Three-dimensional landscape reconstruction method based on multi-view photogrammetry

Abstract

The invention provides a three-dimensional landscape reconstruction method based on multi-view photogrammetry, and relates to the technical field of three-dimensional environment modeling, and the method comprises the steps: collecting terrains, buildings and vegetation through a high-precision sensor, and generating an initial point cloud data set; performing noise reduction and filtering processing on the point cloud data set; a Delaunay triangularization technology is adopted to convert the purified point cloud data into a three-dimensional triangular mesh model; performing real-time rendering of illumination and materials on the basis of the three-dimensional model, calculating the reflectivity of the materials by using a physics-based rendering technology, and integrating the three-dimensional model with data of a geographic information system to generate a detailed elevation model; analyzing features, including gradient and azimuth angle, of the landscape area by using a geographic information system; and according to design parameters input by a user, the design effect of the three-dimensional model is updated and fed back in real time, and the visibility is optimized by adopting a multi-scale rendering strategy at different visual angles. According to the method, precise landscape three-dimensional reconstruction and optimization can be realized, and the design efficiency and the visualization effect are improved.

Description

Technical Field

[0001] The invention relates to the technical field of three-dimensional environment modeling, and in particular to a three-dimensional landscape reconstruction method based on multi-view photogrammetry. Background Art

[0002] 3D landscape reconstruction technology has been widely used in urban planning, environmental monitoring, cultural heritage protection and other fields. By converting 2D image data into 3D models, these technologies provide researchers and engineers with more intuitive and accurate analysis tools. However, the existing 3D landscape reconstruction technology still faces many challenges in its application, especially in the field of multi-view photogrammetry, where there are many technical problems that need to be solved.

[0003] First of all, data fusion and accuracy are one of the main challenges in current 3D reconstruction technology. Multi-view photogrammetry relies on a large amount of image data collected from different angles. Due to differences in viewing angles, lighting, resolution and other factors, there are often significant distortions and differences between images. When performing feature point matching and 3D point cloud generation, existing technologies often lack accuracy due to these differences, and the final generated 3D model may be distorted, inaccurate or incomplete. In addition, when dealing with complex scenes, such as detailed structures of buildings or lush areas of vegetation, this imprecision problem will be more prominent, seriously affecting the application value of the model.

[0004] Secondly, the problem of computational efficiency is also one of the bottlenecks that current technology is difficult to break through. The processing flow of multi-view photogrammetry, including feature point extraction, matching, 3D point cloud generation and 3D model construction, is a computationally intensive task. Especially in large-scale scenes, processing large amounts of image data and point cloud data requires a lot of computing resources and time. Existing computing methods and algorithms are not sufficiently optimized, resulting in excessively high time costs in the reconstruction process, making it difficult to meet the requirements for real-time and efficiency in practical applications. This problem is particularly evident in scenes such as large-scale urban landscapes, complex terrains or real-time dynamic monitoring.

[0005] In addition, the problem of insufficient model details also affects the application effect of 3D landscape reconstruction technology. In existing technologies, due to the limitations of computing power and algorithm accuracy, the generated 3D models are often unsatisfactory in terms of detail performance. Areas with rich details, such as fine structures of buildings and complex natural landscapes, are often unable to be effectively captured and restored. This not only affects the visual quality of the model, but also limits the use of the model in application scenarios such as refined design and precise analysis.

[0006] Finally, the difficulty of processing multi-view data is also a major challenge in existing technologies. In the process of fusing multi-view images, problems such as occlusion, overlapping areas, and perspective differences must be handled to ensure that the generated 3D model is visually and geometrically consistent. However, current methods are often unable to deal with these problems, and the generated model may appear incoherent or unnatural, affecting the final display effect.

[0007] In summary, although multi-view photogrammetry technology provides an effective means for 3D landscape reconstruction, it still has obvious deficiencies in data fusion, computational efficiency, model details and multi-view processing. With the continuous growth of application demand, it is urgent to further optimize and improve the existing technology to overcome these challenges and achieve more efficient and accurate 3D landscape reconstruction. Summary of the invention

[0008] In order to solve the technical problems in the prior art such as insufficient data fusion and accuracy, low computational efficiency, insufficient model details and difficulty in multi-view data processing, the present invention provides a three-dimensional landscape reconstruction method based on multi-view photogrammetry.

[0009] The technical solution provided by the present invention is as follows:

[0010] The present invention provides a three-dimensional landscape reconstruction method based on multi-view photogrammetry, comprising:

[0011] S1. Collecting raw data through high-precision sensors to obtain an initial point cloud data set D, wherein the high-precision sensors include laser radar and stereo cameras, and the raw data include terrain, buildings and vegetation;

[0012] S2. Perform denoising and filtering on the point cloud data set D to obtain the purified point cloud data D'. The denoising process uses the following formula:

[0013]

[0014] Among them, α is the adjustment coefficient, d i is the coordinate of the noise point, N is the number of noise points;

[0015] S3. Generate a 3D model of the purified point cloud through Delaunay triangulation to obtain a triangular mesh M:

[0016] M=Del(D')

[0017] Among them, Del(D') represents the Delaunay triangulation processing of point cloud data;

[0018] S4. Based on the 3D model, real-time rendering of lighting and materials is performed. The lighting model is calculated using the following formula:

[0019] I=I a +I d ·(N·L)+I s ·(R·V) n

[0020] Among them, I is the final illumination intensity of the pixel, I a is the ambient light intensity, I d is the diffuse reflected light intensity, I s is the intensity of specular reflection light, N is the normal vector, L is the light source direction vector, R is the reflection vector, V is the sight vector, and n is the specular reflection index;

[0021] S5. Using physically based rendering technology, the material reflectivity is calculated as:

[0022] R=F0+(1-F0)·(1-cosθ) 5

[0023] Among them, R is the reflectivity, F0 is the intrinsic reflectivity of the material, and θ is the incident angle of the light;

[0024] S6. Integrate the 3D model with the geographic data G in the GIS system to generate an elevation model Z(x,y):

[0025]

[0026] in, is the difference weight, d i is the distance between the point (x, y) and the i-th known point, P is the distance index, Z i is the elevation of the known point;

[0027] S7. Analyze the characteristics of the landscape area through GIS data, wherein the characteristics include slope S and azimuth A:

[0028]

[0029] in, and is the directional derivative of elevation;

[0030] S8. Adjust the design parameters based on the user input and provide real-time feedback on the updated design effect. Set the update formula of the view matrix V to:

[0031] V=C·W

[0032] Among them, C is the camera matrix and W is the world matrix;

[0033] S9. To improve computational efficiency, a multi-scale rendering strategy is used under different viewing angles. The formula for calculating visibility is:

[0034]

[0035] Among them, V ij is the visibility between pixels i and j, d ij is the distance between two points, k is the smoothing coefficient, and d0 is the threshold distance.

[0036] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0037] (1) In the present invention, by optimizing the feature matching and point cloud generation process in multi-view photogrammetry, the problem of insufficient accuracy caused by factors such as perspective differences and illumination changes in traditional technologies is effectively overcome. By introducing adaptive filtering and multi-scale rendering strategies, the present method can accurately restore the details of the landscape in complex scenes, such as the fine structure of buildings and subtle changes in natural landscapes, significantly improving the accuracy and detail expression of the three-dimensional model, and ensuring that the reconstruction results are more realistic and lifelike.

[0038] (2) In the present invention, by introducing an optimization algorithm for multi-view data processing and a level of detail (LOD) adjustment strategy, redundant operations and consumption of computing resources in the calculation process are reduced. Compared with the existing technology, this method can significantly shorten the reconstruction time when processing large-scale scenes and achieve more efficient three-dimensional landscape reconstruction. At the same time, this method also performs real-time optimization for dynamic scenes, which can meet the needs of real-time update and rapid response, and provides technical support for the rapid reconstruction of large-scale scenes.

[0039] (3) In the present invention, in the process of fusion of multi-view image data, through the improved illumination model, material rendering technology and smooth transition strategy of perspective switching, the problems of occlusion, overlapping area processing and perspective difference in multi-view data fusion are effectively solved. The generated three-dimensional model is not only more coherent and natural in visual effect, but also significantly improved in geometric consistency. It is suitable for many fields such as urban planning, environmental analysis, cultural heritage protection, etc., which greatly expands the application scope and value of three-dimensional landscape reconstruction technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 A schematic flow chart of a method for reconstructing a three-dimensional landscape based on multi-view photogrammetry provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0043] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0044] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0045] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are consistent.

[0046] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0047] Reference Manual Attached Figure 1 , showing a schematic flow chart of a three-dimensional landscape reconstruction method based on multi-view photogrammetry provided by an embodiment of the present invention.

[0048] The embodiment of the present invention provides a three-dimensional landscape reconstruction method based on multi-view photogrammetry, and the processing flow may include the following steps:

[0049] S1. Collecting raw data through high-precision sensors to obtain an initial point cloud data set D. The high-precision sensors include laser radar and stereo cameras. The raw data include terrain, buildings and vegetation.

[0050] It is understandable that high-precision sensors (such as lidar and stereo cameras) are used to collect raw data, which includes detailed information on terrain, buildings, and vegetation. This is the basic data for 3D landscape reconstruction, ensuring the accuracy and integrity of subsequent processing.

[0051] S2. Perform denoising and filtering on the point cloud data set D to obtain the purified point cloud data D'. The denoising process uses the following formula:

[0052]

[0053] Among them, α is the adjustment coefficient, d i are the coordinates of the noise points, and N is the number of noise points.

[0054] It can be understood that removing noise and redundant information in point cloud data ensures that the purified point cloud data is smoother and more accurate, thereby providing high-quality data input for the generation of a three-dimensional model.

[0055] S3. Generate a 3D model of the purified point cloud through Delaunay triangulation to obtain a triangular mesh M:

[0056] M=Del(D')

[0057] Among them, Del(D') represents the Delaunay triangulation processing of point cloud data.

[0058] It can be understood that converting the purified point cloud data into a triangular mesh model is a key step in 3D landscape reconstruction, ensuring that the generated 3D model has a good geometric structure.

[0059] S4. Based on the 3D model, real-time rendering of lighting and materials is performed. The lighting model is calculated using the following formula:

[0060] I=I a +I d ·(N·L)+I s ·(R·V) n

[0061] Among them, I is the final illumination intensity of the pixel, I a is the ambient light intensity, I d is the diffuse reflected light intensity, I s is the intensity of specular reflection light, N is the normal vector, L is the light source direction vector, R is the reflection vector, V is the sight vector, and n is the specular reflection index.

[0062] It can be understood that the final illumination intensity of the pixel is calculated based on factors such as ambient light, diffuse light, and specular light to achieve realistic rendering of the three-dimensional model.

[0063] S5. Using physically based rendering technology, the material reflectivity is calculated as:

[0064] R=F0+(1-F0)·(1-cosθ) 5

[0065] Among them, R is the reflectivity, F0 is the intrinsic reflectivity of the material, and θ is the incident angle of the light.

[0066] It can be understood that by calculating the reflectivity of the material and simulating the reflective characteristics of different materials under light, the realism and refinement of the rendering can be improved.

[0067] S6. Integrate the 3D model with the geographic data G in the GIS system to generate an elevation model Z(x,y):

[0068]

[0069] in, is the difference weight, d i is the distance between the point (x, y) and the i-th known point, P is the distance index, Z i is the elevation of the known point.

[0070] It is understood that by integrating the 3D model with data from a geographic information system (GIS), a more accurate elevation model is generated, which enhances the realism and details of the terrain representation.

[0071] S7. Analyze the characteristics of the landscape area through GIS data, wherein the characteristics include slope S and azimuth A:

[0072]

[0073] in, and is the directional derivative of elevation.

[0074] It is understandable that the use of GIS data to analyze characteristics such as slope and azimuth of landscape areas can help to carry out more detailed landscape design and terrain analysis.

[0075] S8. Adjust the design parameters based on the user input and provide real-time feedback on the updated design effect. Set the update formula of the view matrix V to:

[0076] V=C·W

[0077] Among them, C is the camera matrix and W is the world matrix.

[0078] It can be understood that by updating the view matrix, the user's perspective is dynamically adjusted to reflect changes in design parameters in real time, ensuring that the user can instantly see the effects of the design adjustments.

[0079] S9. To improve computational efficiency, a multi-scale rendering strategy is used under different viewing angles. The formula for calculating visibility is:

[0080]

[0081] Among them, V ij is the visibility between pixels i and j, d ijis the distance between two points, k is the smoothing coefficient, and d0 is the threshold distance.

[0082] It can be understood that through the multi-scale rendering strategy, the rendering accuracy and details of the model are dynamically adjusted to ensure efficient rendering effects at different viewing angles.

[0083] In a possible implementation manner, the S2 specifically includes:

[0084] The denoising process of the point cloud data set D further includes adaptive filtering of the point cloud density, and the calculation formula of the filtered point cloud data set D" is as follows:

[0085]

[0086] Among them, β is the adaptive filter coefficient, M is the local point cloud density, and q is the exponential parameter, which further improves the smoothness and accuracy of the point cloud data.

[0087] It should be noted that in the process of denoising the point cloud data set, adaptive filtering is introduced, and the filter strength is adjusted by using the change of point cloud density. By calculating the adaptive filter coefficient and combining the local point cloud density, noise points can be effectively removed while retaining the details of high-density areas. This implementation method improves the smoothness and accuracy of the point cloud data, ensuring that the basic data for subsequent modeling is more reliable.

[0088] In a possible implementation manner, the S3 specifically includes:

[0089] The three-dimensional model also includes optimizing Delaunay triangulation by constraint conditions to reduce the number of thin triangles. The optimized triangular mesh calculation formula is:

[0090]

[0091] Among them, A k is the area of ​​the kth triangle, P k is the perimeter of the kth triangle, and K is the total number of triangles, thereby generating a triangular mesh that is more consistent with the real terrain.

[0092] It should be noted that in the process of generating a three-dimensional model, the Delaunay triangulation is optimized by introducing constraints to reduce the number of thin triangles. Through the optimized triangular mesh calculation formula, a mesh structure that is more in line with the actual terrain can be generated. This implementation method is intended to improve the quality of the triangular mesh, make the generated three-dimensional model closer to the actual terrain, and reduce the model instability problem caused by thin triangles.

[0093] In a possible implementation manner, the S4 specifically includes:

[0094] The lighting model further includes an adjustment formula based on dynamic ambient lighting:

[0095]

[0096] Among them, I a0 is the initial ambient light intensity, γ is the ambient light adjustment coefficient, L is the number of ambient light sources, I j and d j are the light intensity of the jth light source and its distance from the pixel point respectively.

[0097] It should be noted that the dynamic changes of ambient light are taken into account during the rendering process of the 3D model. By adjusting the ambient light intensity, combined with the number and distance of light sources, the lighting effect is adjusted in real time. This implementation makes the rendering effect more realistic, and can simulate different light and shadow effects produced by changes in ambient light sources, thereby enhancing the visual performance of the 3D model in different environments.

[0098] In a possible implementation manner, the light source direction vector specifically includes:

[0099] The light source direction vector L is dynamically updated based on the sun position, and the calculation formula for the sun position is:

[0100] L(t)=(cosφ·cosδ, sinφ·cosδ, sinδ)

[0101] Among them, φ is the solar hour angle, δ is the solar declination, and t is the time parameter, which realizes the dynamic simulation of sunlight.

[0102] It should be noted that by introducing dynamic calculation of the sun's position, the light source direction vector can change over time, simulating the dynamic changes of sunlight in real time. This implementation ensures that the 3D model can accurately reflect the direction and intensity of sunlight at different time periods or different geographical locations, enhancing the simulation effect of natural lighting and making the model closer to reality.

[0103] In a possible implementation manner, the S5 specifically includes:

[0104] The calculation formula of the material reflectivity R further includes the roughness influence factor:

[0105] R=F0+(1-F0)·(1-cosθ) 5 ·(1-ρ)

[0106] Among them, represents the roughness factor of the material surface, which further improves the realism of material rendering.

[0107] It should be noted that in the calculation of material reflectivity, the effect of the roughness of the material surface on the reflected light is taken into account. By introducing the roughness factor, the light reflection characteristics of the material can be more finely controlled, making the performance of different materials more realistic under different lighting conditions. This implementation method improves the realism of material rendering and makes the material performance in the three-dimensional model richer and more delicate.

[0108] In a possible implementation manner, the S6 specifically includes:

[0109] The elevation model generation further considers the slope weighting of the terrain, and the weight formula is:

[0110]

[0111] Among them, λ is the slope weighting coefficient, which is used to enhance the expression of terrain details.

[0112] It should be noted that when generating the elevation model, the weighting factor of the terrain slope is considered to enhance the detailed expression of the terrain. Slope weighting can make the elevation value of the steep area more prominent, thereby more accurately reflecting the terrain characteristics. This implementation method enhances the expressiveness of the elevation model in complex terrain and makes the model more accurate in visual and data analysis.

[0113] In a possible implementation manner, the S7 specifically includes:

[0114] The slope calculation further considers the terrain roughness η, and the updated slope calculation formula is:

[0115]

[0116] Among them, η is the terrain roughness parameter, which is used to accurately calculate the slope of complex terrain.

[0117] It should be noted that in the slope calculation, a terrain roughness parameter is added to improve the accuracy for complex terrain. By considering the roughness of the terrain, the slope can be calculated more accurately, especially in complex and rugged terrain. This implementation ensures that the slope calculation results are more reliable and accurate under terrain conditions of varying complexity.

[0118] In a possible implementation manner, the S8 specifically includes:

[0119] The view matrix update further considers the smooth transition of the user's perspective, and the updated view matrix formula is:

[0120] V new =V old +ξ·(V target -V old )

[0121] Among them, ξ is the smooth transition coefficient, V target is the target view matrix, V old This is the old view matrix, which ensures smoothness of perspective switching.

[0122] It should be noted that when the user adjusts the viewing angle or design parameters, the view matrix will be updated to consider smooth transition to ensure the naturalness of the viewing angle switching. The introduction of this implementation method can avoid the abruptness in the viewing angle switching process, make the user experience smoother, and enhance the interactivity and comfort of the system.

[0123] In a possible implementation manner, the S9 specifically includes:

[0124] The multi-scale rendering strategy further includes perspective-dependent level of detail (LOD) adjustment, and the LOD update formula is:

[0125]

[0126] Among them, κ is the control parameter, d ij is the display distance of the model details under the current viewing angle, and d0 is the reference distance, which is used to ensure the rendering efficiency and effect under different viewing distances.

[0127] It should be noted that in the multi-scale rendering strategy, the perspective-dependent level of detail (LOD) adjustment enables the model to dynamically change the rendering precision at different viewing distances. Through the LOD update formula, the system can optimize rendering efficiency and reduce unnecessary computational burden without sacrificing visual effects. This implementation ensures a balance between rendering effects and computational efficiency in large-scale scenes, allowing the system to render complex scenes in real time and efficiently.

[0128] In a possible embodiment, a city plans to digitally manage and optimize the design of a large urban park, and needs to establish a high-precision three-dimensional landscape model for further landscape design, maintenance and public display. The urban park covers a large area, has complex terrain, and contains multiple buildings, vegetation, lakes and artificial landscapes. In order to achieve this goal, it is decided to use the above method to build a three-dimensional landscape reconstruction and visualization system for the park.

[0129] Initial point cloud data collection:

[0130] Equipment and environment: The entire city park was scanned using the laser radar and stereo camera carried by the drone. The laser radar was used to collect high-precision point cloud data of the park's terrain, while the stereo camera was used to capture texture information on buildings, vegetation, and lake surfaces.

[0131] Data collection: Through multiple aerial photography, a point cloud dataset D covering the entire park is obtained, which contains information such as terrain elevation, building outlines, vegetation height and density, etc.

[0132] Point cloud data denoising and filtering:

[0133] Processing process: For the collected initial point cloud data set D, noise reduction is first performed, and the abnormal data points are removed using the formula of adjustment coefficient, noise point coordinates and number of noise points.

[0134] Adaptive filtering: The point cloud data set is further adaptively filtered. Through adaptive filtering of local point cloud density, the smoothness and accuracy of the point cloud data are further improved, ensuring that the purification processing result of the point cloud data set D is more accurate.

[0135] 3D model generation:

[0136] Delaunay triangulation: Using Delaunay triangulation technology, the purified point cloud data is converted into a three-dimensional triangular mesh model M. The model accurately reflects the terrain changes and the structures of major buildings in the urban park.

[0137] Constraint optimization: In the generation process, in order to reduce the number of thin triangles, Delaunay triangulation optimized by constraints is used to ensure the stability of the triangular mesh and the rationality of the structure.

[0138] Real-time lighting and material rendering:

[0139] Lighting model: Based on the 3D model, the scene is rendered in real time using a dynamic ambient lighting adjustment formula to simulate the lighting effects of the park at different times and under different weather conditions. The rendering model can be updated in real time to reflect the changes in the angle of the sun at different times of the day.

[0140] Material rendering: We further use physically based rendering technology to render the buildings, vegetation and water surfaces in the model. Taking into account the roughness of the material surface, the roughness influence factor is used to make the reflection characteristics of each material more realistic.

[0141] Elevation model and GIS integration:

[0142] GIS data fusion: Integrate the generated 3D model with the geographic data G in the GIS system to build a high-precision elevation model that includes terrain, buildings, and vegetation. This model not only shows the topographic changes of the park, but also combines GIS data to show the geographical features of the region, such as water system distribution, road network, etc.

[0143] Slope weighting: The slope weighting of the terrain was taken into account during the generation of the elevation model, which enhanced the representation of the complex terrain inside the park, especially in areas with drastic terrain changes such as hills and lake edges.

[0144] Landscape characteristics analysis:

[0145] Feature calculation: The GIS system analyzes the slope and azimuth of different areas in the park and generates a detailed terrain analysis report. Combined with the terrain roughness parameter, the slope calculation is further refined to ensure accuracy under complex terrain conditions.

[0146] Design adjustments: Based on the analysis results, landscape designers can propose optimized design solutions based on the characteristics of different areas, such as setting up steps or ramps in areas with larger slopes.

[0147] Real-time design parameter adjustment:

[0148] User interaction: Landscape designers can adjust design parameters in real time through the user interface, such as adjusting the distribution density of plants, adding or removing buildings, etc. The system updates the view matrix in real time based on user input and ensures the naturalness of perspective switching through smooth transitions.

[0149] Dynamic feedback: Each time the design parameters are adjusted, the system will immediately provide feedback on the updated 3D model and rendering effect, allowing designers to intuitively see the adjusted effect and further optimize the design plan.

[0150] Multi-scale rendering strategy:

[0151] Rendering efficiency: To improve computing efficiency, the system dynamically adjusts the model's level of detail (LOD) according to changes in the user's perspective. When the viewing distance is far, the model's details are reduced; when observed at a close distance, the model's details are increased to ensure a balance between rendering effect and efficiency.

[0152] Visibility calculation: Through the visibility calculation formula, the system automatically selects and displays key landscape elements at different viewing angles, ensuring that users always pay attention to important landscape details when browsing the 3D model.

[0153] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0154] (1) In the present invention, by optimizing the feature matching and point cloud generation process in multi-view photogrammetry, the problem of insufficient accuracy caused by factors such as perspective differences and illumination changes in traditional technologies is effectively overcome. By introducing adaptive filtering and multi-scale rendering strategies, the present method can accurately restore the details of the landscape in complex scenes, such as the fine structure of buildings and subtle changes in natural landscapes, significantly improving the accuracy and detail expression of the three-dimensional model, and ensuring that the reconstruction results are more realistic and lifelike.

[0155] (2) In the present invention, by introducing an optimization algorithm for multi-view data processing and a level of detail (LOD) adjustment strategy, redundant operations and consumption of computing resources in the calculation process are reduced. Compared with the existing technology, this method can significantly shorten the reconstruction time when processing large-scale scenes and achieve more efficient three-dimensional landscape reconstruction. At the same time, this method also performs real-time optimization for dynamic scenes, which can meet the needs of real-time update and rapid response, and provides technical support for the rapid reconstruction of large-scale scenes.

[0156] (3) In the present invention, in the process of fusion of multi-view image data, through the improved illumination model, material rendering technology and smooth transition strategy of perspective switching, the problems of occlusion, overlapping area processing and perspective difference in multi-view data fusion are effectively solved. The generated three-dimensional model is not only more coherent and natural in visual effect, but also significantly improved in geometric consistency. It is suitable for many fields such as urban planning, environmental analysis, cultural heritage protection, etc., which greatly expands the application scope and value of three-dimensional landscape reconstruction technology.

[0157] The above contents are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0158] There are a few points to note:

[0159] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0160] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0161] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0162] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A three-dimensional landscape reconstruction method based on multi-view photogrammetry, characterized in that: include: S1. Collecting raw data through high-precision sensors to obtain an initial point cloud data set D, wherein the high-precision sensors include laser radar and stereo cameras, and the raw data include terrain, buildings and vegetation; S2. Perform denoising and filtering on the point cloud data set D to obtain the purified point cloud data D'. The denoising process uses the following formula: Among them, α is the adjustment coefficient, d i is the coordinate of the noise point, N is the number of noise points; S3. Generate a 3D model of the purified point cloud through Delaunay triangulation to obtain a triangular mesh M: M=Del(D') Among them, Del(D') represents the Delaunay triangulation processing of point cloud data; S4. Based on the 3D model, real-time rendering of lighting and materials is performed. The lighting model is calculated using the following formula: I=I a +I d ·(N·L)+I s ·(R·V) n Among them, I is the final illumination intensity of the pixel, I a is the ambient light intensity, I d is the diffuse reflected light intensity, I s is the intensity of specular reflection light, N is the normal vector, L is the light source direction vector, R is the reflection vector, V is the sight vector, and n is the specular reflection index; S5. Using physically based rendering technology, the material reflectivity is calculated as: R=F0+(1-F0)·(1-cosθ) 5 Among them, R is the reflectivity, F0 is the intrinsic reflectivity of the material, and θ is the incident angle of the light; S6. Integrate the 3D model with the geographic data G in the GIS system to generate an elevation model Z(x,y): in, is the difference weight, d i is the distance between the point (x, y) and the i-th known point, P is the distance index, Z i is the elevation of the known point; S7. Analyze the characteristics of the landscape area through GIS data, wherein the characteristics include slope S and azimuth A: in, and is the directional derivative of elevation; S8. Adjust the design parameters based on the user input and provide real-time feedback on the updated design effect. Set the update formula of the view matrix V to: V=C·W Among them, C is the camera matrix and W is the world matrix; S9. To improve computational efficiency, a multi-scale rendering strategy is used under different viewing angles. The formula for calculating visibility is: Among them, V ij is the visibility between pixels i and j, d ij is the distance between two points, k is the smoothing coefficient, and d0 is the threshold distance.

2. The three-dimensional landscape reconstruction method based on multi-view photogrammetry according to claim 1, characterized in that: The S2 specifically includes: The denoising process of the point cloud data set D further includes adaptive filtering of the point cloud density, and the calculation formula of the filtered point cloud data set D" is as follows: Among them, β is the adaptive filter coefficient, M is the local point cloud density, and q is the exponential parameter, which further improves the smoothness and accuracy of the point cloud data.

3. The three-dimensional landscape reconstruction method based on multi-view photogrammetry according to claim 1, characterized in that: The S3 specifically includes: The three-dimensional model also includes optimizing Delaunay triangulation by constraint conditions to reduce the number of thin triangles. The optimized triangular mesh calculation formula is: Among them, A k is the area of ​​the kth triangle, P k is the perimeter of the kth triangle, and K is the total number of triangles, thereby generating a triangular mesh that is more consistent with the real terrain.

4. The three-dimensional landscape reconstruction method based on multi-view photogrammetry according to claim 1, characterized in that: The S4 specifically includes: The lighting model further includes an adjustment formula based on dynamic ambient lighting: Among them, I a0 is the initial ambient light intensity, γ is the ambient light adjustment coefficient, L is the number of ambient light sources, I j and d j are the light intensity of the jth light source and its distance from the pixel point respectively.

5. The method for 3D landscape reconstruction based on multi-view photogrammetry according to claim 4, characterized in that: The light source direction vector specifically includes: The light source direction vector L is dynamically updated based on the sun position, and the calculation formula for the sun position is: L(t)=(cosφ·cosδ, sinφ·cosδ, sinδ) Among them, φ is the solar hour angle, δ is the solar declination, and t is the time parameter, which realizes the dynamic simulation of sunlight.

6. The method for 3D landscape reconstruction based on multi-view photogrammetry according to claim 1, characterized in that: The S5 specifically includes: The calculation formula of the material reflectivity R further includes the roughness influence factor: R=F0+(1-F0)·(1-cosθ) 5 ·(1-p) Among them, represents the roughness factor of the material surface, which further improves the realism of material rendering.

7. The method for 3D landscape reconstruction based on multi-view photogrammetry according to claim 1, characterized in that: The S6 specifically includes: The elevation model generation further considers the slope weighting of the terrain, and the weight formula is: Among them, λ is the slope weighting coefficient, which is used to enhance the expression of terrain details.

8. The method for 3D landscape reconstruction based on multi-view photogrammetry according to claim 1, characterized in that: The S7 specifically includes: The slope calculation further considers the terrain roughness η, and the updated slope calculation formula is: Among them, η is the terrain roughness parameter, which is used to accurately calculate the slope of complex terrain.

9. The method for 3D landscape reconstruction based on multi-view photogrammetry according to claim 1, characterized in that: The S8 specifically includes: The view matrix update further considers the smooth transition of the user's perspective, and the updated view matrix formula is: V new =V old +ξ·(V target -V old ) Among them, ξ is the smooth transition coefficient, V target is the target view matrix, V old This is the old view matrix, which ensures smoothness of perspective switching.

10. The method for 3D landscape reconstruction based on multi-view photogrammetry according to claim 1, characterized in that: The S9 specifically includes: The multi-scale rendering strategy further includes perspective-dependent level of detail (LOD) adjustment, and the LOD update formula is: Among them, κ is the control parameter, d ij is the display distance of the model details under the current viewing angle, and d0 is the reference distance, which is used to ensure the rendering efficiency and effect under different viewing distances.

Images