Rapid three-dimensional building modeling method based on single remote sensing image

Abstract

The invention relates to the technical field of geographic information live-action three-dimensional modeling, and particularly discloses a building rapid three-dimensional modeling method based on a single remote sensing image. According to the method, rapid three-dimensional building modeling is carried out through a single remote sensing image, accurate geographic coordinates and technical parameters such as the inner orientation and the outer orientation of a photo are not needed, and the technical threshold and the cost input of modeling are greatly reduced; multiple algorithms are used for image segmentation and three-dimensional model construction, the automation degree is high, the modeling period is greatly shortened, the modeling speed is high, and manual intervention is not needed; the building shadow length is counted, the elevation is assigned in proportion, the height of the building is accurately calculated by combining the solar elevation angle, the remote sensing equipment elevation angle and the azimuth angle, the precision of the model is improved, three-dimensional modeling data can be lightened to a certain extent, the structure is simple, and the data size is one fifth of that of a normal three-dimensional model. And the data redundancy of the three-dimensional model is greatly reduced.

Description

Technical Field

[0001] The invention relates to the technical field of geographic information real-scene three-dimensional modeling, and in particular to a method for rapid three-dimensional building modeling based on a single remote sensing image. Background Art

[0002] With the proposal of real-life 3D China, the use of multi-resolution, multi-phase, multi-source earth observation data, drones and other applications to build real-life 3D has become a hot technology. Photogrammetry technology uses multi-view sensors on aircraft to image from multiple angles such as vertical and oblique, which can not only obtain ground elevation information but also terrain texture features. In recent years, high-resolution optical satellite remote sensing technology has been developed, and the images obtained have high accuracy after radiation and geometric correction, which can provide an important data source for building high-precision 3D real geographic scene models for the digital earth.

[0003] However, it is difficult to quickly realize large-scale 3D scene modeling based on commonly used software platforms such as SketchUp and 3dMax, and it requires precise parameter settings and spatial 3D calculation programs. The existing methods are slow and require manual modification before they can be actually used. In addition, the 3D model data established by the existing methods is redundant, and the large amount of 3D model data is not conducive to the browsing and use of large 3D model scenes. Due to the massive characteristics of 3D data, how to lightweight image data and quickly establish 3D models is the current technical bottleneck.

[0004] In order to solve the above defects, a technical solution is now provided. Summary of the invention

[0005] The purpose of the present invention is to provide a method for rapid three-dimensional building modeling based on a single remote sensing image to solve the problems raised in the above background.

[0006] The purpose of the present invention can be achieved by the following technical solution: A method for rapid three-dimensional building modeling based on a single remote sensing image comprises the following steps:

[0007] Step 1: Remote sensing image acquisition: remote sensing images corresponding to the target area are acquired by remote sensing equipment, wherein the remote sensing equipment includes a satellite or a drone for acquiring high-quality remote sensing images.

[0008] Step 2: Remote sensing image preprocessing: Perform denoising and contrast enhancement preprocessing operations on the remote sensing images corresponding to the target area.

[0009] Step 3: Remote sensing image segmentation: Segment the remote sensing image corresponding to the target area to extract the building's outer contour, building side texture and corresponding shadow area.

[0010] Step 4: Count the lengths of building shadows and assign heights: Count the lengths of all building shadows and sort them. Take the shortest building shadow as the height of one layer, and assign elevations to other shadows that are multiples of it in equal proportion, to get the relative elevation of the lowest building among all buildings.

[0011] Step 5: 3D model construction: 3D models of all buildings in the target area are constructed to generate 3D models of all buildings in the target area.

[0012] Step 6: Model optimization and adjustment: Optimize and adjust the generated three-dimensional model to obtain an optimized three-dimensional model.

[0013] Step 7. Output: Use the optimized 3D model for subsequent analysis, display or decision support.

[0014] Furthermore, the preprocessing operation of performing denoising and contrast enhancement on the remote sensing image corresponding to the target area includes:

[0015] De-noising can use median filtering or Gaussian filtering methods to remove noise in the image, and contrast enhancement can be achieved through histogram equalization or contrast stretching methods to improve the accuracy of subsequent processing.

[0016] Furthermore, the remote sensing image corresponding to the target area is segmented, and the specific processing method is as follows:

[0017] After identifying the buildings in the target area, the Canny edge detection algorithm is used to detect the edge information of the buildings. The edge points obtained by edge detection are used as seed points, and regions are generated according to the similarity of pixels. The contour area of ​​the buildings is gradually expanded to form the contour area of ​​the buildings. At the same time, the trained convolutional neural network (CNN) learning model is used to automatically extract the segmentation results of the buildings, including the outer contour of the buildings, side textures and shadow areas.

[0018] Furthermore, the lengths of all building shadows are counted and sorted, the shortest building shadow is used as the height of one layer, and other shadows are multiples of it and are assigned elevations in equal proportion, so as to obtain the relative elevation of the lowest building of all buildings. The specific processing method is as follows:

[0019] For each identified building shadow area, the pixel coordinate calculation method is used to measure the shadow length. The shadow can be regarded as a series of continuous pixel points. The shadow length is determined by calculating the distance between these pixel points in the image coordinate system, thereby obtaining the shadow length of all buildings in the target area.

[0020] Sort all the shadow lengths of buildings in the target area from long to short to obtain a sorted set of all the shadow lengths of buildings in the target area, and extract the shortest shadow length of the building from it, mark it as L1, and use L1 to calculate the shortest building height H1. For other building shadow lengths L i , according to its proportional relationship with the shortest building shadow length L1 i =L i / L1, calculate the corresponding height H i =r i ×H1, where i represents the number of all buildings except the building with the shortest shadow length, i=1, 2, ..., m, where m represents a positive integer;

[0021] Thus, the relative elevation of all buildings based on the lowest building is obtained.

[0022] Furthermore, the specific calculation formula of the shortest building height H1 is as follows:

[0023] Among them, β is the solar altitude angle, that is, the angle between the direction of incident sunlight and the horizon, θ is the altitude angle of the remote sensing device, γ is the angle between the solar azimuth and the building azimuth, and its value is equal to the solar azimuth minus the building azimuth, α is the angle between the remote sensing device azimuth and the building azimuth, and its value is equal to the remote sensing device azimuth minus the building azimuth.

[0024] Furthermore, the three-dimensional model of all buildings in the target area is constructed, and the specific execution steps are as follows:

[0025] 501: Contour line stretching: Convert all extracted building outer contour lines into three-dimensional coordinate form, stretch the contour lines along the vertical direction according to the calculated elevation value, and for each contour vertex, calculate the coordinates of the stretched vertex according to its corresponding elevation value.

[0026] 502: Texture Mapping: The extracted texture image of the building side is preprocessed to match the size and direction of the polyhedron side. Then, the texture coordinates of the pixels in the triangle patch are determined by calculating the UV coordinates of the vertices and using interpolation. Next, bilinear filtering is used for texture mapping to accurately paste the texture image to the side and top of the polyhedron to avoid texture aliasing and make the model present a realistic texture appearance. At the same time, the top of the polyhedron is mapped to match its top texture with the real environment.

[0027] 503: Model positioning: Based on the relative height and location information of buildings extracted from remote sensing images, quickly locate and establish a real three-dimensional model on the remote sensing image map.

[0028] Furthermore, the optimization and adjustment process of the generated three-dimensional model specifically includes the following steps:

[0029] 601: Vertex optimization adjustment: Check the vertices in the 3D model, remove duplicate vertices, merge and simplify the vertices.

[0030] 602: Edge and face optimization: For coplanar polygons, you can merge them into a larger polygon to reduce the number of faces. At the same time, remove invisible faces, that is, those that are blocked by other objects and will not be displayed during rendering. The visibility of the face can be determined by the line of sight and the normal vector of the polygon.

[0031] 603: Rendering efficiency optimization adjustment: Use level of detail (LOD) technology to dynamically adjust the model's detail level according to the distance between the model and the observer. When the model is far from the viewpoint, use a simplified low-detail model for rendering; when the model is close to the viewpoint, gradually switch to a high-detail model. This can greatly improve rendering efficiency while ensuring visual effects.

[0032] Beneficial effects of the present invention:

[0033] The present invention provides a method for rapid 3D modeling of buildings based on a single remote sensing image, which does not require precise geographic coordinates, the interior and exterior orientations of the photo and other technical parameters, greatly reducing the technical threshold and cost investment for modeling; uses a variety of algorithms for image segmentation and 3D model construction, has a high degree of automation, greatly shortens the modeling cycle, has a fast modeling speed and does not require human intervention; counts the length of the building's shadow and assigns elevation in proportion, combines the solar altitude angle, the altitude angle of the remote sensing device and the azimuth angle to accurately calculate the building height, improves the accuracy of the model, and to a certain extent the 3D modeling data can be lightweight and simple in structure, which is one-fifth of the normal 3D model data volume, greatly reducing the 3D model data redundancy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The present invention will be further described below in conjunction with the accompanying drawings.

[0035] Figure 1 The present invention is a flowchart of a method for rapid three-dimensional building modeling based on a single remote sensing image. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] See also Figure 1 As shown, the present invention is a method for rapid three-dimensional building modeling based on a single remote sensing image, comprising the following steps:

[0038] Step 1: remote sensing image acquisition: remote sensing images corresponding to the target area are acquired by remote sensing equipment, wherein the remote sensing equipment includes a satellite or a drone for acquiring high-quality remote sensing images;

[0039] It should be noted that during the remote sensing image acquisition process, when the remote sensing equipment selects a satellite, the satellite's orbit, sensor type and resolution parameters must be considered. Different satellite orbits and sensors will affect the coverage, resolution and spectral information of the remote sensing image. Select a high-resolution satellite sensor to obtain clear building details.

[0040] When choosing a drone as a remote sensing device, it is necessary to set the appropriate flight altitude, route planning and camera parameters. The flight altitude determines the resolution and coverage of the image. The route planning should ensure complete coverage of the target area. Camera parameters such as focal length and aperture will affect the quality and clarity of the image.

[0041] Ensure that remote sensing images are not affected by cloud obstruction or atmospheric interference during the acquisition process. Cloud obstruction will cause information loss in some areas, and atmospheric interference will cause image blur or color aberration.

[0042] Collect remote sensing images from multiple time periods so that the best quality image can be selected in subsequent processing, or the image quality can be improved by fusing images from multiple time periods.

[0043] Step 2: Remote sensing image preprocessing: Perform denoising and contrast enhancement preprocessing operations on the remote sensing image corresponding to the target area, including:

[0044] Denoising can use median filtering or Gaussian filtering methods to remove noise in the image, and enhancement can be performed through histogram equalization or contrast stretching methods to improve the accuracy of subsequent processing.

[0045] It should be noted that the median filtering method is specifically: for each pixel, sort the pixel values ​​in its neighborhood and take the middle value as the new value of the pixel. For example, for a 3×3 neighborhood, sort the 9 pixel values ​​from small to large, and take the middle pixel value to replace the value of the central pixel. In this way, isolated noise points can be removed while retaining the edges and details of the image.

[0046] The Gaussian filtering method is specifically: weighted averaging of neighborhood pixels based on Gaussian distribution. The closer the pixel is to the center, the greater the weight; the farther from the center, the smaller the weight. By adjusting the size and standard deviation of the Gaussian kernel, the degree of denoising and the degree of retention of image details can be controlled.

[0047] The histogram equalization method is specifically: by redistributing the grayscale of the image, the number of pixels at each grayscale level is roughly equal. This can make the originally darker or brighter areas clearer and highlight the characteristics of the building. For example, for a remote sensing image, after histogram equalization, the grayscale difference between the building and the background may be more obvious, which is convenient for subsequent segmentation and extraction.

[0048] The contrast stretching method is specifically: by setting the minimum and maximum grayscale values, the original grayscale value range is linearly or nonlinearly stretched. For example, stretching the grayscale value range from [0,100] to [0,255] can make the darker parts of the image brighter and the brighter parts brighter, thereby highlighting the outline and texture of the building.

[0049] Step 3: Remote sensing image segmentation: Segment the remote sensing image corresponding to the target area to extract the building's outer contour, building side texture and corresponding shadow area. The specific processing method is as follows:

[0050] After identifying the buildings in the target area, the Canny edge detection algorithm is used to detect the edge information of the buildings. The edge points obtained by edge detection are used as seed points, and regions are generated according to the similarity of pixels. The contour area of ​​the buildings is gradually expanded to form the contour area of ​​the buildings. At the same time, the trained convolutional neural network (CNN) learning model is used to automatically extract the segmentation results of the buildings, including the outer contour of the buildings, side textures and shadow areas.

[0051] It should be noted that the specific method of the Canny edge detection algorithm is: by calculating the gradient amplitude and direction of the image, the gradient amplitude reflects the intensity of the change of the pixel gray value, and the direction indicates the direction in which the gray value changes fastest; then through non-maximum suppression, the local maximum value in the gradient direction is retained, and other values ​​are set to 0; finally, double threshold detection and edge connection are used to determine the final edge. For example, a high threshold is used to determine strong edge pixels, and a low threshold is used to connect weak edge pixels around strong edge pixels to form a complete edge contour. In building edge detection, the threshold parameter can be adjusted according to the grayscale difference between the building and the background to accurately extract the edge of the building.

[0052] Select seed points, usually pixels inside buildings or with obvious building features are selected as seeds. Then, according to the similarity criteria of pixels (such as gray value, color, texture), the pixels that meet the conditions in the neighborhood are merged into the area where the seeds are located. In the process of regional growth, it is necessary to set appropriate growth criteria and stop conditions. For example, set a gray value difference threshold. If the gray value difference between the neighborhood pixel and the seed pixel is within the threshold range, the neighborhood pixel is merged into the region; when the region no longer meets the growth conditions, stop growing. By repeating this process until there are no pixels that meet the conditions to be merged, the building area is segmented.

[0053] In the process of region growing, we can combine the prior knowledge of buildings, such as the shape and size of buildings, to constrain and adjust the growth process. For example, if the building is roughly rectangular or polygonal, we can limit the direction and range of region growing to make it more consistent with the actual shape of the building.

[0054] For the training of deep learning models, data enhancement techniques such as rotation, flipping, and scaling can be used to increase the diversity of data and improve the generalization ability of the model. At the same time, the hyperparameters of the model, such as learning rate, convolution kernel size, and number of layers, can be adjusted to optimize the performance of the model.

[0055] Step 3: Count the length of building shadows and assign heights: Count the lengths of all building shadows and sort them. Take the shortest building shadow as the height of one layer, and assign elevations to other shadows that are multiples of it in equal proportion. Get the relative elevation of the lowest building among all buildings. The specific processing method is as follows:

[0056] For each identified building shadow area, the pixel coordinate calculation method is used to measure the shadow length. The shadow can be regarded as a series of continuous pixel points. The shadow length is determined by calculating the distance between these pixel points in the image coordinate system. For example, for horizontal shadows, the sum of the horizontal coordinate differences of the shadow pixel points can be calculated; for oblique shadows, the Euclidean distance formula can be used to calculate the distance between adjacent pixel points, and then the sum is obtained to obtain the shadow length; thus, the shadow lengths of all buildings in the target area are obtained;

[0057] It should be noted that when calculating the shadow length, the spatial resolution of the remote sensing image needs to be considered. According to the conversion relationship between the spatial resolution of the image and the pixel coordinates, the pixel length is converted to the actual geographic length. For example, if the spatial resolution is 1 meter / pixel, then the pixel length of the shadow multiplied by 1 meter / pixel can get the actual length of the shadow.

[0058] Sort all the shadow lengths of buildings in the target area from long to short to obtain a sorted set of all the shadow lengths of buildings in the target area, and extract the shortest shadow length of the building from it, mark it as L1, and use L1 to calculate the shortest building height H1. For other building shadow lengths L i , according to its proportional relationship with the shortest building shadow length L1 i =L i / L1, calculate the corresponding height H i =r i ×H1, where i represents the number of all buildings except the building with the shortest shadow length, i=1, 2, ..., m, where m represents a positive integer;

[0059] Specifically, the calculation formula for the shortest building height H1 is as follows:

[0060] Among them, β is the solar altitude angle, that is, the angle between the direction of incident sunlight and the horizon, θ is the altitude angle of the remote sensing equipment, γ is the angle between the solar azimuth and the building azimuth, and its value is equal to the solar azimuth minus the building azimuth, α is the angle between the satellite azimuth and the building azimuth, and its value is equal to the remote sensing equipment azimuth minus the building azimuth.

[0061] Step 5: 3D model construction: 3D models of all buildings in the target area are constructed to generate 3D models of all buildings in the target area. The specific steps are as follows:

[0062] 501: Contour stretching: Convert all the extracted building contour lines into three-dimensional coordinate form, stretch the contour lines in the vertical direction according to the calculated elevation value, and calculate the vertex coordinates after stretching for each contour vertex according to its corresponding elevation value. Assuming that the original contour vertex coordinates are (x, y) and the elevation is h, the vertex coordinates after stretching are (x, y, h). During the stretching process, it is necessary to keep the topological structure of the contour line unchanged, that is, the connection relationship between adjacent vertices remains unchanged. At the same time, in order to improve the precision and accuracy of the model, the curve fitting method can be used to further optimize the contour line. For example, the Bezier curve is used to fit the contour line to make the stretched polyhedron smoother and more natural.

[0063] 502: Texture Mapping: Preprocess the extracted texture image of the building side, including adjusting the image size, orientation and color to match the size and orientation of the polyhedron side. Then, calculate the UV coordinates of the vertices and use interpolation to determine the texture coordinates of the pixels in the triangle patch. Then, use bilinear filtering for texture mapping to accurately paste the texture image to the side and top of the polyhedron to avoid texture aliasing and make the model present a real texture appearance. At the same time, map the top of the polyhedron so that its top texture matches the real environment.

[0064] 503: Model positioning: Based on the relative height and location information of buildings extracted from remote sensing images, quickly locate and establish a real three-dimensional model on the remote sensing image map.

[0065] Step 6: Model optimization and adjustment: Optimize and adjust the generated 3D model to obtain an optimized 3D model, which specifically includes the following steps:

[0066] 601: Vertex optimization and adjustment: Check the vertices in the 3D model and remove duplicate vertices. Duplicate vertices will not only increase the data volume of the model, but may also cause rendering errors. You can determine whether they are duplicated by comparing the coordinate values ​​of the vertices and only retain unique vertices; merge and simplify vertices. For example, for some vertices that are close to each other, you can merge them into one vertex and adjust the structure of the edges and faces connected to them. This can reduce the number of vertices and improve the rendering efficiency of the model.

[0067] 603: Rendering efficiency optimization adjustment: Use level of detail (LOD) technology to dynamically adjust the model's detail level according to the distance between the model and the observer. When the model is far from the viewpoint, use a simplified low-detail model for rendering; when the model is close to the viewpoint, gradually switch to a high-detail model. This can greatly improve rendering efficiency while ensuring visual effects.

[0068] Step 7. Output: Use the optimized 3D model for subsequent analysis, display or decision support.

[0069] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A method for rapid three-dimensional building modeling based on a single remote sensing image, characterized in that: The following steps are involved: Step 1: Remote sensing image acquisition: remote sensing images corresponding to the target area are acquired by remote sensing equipment to obtain remote sensing images corresponding to the target area; Step 2: Remote sensing image preprocessing: performing denoising and contrast enhancement preprocessing operations on the remote sensing image corresponding to the target area; Step 3: Remote sensing image segmentation: Segment the remote sensing image corresponding to the target area to extract the building's outer contour, building side texture and corresponding shadow area; Step 4: Count the length of building shadows and assign heights: Count the lengths of all building shadows and sort them. Take the shortest building shadow as the height of one layer, and assign elevations to other shadows that are multiples of it in equal proportions to obtain the relative elevation of the lowest building among all buildings. Step 5: 3D model construction: construct 3D models of all buildings in the target area to generate 3D models of all buildings in the target area; Step 6: Model optimization and adjustment: Optimize and adjust the generated three-dimensional model to obtain an optimized three-dimensional model; Step 7. Output: Use the optimized 3D model for subsequent analysis, display or decision support.

2. The method for rapid three-dimensional building modeling based on a single remote sensing image according to claim 1, characterized in that: The preprocessing operation of denoising and contrast enhancement of the remote sensing image corresponding to the target area includes: De-noising uses median filtering or Gaussian filtering methods to remove noise from the image, and contrast enhancement is achieved through histogram equalization or contrast stretching methods.

3. The method for rapid three-dimensional building modeling based on a single remote sensing image according to claim 1, characterized in that: The remote sensing image corresponding to the target area is segmented, and the specific processing method is as follows: After identifying the buildings in the target area, the Canny edge detection algorithm is used to detect the edge information of the buildings. The edge points obtained by edge detection are used as seed points, and regions are generated according to the similarity of pixels. The contour area of ​​the buildings is gradually expanded to form the contour area of ​​the buildings. At the same time, the trained convolutional neural network learning model is used to automatically extract the segmentation results of the buildings, including the outer contour of the buildings, side textures and shadow areas.

4. The method for rapid three-dimensional building modeling based on a single remote sensing image according to claim 1, characterized in that: The lengths of all building shadows are counted and sorted, with the shortest building shadow as the height of one layer, and other shadows that are multiples of it are assigned elevations in equal proportion, to obtain the relative elevation of the lowest building of all buildings. The specific processing method is as follows: For each identified building shadow area, the pixel coordinate calculation method is used to measure the shadow length. The shadow is regarded as a series of continuous pixel points. The shadow length is determined by calculating the distance between these pixel points in the image coordinate system, thereby obtaining the shadow length of all buildings in the target area. Sort all the shadow lengths of buildings in the target area from long to short to obtain a sorted set of all the shadow lengths of buildings in the target area, and extract the shortest shadow length of the building from it, mark it as L1, and use L1 to calculate the shortest building height H1. For other building shadow lengths L i , according to its proportional relationship with the shortest building shadow length L1 i =L i / L1, calculate the corresponding height H i =r i ×H1, where i represents the number of all buildings except the building with the shortest shadow length, i=1, 2, ..., m, where m represents a positive integer; This gives the relative elevation of all buildings based on the lowest building.

5. The method for rapid three-dimensional building modeling based on a single remote sensing image according to claim 4, characterized in that: The specific calculation formula of the shortest building height H1 is as follows: Among them, β is the solar altitude angle, that is, the angle between the direction of incident sunlight and the horizon, θ is the altitude angle of the remote sensing device, γ is the angle between the solar azimuth and the building azimuth, and its value is equal to the solar azimuth minus the building azimuth, α is the angle between the remote sensing device azimuth and the building azimuth, and its value is equal to the remote sensing device azimuth minus the building azimuth.

6. The method for rapid three-dimensional building modeling based on a single remote sensing image according to claim 1, characterized in that: The specific steps of constructing the three-dimensional model of all buildings in the target area are as follows: 501: Contour line stretching: convert the extracted outer contour lines of all buildings into three-dimensional coordinate form, stretch the contour lines along the vertical direction according to the calculated elevation value, and for each contour vertex, calculate the coordinates of the stretched vertex according to its corresponding elevation value; 502: Texture mapping: The extracted texture image of the building side is preprocessed to match the size and direction of the polyhedron side. Then, the texture coordinates of the pixels in the triangle patch are determined by calculating the UV coordinates of the vertices and using the interpolation method. Then, bilinear filtering is used for texture mapping to accurately paste the texture image to the side and top of the polyhedron. At the same time, the top of the polyhedron is mapped so that the top texture matches the real environment. 503: Model positioning: Based on the relative height and location information of buildings extracted from remote sensing images, quickly locate and establish a real three-dimensional model on the remote sensing image map.

7. The method for rapid three-dimensional building modeling based on a single remote sensing image according to claim 1, characterized in that: The optimization and adjustment process of the generated three-dimensional model specifically includes the following steps: 601: Vertex optimization adjustment: Check the vertices in the 3D model, remove duplicate vertices, merge and simplify the vertices; 602: Edge and face optimization adjustment: For coplanar polygons, merge them into a larger polygon, reduce the number of faces, and remove invisible faces; 603: Rendering efficiency optimization adjustment: Use the level of detail technology to dynamically adjust the model's detail level according to the distance between the model and the observer. When the model is far away from the viewpoint, use a simplified low-detail model for rendering. When the model is close to the viewpoint, gradually switch to a high-detail model.

Images