A two-dimensional drawing automatic generation method and system based on three-dimensional point cloud

Abstract

The application provides a kind of two-dimensional drawing automatic generation method and system based on three-dimensional point cloud.The method includes obtaining original three-dimensional space data from multiple data sources, the data sources include laser scanning equipment, near-ground photogrammetry equipment and unmanned aerial vehicle oblique photography equipment;Original three-dimensional space data is fused by registration, and a three-dimensional point cloud model is constructed;According to the preset cutting plane, the three-dimensional point cloud model is cut to generate the corresponding depth image;Gradient operation is carried out on the depth image, and the feature line is extracted by Canny edge detection;According to the gradient amplitude of the feature line, a threshold mask is set, the feature line is classified according to the preset rule, and the classification result is filtered and matched based on the feature line obtained by edge detection, to generate the corresponding feature line layer and classified output;Based on different layer data, two-dimensional drawing is generated, and the corresponding relationship between two-dimensional drawing and original three-dimensional point cloud model is established, so that two-dimensional drawing can be traced back to any position of three-dimensional model.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional surveying and mapping technology, and in particular to a method and system for automatically generating two-dimensional drawings based on three-dimensional point clouds. Background Technology

[0002] With the development of laser scanning, near-ground photogrammetry, and UAV oblique photogrammetry, acquiring high-precision 3D spatial data has become an important means of architectural surveying and engineering design. However, existing technologies still face many problems in converting multi-source 3D point cloud data into high-precision 2D drawings. Differences in coordinate systems, sampling densities, and coverage among multi-source data lead to complex point cloud fusion and registration processes that are prone to errors. Traditional 2D drawing generation relies on manual or semi-automated operations, which are inefficient and prone to introducing errors. Furthermore, the accuracy of contour feature extraction is insufficient, making it difficult to fully reflect spatial details, and the generated 2D drawings lack a direct correspondence with the original 3D model, limiting the efficiency of subsequent analysis and modification. Therefore, there is an urgent need for a technical solution that can efficiently process multi-source 3D data, generate high-precision 2D drawings, and establish a correspondence between 2D drawings and 3D models. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for automatically generating two-dimensional drawings based on three-dimensional point clouds, so as to realize the fusion of multi-source three-dimensional spatial data, accurate feature line extraction, and traceable correspondence between two-dimensional drawings and three-dimensional models, thereby improving the efficiency and accuracy of two-dimensional drawing generation.

[0004] In a first aspect, the present invention provides a method for automatically generating two-dimensional drawings based on three-dimensional point clouds, comprising the following steps: Acquire raw 3D spatial data from multiple data sources, including laser scanning equipment, near-ground photogrammetry equipment, and UAV oblique photogrammetry equipment; The original three-dimensional spatial data is registered and fused to construct a three-dimensional point cloud model; The three-dimensional point cloud model is cropped according to a preset cutting plane to generate a corresponding depth image; Gradient calculations are performed on the depth image, and feature lines are extracted using Canny edge detection. A threshold mask is set according to the gradient magnitude of the feature lines, the feature lines are classified according to a preset rule, and the classification results are filtered and matched based on the feature lines obtained by Canny edge detection to generate corresponding feature line layers and output them in categories. Two-dimensional drawings are generated based on data from different layers, and a correspondence is established between the two-dimensional drawings and the original three-dimensional point cloud model, so that the two-dimensional drawings can be traced back to any position in the three-dimensional model.

[0005] Optionally, the specific steps for obtaining raw three-dimensional spatial data from multiple data sources include: Multiple scanning stations are set up in the target garden area using a 3D laser scanning device. The height and rotation angle of the scanner are adjusted at each scanning station to perform a comprehensive scan of the target object's surface to obtain point cloud data with overlapping areas. Images are acquired on the ground and indoors using near-ground photogrammetry equipment to obtain image data for 3D registration; Aerial photography is carried out using drone oblique photography equipment along preset routes, including orthophoto routes and multi-directional oblique routes. The gimbal angle, heading overlap rate and lateral overlap rate are set to cover the target area from multiple angles in the air, forming continuous and registrable image data. The original three-dimensional spatial data is obtained by combining three-dimensional laser scanning data, near-ground photogrammetry data, and UAV oblique photogrammetry data.

[0006] Optionally, the specific steps for registering and fusing the original three-dimensional spatial data to construct a three-dimensional point cloud model include: Point cloud registration is performed based on multi-source data. Laser scanned point clouds are used as reference data. Control points are determined based on feature points in photogrammetric images. The point clouds are then manually or automatically aligned using these control points to reduce deviations caused by seasonal changes, plant growth, and differences in furnishings. After registration is completed, actual size data is obtained at key locations and compared with the measured values ​​of the 3D point cloud model to calculate the error rate. The point cloud model is then fine-tuned to keep the overall error within a preset range. A triangular network model is constructed based on the registered point cloud data. A three-dimensional mesh structure is established through vertices, edges, and faces. The three-dimensional mesh model is then texture-mapped using multi-source image data, enabling the three-dimensional mesh model to realistically reflect the surface morphology and detailed features of the target area.

[0007] Optionally, the specific steps of cropping the 3D point cloud model according to a preset cutting plane to generate a corresponding depth image include: The completed 3D point cloud model is trimmed, and an orthophoto is generated according to a preset spatial region or key structural range. The orthophoto has a scale bar. A depth image is generated based on the orthophoto, and the pixel size of the depth image is set. The depth value in the depth image represents the distance of the target object surface relative to the observation reference plane. The depth image is subjected to grayscale mapping processing to convert the depth value into image grayscale data, and a multi-level depth map is generated according to the characteristics of the target structure.

[0008] Optionally, the specific steps of performing gradient calculation on the depth image include: The depth image is normalized so that the depth values ​​are mapped to a preset range; The normalized depth image is filtered to remove noise. Gradient calculations are performed on the filtered depth image to obtain the gradient magnitude information of each pixel and generate the corresponding gradient magnitude map.

[0009] Optionally, the specific steps for extracting feature lines using Canny edge detection further include: Non-maximum suppression processing is performed based on the gradient magnitude map. The gradient magnitude of each pixel is compared with that of the adjacent pixels along the gradient direction, and local maxima are retained to refine the edge response. Based on the non-maximum suppression processing results, a first threshold and a second threshold are set, and the gradient magnitude is determined by dual thresholds. The first threshold is greater than the second threshold. Pixels greater than the first threshold are determined as first-class edge pixels, and pixels greater than or equal to the second threshold and not greater than the first threshold are determined as second-class edge pixels. Non-edge pixels less than the second threshold are removed. Perform connectivity analysis on the second type of edge pixels, and retain and merge the second type of edge pixels that are connected to the first type of edge pixels to generate continuous edge lines; The continuous edge lines are output as feature lines.

[0010] Optionally, the specific steps of setting a threshold mask based on the gradient magnitude of the feature lines, classifying the feature lines according to preset rules, filtering the classification results based on the feature lines obtained by Canny edge detection, generating corresponding layer data, and outputting the data include: Based on the gradient magnitude map, multiple threshold intervals are determined, and classification rules are established according to the gradient magnitude ranges corresponding to different types of structural elements. The pixels in the gradient magnitude map are marked according to the threshold range to generate three types of feature line masks; The feature line mask is applied to the feature line to classify and identify the feature line, so that different categories of feature lines form corresponding classification results; Based on the feature lines obtained by the Canny edge detection, the classification results are filtered and matched to retain the valid classification data corresponding to the feature lines. Based on the classification results, corresponding layer data is generated and output in layers according to the classification categories.

[0011] Optionally, the specific steps of generating a two-dimensional drawing based on the layer data and establishing a correspondence between the two-dimensional drawing and the original three-dimensional point cloud model, so that the two-dimensional drawing can be traced back to any position in the three-dimensional model, include: Based on the three-dimensional point cloud model and the set of layered feature lines, a projection plane is determined, and the three-dimensional information in the three-dimensional point cloud model is projected onto the projection plane to generate a corresponding two-dimensional view. According to the user-preset cutting plane parameters, the three-dimensional point cloud model is sliced ​​along the specified direction to extract the geometric information corresponding to the cutting plane, and a two-dimensional cutting image is generated by combining the layered feature line set. In the generated 2D drawing, a correspondence is established between each drawing element and the original 3D point cloud data to achieve a traceable mapping from the 2D drawing to the original 3D point cloud data. The generated two-dimensional drawings are annotated, scaled, and standardized to obtain two-dimensional drawings that conform to preset drawing specifications.

[0012] Optionally, the method further includes the following steps: Match the current 2D drawings with historical 2D drawings by corresponding points or regions, and classify and compare buildings, rockeries, flower beds and other garden components. Based on the matching results, the geometric differences and spatial position changes between different two-dimensional drawings are calculated to form quantitative analysis data, which includes position offset, area change and shape difference. Based on the quantitative analysis data, information on changes in the garden layout is generated, and newly added, removed, or adjusted components are identified to achieve dynamic updating and traceable management of garden space information.

[0013] Secondly, the present invention provides an automatic two-dimensional drawing generation system based on three-dimensional point clouds, comprising: Multiple data acquisition units are used to acquire raw three-dimensional spatial data from multiple data sources, including laser scanning equipment, near-ground photogrammetry equipment, and UAV oblique photogrammetry equipment. A data processing unit, connected to the multiple data acquisition units, is used to register and fuse the original three-dimensional spatial data to construct a three-dimensional point cloud model. The slicing unit, connected to the data processing unit, is used to slice the three-dimensional point cloud model according to a preset slicing plane to generate a corresponding depth image; The feature extraction unit, connected to the slicing unit, is used to perform gradient calculation on the depth image, extract feature lines through Canny edge detection, set a threshold mask according to the gradient magnitude of the feature lines, classify the feature lines according to preset rules, and perform filtering and matching processing on the classification results based on the feature lines obtained by Canny edge detection to generate corresponding feature line layers and output them in categories. The two-dimensional drawing generation unit, connected to the feature extraction unit, is used to generate two-dimensional drawings based on different layer data and establish a correspondence between the two-dimensional drawings and the original three-dimensional point cloud model, so that the two-dimensional drawings can be traced back to any position of the three-dimensional model.

[0014] According to the present invention, multi-source three-dimensional spatial data from laser scanning, near-ground photogrammetry, and UAV oblique photography can be efficiently integrated. After precise registration and fusion, a high-precision three-dimensional point cloud model can be constructed. Through preset cutting plane and depth image processing, feature lines can be extracted in layers and two-dimensional drawings can be generated. The generated two-dimensional drawings can establish a traceable correspondence with the three-dimensional model, thereby significantly improving the accuracy and efficiency of two-dimensional drawing generation and reducing the workload of manual surveying and drawing.

[0015] Furthermore, by combining multi-level feature line extraction, sectioning and clipping with triangular mesh modeling, high-precision two-dimensional drawings of complex garden areas and target components can be generated. These drawings can be compared and analyzed with historical drawings to assess layout changes, enabling quantitative monitoring and traceable management of the addition, removal, and adjustment of garden components. Simultaneously, this method supports automatic annotation, scale adjustment, and drawing standardization, providing accurate, intuitive, and efficient digital references for garden planning, maintenance, and heritage protection, thus enhancing the scientific rigor and reliability of overall spatial information management and decision-making.

[0016] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below. Attached Figure Description

[0017] Figure 1 A schematic flowchart of an automatic two-dimensional drawing generation method based on three-dimensional point cloud according to an embodiment of the present invention is shown; Figure 2 It shows Figure 1 A schematic flowchart of step S100 for acquiring raw 3D spatial data from multiple data sources; Figure 3 It shows Figure 1 A schematic flowchart of step S200, which involves registering and fusing the original three-dimensional spatial data to construct a three-dimensional point cloud model; Figure 4It shows Figure 1 A schematic flowchart of step S300, which involves cropping the 3D point cloud model according to a preset cutting plane to generate a corresponding depth image. Figure 5 A structural block diagram of a two-dimensional drawing automatic generation system based on three-dimensional point clouds according to an embodiment of the present invention is shown. Detailed Implementation

[0018] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, it should be noted that, for ease of description, only the parts relevant to this application are shown in the accompanying drawings, not the entire structure. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0019] The terms “comprising” and “having”, and any variations thereof, used in this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] Figure 1 A schematic flowchart illustrating an automatic generation method for two-dimensional drawings based on three-dimensional point clouds according to an embodiment of the present invention is shown. Figure 1 As shown, the method for automatically generating two-dimensional drawings based on three-dimensional point clouds includes: Step S100: Obtain raw 3D spatial data from multiple data sources, including laser scanning equipment, near-ground photogrammetry equipment, and UAV oblique photogrammetry equipment.

[0022] In this step, data is acquired from multiple angles, heights, and orientations of the target area using 3D laser scanning equipment, near-ground photogrammetry equipment, and UAV oblique photogrammetry equipment. The 3D laser scanning equipment deploys multiple scanning stations, adjusting the scanner height and rotation angle to cover the target object's surface, achieving complete coverage and overlap of point cloud data. Near-ground photogrammetry equipment captures high-overlap photos on the ground and indoors, providing accurate images for subsequent registration and texture supplementation. UAV oblique photogrammetry performs orthophoto aerial photography and multi-directional oblique aerial photography along a preset flight path, setting the gimbal angle, heading overlap rate, and lateral overlap rate to achieve continuous aerial coverage. Through this multi-source data acquisition, a rich set of original 3D spatial data is formed, providing a reliable foundation for constructing a high-precision 3D point cloud model while ensuring complete acquisition of the target area's surface and complex structural details.

[0023] Step S200: Register and fuse the original three-dimensional spatial data to construct a three-dimensional point cloud model.

[0024] In this step, 3D data from different sources undergoes coordinate unification and spatial alignment. Using laser-scanned point clouds as the primary reference, control points are set based on feature points in the photogrammetric images to manually or automatically align the point clouds, reducing deviations caused by seasonal changes, plant growth, and differences in furnishings. After registration, the actual dimensions are compared with the model's measured values ​​through distance measurements at key locations. Combined with software optimization or manual fine-tuning, this ensures that the overall error is controlled within a preset range. This step generates a high-precision, seamless 3D point cloud model, providing a stable and accurate spatial data foundation for subsequent depth cutting, feature extraction, and 2D drawing generation, thus improving the accuracy and reliability of the entire method.

[0025] Step S300: According to the preset cutting plane, the three-dimensional point cloud model is cropped to generate the corresponding depth image.

[0026] In this step, a local orthophoto of the 3D point cloud model is generated based on the cutting plane parameters preset by the user or the system. The distance from each point to the reference plane is calculated to generate depth information, which is then converted into a processable depth image through grayscale mapping. This step not only extracts key geometric information but also provides a clear and detailed data foundation for subsequent gradient calculations and feature line extraction, ensuring the spatial authenticity and traceability of the 2D drawing.

[0027] Step S400: Perform gradient calculation on the depth image and extract feature lines using Canny edge detection.

[0028] In this step, gradient operations are performed on the depth image to calculate the local depth change rate, followed by feature lines extracted using the Canny edge detection algorithm. During processing, Gaussian filtering is applied to remove noise, non-maximum suppression is used to refine edges, and a dual-threshold strategy is employed to distinguish between high-response and low-response edges, forming a preliminary candidate set of feature lines. This process accurately identifies key contours and surface variations in the target region, providing a precise line basis for hierarchical contour classification and 2D drawing generation, while maintaining the integrity and continuity of edges in complex areas.

[0029] Step S500: Set a threshold mask based on the gradient magnitude of the feature lines, classify the feature lines according to preset rules, and filter and match the classification results based on the feature lines obtained by Canny edge detection to generate corresponding feature line layer data and output the classification.

[0030] In this step, the gradient magnitude of each feature line extracted from the depth image is first calculated. Based on this, according to a pre-defined layering rule, the feature lines are divided into three levels: thick, medium, and thin lines. Multi-level threshold masks are then constructed based on the gradient magnitude ranges corresponding to different levels, allowing for the differentiation of contour information with different spatial scales and structural attributes. Subsequently, the layering results are compared with the feature lines obtained through Canny edge detection in the previous stage. By constraining edge continuity and response consistency, adjustments are made to address any breaks, misclassifications, or redundant line segments that may occur during the layering process, ensuring that the feature lines at each level maintain consistency and integrity in their spatial distribution.

[0031] After completing the screening and matching process, the outlines of each level are processed in a targeted manner based on the structural characteristics of different objects in the target area. Specifically, for structures with clear boundaries, thicker lines are enhanced to highlight boundary precision; for areas with complex shapes and rich details, midline levels are retained to balance overall form and local features; and for textured or overlapping areas, thinner lines are used to suppress interference from irrelevant details. Finally, based on the feature line results after layering and screening and matching, corresponding multi-layer data is generated and output, ensuring clear distinction between each level of information in the two-dimensional representation. This provides a well-structured and hierarchically distinct data foundation for subsequent two-dimensional drawing generation.

[0032] Step S600: Generate two-dimensional drawings based on different layer data, and establish the correspondence between the two-dimensional drawings and the original three-dimensional point cloud model, so that the two-dimensional drawings can be traced back to any position of the three-dimensional model.

[0033] In this step, layered feature lines are projected onto any user-specified 2D plane, and 2D drawings are generated by combining the cutting plane and the geometric information of the 3D model. During the generation process, the correspondence between each drawing element and the 3D point cloud data is preserved, allowing the 2D drawings to be traced back to any position in the original 3D model. Simultaneously, automatic annotation, scale adjustment, and drawing standardization are performed to ensure the 2D drawings conform to high-precision engineering drawing standards. This step not only completes the generation of 2D drawings but also achieves an efficient mapping between 3D spatial data and 2D representation, providing accurate and traceable digital references for subsequent garden maintenance, planning, and heritage protection.

[0034] According to the above embodiments, multi-source three-dimensional spatial data from laser scanning, near-ground photogrammetry, and UAV oblique photography can be efficiently integrated. After precise registration and fusion, a high-precision three-dimensional point cloud model is constructed. Through preset cutting plane and depth image processing, feature lines are extracted in layers and two-dimensional drawings are generated. The generated two-dimensional drawings can establish a traceable correspondence with the three-dimensional model, thereby significantly improving the accuracy and efficiency of two-dimensional drawing generation and reducing the workload of manual surveying and drawing.

[0035] Figure 2 It shows Figure 1 A schematic flowchart illustrating step S100, which involves acquiring raw 3D spatial data from multiple data sources. Figure 2 As shown, step S100 specifically includes: Step S110: Multiple scanning stations are set up in the target garden area using a 3D laser scanning device. The height and rotation angle of the scanner are adjusted at each scanning station to perform a coverage scan on the surface of the target object to obtain point cloud data with overlapping areas.

[0036] In this step, a systematic deployment of laser scanning points is carried out across the target garden area, achieving full coverage through multiple laser scanning stations. The installation height and rotation angle of each scanning station are adjusted according to the terrain, building structure, and vegetation height to ensure that the scanning beam fully covers the surfaces of various target objects. By rationally planning the spacing and overlap of scanning points, a continuous, gap-free point cloud dataset is formed. During this process, scanning parameters can be dynamically adjusted to adapt to complex terrain or occluded areas, ensuring high-density data acquisition. This step obtains complete and accurate three-dimensional spatial information, providing a reliable foundation for constructing a high-precision point cloud model, and also providing spatial reference for subsequent multi-source data fusion, ensuring the geometric integrity and accuracy of the model.

[0037] Step S120: Images are acquired on the ground and indoors using a near-ground photogrammetry device to obtain image data for three-dimensional registration.

[0038] In this step, photogrammetric equipment is deployed on the ground and indoors of the target area to capture high-overlapping photographs covering the entire scene. Lighting conditions, shooting angles, and camera calibration are considered during the capture process to ensure the geometric accuracy and integrity of the texture information in the images. The resulting image data is used for subsequent registration of the point cloud with the images. Control point and corner point matching is used to eliminate shooting deviations, and the data is also used to supplement texture information for the point cloud model. This step enhances the surface representation of the 3D point cloud model, giving it not only accurate geometric structure but also realistic texture details, providing a reliable data foundation for subsequent feature extraction and 2D drawing generation.

[0039] Step S130: Aerial photography is carried out using a drone oblique photography device along a preset route, including orthophoto routes and multi-directional oblique routes. The gimbal angle, heading overlap rate, and lateral overlap rate are set to ensure that the target area is covered from multiple angles in the air, forming continuous and registrable image data.

[0040] In this step, the UAV flight path is planned, combining orthophoto flight paths and multi-directional tilt flight paths to achieve aerial coverage of the target area. During flight, the gimbal angle, directional overlap rate, and lateral overlap rate are set to ensure that each target object is continuously photographed from multiple angles. The acquired aerial imagery can be used to generate aerial point clouds and 3D texture mappings, and registered with ground scanning and near-ground photogrammetry data to eliminate spatial discrepancies between different sources. This step, through multi-angle and multi-level data acquisition, provides rich spatial information for constructing a complete, continuous, and accurate 3D point cloud model, while enhancing the three-dimensional detail representation of complex objects, providing a more reliable foundation for the generation of high-precision 2D drawings.

[0041] Step S140: The three-dimensional laser scanning data, near-ground photogrammetry data, and UAV oblique photogrammetry data are used together as the original three-dimensional spatial data for the subsequent construction of the three-dimensional point cloud model.

[0042] In this step, data from the aforementioned three sources are aggregated to form a raw 3D spatial dataset. Preprocessing and format standardization of the data provide a directly processable data foundation for subsequent point cloud registration and fusion. Integrating multi-source data fully leverages the geometric accuracy of laser scanning, the textural information of photogrammetry, and the aerial coverage advantages of UAV oblique photography, resulting in a high-precision, highly complete, and information-rich raw 3D dataset. This provides a solid foundation for subsequent 3D point cloud model construction, feature extraction, and 2D drawing generation, while simultaneously improving the spatial resolution and model accuracy of the entire method.

[0043] Figure 3 It shows Figure 1 The flowchart for step S200, which involves registering and fusing the original 3D spatial data to construct a 3D point cloud model, is shown below. Figure 3As shown, step S200 specifically includes: Step S210: Point cloud registration is performed based on multi-source data. The laser scanned point cloud is used as reference data. Control points are determined based on the feature points in the photogrammetric image. The point cloud is then manually or automatically aligned using the control points to reduce deviations caused by seasonal changes, plant growth, and differences in furnishings.

[0044] In this step, the high-precision point cloud obtained by laser scanning is first used as the reference data to ensure the accuracy of the overall 3D geometry. Then, corner points and control points extracted from near-terrestrial photogrammetric images and UAV oblique photogrammetric images are spatially correlated with the laser point cloud. Accurate registration of multi-source data is achieved through manual or automatic alignment algorithms. During the alignment process, potential local deviations introduced by seasonal changes, plant growth, or changes in garden furnishings are considered, and abnormal areas are identified and adjusted using algorithms. This step effectively eliminates spatial inconsistencies between different data sources, ensuring that the final fused point cloud model is geometrically continuous and complete. This provides a reliable foundation for constructing a high-precision 3D model and improves the spatial accuracy and traceability of subsequent 2D drawing generation.

[0045] Step S220: After registration is completed, actual size data is obtained at key locations and compared with the measured values ​​of the 3D point cloud model to calculate the error rate. The point cloud model is then fine-tuned to keep the overall error within a preset range.

[0046] In this step, after point cloud registration, the actual dimensions of the target garden components are acquired at selected key locations. These key locations include architectural components, artificial rockery structures, flower bed boundaries, or other representative spatial feature points. The corresponding physical dimension data is obtained using methods such as laser ranging or measuring tapes. Subsequently, the measured dimensions are compared point-by-point with the corresponding measurements in the registered 3D point cloud model, and the error rate between the two is calculated to reflect the degree of deviation between the model and the real scene. After obtaining the error rate, the point cloud model undergoes accuracy correction processing. Based on the error distribution, manual fine-tuning or automatic software optimization is used to adjust the point cloud coordinates, gradually converging the spatial positional relationships of the model to correspondence with the actual dimensions, thereby eliminating systematic deviations and local offsets. Ultimately, the overall error is controlled within a preset accuracy range of 2% to ensure the consistency and stability of the 3D point cloud model in terms of geometric scale. This step not only ensures that the spatial accuracy of the 3D point cloud model meets the requirements of engineering drawing, but also enhances the correspondence between the model and the actual scene, providing reliable data support for the generation of high-precision 2D drawings, while ensuring the accuracy and repeatability of the model in subsequent feature extraction and sectioning analysis.

[0047] Step S230: Construct a triangular network model based on the registered point cloud data, establish a three-dimensional mesh structure through vertices, edges and faces, and perform texture mapping on the three-dimensional mesh model in combination with multi-source image data, so that the three-dimensional mesh model can realistically reflect the surface morphology and detailed features of the target area.

[0048] In this step, after point cloud registration, a triangular network model is constructed based on the registered point cloud data. By topologically reconstructing the spatial discrete points in the point cloud, the spatial relationships between adjacent points are transformed into a three-dimensional mesh structure composed of vertices, edges, and faces, thereby achieving a discretized representation of the continuous curved surface morphology of the target region. During this process, the mesh generation process is adaptively adjusted according to the point cloud density and spatial distribution characteristics, making the mesh structure of complex curved surface regions more refined, while maintaining a relatively uniform mesh division in relatively flat regions, thus balancing model accuracy and computational efficiency.

[0049] After the 3D mesh structure is constructed, multi-source image data is introduced to perform texture mapping processing on the mesh model. Color information from the 2D images is spatially matched with the 3D mesh surface, and texture information is accurately mapped to the corresponding mesh cells according to projection relationships, thereby supplementing the model's visual detail representation. Through this texture mapping process, the 3D mesh model not only possesses the ability to represent geometric structures but also has the ability to represent realistic surface materials and visual features, thus more accurately reflecting the actual shape and detailed features of the target area and providing more consistent basic model data for subsequent 2D representation and feature extraction processing.

[0050] Figure 4 It shows Figure 1 A schematic flowchart illustrating step S300, which involves cropping the 3D point cloud model according to a preset cutting plane to generate a corresponding depth image. (Example) Figure 4 As shown, step S300 specifically includes: Step S310: The completed 3D point cloud model is cut and an orthographic projection is generated according to the preset spatial region or key structural range. The orthographic projection has a scale.

[0051] In this step, the complete 3D point cloud model is spatially truncated based on user-preset cutting plane parameters or the key structural range of the target garden area, generating an orthographic projection with a scale. The truncating process, based on the XYZ coordinate range or specified geometric boundaries, removes point clouds from non-target areas, retaining only the point set containing buildings, artificial hills, flower beds, trees, and other important landscape elements. This operation not only reduces the amount of data required for subsequent processing and improves computational efficiency but also ensures that the generated depth image and 3D mesh model only reflect the core information of the target area, laying the foundation for accurate modeling and high-quality 2D drawing generation, while avoiding redundant data interfering with feature extraction.

[0052] Step S320: Generate a depth image based on orthophoto projection and set the pixel size of the depth image. The depth value in the depth image represents the distance from the surface of the target object to the observation reference plane.

[0053] In this step, the set of points of interest obtained in step S310 is projected onto a preset reference plane, and the vertical distance from each point to the reference plane is calculated to generate a depth image. During the generation process, the depth values ​​first undergo preliminary grayscale mapping, converting the distance information into grayscale values ​​for visualization and subsequent algorithm processing. After grayscale mapping, different grayscale layers in the image correspond to different spatial depths on the object's surface, making the structural morphology, undulations, and layers clearly discernible. This step effectively converts three-dimensional spatial information into a two-dimensional representation, providing accurate input data for subsequent edge detection, feature line extraction, and two-dimensional drawing generation, while ensuring the continuity and spatial accuracy of the depth information.

[0054] Step S330: Perform grayscale mapping processing on the depth image, convert the depth value into image grayscale data, and generate a multi-level depth map based on the characteristics of the target structure.

[0055] In this step, after generating the depth image and establishing the correspondence between pixels and spatial depth, the depth image undergoes further grayscale mapping processing. Each depth value is converted into a corresponding grayscale value according to a unified mapping rule, allowing different spatial heights or depth levels to be expressed in the two-dimensional image at different grayscale levels, thus forming a grayscale depth representation result with continuously changing characteristics. This processing method transforms depth data, which originally only represented distance information, into an image data structure that is easy for computer vision processing, providing a stable input foundation for subsequent feature extraction. Based on this, the depth image is layered according to the spatial composition characteristics of the target structure. According to different cutting depth ranges and structural undulations, continuous depth information is divided into multiple levels of depth representation maps, forming a multi-layered depth map structure. Different levels of depth maps correspond to structural information at different spatial scales in the target area, allowing different objects such as building components, terrain undulations, and vegetation cover to be hierarchically distinguished in their representation. Through this multi-layered representation method, depth information in two-dimensional space simultaneously possesses the ability to represent both overall contours and local details, thus providing a clearer and more structurally distinct data foundation for subsequent edge detection, contour extraction, and two-dimensional drawing generation.

[0056] In some embodiments, step S400 above, which involves performing gradient calculations on the depth image and extracting feature lines using Canny edge detection, specifically includes: Step S410: Normalize the depth image to map the depth values ​​to a preset range.

[0057] In this step, the input depth image is normalized, mapping the original depth values ​​to a preset numerical range, such as [0,1], to ensure that depth data from different sources and at different scales are numerically consistent. This processing eliminates the impact of differences in acquisition devices and scene scale variations on subsequent calculations, providing a unified processing basis for depth information and a stable input for gradient calculation.

[0058] Step S420: Filter the normalized depth image to remove noise.

[0059] In this step, the normalized depth image is filtered, preferably using Gaussian filtering, to smooth high-frequency noise and local abnormal fluctuations in the image. By weighted averaging of neighboring pixels, the interference of noise on gradient calculation can be effectively reduced, avoiding false edges or edge breaks in subsequent edge detection, thereby improving the overall stability of edge extraction.

[0060] Step S430: Perform gradient calculation based on the filtered depth image to obtain the gradient magnitude information of each pixel and generate the corresponding gradient magnitude map.

[0061] In this step, gradient calculations are performed based on the filtered depth image. The Sobel operator is used to calculate the pixel change rate in both the horizontal and vertical directions to obtain the gradient magnitude of each pixel, and a gradient magnitude map is generated accordingly. This gradient magnitude map can reflect the surface morphological changes of the target area. Areas with larger gradient magnitudes correspond to structural boundaries or abrupt morphological changes, thus providing a basis for subsequent edge detection and forming a terrain gradient map to characterize spatial undulation features.

[0062] Step S440: Non-maximum suppression processing is performed based on the gradient magnitude map. The gradient magnitude of each pixel is compared with that of the adjacent pixels along the gradient direction, and local maxima are retained to refine the edge response.

[0063] In this step, non-maximum suppression is performed based on the obtained gradient magnitude map. The gradient magnitude of each pixel along its gradient direction is compared with that of its neighboring pixels, and only pixels corresponding to local maxima are retained. This refines the edge response and removes wide edge bands. Through this processing, the original blurry banded structure of the edge converges into a single-pixel-level fine line structure, improving edge localization accuracy.

[0064] Step S450: Based on the non-maximum suppression processing result, set a first threshold and a second threshold, and perform dual threshold determination on the gradient magnitude. The first threshold is greater than the second threshold. Pixels greater than the first threshold are determined as first-class edge pixels, and pixels greater than or equal to the second threshold and not greater than the first threshold are determined as second-class edge pixels. Non-edge pixels less than the second threshold are removed.

[0065] In this step, two different thresholds are set based on the non-maximum suppression processing results to classify and determine the gradient magnitude. The first threshold is used to identify first-class edge pixels, and the second threshold is used to identify second-class edge pixels, removing non-edge pixels below the second threshold. Specifically, the first threshold corresponds to a higher gradient response level, which can be used to identify strong edge pixels with clear structural boundaries and significant changes; the second threshold corresponds to a relatively lower gradient response level, which is used to identify weak edge pixels that may constitute boundaries but have a weak response. This dual-threshold classification method allows for preliminary screening of potential boundary information while preserving the main structural outline, thus achieving a balance between edge integrity and noise suppression. Preferably, the two thresholds can be adaptively set according to the gradient magnitude distribution to adapt to differences in structural complexity under different scenarios.

[0066] Step S460: Perform connectivity analysis on the second type of edge pixels, and retain and merge the second type of edge pixels that are connected to the first type of edge pixels to generate continuous edge lines.

[0067] In this step, connectivity analysis is performed on the second type of edge pixels. Edge pixels that are spatially connected to the first type of edge pixels are retained and fused. In this process, the first type of edge pixels are considered strong edge pixels, used to form the main structural outline, while the second type of edge pixels are considered weak edge pixels, used to supplement details or connect broken boundaries. By judging the spatial connectivity between weak and strong edge pixels, only the weak edge portions that form continuous paths with the strong edges are retained, thereby removing isolated noise points or invalid edge responses. Through this fusion mechanism, the edge structure maintains continuity while avoiding the introduction of redundant information, ultimately obtaining a structurally complete and accurately expressed edge contour.

[0068] Step S470: Output continuous edge lines as feature lines.

[0069] In this step, the continuous edge lines output after connectivity fusion are used as a set of feature lines. This set of feature lines can fully reflect the various structural boundaries and morphological changes in the target area, providing basic data for subsequent feature line layering processing based on gradient threshold masks, and providing a clear, continuous and structurally accurate contour expression for the two-dimensional drawing generation process.

[0070] In one embodiment, the specific steps of step S500 above—setting a threshold mask based on the gradient magnitude of the feature lines, classifying the feature lines according to preset rules, filtering and matching the classification results based on the feature lines obtained by Canny edge detection, generating the corresponding feature line layer, and outputting the classification—include: Step S510: Determine multiple threshold intervals based on the gradient magnitude map, and establish classification rules according to the gradient magnitude ranges corresponding to different types of structural elements.

[0071] In this step, the gradient response characteristics of different spatial structures are analyzed based on the gradient amplitude map, multiple gradient amplitude intervals are divided, and classification rules are established in conjunction with the drawing specifications of garden components. Specifically, the outlines of roofs, walls, artificial hills, flower beds, etc., are defined as outer contour lines; steps, stacked stones, doorways, etc., are defined as secondary contour lines; and surface textures of stones, ground cover textures, and details of doors and windows are defined as texture lines. Based on this, the gradient amplitude range of the corresponding region is extracted according to the distribution characteristics of each level of elements in the gradient amplitude map, and the threshold intervals corresponding to each level are determined by statistically analyzing their maximum and minimum values, thus forming a classification basis that matches the actual structure and providing a foundation for subsequent mask generation.

[0072] Step S520: Mark the pixels in the gradient magnitude map according to the threshold range to generate feature line masks of three types.

[0073] In this step, based on the threshold range determined in step S510, pixels in the gradient magnitude map are judged and marked point by point to generate multi-class feature line masks. Specifically, pixels with gradient magnitudes greater than the upper threshold are marked as outer contour line masks, pixels with gradient magnitudes between the two thresholds are marked as secondary contour line masks, and pixels with gradient magnitudes lower than the lower threshold are marked as texture line masks. Through the above processing, multiple clearly defined and mutually distinguishable mask regions are constructed in the same gradient magnitude map, enabling clear differentiation of different structural features at the data level.

[0074] Step S530: Apply the feature line mask to the feature lines to classify and label the feature lines, so that feature lines of different categories form corresponding classification results.

[0075] In this step, the feature line mask generated in step S520 is applied to the extracted feature lines to classify and label them. Specifically, the pixel positions in the feature lines are mapped to the corresponding mask regions, and the feature lines are categorized according to the type of mask they fall into, thus forming a clearly structured classification result for feature lines of different categories. In this way, the transformation from a "set of continuous edges" to a "set of feature lines with semantic hierarchy" can be achieved, so that the contour information includes not only geometric position but also structural attributes.

[0076] Step S540: Based on the feature lines obtained by Canny edge detection, the classification results are filtered and matched to retain the valid classification data corresponding to the feature lines.

[0077] In this step, the classification results obtained in step S530 are filtered and matched based on the feature lines obtained from Canny edge detection. Specifically, the Canny edge detection results are used as constraints to match and verify the classified feature lines, eliminating classification results that do not correspond to edge responses, and compensating for edge continuity to ensure that the classified feature lines are spatially consistent with the real edges. This process essentially uses edge detection results to constrain and correct the classification results obtained based on gradient thresholding, thereby avoiding misclassification or omissions caused by simple thresholding.

[0078] Step S550: Generate corresponding layer data based on the classification results, and output the data in layers according to the classification categories.

[0079] In this step, corresponding layer data is generated based on the filtered and matched classification results, and then output in layers according to different categories. Specifically, the outer contour lines, secondary contour lines, and texture lines are each constructed as independent layers and output in image or vector form, giving each type of feature line an independent expressive capability in the 2D drawing. Through layered output, information from different structural levels can be flexibly accessed and combined in subsequent drawing generation and visualization processes, thereby ensuring both the integrity of the overall structural expression and improving the precision of detail presentation.

[0080] In one embodiment, the specific steps of step S600 above—generating two-dimensional drawings based on different layer data and establishing a correspondence between the two-dimensional drawings and the original three-dimensional point cloud model, so that the two-dimensional drawings can be traced back to any position in the three-dimensional model—include: Step S610: Based on the 3D point cloud model and the set of layered feature lines, determine the projection plane and project the 3D information in the 3D point cloud model onto the projection plane to generate the corresponding 2D view.

[0081] In this step, based on the constructed 3D point cloud model and the set of hierarchical feature lines, the projection plane for 2D representation is first determined. This projection plane can be set according to different application requirements, such as using horizontal projection to generate a plan view or vertical projection to generate a elevation view. During the projection process, the spatial coordinate information in the 3D point cloud data is mapped to the 2D coordinate system according to projection rules, and the hierarchical feature lines are simultaneously projected to their corresponding positions, thus forming a 2D view containing structural outlines and hierarchical information. By preserving the hierarchical attributes of different types of feature lines during the projection process, the 2D view not only possesses geometric form but also reflects the primary and secondary relationships of the structure, thereby providing semantically distinct basic data for subsequent drawing generation.

[0082] Step S620: According to the user-preset cutting plane parameters, the three-dimensional point cloud model is sliced ​​along the specified direction to extract the geometric information corresponding to the cutting plane, and a two-dimensional cutting map is generated by combining the layered feature line set.

[0083] In this step, the 3D point cloud model is sliced ​​along a specified direction according to the user-preset cutting plane parameters. Specifically, by setting the spatial position and direction of the cutting plane, the point cloud model is locally truncated, and geometric information near the cutting plane is extracted, including structural boundaries, component outlines, and internal spatial relationships. Based on this, the geometric information obtained from the cutting is fused with the corresponding layered feature lines to generate a 2D sectional view. This sectional view can clearly reflect the internal structure and spatial hierarchy of the target area, further enhancing the drawing's ability to express complex spatial relationships compared to a 2D view based solely on external projection.

[0084] Step S630: In the generated two-dimensional drawing, establish a correspondence between each drawing element and the original three-dimensional point cloud data to achieve a traceable mapping from the two-dimensional drawing to the original three-dimensional point cloud data.

[0085] In this step, a correspondence is established between each element in the generated 2D drawing and the original 3D point cloud data. Specifically, by recording the source location of each element in the 2D drawing and its coordinate mapping relationship in 3D space, each line segment or graphic can be associated with the corresponding point set or spatial region in the point cloud model. Based on this mapping relationship, when data tracing or detailed verification is required, the corresponding position in the 3D model can be directly located from the 2D drawing, achieving rapid reverse lookup from 2D to 3D. This mechanism makes the generated 2D drawing no longer a static result, but a dynamic expression that is linked with the 3D data, improving the flexibility and accuracy of data utilization.

[0086] Step S640: The generated two-dimensional drawing is annotated, scaled, and standardized to obtain a two-dimensional drawing that conforms to the preset drawing specifications.

[0087] In this step, the generated 2D drawings undergo further graphic processing to meet engineering drawing specifications. Specifically, various elements in the drawings are automatically labeled, and dimension information, structural descriptions, and necessary annotations are added. Simultaneously, the drawings are uniformly scaled according to a preset ratio to conform to standard scale requirements. Based on this, the drawings are standardized, including line type unification, layer organization, and layout adjustments, ensuring the output conforms to engineering drawing conventions and application needs. Through this step, the final generated 2D drawings, while ensuring data accuracy, possess good readability and engineering application value, and can be directly used in scenarios such as landscape planning, maintenance, and digital archiving.

[0088] In one embodiment, the automatic generation method for two-dimensional drawings based on three-dimensional point clouds of the present invention further includes the following steps: The current 2D drawing is matched with historical 2D drawings by corresponding points or corresponding areas, and the buildings, rockeries, flower beds and other garden components are classified and compared.

[0089] In this step, the system first spatially aligns the currently generated 2D drawing with historical 2D drawings. Through feature point matching, key area identification, or manual-assisted positioning, corresponding areas in the two drawings are precisely matched. Subsequently, the system categorizes garden components (such as buildings, artificial hills, flower beds, roads, or vegetation cover) and compares each type of target independently to distinguish information at different spatial levels. This process allows for a visual representation of the correspondence between historical and current garden spaces on a 2D plane, providing a foundation for accurately quantifying structural changes and dynamic updates, and effectively avoiding matching deviations caused by drawing scale, projection methods, or recording errors.

[0090] Based on the matching results, the geometric differences and spatial position changes between different two-dimensional drawings are calculated to form quantitative analysis data, which includes position offset, area change and shape difference.

[0091] In this step, based on the aforementioned matching results, the system extracts the geometric features of each garden component, such as vertex coordinates, outline length, area, and curvature information, and calculates the differences between historical and current drawings. Through comparative analysis, quantitative data is generated, including component position offsets, area increases or decreases, and the degree of morphological change. Simultaneously, it can identify local changes or overall movement trends in component morphology. This operation not only provides accurate data support for subsequent change analysis but also enables garden managers to scientifically assess the actual impact of spatial layout adjustments, thereby achieving more reliable and objective spatial management and maintenance strategies.

[0092] Based on quantitative analysis data, information on changes in garden layout is generated, and newly added, removed, or adjusted components are identified to achieve dynamic updates and traceable management of garden space information.

[0093] In this step, the system transforms the calculated geometric differences and spatial changes into a visual report, including a list of changed components, change types (addition, removal, adjustment), location offset annotations, and area change diagrams. By linking with the original 2D drawings and 3D point cloud model, each change point or area can be traced back to the 3D model, ensuring data traceability. Simultaneously, the report can automatically generate standardized documents or interactive views for use in landscape planning, maintenance, and record updates, reducing errors from manual statistics and recording, and improving landscape management efficiency and information transparency.

[0094] According to the above embodiments, by combining multi-level feature line extraction, sectioning and clipping, and triangular mesh modeling, high-precision two-dimensional drawings of complex garden areas and target components can be generated. These drawings can be compared and analyzed with historical drawings to assess layout changes, enabling quantitative monitoring and traceable management of the addition, removal, and adjustment of garden components. Simultaneously, this method supports automatic annotation, scale adjustment, and drawing standardization, providing accurate, intuitive, and efficient digital references for garden planning, maintenance, and heritage protection, thereby enhancing the scientific rigor and reliability of overall spatial information management and decision-making.

[0095] This invention also provides an automatic two-dimensional drawing generation system based on three-dimensional point clouds, such as... Figure 5 As shown, the system includes multiple data acquisition units 101, a data processing unit 102, a sectioning unit 103, a feature extraction unit 104, and a 2D drawing generation unit 105. The multiple data acquisition units 101 acquire raw 3D spatial data from multiple data sources, including laser scanning equipment, near-ground photogrammetry equipment, and UAV oblique photogrammetry equipment. The data processing unit 102, connected to the multiple data acquisition units 101, registers and fuses the raw 3D spatial data to construct a 3D point cloud model. The sectioning unit 103, connected to the data processing unit 102, sections the 3D point cloud model according to a preset sectioning plane to generate a corresponding depth image. The feature extraction unit 104, connected to the sectioning unit 103, performs gradient calculations on the depth image, extracts feature lines using Canny edge detection, sets a threshold mask based on the gradient magnitude of the feature lines, classifies the feature lines according to preset rules, and filters and matches the classification results based on the feature lines obtained from Canny edge detection, generating corresponding feature line layers and outputting them in categories. The two-dimensional drawing generation unit 105 is connected to the feature extraction unit 104 and is used to generate two-dimensional drawings based on different layer data and establish the correspondence between the two-dimensional drawings and the original three-dimensional point cloud model, so that the two-dimensional drawings can be traced back to any position of the three-dimensional model.

[0096] In the above-mentioned automatic generation system of two-dimensional drawings based on three-dimensional point clouds, the specific implementation method refers to the relevant content of the embodiments in the above method, and will not be repeated here.

[0097] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0098] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for automatically generating two-dimensional drawings based on three-dimensional point clouds, characterized in that, Includes the following steps: Acquire raw 3D spatial data from multiple data sources, including laser scanning equipment, near-ground photogrammetry equipment, and UAV oblique photogrammetry equipment; The original three-dimensional spatial data is registered and fused to construct a three-dimensional point cloud model; The three-dimensional point cloud model is cropped according to a preset cutting plane to generate a corresponding depth image; Gradient calculations are performed on the depth image, and feature lines are extracted using Canny edge detection. A threshold mask is set according to the gradient magnitude of the feature lines, the feature lines are classified according to a preset rule, and the classification results are filtered and matched based on the feature lines obtained by Canny edge detection to generate corresponding feature line layers and output them in categories. Two-dimensional drawings are generated based on data from different layers, and a correspondence is established between the two-dimensional drawings and the original three-dimensional point cloud model, so that the two-dimensional drawings can be traced back to any position in the three-dimensional model.

2. The method for automatically generating two-dimensional drawings according to claim 1, characterized in that, The specific steps for obtaining raw 3D spatial data from multiple data sources include: Multiple scanning stations are set up in the target garden area using a 3D laser scanning device. The height and rotation angle of the scanner are adjusted at each scanning station to perform a comprehensive scan of the target object's surface to obtain point cloud data with overlapping areas. Images are acquired on the ground and indoors using near-ground photogrammetry equipment to obtain image data for 3D registration; Aerial photography is carried out using drone oblique photography equipment along preset routes, including orthophoto routes and multi-directional oblique routes. The gimbal angle, heading overlap rate and lateral overlap rate are set to cover the target area from multiple angles in the air, forming continuous and registrable image data. The original three-dimensional spatial data is obtained by combining three-dimensional laser scanning data, near-ground photogrammetry data, and UAV oblique photogrammetry data.

3. The method for automatically generating two-dimensional drawings according to claim 2, characterized in that, The specific steps for registering and fusing the original three-dimensional spatial data to construct a three-dimensional point cloud model include: Point cloud registration is performed based on multi-source data. Laser scanned point clouds are used as reference data. Control points are determined based on feature points in photogrammetric images. The point clouds are then manually or automatically aligned using these control points to reduce deviations caused by seasonal changes, plant growth, and differences in furnishings. After registration is completed, actual size data is obtained at key locations and compared with the measured values ​​of the 3D point cloud model to calculate the error rate. The point cloud model is then fine-tuned to keep the overall error within a preset range. A triangular network model is constructed based on the registered point cloud data. A three-dimensional mesh structure is established through vertices, edges, and faces. The three-dimensional mesh model is then texture-mapped using multi-source image data, enabling the three-dimensional mesh model to realistically reflect the surface morphology and detailed features of the target area.

4. The method for automatically generating two-dimensional drawings according to claim 3, characterized in that, The specific steps for cropping the 3D point cloud model according to a preset cutting plane to generate a corresponding depth image include: The completed 3D point cloud model is trimmed, and an orthophoto is generated according to a preset spatial region or key structural range. The orthophoto has a scale bar. A depth image is generated based on the orthophoto, and the pixel size of the depth image is set. The depth value in the depth image represents the distance of the target object surface relative to the observation reference plane. The depth image is subjected to grayscale mapping processing to convert the depth value into image grayscale data, and a multi-level depth map is generated according to the characteristics of the target structure.

5. The method for automatically generating two-dimensional drawings according to claim 4, characterized in that, The specific steps for performing gradient calculations on the depth image include: The depth image is normalized so that the depth values ​​are mapped to a preset range; The normalized depth image is filtered to remove noise. Gradient calculations are performed on the filtered depth image to obtain the gradient magnitude information of each pixel and generate the corresponding gradient magnitude map.

6. The method for automatically generating two-dimensional drawings according to claim 5, characterized in that, The specific steps for extracting feature lines using Canny edge detection also include: Non-maximum suppression processing is performed based on the gradient magnitude map. The gradient magnitude of each pixel is compared with that of the adjacent pixels along the gradient direction, and local maxima are retained to refine the edge response. Based on the non-maximum suppression processing results, a first threshold and a second threshold are set, and the gradient magnitude is determined by dual thresholds. The first threshold is greater than the second threshold. Pixels greater than the first threshold are determined as first-class edge pixels, and pixels greater than or equal to the second threshold and not greater than the first threshold are determined as second-class edge pixels. Non-edge pixels less than the second threshold are removed. Perform connectivity analysis on the second type of edge pixels, and retain and merge the second type of edge pixels that are connected to the first type of edge pixels to generate continuous edge lines; The continuous edge lines are output as feature lines.

7. The method for automatically generating two-dimensional drawings according to claim 6, characterized in that, The specific steps of setting a threshold mask based on the gradient magnitude of the feature lines, classifying the feature lines according to preset rules, filtering and matching the classification results based on the feature lines obtained by Canny edge detection, generating corresponding feature line layers, and outputting the classifications include: Based on the gradient magnitude map, multiple threshold intervals are determined, and classification rules are established according to the gradient magnitude ranges corresponding to different types of structural elements. The pixels in the gradient magnitude map are marked according to the threshold range to generate three types of feature line masks; The feature line mask is applied to the feature line to classify and identify the feature line, so that different categories of feature lines form corresponding classification results; Based on the feature lines obtained by the Canny edge detection, the classification results are filtered and matched to retain the valid classification data corresponding to the feature lines. Based on the classification results, corresponding layer data is generated and output in layers according to the classification categories.

8. The method for automatically generating two-dimensional drawings according to claim 7, characterized in that, The specific steps for generating two-dimensional drawings based on data from different layers and establishing a correspondence between the two-dimensional drawings and the original three-dimensional point cloud model, so that the two-dimensional drawings can be traced back to any position in the three-dimensional model, include: Based on the three-dimensional point cloud model and the set of layered feature lines, a projection plane is determined, and the three-dimensional information in the three-dimensional point cloud model is projected onto the projection plane to generate a corresponding two-dimensional view. According to the user-preset cutting plane parameters, the three-dimensional point cloud model is sliced ​​along the specified direction to extract the geometric information corresponding to the cutting plane, and a two-dimensional cutting image is generated by combining the layered feature line set. In the generated 2D drawing, a correspondence is established between each drawing element and the original 3D point cloud data to achieve a traceable mapping from the 2D drawing to the original 3D point cloud data. The generated two-dimensional drawings are annotated, scaled, and standardized to obtain two-dimensional drawings that conform to preset drawing specifications.

9. The method for automatically generating two-dimensional drawings according to claim 8, characterized in that, The method further includes the following steps: Match the current 2D drawings with historical 2D drawings by corresponding points or regions, and classify and compare buildings, rockeries, flower beds and other garden components. Based on the matching results, the geometric differences and spatial position changes between different two-dimensional drawings are calculated to form quantitative analysis data, which includes position offset, area change and shape difference. Based on the quantitative analysis data, information on changes in the garden layout is generated, and newly added, removed, or adjusted components are identified to achieve dynamic updating and traceable management of garden space information.

10. A system for automatically generating two-dimensional drawings based on three-dimensional point clouds, characterized in that, include: Multiple data acquisition units are used to acquire raw three-dimensional spatial data from multiple data sources, including laser scanning equipment, near-ground photogrammetry equipment, and UAV oblique photogrammetry equipment. A data processing unit, connected to the multiple data acquisition units, is used to register and fuse the original three-dimensional spatial data to construct a three-dimensional point cloud model. The slicing unit, connected to the data processing unit, is used to slice the three-dimensional point cloud model according to a preset slicing plane to generate a corresponding depth image; The feature extraction unit, connected to the slicing unit, is used to perform gradient calculation on the depth image, extract feature lines through Canny edge detection, set a threshold mask according to the gradient magnitude of the feature lines, classify the feature lines according to preset rules, and perform filtering and matching processing on the classification results based on the feature lines obtained by Canny edge detection to generate corresponding feature line layers and output them in categories. The two-dimensional drawing generation unit, connected to the feature extraction unit, is used to generate two-dimensional drawings based on different layer data and establish a correspondence between the two-dimensional drawings and the original three-dimensional point cloud model, so that the two-dimensional drawings can be traced back to any position of the three-dimensional model.

Images