Building feature line extraction method based on dense point cloud

Abstract

The invention provides a dense point cloud-based building feature line extraction method, and belongs to the technical field of building feature line extraction, and the method comprises the following steps: carrying out the data preprocessing of point cloud data, carrying out the edge detection and corner detection of the preprocessed point cloud data, and carrying out the linear segment detection and curve segment detection of feature detection data, and performing line segment connection and removal processing on the line segment detection data, establishing a mapping relation table of optimized data, performing three-dimensional data model reconstruction on the building, and generating a three-dimensional model of the building. The cloud point data is added possibly at the beginning, errors and defects caused by environmental influence can be avoided, raindrop data are richer and more perfect, then the cloud point data are denoised, summary recognition of an initial model is achieved, irrelevant points are removed in the later period, follow-up data processing is more accurate, and the accuracy of raindrop data processing is improved. And through edge and curve identification, the boundary of the model can be better planned, and the subsequent processing efficiency is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of building feature line extraction, and in particular to a building feature line extraction method based on dense point cloud. Background Art

[0002] Buildings are prominent features in urban areas, and extracting them from scenes has long been a research focus and hotspot for scholars both domestically and internationally. Oblique aerial imagery offers the advantage of visible, unobstructed facades, providing information not only on building rooftops but also on building facades. Therefore, research on extracting buildings using oblique aerial imagery has significant research value and social benefits.

[0003] The Chinese patent with the announcement number CN111652241B discloses a building contour extraction method that fuses image features and dense matching point cloud features. The building contour extraction method includes: step 1, generating a three-dimensional dense image matching (DIM) point cloud and an orthophoto from an aerial image; step 2, fusing the DIM point cloud and the aerial image to detect the mask of a single building; step 3, detecting rough building contour lines based on line matching; step 4, regularizing the building boundary by fusing the matching lines and the contour lines of the rough building mask. This method only extracts from the acquired cloud point data, but since the point cloud data is directly obtained from horizontal photography, the point cloud data is often not perfect due to some occlusion or image pixel influence when the horizontal image is acquired, and is not accurate enough in the subsequent feature extraction and building restoration. Therefore, it is necessary to design a building feature line extraction method based on dense point clouds. Summary of the Invention

[0004] The purpose of the present invention is to provide a building feature line extraction method based on dense point cloud to solve the technical problem that the existing building feature line extraction and building three-dimensional restoration data are not accurate enough.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A method for extracting building feature lines based on dense point clouds, comprising the following steps:

[0007] Step 1: Preprocess the point cloud data, including point cloud data denoising and point cloud data segmentation, to obtain preprocessed data;

[0008] Step 2: Perform edge detection and corner detection on the pre-processed point cloud data to obtain feature detection data;

[0009] Step 3: Perform straight line segment detection and curve segment detection on the feature detection data to obtain line segment detection data;

[0010] Step 4: Perform line segment connection and removal processing on the line segment detection data to obtain optimized data;

[0011] Step 5: Create a mapping table for optimized data and create an array to put data features of the same type into the same array;

[0012] Step 6: Reconstruct the three-dimensional data model of the building to generate a three-dimensional model of the building.

[0013] Furthermore, in step 1, before preprocessing the point cloud data, the point cloud data is supplemented and added, and the point cloud data in the image is density detected, and then the general image recognition is used to preliminarily predict the preliminary imaging summary points in the image of the point cloud data, and then the cloud point data is added to the attachment of the imaging summary points to obtain the gain point cloud data, which makes the point cloud data of the image more complete and the restored imaging model more accurate.

[0014] Furthermore, the specific process of point cloud data denoising in step 1 is as follows: input gain point cloud data, and mark the image structure model of the building at each perspective and the plane structure model of the objects around the building, set the threshold for determining inliers, the maximum number of iterations, the minimum number of data points required for fitting the model, and the threshold for the number of inliers, randomly obtain several points from the gain point cloud data, and randomly fit a model to the obtained points. For each point in the gain point cloud data, calculate the distance from each point to the fitting model. If the distance is less than the threshold for determining inliers, mark the point as an inlier, otherwise mark it as an outlier, and then set a loop of for iin range(N), where N is the number of loops, that is, the number of fitted models. When the number of inliers found in the current iteration is greater than the threshold for the number of inliers, update the best model and record the current set of inliers. Repeat N times to find several models that meet the requirements, that is, when the number of inliers reaches the threshold for the number of inliers, all remaining points outside the model are deleted.

[0015] Furthermore, the specific process of point cloud data segmentation in step 1 is as follows: select one or several seed points from the denoised point cloud, define the criteria for neighboring points to be added to the current area, the criteria are the similarity of curvature, normal vector and distance, and starting from the seed point, gradually add neighboring points that meet the growth criteria to the current area. When there are no more neighboring points that meet the criteria to be added, stop expanding and select new seed points until all points are assigned to a certain area.

[0016] Furthermore, the specific process of step 2 is: calculate the curvature of each point in the point cloud, compare the curvatures between neighboring points, and if the change in curvature exceeds a preset first threshold, the point is considered to be an edge point; if the change in curvature between edge points exceeds a preset second threshold, the point is considered to be a corner point.

[0017] Furthermore, the specific process of step 3 is: setting the parameters of the Hough transform, distance resolution rho, angular resolution theta, threshold threshold, minimum line segment length minLineLength and maximum line segment gap maxLineGap, using the cv2.HoughLinesP function to detect straight line segments on the edge image, parsing the straight line segment information returned by the Hough transform, and drawing the straight line segments, and then analyzing and clustering the straight line segments, converting the pixel value of the image from 255 to 1, extracting the outer contour of the target, and then gradually eroding the contour until only the skeleton remains, and then converting the pixel value of the skeleton image from 1 back to 255, removing isolated points, and generating several isolated points during the skeleton extraction process. Isolated points are not part of the skeleton, and morphological opening operations are used to remove these isolated points, using the median filter method to smooth the skeleton, and connecting the breakpoints of the skeleton by the breakpoint extension method, using the cv2.findContours function to detect the contours in the skeleton image, and summarizing the contours as the curves in the skeleton.

[0018] Furthermore, the specific process of step 5 is as follows: establishing mapping relationship rules, including that lines cannot overlap and must be connected, performing feature classification, including wall lines, roof lines, and door and window lines, adding additional attribute fields to feature lines to store classification information, performing feature classification, checking the accuracy of the classification, and performing manual adjustments when errors are found.

[0019] Furthermore, the specific process of step 6 is as follows: first, polygon construction is performed, and characteristic lines connected by mapping and topology are used to form a number of closed polygons. The closed polygons represent the external outline of the building. If the characteristic lines include information about the internal structure of walls and doors and windows, it is necessary to construct a number of internal polygons to express the internal layout of the building. The outline of the building is optimized at right angles. According to the main direction and geometric characteristics of the building, the outline is regularized to eliminate unnecessary details and irregular shapes. Combined with the height information of the point cloud, the two-dimensional outline is expanded into a three-dimensional model.

[0020] The present invention has the following beneficial effects due to the adoption of the above technical solution:

[0021] The present invention adds possibilities to cloud point data at the beginning, which can avoid errors and defects in shooting due to environmental influences, and make rain point data richer and more complete. Then, by denoising the cloud point data in the later stage, the initial model's outline recognition is realized, and irrelevant points are removed, making subsequent data processing more accurate. Then, through edge and curve recognition, the model's boundary can be better planned, and the subsequent processing efficiency is higher. At the same time, the curve is selected based on the edge, which speeds up the processing speed and makes the extraction of building feature lines more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and by way of preferred embodiments. However, it should be noted that many of the details listed in this specification are merely provided to help the reader gain a thorough understanding of one or more aspects of the present invention, and these aspects of the present invention can be practiced even without these specific details.

[0024] like Figure 1 As shown, a building feature line extraction method based on dense point cloud includes the following steps:

[0025] Step 1: Preprocess the point cloud data, including denoising and segmenting, to obtain preprocessed data. The specific process of point cloud denoising is as follows: input the gain point cloud data, annotate the image structure model of the building from each perspective, and the planar structure model of the objects surrounding the building. Set the threshold for determining inliers, the maximum number of iterations, the minimum number of data points required to fit the model, and the threshold for the number of inliers. Randomly obtain several points from the gain point cloud data and fit a model to these points. For each point in the gain point cloud data, calculate the distance from each point to the fitted model. If the distance is less than the threshold for determining inliers, mark the point as an inlier; otherwise, mark it as an outlier. Then, set a loop called for iin range(N), where N is the number of loops, i.e., the number of fitted models. If the number of inliers found in the current iteration exceeds the threshold for the number of inliers, update the best model and record the current set of inliers. Repeat N times until several models that meet the requirements are found. When the number of inliers reaches the threshold for the number of inliers, all remaining points outside the model are deleted.

[0026] During preprocessing, the floor plan diagrams of houses in all directions and shape diagrams of trees, etc. are used as the basic model diagrams for recognition, and then in the subsequent recognition process, trees and buildings can be effectively distinguished.

[0027] Before preprocessing the point cloud data, the point cloud data is supplemented and added. The density of the point cloud data in the image is detected, and then the image is recognized to preliminarily predict the preliminary imaging summary points in the image. Then, cloud point data is added to the attachment of the imaging summary points to obtain gain point cloud data, which makes the point cloud data of the image more complete and the restored imaging model more accurate.

[0028] The specific process of point cloud data segmentation is as follows: select one or several seed points from the denoised point cloud, define the criteria for neighboring points to join the current area, the criteria are the similarity of curvature, normal vector and distance, and gradually add neighboring points that meet the growth criteria to the current area starting from the seed point. When there are no more neighboring points that meet the criteria to be added, stop expanding and select new seed points until all points are assigned to a certain area.

[0029] Step 2: Perform edge and corner detection on the preprocessed point cloud data to obtain feature detection data. Calculate the curvature of each point in the point cloud and compare the curvatures of neighboring points. If the change in curvature exceeds a preset first threshold, the point is considered an edge point. If the change in curvature between edge points exceeds a preset second threshold, the point is considered a corner point.

[0030] Step 3: Perform straight line segment detection and curve segment detection on the feature detection data to obtain line segment detection data; set the Hough transform parameters distance resolution rho, angular resolution theta, threshold threshold, minimum line segment length minLineLength and maximum line segment gap maxLineGap, use cv2.HoughLinesP function to perform straight line segment detection on the edge image, parse the straight line segment information returned by the Hough transform, and draw the straight line segment, then analyze and cluster the straight line segment, convert the pixel value of the image from 255 to 1, extract the outer contour of the target, and then gradually erode the contour until only the skeleton remains, then convert the pixel value of the skeleton image from 1 back to 255, remove isolated points, several isolated points will be generated during the skeleton extraction process, and the isolated points are not part of the skeleton. Use morphological opening operation to remove these isolated points, use median filtering to smooth the skeleton, and connect the breakpoints of the skeleton by the appropriate breakpoint extension method, use cv2.findContours function to detect the contours in the skeleton image, and summarize the contours as the curves in the skeleton.

[0031] Step 4: Connect and remove line segments from the detected segment data to obtain optimized data. After line segment detection, further processing is required, including connecting and removing redundant or erroneous segments. This is crucial for improving the overall quality of line segment detection, especially when building higher-level visual understanding, such as building outlines or road networks. Line segment connection aims to merge broken or incomplete segments into complete and coherent lines. This is particularly useful for reconstructing complete structures from scattered segments.

[0032] Step 5: Create a mapping table for optimized data and create an array to store data features of the same type. Establish mapping rules, including that lines cannot overlap and must be connected. Classify features, including wall lines, roof lines, and door and window lines. Add additional attribute fields to feature lines to store classification information. Perform feature classification and check classification accuracy. Manually adjust any errors.

[0033] Step 6: Reconstruct the 3D data model of the building to generate a 3D model of the building. First, polygon construction is performed, using feature lines connected by mapping and topology to form a number of closed polygons. These closed polygons represent the building's external outline. If the feature lines include information about the internal structure of walls, doors, and windows, several internal polygons need to be constructed to represent the building's internal layout. The building's outline is optimized for right angles and regularized based on the building's main direction and geometric characteristics, eliminating unnecessary details and irregular shapes. Combined with the height information of the point cloud, the 2D outline is expanded into a 3D model.

[0034] By initially adding possibilities to the cloud point data, errors and defects in shooting due to environmental influences can be avoided, making the rain point data richer and more complete. Then, by denoising the cloud point data in the later stage, the initial model's outline recognition can be achieved, and irrelevant points can be removed, making subsequent data processing more accurate. Then, through edge and curve recognition, the model's boundaries can be better planned, and subsequent processing efficiency is higher. At the same time, the curve is selected based on the edge, which speeds up the processing speed and makes the extraction of building feature lines more accurate.

[0035] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A building feature line extraction method based on dense point cloud, characterized by: The method comprises the following steps: Step 1: Preprocess the point cloud data, including point cloud data denoising and point cloud data segmentation, to obtain preprocessed data; Step 2: Perform edge detection and corner detection on the pre-processed point cloud data to obtain feature detection data; Step 3: Perform straight line segment detection and curve segment detection on the feature detection data to obtain line segment detection data; Step 4: Perform line segment connection and removal processing on the line segment detection data to obtain optimized data; Step 5: Create a mapping table for optimized data and create an array to put data features of the same type into the same array; Step 6: Reconstruct the three-dimensional data model of the building to generate a three-dimensional model of the building.

2. The method for extracting building feature lines based on dense point clouds according to claim 1, wherein: In step 1, before preprocessing the point cloud data, the point cloud data is supplemented and added. The density of the point cloud data in the image is detected, and then the image is recognized to preliminarily predict the preliminary imaging summary points in the image. Then, cloud point data is added to the attachment of the imaging summary points to obtain gain point cloud data, which makes the point cloud data of the image more complete and the restored imaging model more accurate.

3. The method for extracting building feature lines based on dense point clouds according to claim 1, wherein: The specific process of point cloud data denoising in step 1 is as follows: input the gain point cloud data, and mark the image structure model of the building from each perspective and the plane structure model of the objects around the building, set the threshold for determining inliers, the maximum number of iterations, the minimum number of data points required to fit the model, and the threshold for the number of inliers, randomly obtain several points from the gain point cloud data, and randomly fit a model to the obtained points. For each point in the gain point cloud data, calculate the distance from each point to the fitted model. If the distance is less than the threshold for determining inliers, mark the point as an inlier, otherwise mark it as an outlier. Then set a loop of for iin range(N), where N is the number of loops, that is, the number of fitted models. When the number of inliers found in the current iteration is greater than the threshold for the number of inliers, update the best model and record the current set of inliers. Repeat N times to find several models that meet the requirements, that is, when the number of inliers reaches the threshold for the number of inliers, delete all remaining points outside the model.

4. The method for extracting building feature lines based on dense point clouds according to claim 1, wherein: The specific process of point cloud data segmentation in step 1 is as follows: select one or several seed points from the denoised point cloud, define the criteria for neighboring points to be added to the current area, the criteria are the similarity of curvature, normal vector and distance, and gradually add neighboring points that meet the growth criteria to the current area starting from the seed point. When there are no more neighboring points that meet the criteria to be added, stop expanding and select new seed points until all points are assigned to a certain area.

5. The method for extracting building feature lines based on dense point clouds according to claim 1, wherein: The specific process of step 2 is: calculate the curvature of each point in the point cloud, compare the curvature between neighboring points, and if the change in curvature exceeds a preset first threshold, the point is considered to be an edge point; if the change in curvature between edge points exceeds a preset second threshold, the point is considered to be a corner point.

6. The method for extracting building feature lines based on dense point clouds according to claim 1, wherein: The specific process of step 3 is: set the parameters of Hough transform distance resolution rho, angular resolution theta, threshold threshold, minimum line segment length minLineLength and maximum line segment gap maxLineGap, use cv2.HoughLinesP function to detect straight line segments on the edge image, parse the straight line segment information returned by Hough transform, draw straight line segments, and then analyze and cluster the straight line segments, convert the pixel value of the image from 255 to 1, extract the outer contour of the target, and then gradually erode the contour until only the skeleton remains, and then convert the pixel value of the skeleton image from 1 back to 255, remove isolated points, and generate several isolated points during the skeleton extraction process. Isolated points are not part of the skeleton. Use morphological opening operation to remove these isolated points, use median filtering to smooth the skeleton, and connect the breakpoints of the skeleton by the appropriate breakpoint extension method. Use cv2.findContours function to detect the contours in the skeleton image, and summarize the contours as the curves in the skeleton.

7. The method for extracting building feature lines based on dense point clouds according to claim 1, wherein: The specific process of step 5 is as follows: establish mapping relationship rules, including that lines cannot overlap and must be connected, perform feature classification, including wall lines, roof lines, and door and window lines, add additional attribute fields to feature lines to store classification information, perform feature classification, check the accuracy of the classification, and make manual adjustments when errors are found.

8. The method for extracting building feature lines based on dense point clouds according to claim 1, wherein: The specific process of step 6 is as follows: first, polygon construction is performed, and feature lines connected by mapping and topology are used to form several closed polygons. The closed polygons represent the external outline of the building. If the feature lines include information about the internal structure of walls, doors and windows, several internal polygons need to be constructed to express the internal layout of the building. The outline of the building is optimized for right angles. According to the main direction and geometric characteristics of the building, the outline is regularized to eliminate unnecessary details and irregular shapes. Combined with the height information of the point cloud, the two-dimensional outline is expanded into a three-dimensional model.

Images