DEM (digital elevation model) refinement processing method based on pixel-level dense matching point cloudTechnical Field
The invention relates to the technical field of topography measurement, in particular to a DEM (digital elevation model) refinement processing method based on pixel-level dense matching point cloud.
Background
The Digital Elevation Model (DEM) is a digital reflection of the actual surface elevation fluctuation, and is widely used as basic data for geochemical analysis and modeling in the fields of geomorphic geology, watershed hydrology, digital twinning and the like.
The topography measurement is an important content of river course survey work, and relates to underwater topography measurement and land topography measurement, the traditional land topography measurement usually adopts RTK, total station and other equipment, the field acquires discrete characteristic points, the inner industry generates DLG by means of points, lines and surfaces, the DEM constructed by the method has lower definition degree, and the change of the ground and the landform cannot be reflected in detail; with the development of unmanned aerial vehicle aerial survey technology, aerial survey operation efficiency is high, data acquisition is digitalized, the airborne LIDAR technology is a main means for three-dimensional information acquisition, high-precision ground point cloud can be effectively acquired based on the characteristic of multiple echoes of the airborne LIDAR technology, high-precision DEM is obtained through point cloud classification and denoising, but the high-precision DEM cannot be widely applied to production due to high price, the low-altitude photogrammetry technology is widely applied to land topography measurement at present due to the ultrahigh cost performance of the low-altitude photogrammetry technology, but pixel-level dense matching point cloud generated based on an orthophoto contains a large number of noise points and redundancy, and the precision for generating the DEM through a traditional processing method is not high.
In the current research of image point clouds, the application of unmanned aerial vehicle image point clouds in large scale DEM data production is studied, the research of extracting refined point clouds by carrying out triangular mesh interpolation and constraint on the image point clouds is also studied, the research of effectively eliminating noise of the image point clouds by an iterative median filtering algorithm is also studied, the research of effectively filtering non-dense vegetation point clouds based on a cloth simulation algorithm is also studied, the DEM precision is effectively improved by an improved local maximum value algorithm on the filtering of the dense vegetation point clouds, and in combination with production practice, the DEM refined processing method based on orthographic image pixel-level dense matching point clouds is provided.
Disclosure of Invention
(One) solving the technical problems
Aiming at the defects of the prior art, the invention provides a DEM (digital elevation model) refinement processing method based on a pixel-level dense matching point cloud, which solves the problem that the precision of DEM generated by the traditional processing method is not high.
(II) technical scheme
In order to achieve the above purpose, the present invention provides the following technical solutions: a DEM (digital elevation model) refinement processing method based on pixel-level dense matching point cloud specifically comprises the following steps:
Step one, source data screening: after point cloud data are acquired by utilizing orthographic images and pixel-level dense matching, analyzing the point cloud data with different image proportions, and selecting proper point cloud data as source data;
Step two, classifying point clouds: after the source data in the first step are primarily screened, extracting through an orthographic image aiming at characteristic lines with consistent elevation values, uniformly assigning values, and acquiring the characteristic lines with uncertain elevation values by adopting a moving section window method and setting a window moving distance D;
Step three, classification limiting: estimating according to the actual point cloud density, setting the boundary range of the buffer area to be larger than the average distance of the point cloud, and reserving an error range between the elevation of the ground point and the elevation of the water edge;
Step four, denoising ground points: and (3) establishing an irregular triangular net by using the classified ground points, judging whether the water edge passes through the triangle according to the elevation value of the vertex of the initial triangle, acquiring the intersection points (xj,yj) of the water edge and two sides of the triangle, forming a new water edge by connecting the intersection points, and removing the ground points on the water surface by using the new water edge.
The invention is further provided with: the orthographic image in the first step is an orthographic image of the unmanned aerial vehicle, a collineation equation of a single image is used as a base of adjustment, the positions and the postures of the images are adjusted to enable light rays of the same-name points of the whole model to reach the most favorable intersection, and then the model is fused into a required coordinate system to obtain three-dimensional coordinates (X, Y, Z) of the ground points.
The invention is further provided with: the pixel-level dense matching is a process of identifying homonymous points between two or more images through pixel-by-pixel matching, wherein the determination of homonymous points is based on a matching measure, and rough difference points generated in the matching process are deleted, so that a pixel-level dense matching point cloud based on an orthographic image is obtained.
The invention is further provided with: in the first step, the different image proportions are 1/2 and 1/4 respectively.
The invention is further provided with: and in the second step, the primary screening of the source data comprises screening out low points and isolated points in the point cloud data.
The invention is further provided with: the characteristic line in the second step also comprises a ground feature with obvious characteristics, including but not limited to a water edge line and a ridge line.
The invention is further provided with: and in the second step, the orthographic image is extracted by extracting three-dimensional coordinates of the ground points in the source data.
The invention is further provided with: the calculation formula of xj,yj in the fourth step is as follows:
(III) beneficial effects
The invention provides a DEM (digital elevation model) refinement processing method based on a pixel-level dense matching point cloud. The beneficial effects are as follows:
(1) According to the DEM refining processing method based on the pixel-level dense matching point cloud, the DEM is refined by starting from the selection and classification of the point cloud and the denoising of the ground point, the influence of the image proportion on the quantity and the quality of the point cloud is eliminated, the effect of classifying the point cloud corresponding to a buffer area is enhanced by using the characteristic line, and then the water-side line is quickly searched by using the irregular triangular net to reject the ground point on the water surface, so that the DEM refining degree is quickly and accurately improved.
(2) According to the DEM refinement processing method based on the pixel-level dense matching point clouds, classification influence is eliminated through the relation between different image proportions and the total point clouds, meanwhile, selection of the pixel-level dense matching point clouds is effectively guided, and in a point cloud classification link, the effect of classification of ground points is effectively improved by extracting the characteristic lines of a buffer area, so that the precision of the DEM is improved.
(3) According to the DEM refining processing method based on the pixel-level dense matching point cloud, the new water edge line is quickly searched by means of establishing an irregular triangular net aiming at the denoising of the ground points, and the new water edge line is utilized for denoising the ground points on the water surface, so that the working efficiency of the point cloud refining is improved.
Drawings
FIG. 1 is a schematic diagram of a method for rapidly searching water edges by using an irregular triangular network according to the present invention;
FIG. 2 is a schematic diagram of a moving section window method of the present invention;
FIG. 3 is a schematic diagram of the classification of ground points based on buffer analysis feature lines according to the present invention;
FIG. 4 is an orthographic view of a region in an embodiment of the invention;
FIG. 5 is a chart of 1/2 image proportional point cloud classification statistics in an embodiment of the present invention;
FIG. 6 is a 1/4 image proportional point cloud classification statistical chart in an embodiment of the invention;
FIG. 7 is a 1/2 image scale water edge ground point classification diagram in accordance with an embodiment of the present invention;
FIG. 8 is a statistical chart of the classification accuracy of the point cloud in an embodiment of the invention;
FIG. 9 is a classification chart of step ground points in an embodiment of the invention;
FIG. 10 is a diagram of a classification of measured step ground points according to the present invention;
FIG. 11 is a diagram showing classification of measured water edge ground points according to the present invention;
FIG. 12 is a chart of statistical table of measured point cloud classification according to the present invention;
FIG. 13 is a chart of statistics of the classification accuracy of the measured point cloud according to the present invention;
FIG. 14 is a schematic view of a surface point of a water surface without culling in an embodiment of the present invention;
FIG. 15 is a schematic view of ground points on the water surface after being removed in an embodiment of the present invention;
Fig. 16 is a statistical chart of point clouds after the water surface and ground points are removed in the embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1-16, the embodiment of the present invention provides a technical solution: a DEM (digital elevation model) refinement processing method based on pixel-level dense matching point cloud specifically comprises the following steps:
Step one, source data screening: after point cloud data are acquired by utilizing orthographic images and pixel-level dense matching, analyzing the point cloud data with the image proportion of 1/2 and 1/4, and selecting proper point cloud data as source data;
Step two, classifying point clouds: screening out low points and isolated points in the source data in the first step, extracting through an orthographic image aiming at characteristic lines with consistent elevation values, uniformly assigning values, and acquiring the characteristic lines with uncertain elevation values by adopting a moving section window method and setting a window moving distance D;
Step three, classification limiting: estimating according to the actual point cloud density, setting the boundary range of the buffer area to be larger than the average distance of the point cloud, and reserving an error range between the elevation of the ground point and the elevation of the water edge;
Step four, denoising ground points: an irregular triangular network is established by using the classified ground points, as shown in fig. 1, according to the elevation value of the initial triangle vertex, whether the water edge line passes through the triangle is judged, and the intersection point of the water edge line and two sides of the triangle is obtained (the calculation formula of xj,yj),xj,yj is as follows:
the connecting line of the intersection points forms a new water edge line, and the new water edge line is utilized to remove the ground points on the water surface.
As a detailed description, the conventional ground point classification is to create an irregular triangle network (TIN) by continuously searching for the lowest point circulation to approach the real ground surface, and is particularly sensitive to the low points, so that the removal of the low points and the isolated points is the basis of classification, and the conventional method starts to classify the ground points after the low points and the isolated points are removed, but the method has poor effect on the characteristic obvious areas such as a sharp ridge, a slope, a water edge and the like, and the method for classifying the ground points based on the characteristic line of the buffer area analysis is provided.
As shown in fig. 2 and fig. 3, the default point is a point cloud for removing low points and isolated points, the characteristic lines can be features with obvious characteristics, such as water edges, ridge lines and the like, the characteristic lines with consistent elevation values are extracted through DOM, assigned uniformly, the characteristic lines with uncertain elevation values are obtained by adopting a moving section window method, as shown in fig. 2, the characteristic lines are obtained by setting a window moving distance D, and the fineness of the characteristic lines is determined by the value of the D.
Embodiment 1,
In order to verify the effectiveness of the DEM refinement treatment method, a certain area downstream of a Danjiang dam is selected for analysis, the area of the area is about 0.3 square kilometer, the east side is close to Han river, the ground feature of the area is rich, as shown in fig. 4, 4 image control points are uniformly distributed in the area, a DJI M RTK unmanned aerial vehicle is adopted to carry ZenmuseP1 camera, GNSS real-time differential positioning is adopted for aerial survey, the focal length of the camera is 35mm, and the resolution is 8192 x 5460; heading and side overlap were set to 80% and 60%, respectively, with a altitude of 200 meters, and 312 photographs were taken.
And (3) selecting a point cloud: image data acquired by unmanned aerial vehicle field industry are resolved, image proportions are set to be 1/2,1/4, point density and the like are kept unchanged in a point cloud generation link, las point cloud data based on different image proportions are generated, then point cloud classification is carried out, and classification statistical information is shown in fig. 5 and fig. 6.
As can be seen from fig. 5 and fig. 6, the total point cloud, the building points and the high vegetation points generated by the 1/2 image proportion are about 4 times of the 1/4 image proportion, and the classified ground points are more than 2 times of the 1/4 image proportion, which indicates that the image proportion plays an important role in generating the point cloud.
According to 304 real-time RTK points on site, the real-time RTK points respectively comprise a bare dew point, a water edge point, a ridge line point and a vegetation coverage point, various types of comparison measurement errors are shown in an attached figure 8, the errors in the total point cloud of 1/2 image proportion are larger than the errors in the point cloud of 1/4 image proportion, the analysis shows that the errors in the classification process of the point cloud of 1/2 image proportion are overlarge because part of water edge areas do not obtain ground points, and the accuracy of other classification points based on 1/2 image proportion is obviously higher than that of the classification points of 1/4 image proportion, but the accuracy of the errors in the vegetation coverage area and the ridge line area is generally lower.
And (3) classifying point clouds: the traditional classification method starts classifying the ground points after the low points and the isolated points are removed, but the method has poor effect on the characteristic obvious areas such as water edges, sharp ridges, slopes and the like, and as shown in fig. 7 and 9, the method for classifying the ground points based on the characteristic line of the buffer area analysis is provided, and the ground points are classified based on the method, wherein the parameters are as follows: the moving distance of the section window is set to be 0.2 m, the average distance between the points is about 8.5cm due to the fact that the point density is 138.649 per square meter, the boundary of the buffer area is set to be 10cm, the height error range is +/-10 cm, the classification effect is as shown in fig. 10 and 11, and point clouds generated by 1/2 image proportion are selected as source data of point cloud classification.
It is obvious from fig. 10 and 11 that the ground points at the feature line are effectively classified, the statistical information after classification is shown in fig. 12, and the statistical information also shows that the number of the ground points, the construction points and the vegetation coverage points is greatly changed, because the feature line ground points are classified based on the new method, the feature line ground points effectively participate in the classification of the whole ground points, the classification precision is shown in fig. 13, and the point cloud precision after classification by the new method is effectively improved.
Denoising ground points: after the classification of the point cloud is finished, noise points and misplaced points still exist, as can be seen from fig. 12, the minimum elevation value of the ground points is 33.47 m, which is far lower than the elevation of the water surface by 87.15 m, more erroneous point cloud data can be generated because the water surface is a weak texture area, the water edge line is quickly determined by adopting a method of searching the water edge line based on an irregular triangular net, the water surface point cloud data is removed, after the water surface noise points are removed, the land rough difference points are removed by adopting visual inspection, finally, the basic removal of the underwater noise points can be obviously found by comparing fig. 14 with fig. 15, and as shown in fig. 16, the minimum elevation value of the ground points is 80.09 m, which also shows that the noise points are effectively removed.
In summary, in combination with production practice, the method for processing the DEM based on the pixel-level dense matching point cloud is provided from the selection, classification and surface point denoising of the point cloud, the influence of the image proportion on the quantity and quality of the point cloud is analyzed based on the pixel-level dense matching point cloud of a certain area downstream of a Danjiang mouth dam, the effect of the characteristic line surface point extraction method based on buffer analysis on the point cloud classification is verified, the effect of rapidly searching water edges to remove the surface points based on an irregular triangle network is provided, and the result shows that the processing method rapidly and accurately improves the DEM refinement degree.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.