High-precision DEM construction method based on point cloud model

Abstract

The invention discloses a high-precision DEM construction method based on a point cloud model. The method comprises the following steps: step 1, performing grid division on urban point cloud data, dynamically adjusting the grid size and the neighborhood range based on a weighted gradient value, and calculating the weighted gradient value through the height difference and the distance between the lowest points of a central grid and a neighborhood grid; 2, clustering the urban point clouds by using DBSCAN (Density Based Spatial Clustering of Applications with Noise) and combining the features of color, intensity and texture of the point clouds; 3, taking the extracted ground feature line as a constraint during point cloud filtering, and determining the boundary of a ground point and a non-ground point; step 4, based on the ground feature line constraint in the step 3, executing multi-scale adaptive filtering, and removing non-ground points; 5, taking the extracted ground feature line as a constraint during point cloud filtering, and determining the boundary of a ground point and a non-ground point; and step 6, aiming at different partitions, respectively using a nearest neighbor interpolation up-sampling algorithm to construct a TIN (Triangulated Irregular Network) method to construct an urban high-precision DEM (Digital Elevation Model).

Description

Technical Field

[0001] The present invention relates to the field of point cloud model construction, and particularly to a high-precision DEM construction method based on a point cloud model. Background Technique

[0002] Urban scenes are large in scale, with tall buildings standing in great numbers, streets crisscrossing, and artificial facilities being dense and complex. The topography and landforms are characterized by fragmentation and rich detailed structures. Human activities have significantly changed the surface morphology, forming a new type of landform landscape where abrupt terrains and gradual terrains coexist in an interlaced manner. Traditional DEM construction methods focus on the continuous expression of the surface morphology and are difficult to meet the DEM accuracy requirements in urban areas. Currently, using a laser scanner to scan to obtain point cloud data and generate high-precision DEM data is one of the mainstream methods for obtaining high-precision urban DEM data. Point cloud is a collection of points in three-dimensional space, retaining geometric information such as three-dimensional coordinates, textures, and colors, and has become an important means for high-precision urban terrain modeling.

[0003] In recent years, how to construct a high-precision urban DEM has become an urgent problem to be solved in related research fields such as urban planning and urban surface process simulation. In a large-scale urban point cloud scene, constructing a high-precision DEM needs to meet the requirements of both efficiency and accuracy. Currently, most of the related research on urban DEM construction focuses on specific regions or emphasizes the solution of local problems, such as the construction of road DEM and the construction of DEM in waterlogging-prone areas. Common methods for constructing urban DEM using point clouds mainly include elevation grids, ordinary Kriging models, and Delaunay interpolation triangular meshes, etc.

[0004] In summary, the accuracy and currency of a high-precision urban DEM can effectively improve the working level of urban planning, design, and management. However, due to the dense artificial facilities and fragmented topography and landforms in cities, there are mostly problems in the construction of DEM, such as incomplete construction of the fine urban surface morphology, difficult to accurately remove surface coverings, and attribute loss. How to construct a high-precision urban DEM has become an urgent problem to be solved in related research fields such as urban planning and urban surface process simulation. Summary of the Invention

[0005] The purpose of the present invention is to provide a high-precision DEM construction method based on a point cloud model to solve the problems raised in the above background technique.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A high-precision DEM construction method based on a point cloud model, the specific steps are as follows:

[0007] S1. Divide the urban point cloud data into grids, dynamically adjust the grid size and neighborhood range based on the weighted slope value, and calculate the weighted slope value through the height difference and distance between the central grid and the lowest point of the neighborhood grid;

[0008] S2. Adopt an improved DBSCAN clustering algorithm, and combine the color, intensity, and texture features of the point cloud to perform multi-feature fusion clustering on the urban point cloud, so as to distinguish features and ground points with similar elevations but different attributes;

[0009] S3. Extract the feature normal of the point cloud based on principal component analysis (PCA), solve the normal direction through the covariance matrix of the neighborhood point set, and combine the view point constraint to determine the normal orientation, and extract the feature line of the feature as the segmentation boundary between the ground and non-ground points;

[0010] S4. Based on the constraint of the feature line of the feature in step 3, perform multi-scale adaptive filtering, use the ground seed points to construct a quadratic surface to fit the initial ground reference surface, dynamically adjust the height difference threshold through the normal distribution model, and remove non-ground points;

[0011] S5. For the filtered ground point cloud, use a region growing algorithm that combines the normal angle, curvature, and color similarity to perform zoning, and divide the urban terrain into flat surfaces, natural undulations, and artificial undulation regions;

[0012] S6. For different partitions, use the nearest neighbor interpolation to upsample and encrypt the point cloud and then construct a regular grid DEM, and use the irregular triangular network (TIN) to fit the terrain.

[0013] Preferably, in step S1, dividing the large-scale urban point cloud data set into grids can significantly improve the data processing efficiency while retaining the key spatial features;

[0014] The size of the grid and the neighborhood size are set according to the weighted slope value, and the calculation formula of the weighted slope value is:

[0015]

[0016] Where D i is the distance from the lowest point of the central grid to the lowest point of the i-th neighborhood grid, and Z min is the minimum value of the point cloud coordinates.

[0017] Preferably, in step S2, after the grid division is completed, use the density-based DBSCAN clustering algorithm combined with the point cloud features to iteratively perform point cloud clustering;

[0018] Use the color information and intensity information as auxiliary features for point cloud clustering to improve the distinguishability of features and ground points with different colors and similar elevations.

[0019] Preferably, in step S3, the normal direction is solved by solving the eigenvectors and eigenvalues of the covariance matrix, and the view point constraint is introduced to ensure that the normal orientations are consistent, satisfying: n i *(v p -p i) > 0,

[0020] where n i is the normal of point p i , v p is the preset viewpoint coordinate, and the method of using the K-nearest neighbor classification algorithm to determine the neighborhood point set of a certain point is used to extract the feature lines of ground objects.

[0021] Preferably, in step S4, first, the lowest point in the ground cluster identified by the point cloud clustering result is used as the ground seed point to construct a reference plane for initially fitting the ground quadratically;

[0022] After selecting a sufficient number of seed points, approximate the terrain with these seed points. Then, while determining the height difference threshold for surface fitting through the idea of normal distribution, perform multi-scale adaptive filtering. Use the height difference between the point to be classified and the corresponding grid of the ground reference plane as the criterion to identify ground points. When the height difference of the point to be classified is less than a certain threshold, mark this point as a ground point.

[0023] Preferably, in step S5, based on the ground point cloud after point cloud filtering obtained in step S4, regional segmentation is performed. First, the constraint conditions of the region growing algorithm include: the normal angle is less than the threshold θ_threshold, the curvature is lower than the threshold k_threshold, and the color similarity is higher than the set value, which are used to determine whether adjacent points can belong to the same region;

[0024] Then, select a seed point as the starting point and mark it as visited;

[0025] Starting from the seed point, traverse its neighboring points, and judge whether to add it to the same region according to the growth criterion. If the conditions are met, mark this point as visited and add it to the current region;

[0026] Repeat the above steps to continuously expand the current region until no new points meet the growth conditions, and complete the point cloud regional segmentation.

[0027] Preferably, in step S6, for the urban ground point clouds in different regions after segmentation in step S5, DEM construction is carried out. The method of upsampling the sparse point cloud is adopted to increase the point cloud density and details. On this basis, continuously continuous irregular triangular surfaces are constructed to approximate the real terrain.

[0028] Compared with the prior art, the beneficial effects of the present invention are:

[0029] 1. By extracting the local feature lines of ground objects, the boundaries between ground objects and the ground are clarified and accurately segmented. Combining with the rich spatial geometric information of the point cloud, the influence of ground object occlusion on the separation of ground points and non-ground points is greatly reduced;

[0030] 2. On the basis of accurately separating ground points from non-ground points, slope factors are introduced to set grid neighborhoods of different sizes, realizing the adaptive adjustment of the height difference threshold for point cloud surface fitting, and improving the efficiency of point cloud filtering and the universality of the algorithm;

[0031] 3. Improve the region growing algorithm by combining point cloud normal, curvature and color features, and process the extracted ground point cloud in regions according to the urban surface morphology, improving the overall accuracy of the constructed DEM. Brief Description of the Drawings

[0032] Figure 1 is the principle block diagram of the present invention;

[0033] Figure 2 is the overall work flow chart of the present invention;

[0034] Figure 3 is the flow chart for extracting feature lines of ground objects of the present invention;

[0035] Figure 4 is the point cloud adaptive filtering diagram of the present invention;

[0036] Figure 5 is the point cloud segmentation diagram based on the region growing algorithm of the present invention. Detailed Embodiment

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0038] Please refer to Figures 1 to 5 , the present invention provides a technical solution: a method for constructing a high-precision DEM based on a point cloud model, characterized in that the specific steps are as follows:

[0039] S1. Divide the urban point cloud data into grids, dynamically adjust the grid size and neighborhood range based on the weighted slope value, and calculate the weighted slope value through the height difference and distance between the center grid and the lowest point of the neighborhood grid;

[0040] First, divide the large-scale urban point cloud data set into grids, which can significantly improve the data processing efficiency while retaining key spatial features. In this example, the grid size and neighborhood size are set according to the weighted slope value.

[0041] That is, the calculation formula

[0042] W i is the weight factor, Di is the distance from the lowest point of the central grid to the lowest point of the i-th neighboring grid, S i is the slope value calculated from the lowest point of the central grid and the lowest point of the i-th neighboring grid, Z min is the minimum value of the point cloud coordinates, and the calculated W i and S i are substituted into the calculation formula to calculate the weighted slope value.

[0043] S2. An improved DBSCAN clustering algorithm is adopted to perform multi-feature fusion clustering on urban point clouds by combining the color, intensity, and texture features of the point clouds, so as to distinguish objects with similar elevations but different attributes from ground points;

[0044] After the grid division is completed, a density-based clustering algorithm (DBSCAN) is used to iteratively perform point cloud clustering in combination with point cloud features. The color information and intensity information are used as auxiliary features for point cloud clustering to improve the distinguishability of objects and ground points with different colors and similar elevations.

[0045] S3. Based on principal component analysis (PCA), the normal vectors of point cloud features are extracted. The normal direction is solved through the covariance matrix of the neighborhood point set, and the normal orientation is determined in combination with the view point constraint, and the feature lines of the objects are extracted as the segmentation boundary between the ground and non-ground points;

[0046] For the situation where the objects are complex and the ground points cover each other, the present invention extracts the feature lines of the objects to clarify the boundary between the objects and the ground

[0047] The problem of determining the normal vector of a point on the ground surface is approximately the problem of estimating the normal vector of the tangent plane of the ground surface. Therefore, the estimation of the surface normal vector can be converted into solving the eigenvectors and eigenvalues of the covariance matrix (or PCA - principal component analysis). This covariance matrix is created by querying the neighboring points within a certain range of the point. Specifically, for each point p i , the corresponding covariance matrix C is as follows:

[0048]

[0049]

[0050] In Equation (2), K is the number of neighboring points of point p i , represents the three-dimensional centroid of the nearest neighbor points, λ j is the j-th eigenvalue of the covariance matrix, is the j-th eigenvector.

[0051] PCA can well achieve dimensionality reduction and find the tangent plane, but it cannot determine the normal direction of the tangent plane. The way to solve this problem is to introduce a viewpoint constraint to ensure that when determining the normal direction, it can start from a specific and limited perspective. Assume a viewpoint v is introduced. p To make all the normal vectors n i always face the side of the viewpoint, they need to satisfy the following inequality constraint:

[0052] n i ·(v p -p i ) > 0

[0053] As described above, the normal estimation of a certain point is inseparable from the constraints of its neighboring points. In the present invention, the k-Nearest Neighbor (KNN) classification algorithm is used to determine the neighboring point set of a certain point.

[0054] S4. Based on the ground feature line constraint in step 3, perform multi-scale adaptive filtering, use the ground seed points to construct a quadratic surface to fit the initial ground reference plane, and dynamically adjust the height difference threshold through the normal distribution model to eliminate non-ground points.

[0055] Perform multi-scale point cloud adaptive filtering on the clustered point cloud obtained in the second step and use the ground feature line obtained in step 3 as the point cloud filtering constraint condition. Due to the complexity of the urban point cloud scene, some terrain detail features may not be clustered into point cloud clusters and are ignored. To enhance the recognition ability of ground points at terrain micro differences and achieve the adaptive dynamic adjustment of parameters according to specific terrain features, this example uses an adaptive filtering method to separate the ground point cloud from the object point cloud.

[0056] To further identify the ground points in the terrain detail part, use the lowest point in the ground cluster identified by the point cloud clustering result as the ground seed point and construct a reference plane for quadratic fitting of the initial ground. After selecting enough seed points, use these seed points to fit the approximate terrain, and the quadratic surface fitting equation is as follows:

[0057]

[0058] In the formula, a = (j = 0, 1,..., 5) represents the coefficients of the surface equation, and X i , Y i , Z i represent the three-dimensional coordinate values of the point cloud data.

[0059] After quadratic surface fitting, the difference between the fitted elevation value and the true elevation of ground points is small, and their height differences should show a normal distribution; while the differences of non-ground points are large, disturbing the normal distribution of the height differences of ground points. Therefore, according to the stratification phenomenon of ground points and feature points, a clustering algorithm is used to separate the height difference data of the two, and then the normal distribution characteristics are used to determine the height difference threshold for surface fitting. Considering that in some areas, feature points and ground points are relatively close, such as low vegetation, and their stratification phenomenon is not obvious. In this case, directly calculate the mean value μ2 and the standard deviation σ2 of the height differences, and remove the points with height differences greater than μ2 + σ2 as non-ground points. The normal distribution formula is as follows:

[0060]

[0061] In the above formula, N is the number of points in the point cloud, h i is the elevation value of each point, x is a random variable, μ is the mean value, and σ is the standard deviation.

[0062] Combined with the point cloud feature information in step 2, perform multi-scale adaptive filtering, and use the height difference between the point to be classified and the corresponding grid of the ground reference plane as a criterion to identify ground points. When the height difference of the point to be classified is less than a certain threshold, mark the point as a ground point

[0063] S5. For the filtered ground point cloud, use the region growing algorithm that combines normal vector angle, curvature, and color similarity to divide the urban terrain into flat surfaces, natural undulations, and man-made undulation regions;

[0064] First, set the constraint conditions of the growth criterion as the color, normal vector, and curvature attributes of the point cloud to determine whether adjacent points can belong to the same region; then, calculate the normal vector normal and the curvature curvatures, and sort them in ascending order of curvature. Start growing from the point with the smallest curvature, select the point with the smallest curvature as the initial seed point, compare the neighboring points around the seed point with the seed point, and then judge whether to add them to the same region according to the growth criterion; calculate the color components of the points, calculate the color similarity, and divide the points with high similarity into the same region; at the same time, set the normal vector angle threshold, search for the neighboring points of the current seed point, calculate the angle between the normal vector of the neighboring point and the normal vector of the current seed point, and add the neighboring points with an angle less than the threshold to the current region; set the curvature threshold, check the curvature of each neighboring point, add the neighboring points with a curvature less than the curvature threshold to the seed point sequence, and delete the current seed point, and continue to grow with the new seed point. Repeat the above steps to continuously expand the current region until no new points meet the growth conditions. At this time, the point cloud region segmentation is completed.

[0065] In the point cloud region growing method, the normal vector angle is an important judgment basis. For two points p i and p j , their normal vectors are ni and n j Then, the included angle θ between them can be calculated by the following formula:

[0066] θ = arccos(n i · n j / (‖n i ‖ ‖n j ‖))

[0067] n i · n j / (‖n i ‖ ‖n j ‖)| > cos(θ_threshold)

[0068] k i < k_threshold

[0069] Wherein, n i · n j represents the dot product of the normal vectors n i and n j ; ||n i || and ||n j || respectively represent the magnitudes of the normal vectors n i and n j . θ_threshold is the normal vector included angle threshold, k_threshold is the curvature threshold, and k i is the curvature of the current point

[0070] S6. For different partitions, after upsampling and encrypting the point cloud by nearest neighbor interpolation, a regular grid DEM is constructed, and the terrain is fitted by using a Triangulated Irregular Network (TIN).

[0071] Perform regional segmentation on the urban ground point cloud obtained in the fourth step. In this example, the urban point cloud model is divided according to the similarity of the regional characteristics of the point cloud, that is, the complex urban point cloud data is divided into regions such as flat areas, natural undulating terrains, and artificial undulating terrains.

[0072] As is known by common technical knowledge, the present invention can be implemented by other embodiments that do not depart from its spiritual essence or essential features. Therefore, the above-disclosed embodiments are illustrative in all aspects and not exclusive. All changes within the scope of the present invention or equivalent to the scope of the present invention are encompassed by the present invention.

Claims

1. A high-precision DEM construction method based on a point cloud model, characterized in that: The specific steps are as follows: S1. Divide the urban point cloud data into grids, dynamically adjust the grid size and neighborhood range based on the weighted slope value, and calculate the weighted slope value through the height difference and distance between the central grid and the lowest point of the neighborhood grid; S2. Adopt an improved DBSCAN clustering algorithm, combine the color, intensity and texture features of the point cloud to perform multi-feature fusion clustering on the urban point cloud, and distinguish features and ground points with similar elevations but different attributes; S3. Extract the feature normal of the point cloud based on principal component analysis (PCA), solve the normal direction through the covariance matrix of the neighborhood point set, and determine the normal orientation in combination with the view point constraint, and extract the feature line of the feature as the segmentation boundary between the ground and non-ground points; S4. Based on the constraint of the feature line of the feature in step 3, perform multi-scale adaptive filtering, use the ground seed points to construct a quadratic surface to fit the initial ground reference surface, dynamically adjust the height difference threshold through the normal distribution model, and remove non-ground points; S5. For the filtered ground point cloud, adopt a region growing algorithm that combines the normal angle, curvature and color similarity to perform zoning, and divide the urban terrain into flat straight surfaces, natural undulations and artificial undulation regions; S6. For different partitions, use the nearest neighbor interpolation upsampling to encrypt the point cloud and then construct a regular grid DEM, and use an irregular triangular network (TIN) to fit the terrain.

2. The high-precision DEM construction method based on a point cloud model according to claim 1, characterized in that: In the step S1, dividing the large-scale urban point cloud data set into grids can significantly improve the data processing efficiency while retaining the key spatial features; The size and neighborhood size of the grid are set according to the weighted slope value, and the calculation formula of the weighted slope value is: Among them, D i is the distance from the lowest point of the central grid to the lowest point of the i-th neighboring grid, and Z min is the minimum value of the point cloud coordinates.

3. A high-precision DEM construction method based on a point cloud model according to claim 1, characterized in that: In the step S2, after the grid division is completed, use the density-based DBSCAN clustering algorithm to iteratively perform point cloud clustering in combination with the point cloud features; Use the color information and intensity information as auxiliary features for point cloud clustering to improve the distinguishability of features and ground points with different colors and similar elevations.

4. A high-precision DEM construction method based on a point cloud model according to claim 1, characterized in that: In the step S3, the normal direction is solved by finding the eigenvectors and eigenvalues of the covariance matrix, and a view point constraint is introduced to ensure that the normal directions are consistent, satisfying: n i *(v p -p i )>0, where n i is the normal of point p i , v p is the preset view point coordinate, and the method of using the K-nearest neighbor classification algorithm to determine the neighborhood point set of a certain point is used to extract the feature line of the ground object.

5. A high-precision DEM construction method based on a point cloud model according to claim 1, characterized in that: In the step S4, first, use the lowest point in the ground cluster identified by the point cloud clustering result as the ground seed point to construct a reference surface for the quadratic fitting of the initial ground; After selecting enough seed points, fit the approximate terrain with these seed points. Then, while determining the height difference threshold of the surface fitting through the idea of normal distribution, perform multi-scale adaptive filtering, and use the height difference between the point to be classified and the corresponding grid of the ground reference surface as the criterion to identify the ground points. When the height difference of the point to be classified is less than a certain threshold, mark the point as a ground point.

6. A high-precision DEM construction method based on a point cloud model according to claim 1, characterized in that: In the step S5, based on the ground point cloud after point cloud filtering obtained in step S4, perform regional segmentation. First, the constraint conditions of the region growing algorithm include: the normal angle is less than the threshold θ_threshold, the curvature is lower than the threshold k_threshold, and the color similarity is higher than the set value, which are used to judge whether adjacent points can belong to the same region; Then, select a seed point as the starting point and mark it as visited; Then, starting from the seed point, traverse its neighboring points, and judge whether to add it to the same region according to the growth criterion. If the condition is met, mark the point as visited and add it to the current region; Repeat the above steps to continuously expand the current area until no new points meet the growth conditions, and complete the point cloud region segmentation.

7. A high-precision DEM construction method based on a point cloud model according to claim 1, characterized in that: In step S6, for the urban ground point clouds in different regions after segmentation in step S5, a method of upsampling the sparse point clouds is adopted to increase the point cloud density and details. On this basis, continuously continuous irregular triangular surfaces are constructed to approximate the real terrain.

Images