421, the back-end factor graph can be constructed using optimization libraries such as GTSAM (Georgia Tech Smoothing and Mapping) or Ceres Solver. Among them, the variable nodes can include the pose of the mobile platform (e.g., represented as SE(3) group elements), and the factor nodes can include motion model factors, lidar observation factors, and virtual point cloud observation factors. For the virtual point cloud observation factor, its observation value can be set to the vertical distance measured by the fixed-altitude radar. In step S422, when assigning a covariance matrix to the observations corresponding to the virtual point cloud set, the measurement error model of the altitude-fixed radar can be used. To determine this. For example, the covariance matrix can be a diagonal matrix, where the variance in the vertical direction (z-axis) can be set to... The values ​​are proportional, while the variance in the horizontal direction (x, y axes) can be set to a smaller constant or determined based on other sensor noise models. In step S423, when adaptively adjusting the weights, the distribution density of the virtual point cloud set (or its corresponding vertical distance) can be considered during the factor graph optimization process. ), dynamically adjusting the information matrix (the inverse of the covariance matrix) corresponding to its observed factors. For example, when When the density is small, the distribution density is high, and the vertical elements of the information matrix can be set larger, indicating that the observation is more reliable and has a higher weight. When the density is large, the distribution density is low, the elements of the information matrix are correspondingly reduced, and the weight is decreased. Finally, in step S43, after optimization, all optimized poses and map points can be extracted from the factor map to generate the final enhanced point cloud data. These point cloud data can be visualized using point cloud processing software (such as CloudCompare, MeshLab) or a custom program, generating a closed 3D point cloud model containing the complete sidewall structure, bottom structure, and top structure. This model can be exported to common point cloud formats (such as .pcd, .ply) or mesh model formats (such as .obj, .stl) for import into building information modeling software (such as Revit, ArchiCAD) for subsequent modeling, installation verification, or safety assessment.

[0067] Through the above technical solution, this application effectively solves the problem that conventional synchronous positioning and mapping algorithms are prone to error accumulation and model incompleteness in the elevation direction in vertical narrow spaces. This solution effectively utilizes the elevation constraints brought by the virtual point cloud by performing inter-frame registration and global optimization on the enhanced point cloud data, and designing an adaptive optimization strategy for the virtual point cloud. The virtual point cloud set is introduced as an observation into the backend factor map, and explicit uncertainty modeling is performed. Simultaneously, based on the covariance matrix, the weight of the virtual point cloud set in the backend factor map optimization process is adaptively adjusted according to its distribution density, allowing highly reliable close-range observations to play a greater role and effectively suppressing error accumulation in the elevation direction. Furthermore, by outputting a closed 3D point cloud model containing complete sidewall, bottom, and top structures, the lack of data in the top and bottom regions of the native laser point cloud is compensated for, making the output model more complete and closed. This meets the requirements for model accuracy and completeness in practical applications such as building information modeling, installation verification, or safety assessment.

[0068] Please refer to Figure 3 , Figure 3 This is a vertical space modeling device based on virtual point cloud enhancement in some embodiments of the present invention (the vertical space modeling device based on virtual point cloud enhancement adopts the vertical space modeling method based on virtual point cloud enhancement in the above embodiments, and the specific process is referred to the corresponding steps above). The vertical space modeling device based on virtual point cloud enhancement is integrated into the back-end control device in the form of a computer program, including: The acquisition module 100 is used to acquire environmental point cloud data collected by the mobile platform in the vertical space of the target, and simultaneously acquire the vertical distance information of the mobile platform relative to the reference plane. The generation module 200 is used to generate a set of virtual point clouds located on the height plane corresponding to the vertical distance based on the preset geometric features of the target space cross section and combined with the vertical distance information. The sampling interval between adjacent virtual points in the set of virtual point clouds is dynamically adjusted according to the size of the vertical distance so that the distribution density of the set of virtual point clouds decreases as the vertical distance increases. The fusion module 300 is used to obtain enhanced point cloud data by fusing environmental point cloud data with the adjusted virtual point cloud set; Output module 400 is used to output a three-dimensional structural model of the target in vertical space by performing inter-frame registration and global optimization on the enhanced point cloud data.

[0069] Please refer to Figure 4 , Figure 4This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The present invention provides an electronic device 13, including: a processor 1301 and a memory 1302. The processor 1301 and the memory 1302 are interconnected and communicate with each other via a communication bus 1303 and / or other forms of connection mechanism (not shown). The memory 1302 stores computer-readable instructions executable by the processor 1301. When the electronic device is running, the processor 1301 executes these computer-readable instructions to execute the vertical space modeling method based on virtual point cloud enhancement in any optional implementation of the above embodiments, to achieve the following function: obtaining the vertical position of the mobile platform in the target vertical direction. The system collects environmental point cloud data within the space and simultaneously acquires the vertical distance information of the mobile platform relative to the reference plane. Based on the preset geometric features of the target space cross-section and combined with the vertical distance information, it generates a virtual point cloud set located on the height plane corresponding to the vertical distance. According to the magnitude of the vertical distance, the sampling interval between adjacent virtual points in the virtual point cloud set is dynamically adjusted so that the distribution density of the virtual point cloud set decreases as the vertical distance increases. By fusing the environmental point cloud data with the adjusted virtual point cloud set, enhanced point cloud data is obtained. By performing inter-frame registration and global optimization on the enhanced point cloud data, a three-dimensional structural model of the target vertical space is output.

[0070] This invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it executes the vertical space modeling method based on virtual point cloud enhancement in any optional implementation of the above embodiments to achieve the following functions: acquiring environmental point cloud data collected by a mobile platform in the target vertical space, and simultaneously acquiring the vertical distance information of the mobile platform relative to a reference plane; generating a virtual point cloud set located on the height plane corresponding to the vertical distance based on the preset geometric features of the target space cross section and combined with the vertical distance information, and dynamically adjusting the sampling interval distance between adjacent virtual points in the virtual point cloud set according to the size of the vertical distance so that the distribution density of the virtual point cloud set decreases as the vertical distance increases; obtaining enhanced point cloud data by fusing the environmental point cloud data with the adjusted virtual point cloud set; and outputting a three-dimensional structural model of the target vertical space by performing inter-frame registration and global optimization on the enhanced point cloud data.

[0071] The computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0072] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0073] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0074] Furthermore, in the various embodiments of the present invention, the functional modules can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0075] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0076] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., 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 vertical spatial modeling method based on virtual point cloud enhancement, characterized in that, Includes the following steps: S1. Acquire environmental point cloud data collected by the mobile platform in the vertical space of the target, and simultaneously acquire the vertical distance information of the mobile platform relative to the reference plane; S2. Based on the preset geometric features of the target space cross section and combined with the vertical distance information, a virtual point cloud set located on the height plane corresponding to the vertical distance is generated. According to the magnitude of the vertical distance, the sampling interval distance between adjacent virtual points in the virtual point cloud set is dynamically adjusted so that the distribution density of the virtual point cloud set decreases as the vertical distance increases; the geometric features of the target space cross section are square. S3. Enhanced point cloud data is obtained by fusing the environmental point cloud data with the adjusted virtual point cloud set; S4. By performing inter-frame registration and global optimization on the enhanced point cloud data, output the three-dimensional structural model of the target's vertical space; The specific steps in step S2 include: S21. Based on the magnitude of the vertical distance, determine the corresponding sampling interval distance using a preset linear increasing relationship, so that the sampling interval distance increases as the vertical distance increases; S22. Based on the contour dimensions of the geometric features of the target space cross section, in the height plane corresponding to the vertical distance, perform uniform grid division according to the determined sampling interval distance, and determine the grid intersections as virtual points to obtain the virtual point cloud set; The specific steps in step S21 include: S211. The sampling interval distance is determined according to the following formula: ； ； ； ； in, Let be the sampling interval distance along the x-axis on the height plane corresponding to the vertical distance at time t. Let be the perpendicular distance at time t. The preset x-axis reference sampling interval distance. The preset density attenuation coefficient, Let be the sampling interval distance along the y-axis on the height plane corresponding to the vertical distance at time t. The preset y-axis reference sampling interval distance. The total vertical length of the target vertical space; The specific steps in step S4 include: S41. Perform inter-frame registration on the enhanced point cloud data according to the following step S411: S411. Using the normal distribution transformation algorithm or the iterative nearest point algorithm, feature matching is performed on the enhanced point cloud data of consecutive frames to obtain the pose transformation information of the mobile platform, so as to complete the inter-frame registration. S42. Perform global optimization on the enhanced point cloud data according to the following steps S421-S423: S421. Construct a backend factor graph containing the observations corresponding to the virtual point cloud set; S422. Explicit uncertainty modeling is performed by assigning a covariance matrix to the observations corresponding to the virtual point cloud set; S423. Based on the covariance matrix, the weight of the virtual point cloud set in the back-end factor graph optimization process is adaptively adjusted according to the distribution density of the virtual point cloud set to suppress the accumulation of errors in the elevation direction; S43. Based on the enhanced point cloud data that has completed inter-frame registration and global optimization, output a closed three-dimensional point cloud model containing the complete side wall structure, bottom structure and top structure, and use it as the three-dimensional structural model of the target vertical space.

2. The vertical space modeling method based on virtual point cloud enhancement according to claim 1, characterized in that, The target space cross-sectional geometry is square; The specific steps in step S22 include: S221. The virtual point cloud set is obtained according to the following formula: ； in, Let be the set of virtual point clouds located on the height plane corresponding to the vertical distance at time t. Let be the x-axis coordinate of the i-th virtual point along the x-axis direction on the height plane corresponding to the vertical distance at time t. Let be the y-axis coordinate of the j-th virtual point along the y-axis direction on the height plane corresponding to the vertical distance at time t. The x-axis profile dimension of the square is given. Let be the index of the i-th virtual point along the x-axis on the height plane corresponding to the vertical distance at time t. Let be the y-axis profile dimension of the square. It is the index of the j-th virtual point along the y-axis on the vertical distance plane at time t.

3. The vertical space modeling method based on virtual point cloud enhancement according to claim 2, characterized in that, The specific steps in step S3 include: S31. Calculate the enhanced point cloud data according to the following formula: ； in, For the augmented point cloud data at time t, The data represents the environmental point cloud data collected at time t.

4. A vertical spatial modeling apparatus based on virtual point cloud enhancement for implementing the vertical spatial modeling method based on virtual point cloud enhancement as described in any one of claims 1-3, characterized in that, include: The acquisition module is used to acquire environmental point cloud data collected by the mobile platform in the vertical space of the target, and simultaneously acquire the vertical distance information of the mobile platform relative to the reference plane. The generation module is used to generate a set of virtual point clouds located on the height plane corresponding to the vertical distance, based on the preset geometric features of the target space cross section and the vertical distance information, and to dynamically adjust the sampling interval between adjacent virtual points in the set of virtual point clouds according to the size of the vertical distance, so that the distribution density of the set of virtual point clouds decreases as the vertical distance increases. The fusion module is used to obtain enhanced point cloud data by fusing the environmental point cloud data with the adjusted virtual point cloud set; The output module is used to output a three-dimensional structural model of the target vertical space by performing inter-frame registration and global optimization on the enhanced point cloud data.

5. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, perform the steps of the vertical spatial modeling method based on virtual point cloud enhancement as described in any one of claims 1-3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it performs the steps in the vertical spatial modeling method based on virtual point cloud enhancement as described in any one of claims 1-3.

Images