CN116824379A - Laser point cloud building contour progressive optimization method based on multidimensional features - Google Patents
Laser point cloud building contour progressive optimization method based on multidimensional features Download PDFInfo
- Publication number
- CN116824379A CN116824379A CN202310806742.2A CN202310806742A CN116824379A CN 116824379 A CN116824379 A CN 116824379A CN 202310806742 A CN202310806742 A CN 202310806742A CN 116824379 A CN116824379 A CN 116824379A
- Authority
- CN
- China
- Prior art keywords
- point
- point cloud
- points
- building
- clustering
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 96
- 238000005457 optimization Methods 0.000 title claims abstract description 21
- 230000000750 progressive effect Effects 0.000 title claims abstract description 17
- 238000000605 extraction Methods 0.000 claims abstract description 50
- 238000001914 filtration Methods 0.000 claims description 13
- 230000011218 segmentation Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 8
- 238000005096 rolling process Methods 0.000 claims description 4
- 230000008030 elimination Effects 0.000 claims description 3
- 238000003379 elimination reaction Methods 0.000 claims description 3
- 238000007670 refining Methods 0.000 claims description 3
- 238000011282 treatment Methods 0.000 claims description 3
- 238000012217 deletion Methods 0.000 abstract 1
- 230000037430 deletion Effects 0.000 abstract 1
- 238000005192 partition Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 11
- 238000004364 calculation method Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011284 combination treatment Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
Abstract
The invention discloses a laser point cloud building contour progressive optimization method based on multidimensional features, which adopts layer-by-layer optimization and coarse-to-fine building contour extraction, firstly partitions building point clouds into building vertical point clouds (including trees, building enclosing walls and the like) and top surface point clouds (including building top surfaces, trees and the like), then refines the building vertical point clouds and top surface point clouds by adopting different features according to the characteristics of the vertical point clouds and the top surface point clouds respectively, and rapidly and accurately identifies the building point clouds with different heights and different structural complexity, and can eliminate the interference of non-building point clouds and the influence of vertical point cloud deletion when extracting the contour point clouds in a refined manner, thereby improving the accuracy of contour extraction.
Description
Technical Field
The invention belongs to the technical field of laser radar data processing, and particularly relates to a laser point cloud building contour progressive optimization method based on multidimensional features.
Background
With the continuous development of new technologies such as smart city and digital twin, the rapid and accurate extraction of building outlines becomes a research hotspot. The airborne laser radar LiDAR technology can rapidly acquire high-precision three-dimensional point cloud data of the earth surface, can accurately acquire geometric topological relations among the earth features, is an economic and reliable earth surface three-dimensional space data acquisition technology, is an important means for acquiring an urban three-dimensional building model, and has a wide application prospect. Therefore, research on how to extract buildings from Lidar point cloud data rapidly and accurately has important practical significance.
Building contour extraction research based on original point clouds mainly focuses on how to quickly and accurately identify building point clouds with different heights and different structural complexity, and meanwhile interference of tree point clouds and the like is avoided.
At present, the contour extraction of a building is mostly carried out by only depending on one or a combination of a plurality of space geometric features such as height difference, projection density, normal vector, curvature and the like of the point cloud, and then clustering is carried out to obtain the point cloud of the top surface of the building, but the problems of missed detection of the low building, partial missing of the point cloud of the top surface or the vertical surface of the building, wrong separation of adjacent buildings and trees and the like easily exist in the extraction process. In addition, the point cloud information can be converted into an image, and then the contour extraction is carried out by utilizing an image processing algorithm, so that the problems that the grid resolution scale is difficult to determine, the boundary position extraction is inaccurate and the like are also existed. In summary, there are many problems to be solved in the current method for rapidly extracting the LOD1 (Levels of Detai l) level three-dimensional model of the building by using the original point cloud.
The Chinese patent with application number 202210672480.0 discloses an urban building attribute extraction method based on airborne laser point clouds, which mainly comprises the steps of calculating geometric features, height features and attribute features of each non-ground laser point respectively and marking the non-ground point clouds as building points and vegetation points point by utilizing a random forest classifier; then, giving a three-dimensional distance threshold value, a height distance threshold value and an intensity difference threshold value, and carrying out region growth on the laser points marked as the buildings to obtain a plurality of building points Yun Julei; and carrying out plane fitting on each building point cloud cluster to obtain building structure height and contour information of the building structure. According to the method, on one hand, the random forest classifier is used for marking the non-ground point cloud point by point, manual intervention is needed in the marking process, the processing efficiency is greatly reduced, and meanwhile, the factor of artificial interference is increased; on the other hand, the method obtains the complete point cloud of the building category, and under the requirement that only the LOD1 building outline needs to be extracted, the redundancy of data is probably caused, and the algorithm efficiency of subsequent clustering and outline extraction is reduced.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a laser point cloud building contour progressive optimization method based on multidimensional features, which aims to solve the problems of low data processing efficiency and poor extraction accuracy of the prior method.
In order to solve the technical problems, the invention adopts a technical scheme that: the laser point cloud building contour progressive optimization method based on the multidimensional features comprises the following steps:
(1) Separating the non-facade point cloud: filtering the original airborne LiDAR point cloud data, and separating the ground points from the non-ground points to obtain a point cloud of the non-ground points;
(2) Pretreatment: calculating the relative height of each point in the non-ground point cloud, and eliminating points with the relative height smaller than a set elevation threshold value;
(3) Rough segmentation: calculating the normal vector nz of each point in the non-ground point cloud after the elimination processing, and performing preliminary segmentation on the building vertical point cloud and the top surface point cloud by utilizing a normal vector threshold value, so as to realize the rough segmentation on the building vertical point cloud and the top surface point cloud;
(4) And (3) refining and extracting the vertical point cloud: performing two-dimensional projection of an XOY plane on the roughly divided building vertical point cloud, determining a projection range by taking a corresponding projection point as a circle center and R2 as a radius, calculating the projection density of each point, simultaneously calculating the height difference of the point cloud in the projection range of each point, thus obtaining the density and the height difference of each point, and performing refined extraction of the vertical point cloud by utilizing a density threshold value and a height difference threshold value to eliminate tree point cloud interference;
(5) Top point cloud clustering: when the building vertical point cloud is extracted finely, firstly filtering and denoising the top point cloud, and then clustering each point in the processed top point cloud to obtain a plurality of clustering objects;
(6) Fine extraction of top point cloud: calculating flatness of each point in the top point cloud and the projection area of the clustering objects of the top point cloud on the XOY plane, carrying out refined extraction of the top point cloud by utilizing a projection area threshold, a flatness threshold and a flatness point cloud number occupation ratio threshold, and removing the clustering objects which do not meet the conditions, so as to obtain the top point cloud of the building which does not contain other category interference;
(7) And (3) top point cloud filling: extracting boundary points of the refined top surface point cloud, searching building vertical surface points near the boundary points of the top surface point cloud, and merging the searched vertical surface points into the top surface point cloud, thereby completing the extraction of the point cloud of the top surface and part of the side surfaces of the complete building;
(8) Building contour extraction: and (3) performing two-dimensional projection of an XOY plane on the point cloud obtained in the previous step, extracting and merging boundary points by using an alpha-shape method, performing linear fitting and regularization and merging treatment by using a RANSAC method, and finally obtaining the two-dimensional contour of the building.
In another embodiment of the present invention, the calculated radius selected in the step (3) when calculating the normal vector of each point is R1, the value of R1 is preferably 5.2m, the normal vector threshold includes a high limit value and a low limit value, wherein the low limit value is preferably-0.3, the high limit value is preferably 0.3, that is, the point where the normal vector meets-0.3 < nz <0.3 is a vertical point, and otherwise, is a top point.
In another embodiment of the present invention, the calculating of the projection density of each point in the step (4) is performed by a method of building a KD-tree, and the value of R2 is preferably 0.1m; the projection density is the number of points in the projection range, the density threshold is preferably 3, the height difference threshold is preferably 5m, and the points meeting the projection density of >3 and the height difference of >5m are the vertical points meeting the conditions, so as to be reserved; the points which do not meet the above conditions are interference points and are removed.
In another embodiment of the invention, the KD-tree method is as follows:
calculating the variance of each dimension in the k-dimensional data set, and selecting the dimension k with the largest variance from the variances:
II, arranging the data in the dimension k from small to large to obtain a data setWherein->N k For the number of data in dimension k, the median m in dimension k is calculated:
III, setting the threshold value as the median value m obtained in the previous step to obtain two sets K sub_low And K sub_high And creating a tree node for storage, the set satisfying the requirement of the following formula (3):
IV, repeating the steps until all subsets can not be divided any more; if a subset can not be divided any more, the data in the subset is stored in the leaf node without the child node.
In another embodiment of the present invention, the top point cloud clustering method in step (5) adopts an euro-type clustering or region growing clustering method.
In another embodiment of the present invention, the method of European clustering is as follows: for a certain point P in space, k points closest to the P point are found through a KD tree neighbor search method, the points with the distance smaller than a set threshold value are clustered into a set Q ', and if the number of elements in the set Q' is not increased any more, the whole clustering process is finished; otherwise, the above process should be repeated by selecting points other than the point P in the set Q 'until the number of elements in Q' is no longer increased.
In another embodiment of the present invention, the projection area threshold in the step (6) is preferably 30 square meters, the flatness threshold is preferably 0.5, the number of the flatness point clouds is 30% of the threshold, that is, the objects satisfying the projection area >30 square meters and (the point number ratio of the flatness > 0.5) >30% are the clustering objects meeting the condition, and the clustering objects are reserved; objects which do not meet the conditions are interference clustering objects, and the objects are eliminated.
In another embodiment of the present invention, the step (7) uses an alpha-shape method when extracting boundary points of the top point cloud; the building elevation point search is realized by establishing a KD tree in the refined elevation point cloud and searching building elevation points near the boundary point of the top elevation point cloud by taking 1m as a radius.
In another embodiment of the present invention, the method for boundary point extraction using alpha-shape is as follows:
(1) for any point p (x, y), rolling the circle radius alpha, and searching a point set Q of all points within 2 alpha from the point p in the point cloud;
selecting a point p in Q 1 (x 1 ，y 1 ) Calculating the center coordinates according to the coordinates of the two points and the radius alpha, wherein the center p 2 (x 2 ，y 2 ) And p 3 (x 3 ，y 3 ) To pass through p and p 1 The coordinates of the centers of circles of two points and with the radius alpha are calculated respectively, and the coordinate formulas are as follows:
wherein the method comprises the steps ofS 2 ＝(x-x 1 ) 2 +(y-y 1 ) 2
(2) P removal in Q-Point concentration 1 After the point, calculate the remaining point to p 2 And p 3 Distance of the points. If the distances from all points to the two points are larger than the radius alpha, indicating that p is a boundary point; if the rest points to p 2 Or p 3 If the point distance is not all greater than alpha, traversing all points in the point set Q to be p points in turn, if a certain point meets the two conditions, indicating the point as a boundary point, ending the judgment of the point, judging the next point, and if all the adjacent points in Q do not have p 1 Such points indicate that p is a non-boundary point.
In another embodiment of the present invention, the method of straight line fitting using the RANSAC method is as follows:
a. randomly selecting two points, and determining a straight line l by the two points;
b. setting a distance threshold t, determining a data set S (l) with the geometric distance smaller than t from a straight line l, and calling the data set S (l) as a consistent set of the straight line l;
c. repeating the random selection for several times to obtain a straight line l 1 ，l 2 ，...，l n And corresponding consistent set S (l 1 )，S(l 2 )，…，S(l n )；
d. The geometric distance is used to find the best fit line for the largest consistent set as the best fit line for the data points.
The beneficial effects of the invention are as follows: the laser point cloud building contour progressive optimization method based on the multidimensional features adopts layer-by-layer optimization and coarse-to-fine building contour extraction, firstly, building point clouds are divided into building vertical point clouds (including trees, building enclosing walls and the like) and top surface point clouds (including building top surfaces, trees and the like), then, according to the characteristics of the vertical point clouds and the top surface point clouds, different features are adopted to refine the building point clouds, building point clouds with different heights and different structural complexity are rapidly and accurately identified, and interference of non-building point clouds and influence of missing of the vertical point clouds can be eliminated while the contour point clouds are extracted in a refined mode, so that the contour extraction efficiency and accuracy are greatly improved.
In addition, the optimization method combines the characteristics of high projection density and large height difference of the building vertical point cloud on the XOY plane to improve the identification precision of the building vertical point cloud.
And secondly, introducing flatness duty ratio in the clustering process of the building top point cloud to judge the class of the clustered objects, so that the recognition accuracy of the building top point cloud is improved.
Drawings
FIG. 1 is a flow chart of the laser point cloud building contour progressive optimization method based on multidimensional features of the present invention;
FIG. 2 is a flow chart of a rough segmentation of a building elevation point cloud and a roof point cloud of the present invention;
FIG. 3 is a flow chart of the refinement and contour extraction of the elevation and roof point clouds of the building of the present invention;
FIG. 4 is a European clustering flow chart of an embodiment of the invention;
FIG. 5 is a schematic diagram of boundary point extraction of an alpha-shape method according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a line fitting process of the RANSAC method according to an embodiment of the invention;
FIG. 7 is a diagram showing the intensity of an original point cloud according to an embodiment of the present invention;
FIG. 8 is a normal vector calculation result height map of a point cloud according to an embodiment of the present invention;
FIG. 9 is a elevation view of a building elevation point cloud after rough segmentation in accordance with an embodiment of the present invention;
FIG. 10 is a top point cloud elevation view of a roughly divided building according to an embodiment of the present invention;
FIG. 11 is a graph of refined extraction results of a building elevation point cloud in accordance with an embodiment of the present invention;
FIG. 12 is a graph of refined extraction results of point clouds on the top surface of a building according to an embodiment of the invention;
FIG. 13 is a graph showing the combined results of building finishing vertical points and roof points in accordance with an embodiment of the present invention;
FIG. 14 is a graph of extraction results of rectangular complex building contours based on Alpha Shapes and RANSAC in accordance with an embodiment of the present invention;
FIG. 15 is a graph of circular building contour extraction results based on Alpha Shapes and RANSAC in accordance with an embodiment of the invention;
FIG. 16 is a graph of complex structure building contour extraction results based on Alpha Shapes and RANSAC in accordance with an embodiment of the present invention.
Detailed Description
In order that the invention may be readily understood, a more particular description thereof will be rendered by reference to specific embodiments that are illustrated in the appended drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and/or" as used in this specification includes any and all combinations of one or more of the associated listed items.
As shown in fig. 1, the method for gradually optimizing the outline of the laser point cloud building based on the multidimensional features of the invention comprises the following steps:
(1) Separating the non-facade point cloud: filtering the original airborne LiDAR point cloud data, and separating the ground points from the non-ground points to obtain a point cloud of the non-ground points;
(2) Pretreatment: calculating the relative height of each point in the non-ground point cloud, and eliminating points with the relative height smaller than a set elevation threshold value;
(3) Rough segmentation: calculating the normal vector nz of each point in the non-ground point cloud after the elimination processing, and performing preliminary segmentation on the building vertical point cloud and the top surface point cloud by utilizing a normal vector threshold value, so as to realize the rough segmentation on the building vertical point cloud and the top surface point cloud;
(4) And (3) refining and extracting the vertical point cloud: performing two-dimensional projection of an XOY plane on the roughly divided building vertical point cloud, determining a projection range by taking a corresponding projection point as a circle center and R2 as a radius, calculating the projection density of each point, simultaneously calculating the height difference of the point cloud in the projection range of each point, thus obtaining the density and the height difference of each point, and performing refined extraction of the vertical point cloud by utilizing a density threshold value and a height difference threshold value to eliminate tree point cloud interference;
(5) Top point cloud clustering: when the building vertical point cloud is extracted finely, firstly filtering and denoising the top point cloud, and then clustering each point in the processed top point cloud to obtain a plurality of clustering objects;
(6) Fine extraction of top point cloud: calculating flatness of each point in the top point cloud and the projection area of the clustering objects of the top point cloud on the XOY plane, and carrying out refined extraction of the top point cloud by utilizing a projection area threshold, a flatness threshold and a flatness point cloud number occupation ratio threshold, so as to obtain the top point cloud of the building which does not contain other types of interference;
(7) And (3) top point cloud filling: extracting boundary points of the refined top surface point cloud, searching building vertical surface points near the boundary points of the top surface point cloud, and merging the searched vertical surface points into the top surface point cloud, thereby completing the extraction of the point cloud of the top surface and part of the side surfaces of the complete building;
(8) Building contour extraction: and (3) performing two-dimensional projection of an XOY plane on the point cloud obtained in the previous step, extracting and merging boundary points by using an alpha-shape method, performing linear fitting and regularization and merging treatment by using a random sampling consistency method RANSAC, and finally obtaining the two-dimensional contour of the building.
The optimization method mainly comprises two parts, namely, firstly, roughly dividing the vertical point cloud and the top point cloud of the building, wherein a specific flow chart is shown in fig. 2, and then, finely processing and extracting contours of the roughly divided top point cloud and the roughly divided top point cloud of the building, and a specific flow chart is shown in fig. 3.
In the step (1), filtering the original airborne LiDAR point cloud data is to separate the ground point and the non-ground point from the point cloud, which is an important premise of the building extraction step. The filtering method is various, such as mathematical morphology filtering, iterative encryption triangular filtering, multi-scale virtual grid filtering, etc., and the cloth filtering method is preferred in this embodiment. Fig. 7 is a diagram showing the intensity of the original point cloud according to the embodiment of the present invention, which includes other types of interference factors such as trees, dense vegetation, and the like, besides buildings.
In the step (2), preprocessing is carried out on the basis of the separated non-ground point cloud, and some short features are removed. Generally, vegetation, short shrubs, and ground short protrusions (such as automobiles, people, etc.) are not too tall, gao Chengda is below 3m, and the building is relatively tall, so the height threshold is set to 3m, and the short ground points can be effectively removed.
After the low ground feature points are removed, the ground feature points mainly comprise buildings, some tall trees and the like. In order to distinguish wall points and top surface points of a refined building, improve contour extraction accuracy and avoid the defect of the wall points, the invention divides the building point cloud into a building vertical point cloud (comprising trees, building enclosing walls and the like) and a top surface point cloud (comprising building top surfaces, trees and the like), and then adopts different characteristics to refine the building vertical point cloud and the top surface point cloud according to the respective characteristics of the vertical point cloud and the top surface point cloud, thereby improving extraction accuracy.
In the step (3), the pre-processed non-ground point cloud is subjected to rough segmentation, mainly by means of normal vector comparison, firstly, the normal vector nz of each point in the non-ground point cloud is calculated, and the calculated radius R1 is selected during calculation, wherein the value of R1 is preferably 5.2m. The normal vector threshold in this embodiment includes a high limit value and a low limit value, wherein the low limit value is preferably-0.3 and the high limit value is preferably 0.3. Comparing the calculated normal vector nz of each point with a set normal vector threshold, if the normal vector is in accordance with-0.3 < nz <0.3, the point is a vertical point, otherwise, the point is a top point. Fig. 8 is a normal vector calculation result height diagram of a building point cloud according to an embodiment of the present invention, fig. 9 is a rough division building vertical point cloud result height diagram, and fig. 10 is a rough division building top point cloud result height diagram, where a non-ground point cloud of a building is roughly divided into two parts, one part is a vertical point cloud (including trees, building fences, etc.), and the other part is a top point cloud (including building top and trees, etc.).
After roughly dividing the top surface point cloud and the vertical surface point cloud, the top surface point cloud and the vertical surface point cloud need to be extracted in a fine mode. In step (4), the fine extraction of the standing point cloud is performed, mainly by using the features of the projection density and the height difference to perform the judgment process, firstly, the two-dimensional projection of the XOY plane is performed on the building standing point cloud after coarse segmentation, the projection range is determined by taking the corresponding projection point as the center of a circle and taking R2 as the radius, the projection density of each point, that is, the number of statistical points in the radius range is calculated, and the value of R2 in this embodiment is preferably 0.1m. Meanwhile, calculating the height difference of the point cloud in the projection radius range of each point, namely counting the maximum value and the minimum value of the heights of all points in the projection radius range, and then calculating the difference value to obtain the height difference of the point. And after the density and the high difference value of each point are obtained, comparing by using a density threshold value and a high difference threshold value. In the embodiment, the density threshold is preferably 3, the height difference threshold is preferably 5m, and the points meeting the projection density of >3 and the height difference of >5m are the vertical points meeting the conditions, so as to be reserved; the points which do not meet the above conditions are interference points and are removed. Through the operation, the refined extraction of the stereoscopic point cloud is realized to eliminate the interference of the stereoscopic point cloud, and the refined extraction result diagram of the stereoscopic point cloud of the building is shown in fig. 11.
Furthermore, in order to improve the calculation efficiency, the invention adopts a KD tree method when calculating the projection density, wherein the KD tree is a binary tree, and k-dimensional data are stored in the tree. Constructing a KD-tree over a K-dimensional dataset represents a division of the K-dimensional space that the K-dimensional dataset forms, i.e., each node in the tree corresponds to a hyper-rectangular region of K-dimensions. The KD tree adopts the idea of divide-and-conquer, namely the whole space is divided into a plurality of small parts, which is a recursion process, the process of repeating root nodes on the data in the left subspace and the right subspace can obtain first-level child nodes, and meanwhile, the space and the data set are further subdivided, and the process is repeated until the space only contains one data point.
The KD-tree method is as follows:
calculating the variance of each dimension in the k-dimensional data set, and selecting the dimension k with the largest variance from the variances:
II, arranging the data in the dimension k from small to large to obtain a data setWherein->N k For the number of data in dimension k, the median m in dimension k is calculated:
III, setting the threshold value as the median value m obtained in the previous step to obtain two sets K sub_low And K sub_high And creating a tree node for storage, the set satisfying the followingRequirements of formula (3):
IV, repeating the steps until all subsets can not be divided any more; if a subset can not be divided any more, the data in the subset is stored in the leaf node without the child node.
After the KD tree data structure is established, the adjacent points are generally searched in the child nodes and the father nodes, so that the extra calculation amount brought by searching the adjacent points can be greatly reduced, and the searching efficiency is improved.
The top surface point cloud can be processed simultaneously while the building vertical point cloud is extracted finely. In step (5), filtering and denoising are performed on the top point cloud, and then clustering is performed on each point in the processed top point cloud to obtain a plurality of clustered objects. The clustering method in the embodiment of the invention adopts European clustering or region growing clustering method.
The Euclidean clustering method is a clustering method based on Euclidean distance measurement. In three dimensions, points (x 1 ，y 1 ，z 1 ) And point (x) 2 ，y 2 ，z 2 ) The Euclidean distance between is defined as d E ，d E The calculation formula of (2) is as follows:the nearest neighbor query method based on KD tree is a precondition for accelerating the European clustering process, and as shown in FIG. 4, which is a European clustering flow chart in the embodiment of the invention, the method for European clustering is as follows: for a certain point P in space, k points closest to the P point are found through a KD tree neighbor search method, the points with the distance smaller than a set threshold value are clustered into a set Q ', and if the number of elements in the set Q' is not increased any more, the whole clustering process is finished; otherwise, the above process should be repeated by selecting points other than the point P in the set Q 'until the number of elements in Q' is no longer increased.
In addition, the region growing clustering method is a conventional method in the field, and the invention is not repeated.
In the step (6), after the top surface point cloud is clustered, the point cloud is extracted in a refined mode, the flatness of each point in the top surface point cloud and the projection area of the clustered objects of the top surface point cloud on the XOY surface are calculated, the projection area threshold is preferably 30%, the flatness threshold is preferably 0.5, and the number of the flatness point clouds is 30%. Comparing the calculated projection area and flatness of each clustered object with corresponding threshold values, and reserving the clustered objects which meet the conditions that the projection area is more than 30 square meters and the number of points with flatness of more than 0.5 is more than 30%; points which do not meet the conditions are interference clustering objects and are removed. And obtaining the top surface point cloud of the building which does not contain other types of interference through the refined extraction. Fig. 12 is a diagram of a result of fine extraction of point clouds on the top surface of a part of a building according to an embodiment of the present invention.
The flatness calculation of the point cloud is calculated by adopting covariance characteristics, and the calculation method is a conventional technology in the field, and is not repeated.
In addition, in the fine extraction of the top surface point cloud, besides the two features of the projection area and the flatness, the projection area can be replaced by the point features, namely the number of points of the clustering object, and the extraction can be judged through the two features of the points and the flatness, wherein the point threshold value is 200-300000.
After the interference points are set by the top point cloud, boundary point extraction can be performed, in step (7), boundary points of the refined top point cloud are extracted, then searching for building vertical points near the boundary points of the top point cloud, and merging the searched vertical points into the top point cloud, so that extraction of the point cloud of the top surface and part of the side surfaces of the complete building is completed, and a graph of merging results of the top surface points and the complete building after the vertical points are complemented is shown in fig. 13.
Preferably, when extracting boundary points of the top point cloud, an Alpha-shape method is utilized, and the Alpha Shapes method is a simple and effective method for rapidly extracting the boundary points, is not influenced by the shape of the boundary points of the point cloud, and can rapidly and accurately extract the boundary points, and the principle is as follows: for any shape of planar point cloud, if a circle with radius alpha is used, the plane point cloud rolls around the circle. If the rolling circle radius alpha is small enough, each point in the point cloud is a boundary point; if properly increased to a certain degree, the device only scrolls on the boundary points, and the track of the scrolling is a point cloud boundary. FIG. 5 is a schematic diagram showing boundary point extraction by the Alpha Shapes method according to the embodiment of the present invention, and the method for extracting boundary points by using Alpha-Shapes is as follows:
(1) for any point p (x, y), rolling the circle radius alpha, and searching a point set Q of all points within 2 alpha from the point p in the point cloud;
selecting a point p in Q 1 (x 1 ，y 1 ) Calculating the center coordinates according to the coordinates of the two points and the radius alpha, wherein the center p 2 9x 2 ，y 2 ) And p 3 (x 3 ，y 3 ) To pass through p and p 1 The coordinates of the centers of circles of two points and with the radius alpha are calculated respectively, and the coordinate formulas are as follows:
wherein the method comprises the steps ofS 2 ＝(x-x 1 ) 2 +(y-y 1 ) 2
(2) P removal in Q-Point concentration 1 After the point, calculate the remaining point to p 2 And p 3 Distance of the points. If the distances from all points to the two points are larger than the radius alpha, indicating that p is a boundary point; if the rest points to p 2 Or p 3 If the point distance is not all greater than alpha, traversing all points in the point set Q to be p points in turn, if a certain point meets the two conditions, indicating the point as a boundary point, ending the judgment of the point, judging the next point, and if all the adjacent points in Q do not have p 1 Such points indicate that p is a non-boundary point.
Further, in order to improve the searching efficiency, the searching of building elevation points is realized by establishing a KD tree in the refined elevation point cloud and searching building elevation points near the boundary point of the elevation point cloud by taking 1m as a radius.
In the step (8), after the top point cloud is supplemented, performing two-dimensional projection of an XOY plane on the supplemented point cloud, then extracting and combining boundary points by using an alpha-shape method, and then performing linear fitting and regularization and combination treatment by using a RANSAC method to finally obtain the two-dimensional contour of the building. Fig. 14-16 are graphs of contour extraction results of rectangular complex buildings, circular buildings and complex structure buildings based on Alpha Shapes and RANSAC according to the embodiment of the invention.
The RANSAC method is to randomly extract a sample set from a model by using an iterative method, search for more supporting interior points and better model parameters, check the extracted sample by using the residual set of the model, iterate for a certain number of times, and finally, when the probability that the selected sample set approaches to the comprehensive understanding is maximum, take the selected sample set as the sample set closest to the comprehensive understanding, and finally, support the correctness of the obtained parameter solution by using the residual set of the sample. The method for straight line fitting by using the RANSAC method is as follows:
a. randomly selecting two points, and determining a straight line l by the two points;
b. setting a distance threshold t, determining a data set S (l) with the geometric distance smaller than t from a straight line l, and calling the data set S (l) as a consistent set of the straight line l;
c. repeat several times random selectionObtaining a straight line l 1 ，l 2 ，...，l n And corresponding consistent set S (l 1 )，S(l 2 )，…，S(l n )；
d. The geometric distance is used to find the best fit line for the largest consistent set as the best fit line for the data points.
As shown in fig. 6, which is a schematic diagram of a linear fitting process of the RANSAC method according to the embodiment of the present invention, two points are randomly selected in a point set, linear parameters formed by the two points are solved, the distance between the rest points in the point set and the linear is calculated, the points with the distance smaller than a set distance threshold are taken as interior points, and the number of the interior points is counted; then randomly selecting two points, counting the number of the inner points, and the like; the point set with the largest number of inner points is the largest consistent set, and finally, the points in the largest consistent set are fitted into a straight line by least square.
The building contour progressive optimization method can quickly and accurately extract the building contour under the condition of effectively avoiding the interference of tree point clouds and the missing of elevation point clouds.
The foregoing description is only illustrative of the present invention and is not intended to limit the scope of the invention, and all equivalent structural changes made by the present invention and the accompanying drawings, or direct or indirect application in other related technical fields, are included in the scope of the present invention.
Claims (10)
1. A laser point cloud building contour progressive optimization method based on multidimensional features is characterized by comprising the following steps:
(1) Separating the non-facade point cloud: filtering the original airborne LiDAR point cloud data, and separating the ground points from the non-ground points to obtain a point cloud of the non-ground points;
(2) Pretreatment: calculating the relative height of each point in the non-ground point cloud, and eliminating points with the relative height smaller than a set elevation threshold value;
(3) Rough segmentation: calculating the normal vector nz of each point in the non-ground point cloud after the elimination processing, and performing preliminary segmentation on the building vertical point cloud and the top surface point cloud by utilizing a normal vector threshold value, so as to realize the rough segmentation on the building vertical point cloud and the top surface point cloud;
(4) And (3) refining and extracting the vertical point cloud: performing two-dimensional projection of an XOY plane on the roughly divided building vertical point cloud, determining a projection range by taking a corresponding projection point as a circle center and R2 as a radius, calculating the projection density of each point, simultaneously calculating the height difference of the point cloud in the projection range of each point, thus obtaining the density and the height difference of each point, and performing refined extraction of the vertical point cloud by utilizing a density threshold value and a height difference threshold value to eliminate tree point cloud interference;
(5) Top point cloud clustering: when the building vertical point cloud is extracted finely, firstly filtering and denoising the top point cloud, and then clustering each point in the processed top point cloud to obtain a plurality of clustering objects;
(6) Fine extraction of top point cloud: calculating flatness of each point in the top point cloud and the projection area of the clustering objects of the top point cloud on the XOY plane, carrying out refined extraction of the top point cloud by utilizing a projection area threshold, a flatness threshold and a flatness point cloud number occupation ratio threshold, and removing the clustering objects which do not meet the conditions, so as to obtain the top point cloud of the building which does not contain other category interference;
(7) And (3) top point cloud filling: extracting boundary points of the refined top surface point cloud, searching building vertical surface points near the boundary points of the top surface point cloud, and merging the searched vertical surface points into the top surface point cloud, thereby completing the extraction of the point cloud of the top surface and part of the side surfaces of the complete building;
(8) Building contour extraction: and (3) performing two-dimensional projection of an XOY plane on the point cloud obtained in the previous step, extracting and merging boundary points by using an alpha-shape method, performing linear fitting and regularization and merging treatment by using a RANSAC method, and finally obtaining the two-dimensional contour of the building.
2. The method for gradually optimizing the contour of a laser point cloud building based on the multidimensional features according to claim 1, wherein the calculated radius selected in the step (3) when the normal vector of each point is calculated is R1, the value of R1 is preferably 5.2m, the normal vector threshold comprises a high limit value and a low limit value, wherein the low limit value is preferably-0.3, the high limit value is preferably 0.3, namely, the point of the normal vector conforming to-0.3 < nz <0.3 is a vertical point, and otherwise, the point is a top point.
3. The method for progressively optimizing the contour of a building by using a laser point cloud based on multi-dimensional features according to claim 2, wherein the projection density of each point is calculated in the step (4) by establishing a KD-tree, and the value of R2 is preferably 0.1m; the projection density is the number of points in the projection range, the density threshold is preferably 3, the height difference threshold is preferably 5m, and the points meeting the projection density of >3 and the height difference of >5m are the vertical points meeting the conditions, so as to be reserved; the points which do not meet the above conditions are interference points and are removed.
4. A multi-dimensional feature-based laser point cloud building contour progressive optimization method according to claim 2 or 3, characterized in that the KD-tree method is as follows:
calculating the variance of each dimension in the k-dimensional data set, and selecting the dimension k with the largest variance from the variances:
II, arranging the data in the dimension k from small to large to obtain a data setWherein the method comprises the steps ofN k For the number of data in dimension k, the median m in dimension k is calculated:
III, setting the threshold value as the median value m obtained in the previous step to obtain two sets K sub_low And K sub_high And creating a tree node for storage, the set satisfying the requirement of the following formula (3):
IV, repeating the steps until all subsets can not be divided any more; if a subset can not be divided any more, the data in the subset is stored in the leaf node without the child node.
5. The multi-dimensional feature-based laser point cloud building contour progressive optimization method of claim 4, wherein the top point cloud clustering method of step (5) adopts an European clustering or region growing clustering method.
6. The multi-dimensional feature-based laser point cloud building contour progressive optimization method of claim 5, wherein the method of European clustering is as follows: for a certain point P in space, k points closest to the P point are found through a KD tree neighbor search method, the points with the distance smaller than a set threshold value are clustered into a set Q ', and if the number of elements in the set Q' is not increased any more, the whole clustering process is finished; otherwise, the above process should be repeated by selecting points other than the point P in the set Q 'until the number of elements in Q' is no longer increased.
7. The progressive optimization method of the laser point cloud building contour based on the multidimensional features according to claim 6, wherein the projection area threshold in the step (6) is preferably 30 square meters, the flatness threshold is preferably 0.5, the number of the flatness point clouds is 30% of the threshold, namely, the objects which simultaneously satisfy the projection area of >30 square meters and (the point number of which the flatness is >0.5 is the ratio of > 30%) are the clustering objects which meet the conditions, and the clustering objects are reserved; objects which do not meet the conditions are interference clustering objects, and the objects are eliminated.
8. The progressive optimization method of the laser point cloud building contour based on the multidimensional features of claim 7, wherein the step (7) is characterized in that an alpha-shape method is utilized when boundary points of the top point cloud are extracted; the building elevation point search is realized by establishing a KD tree in the refined elevation point cloud and searching building elevation points near the boundary point of the top elevation point cloud by taking 1m as a radius.
9. The multi-dimensional feature-based laser point cloud building contour progressive optimization method of claim 8, wherein the method for extracting boundary points by using alpha-shape is as follows:
(1) for any point p (x, y), rolling the circle radius alpha, and searching a point set Q of all points within 2 alpha from the point p in the point cloud;
selecting a point p in Q 1 (x 1 ，y 1 ) Calculating the center coordinates according to the coordinates of the two points and the radius alpha, wherein the center p 2 (x 2 ，y 2 ) And p 3 (x 3 ，y 3 ) To pass through p and p 1 The coordinates of the centers of circles of two points and with the radius alpha are calculated respectively, and the coordinate formulas are as follows:
wherein the method comprises the steps ofS 2 ＝(x-x 1 ) 2 +(y-y 1 ) 2
(2) P removal in Q-Point concentration 1 After the point, calculate the remaining point to p 2 And p 3 Distance of the points. If the distances from all points to the two points are larger than the radius alpha, indicating that p is a boundary point; if the rest points to p 2 Or p 3 If the point distances are not all greater than alpha, traversing all the point rotations in the point set Q as p 1 If a certain point meets the two conditions, indicating the point as a boundary point, ending the judgment of the point, judging the next point, and if all the adjacent points in Q do not have p 1 Such points indicate that p is a non-boundary point.
10. The multi-dimensional feature-based laser point cloud building contour progressive optimization method of claim 9, wherein the method of straight line fitting using RANSAC method is as follows:
a. randomly selecting two points, and determining a straight line l by the two points;
b. setting a distance threshold t, determining a data set S (l) with the geometric distance smaller than t from a straight line l, and calling the data set S (l) as a consistent set of the straight line l;
c. repeating the random selection for several times to obtain a straight line l 1 ，l 2 ，...，l n And corresponding consistent set S (l 1 )，S(l 2 )，...，S(l n )；
d. The geometric distance is used to find the best fit line for the largest consistent set as the best fit line for the data points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310806742.2A CN116824379A (en) | 2023-07-04 | 2023-07-04 | Laser point cloud building contour progressive optimization method based on multidimensional features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310806742.2A CN116824379A (en) | 2023-07-04 | 2023-07-04 | Laser point cloud building contour progressive optimization method based on multidimensional features |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116824379A true CN116824379A (en) | 2023-09-29 |
Family
ID=88121835
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310806742.2A Pending CN116824379A (en) | 2023-07-04 | 2023-07-04 | Laser point cloud building contour progressive optimization method based on multidimensional features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116824379A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117437364A (en) * | 2023-12-20 | 2024-01-23 | 深圳大学 | Method and device for extracting three-dimensional structure of building based on residual defect cloud data |
| CN117437364B (en) * | 2023-12-20 | 2024-04-26 | 深圳大学 | Method and device for extracting three-dimensional structure of building based on residual defect cloud data |
-
2023
- 2023-07-04 CN CN202310806742.2A patent/CN116824379A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117437364A (en) * | 2023-12-20 | 2024-01-23 | 深圳大学 | Method and device for extracting three-dimensional structure of building based on residual defect cloud data |
| CN117437364B (en) * | 2023-12-20 | 2024-04-26 | 深圳大学 | Method and device for extracting three-dimensional structure of building based on residual defect cloud data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112070769B (en) | Layered point cloud segmentation method based on DBSCAN | |
| CN106022381B (en) | Automatic extraction method of street lamp pole based on vehicle-mounted laser scanning point cloud | |
| CN106447676B (en) | A kind of image partition method based on fast density clustering algorithm | |
| CN111598823A (en) | Multi-source mobile measurement point cloud data air-ground integrated fusion method and storage medium | |
| CN110807781B (en) | Point cloud simplifying method for retaining details and boundary characteristics | |
| CN112883878B (en) | Automatic point cloud classification method under transformer substation scene based on three-dimensional grid | |
| CN112132840B (en) | Vehicle-mounted street tree point cloud classification and feature information extraction method | |
| CN113920360A (en) | Road point cloud rod extraction and multi-scale identification method | |
| CN114241217B (en) | Trunk point cloud efficient extraction method based on cylindrical features | |
| CN106874421A (en) | Image search method based on self adaptation rectangular window | |
| CN114387288A (en) | Single standing tree three-dimensional information extraction method based on vehicle-mounted laser radar point cloud data | |
| CN111783722B (en) | Lane line extraction method of laser point cloud and electronic equipment | |
| CN109102538B (en) | Method for extracting land characteristic points and land characteristic lines by using contour line data and weather map analysis method | |
| CN112396641A (en) | Point cloud global registration method based on congruent two-baseline matching | |
| CN116258857A (en) | Outdoor tree-oriented laser point cloud segmentation and extraction method | |
| CN115861247A (en) | High-resolution remote sensing image contour multistage regularization method, system and application | |
| CN115063555A (en) | Method for extracting vehicle-mounted LiDAR point cloud street tree growing in Gaussian distribution area | |
| CN114170149A (en) | Road geometric information extraction method based on laser point cloud | |
| CN107993242B (en) | Method for extracting boundary of missing area based on airborne LiDAR point cloud data | |
| CN112070787B (en) | Aviation three-dimensional point cloud plane segmentation method based on opponent reasoning theory | |
| CN110348478B (en) | Method for extracting trees in outdoor point cloud scene based on shape classification and combination | |
| CN115546372A (en) | Three-dimensional point cloud simplification method based on voxel filtering | |
| CN116824379A (en) | Laser point cloud building contour progressive optimization method based on multidimensional features | |
| CN114782715B (en) | Vein recognition method based on statistical information | |
| CN116051841A (en) | Roadside ground object multistage clustering segmentation algorithm based on vehicle-mounted LiDAR point cloud |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |