Working plane progress identification method and device based on construction scene three-dimensional modeling

Abstract

The invention relates to a working plane progress recognition method and equipment based on construction scene three-dimensional modeling, and the method comprises the steps: obtaining an image set of a target building collected by a tower crane camera in a multi-view manner, arranging ground control points on site, improving a COLMAP frame based on fusion of GCP, and carrying out three-dimensional reconstruction and scale recovery. Obtaining a point cloud site model with a real scale; constructing an adaptive filter, and separating a main building model according to the obtained real-scale point cloud site model; and based on the point cloud distribution characteristics of the construction operation surface, counting the height of the main building, identifying the operation surface, obtaining the height of the operation surface, and evaluating the construction progress. Compared with the prior art, the method has the advantages that the progress of the working plane is accurately evaluated in real time according to the images acquired by the tower crane camera at the construction commanding height in a multi-view manner.

Description

Technical Field

[0001] The present invention relates to the field of computer vision, and in particular to a method and device for identifying the progress of a work surface based on three-dimensional modeling of a construction scene. Background Art

[0002] In recent years, with the vigorous development of the construction industry, construction management has gradually attracted attention. In on-site construction, progress, quality, and safety are the three major management goals. Among them, progress often affects cost, project benefits, and project quality. Therefore, on-site construction progress has always been the focus of engineers. How to efficiently evaluate the construction progress has attracted extensive research. In the process of construction, the height of the construction work surface is an important indicator for measuring the construction progress. Traditional work surface height identification relies on manual labor, which is high in cost and low in precision, and is gradually being replaced by new methods.

[0003] With the development of computer technology, computer vision and 3D reconstruction technology have gradually matured and have been widely used in the civil engineering industry, achieving many results. The 3D reconstruction of the construction site can intuitively reflect the on-site situation and facilitate the extraction of relevant parameters. However, the research on the high-precision extraction of the building working surface is still at the level of low efficiency, low precision and poor adaptability. Our method aims to achieve high-precision extraction of the building working surface.

[0004] The Chinese invention patent applied by Glodon Technology Co., Ltd. in 2021 and published as CN113720283B discloses a device, electronic equipment and system for identifying the height of construction. The device is installed on the boom of a tower crane to locate and measure the distance. The device with data processing capability analyzes the current distance data to determine and visualize the current height of the target building. The overall system has high accuracy, but also places extremely high demands on the equipment and high cost. The module has strict logic and also places high demands on computing power.

[0005] Hangzhou Weiming Xinke Technology Co., Ltd. applied for a Chinese invention patent in 2022 with publication number CN115240136A, which discloses a method for detecting the construction progress of steel structure buildings. Based on on-site construction images, three-dimensional design models, and steel structure building height detection models, the target steel structure building height is obtained, and combined with data such as the signed freight bill, the construction progress of the steel structure building at the installation unit level is automatically identified. However, the method used requires a high amount of data, and requires complete freight bill information and construction drawings, which makes it difficult to implement.

[0006] Shenzhen Ruiyun Technology Co., Ltd. applied for a Chinese invention patent with publication number CN116431037A in 2023, which discloses a 3D reconstruction method based on COLMAP+OPENMVS. The dense point cloud results of COLMAP are combined with OPEMNVS to build a 3D model, reducing energy consumption and time consumption. However, the results of COLMAP 3D reconstruction lack the true scale and need to be restored.

[0007] Hangzhou Yixian Advanced Technology Co., Ltd. applied for a Chinese invention patent with publication number CN115830276A in 2022, which discloses a method of SFM mapping scale recovery based on terminal real scale tracking. The real scale trajectory of the image is calculated by a random consistency similarity transformation algorithm based on the camera model, and the global optimal scale coefficient is obtained based on the weighted average method, and the three-dimensional model of the real scale is obtained, which improves the reliability of scale recovery. However, the overall framework calculation is complicated and requires high computing power.

[0008] The Chinese University of Hong Kong disclosed a three-dimensional building extraction method in 2021, which extracts the building outline from one or more stereo pairs of the building, determines the first height estimate of the extracted building outline based on the rational polynomial coefficients of the stereo pair; obtains scattered points with stable properties from the multi-phase SAR image of the building; determines the second height estimate of the determined scattered points; and combines the first height estimate and the second height estimate to generate a fused height of each of the extracted building outlines. The method is significantly better than the single height extraction method, however, the acquisition of stereo pairs is more difficult at the construction site, and the drone acquisition method is easily blocked. At the same time, the height recognition accuracy for a single building is poor.

[0009] The journal article "A Direction-Based Correction Method for Outcrop SfM-MVS Point Cloud Correction Method Based on Direction Ground Control Plane" published in the Journal of Structural Geology in 2021 mentioned a correction method based on the GCPL direction, which relies on the direction of a pair of non-parallel planes for correction and is ideally more suitable for large-scale outcrops. However, in construction scenarios, the requirements for the height accuracy and range of the working surface are lower than those in geological exploration. The method is expensive and requires high-precision equipment, and is not very applicable to construction scenarios.

[0010] The journal article "Robust Segmentation and Localization of Structural Planes from Photogrammetric Point Clouds at Construction Sites" published in 2020 in Building Automation proposed a bottom-up approach that uses voxelized structures and face-based global clustering to achieve plane segmentation, and uses plane fitting based on a parametric model to extract points of plane segments. The proposed method can segment most structural planes and estimate accurate model parameters and dimensions. The resulting model has good adaptability to staged scenarios, but in actual construction scenarios, there are interferences such as guardrails and scaffolding, and the surface is relatively rough. It cannot be selected as a plane and is difficult to adapt to more complex on-site construction scenarios.

[0011] The journal article "Building Extraction Based on Semi-Supervised Deep Learning with Hyperspectral Remote Sensing Images and Limited Training Samples" published in Computer and Electrical Engineering in 2023 proposed a lightweight semantic segmentation model based on the "encoder-decoder" structure, using MobileNet and combined with multi-scale feature fusion, which has high advancement and accuracy in building extraction. However, construction site images rely on drones and deployed cameras. Acquiring advanced remote sensing images will increase construction costs, and the accuracy itself is not enough to meet on-site construction requirements.

[0012] The journal article "Combining ICESat-2 Photons and Google Earth Satellite Images for Building Height Extraction" published in the International Journal of Applied Earth Observation and Geoinformation in 2023 established an improved shadow-based building height estimation model from the building height annotations inverted from ICESat-2 photons through the U-Net deep learning framework. The model has high estimation accuracy and is suitable for large-scale building height inversion. However, it has low accuracy in extracting single-target building height and is therefore not suitable for construction sites.

[0013] The Chinese invention patent applied by Qingdao Institute of Surveying and Mapping in 2022 and published as CN114764871A discloses a method for extracting urban building attributes based on airborne laser point clouds. It uses a cloth filter algorithm to extract non-ground points, and then obtains relevant information about the building through plane fitting technology after clustering, achieving high accuracy. However, airborne laser point cloud radar is expensive and cannot be widely used on every construction site, and does not meet the requirements of obtaining the height of the working surface in real time.

[0014] How to achieve real-time and accurate identification of the progress of construction work surfaces has become a technical problem that needs to be solved. Summary of the invention

[0015] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method and device for identifying the progress of a work surface based on three-dimensional modeling of a construction scene.

[0016] The purpose of the present invention can be achieved by the following technical solutions:

[0017] According to one aspect of the present invention, a method for identifying the progress of a work surface based on three-dimensional modeling of a construction scene is provided, the method comprising the following steps:

[0018] Step S1, obtaining an image set of the target building captured by a tower crane camera from multiple perspectives, arranging ground control points on site, and implementing three-dimensional reconstruction and scale restoration based on an improved COLMAP framework based on GCP to obtain a point cloud site model with a real scale;

[0019] Step S2, constructing an adaptive filter, identifying the main building of the target building according to the point cloud site model obtained in step S1, and obtaining a main building model;

[0020] Step S3, based on the point cloud distribution characteristics of the construction work surface, count the height of the main building, identify the work surface, obtain the height of the work surface, and realize the construction progress evaluation.

[0021] Preferably, in step S1, the process of obtaining a point cloud site model with a real scale includes:

[0022] Step S1-1, laying out a slide rail at the bottom of the tower crane boom, laying out a camera at the front end of the boom, and adjusting the relative position of the camera on the tower crane boom and the shooting angle of the camera;

[0023] Step S1-2, photographing an image of a target building during the operation of the tower crane to obtain an image set of the target building;

[0024] Step S1-3, arranging ground control points on site, measuring the real coordinates of the ground control points, and identifying the two-dimensional coordinates of the GCPs in the target building image set;

[0025] Step S1-4, detecting and matching the feature points of each image in the image set, obtaining the internal and external parameters of the camera, and estimating the relative position and posture of the camera based on the COLMAP framework;

[0026] Steps 1-5, based on the relative position of the camera and the two-dimensional coordinates of the GCP, use COLMAP's built-in triangulation to calculate the GCP relative spatial coordinates;

[0027] Step S1-6, based on the least squares method, the coordinate transformation matrix between the relative space coordinates and the real coordinates is calculated, the camera posture is changed, and the format is adjusted to conform to the COLMAP format to obtain a point cloud model with a real scale.

[0028] More preferably, the arranging of ground control points on site is performed according to corresponding requirements, including:

[0029] Ground control point distribution requirements: multiple ground control points are located at different heights and have set height differences;

[0030] Ground control point size requirement: The pixel value of the ground control point in the actual image should be greater than the pixel threshold to ensure that the camera can recognize the ground control point;

[0031] Relative angle requirements for ground control points: The angle between the plane normal of the center point of the ground control point and the straight line from the camera to the point is within the set range;

[0032] Ground control point style requirements: The style of the ground control point should be different from the surrounding features.

[0033] Preferably, the step S2 comprises:

[0034] Step S2-1, using a voxel filter to downsample the point cloud of the point cloud site model, filtering out sparse outliers;

[0035] Step S2-2, using a height filter to separate ground points in the point cloud site model;

[0036] Step S2-3, using the DBSCAN clustering algorithm to divide the construction site point cloud into different clusters, and clustering relatively independent ground object blocks into one category;

[0037] Step S2-4, adaptively optimizing the height filter according to the clustering result;

[0038] Step S2-5, setting corresponding height thresholds for different construction scenarios, determining buildings exceeding the height thresholds as main buildings, and extracting the main building model.

[0039] More preferably, in step S2-1, downsampling the point cloud of the point cloud site model using a voxel filter includes:

[0040] Step S2-1-1, dividing the point cloud into voxels;

[0041] Step S2-1-2, calculate the centroid of the non-empty voxel to replace all the points in the voxel to achieve downsampling of the point cloud. The calculation method is as follows: Given a non-empty voxel V containing m points, the centroid P of the voxel cen (x cen ,y cen ,z cen )for:

[0042]

[0043] where x i ,y i and z i are the coordinates of the points inside the voxel, and m is the number of points inside the voxel.

[0044] More preferably, in step S2-2, using a height filter to separate ground points in the point cloud site model comprises the following steps:

[0045] Step S2-2-1, initialize the filter and set the sliding window size;

[0046] Step S2-2-2, setting a height difference threshold, when the height difference between the highest point cloud and the lowest point cloud in the window is less than the height difference threshold, it is determined to be a ground point; otherwise, it is determined to be a ground object point;

[0047] Step S2-2-3, using a sliding window to traverse and filter the site model, separate ground points and object points, and remove ground points;

[0048] Step S2-2-4, calculating the silhouette coefficient and ARI index of the clustering result of step S2-3;

[0049] Step S2-2-5, construct a new evaluation index ShRI through the weighting method to comprehensively measure the clustering effect; based on the genetic algorithm GA, obtain the window size and height difference threshold under the optimal ShRI.

[0050] Preferably, the step S3 comprises the following steps:

[0051] Step S3-1, draw a point cloud quantity-height distribution histogram, and count the elevation distribution of the original construction site point cloud;

[0052] Step S3-2, according to the statistical results of step S3-1, select the elevation range where the point cloud density is higher than the set density threshold as the preliminary working surface height range, segment the point cloud model, and obtain the preliminary working surface point cloud model;

[0053] Step S3-3, processing the preliminary work surface point cloud model according to the work surface structure type;

[0054] Step S3-4, further performing height statistics on the obtained preliminary work surface point cloud model, and identifying the histogram peak based on the AMPD peak recognition algorithm;

[0055] Step S3-5, identifying the height range of the working surface according to the statistical results of step S3-4;

[0056] Step S3-6, according to the result of step S3-5, based on the RANSAC plane fitting algorithm, the local point in the plane fitting peak area is used as the height of the working surface;

[0057] Step S3-7, add plane visualization display, manually verify the visualization results, and output the precise height of the working surface;

[0058] Step S3-8: Import the outputted precise height of the working surface into the construction system database, calculate the corresponding construction progress according to the working surface height by using the interpolation method, and realize the construction progress evaluation.

[0059] More preferably, in the step S3-3, if the working surface is a cast-in-place structure, the segmented point cloud model is Gaussian sampled and denoised; if the working surface is a steel structure, Gaussian sampling is not required.

[0060] More preferably, in step S3-5, identifying the height range of the working surface includes:

[0061] When the target building involves multiple working surfaces, human-computer interaction is used for identification according to different situations, and judgment is made according to the following method: If the working surface of the target building is a combination of steel frame and core tube, according to the original statistical results, the screening algorithm is imported for peak recognition, and two groups of working surface height ranges are output in turn, and step S3-1 is repeated for the two groups of working surface height ranges respectively; if the working surface of the target building is a core tube and an access platform, according to the original statistical results, only the peak ranges of the two groups of working surface heights with relatively low heights are retained as the core tube working surface height range.

[0062] According to another aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the method described above is implemented when the processor executes the program.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1) The present invention aims at a single tower crane construction site. According to the multi-perspective real-time images collected from the construction commanding heights, the ground control points are arranged on site. Based on the improved COLMAP framework of GCP, three-dimensional reconstruction and scale recovery are realized, and the point cloud site modeling of the real scale is quickly performed. Then, an adaptive filter is constructed to separate the main building model. Finally, the main building height is counted according to the height distribution characteristics of the point cloud, the working surface is identified, and the real height of the working surface is obtained, thereby realizing real-time and accurate evaluation of the construction progress.

[0065] 2) The present invention arranges ground control points on site according to set rules, which helps to improve the accuracy of the three-dimensional reconstructed point cloud site model. The site model is traversed and filtered using a sliding window to separate the ground points of the point cloud site model. The height filter is adaptively optimized according to the clustering results, and the window size and height difference threshold under the optimal ShRI are obtained based on the genetic algorithm to evaluate the clustering effect, thereby extracting the main building more accurately.

[0066] 3) The present invention adopts a two-step statistical method, in which the original construction site point cloud elevation distribution is roughly counted to identify the working surface height and obtain a preliminary working surface point cloud model, and the height interval is further accurately counted to perform main building height statistics, and the RANSAC plane fitting algorithm is iterated to obtain the optimal plane, thereby further improving the recognition accuracy of the working surface height.

[0067] 4) The present invention proposes different adaptation methods according to different construction structures and working surface forms, which are suitable for more accurate construction site progress management in multiple scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flow chart of the three-dimensional modeling and work surface progress identification method in the present invention;

[0069] Figure 2 is a schematic diagram of GCP and ground settings in an embodiment of the present invention;

[0070] Figure 3 is a schematic diagram of the relative angle setting of the GCP in an embodiment of the present invention;

[0071] Figure 4 is a schematic diagram of the process of restoring the orientation and scale of a point cloud model in a horizontal plane in an embodiment of the present invention;

[0072] Figure 5 is a schematic diagram of the process of height difference filtering in an embodiment of the present invention;

[0073] Figure 6 This is a schematic diagram of preliminary statistical point cloud distribution in an embodiment of the present invention;

[0074] Figure 7 In the embodiment of the present invention, Figure 6 Schematic diagram of secondary statistical point cloud distribution in Part A;

[0075] Figure 8 Schematic diagram of the pattern of ground control points in an embodiment of the present invention. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0077] This embodiment relates to a method for identifying the progress of a work surface based on three-dimensional modeling of a construction scene. The method is for a general construction scene with a single tower crane, and will not be explained for super-high-rise (with a floor height of more than 40 floors or a height of more than 100 meters) and multi-tower crane scenes. Figure 1 , including the following steps:

[0078] Step S1: The tower crane camera collects images, and based on the improved COLMAP framework that integrates ground control points (GCPs), three-dimensional reconstruction and scale recovery are achieved to obtain a point cloud site model with a real scale.

[0079] Step S2: construct an adaptive filter and identify the main building according to the obtained real model.

[0080] Step S3: Based on the point cloud distribution characteristics of the construction work surface, the main building height is counted, the work surface is identified, the work surface height is obtained, and the construction progress is evaluated.

[0081] In this embodiment, in step S1, the specific steps of collecting images by the tower crane camera are:

[0082] Step S1-1, a slide rail is arranged at the bottom of the tower crane boom so that a camera can be arranged at the front end of the boom and the relative position of the camera on the tower crane boom and the shooting angle of the camera can be adjusted.

[0083] Step S1-2: photographing an image of a target building during the operation of the tower crane to obtain an image set of the target building.

[0084] In this embodiment, an improved COLMAP framework based on GCP is used to achieve 3D reconstruction and scale recovery. COLMAP is a general motion structure (SfM) and multi-view stereo (MVS) pipeline with graphical and command line interfaces, which provides a wide range of functions for the reconstruction of ordered and unordered image collections. Figure 4 , the specific steps are:

[0085] Step S1-3: Arrange ground control points on site according to the requirements of distribution, size, relative angle and style, measure the real coordinates B(x2, y2, z2) of the ground control points, and identify the two-dimensional coordinates of the GCP in the target building image set.

[0086] Step S1-4: Detect and match the feature points of each image, obtain the internal and external parameters of the camera, and estimate the relative position of the camera based on the COLMAP framework.

[0087] Step 1-5: According to the relative position of the camera and the two-dimensional coordinates of the GCP, use the triangulation provided by COLMAP to calculate the relative spatial coordinates A(x1, y1, z1) of the GCP.

[0088] Step S1-6: Based on the least squares fitting method, calculate the coordinate transformation matrix T between the relative space coordinates A (x1, y1, z1) and the real coordinates B (x2, y2, z2), change the camera posture, and obtain a point cloud model with a real scale.

[0089] In step S1-3, "arranging ground control points on site" includes many requirements, such as Figure 2 , including ground control point distribution requirements, ground control point size requirements, ground control point relative angle requirements and ground control point style requirements, specifically:

[0090] Ground control point distribution requirements: Ground control points should be located at different heights with a certain height difference to enhance the reconstruction effect;

[0091] Ground control point size requirement: The pixel value of the ground control point in the actual image should be greater than 30*30 to ensure that the camera can recognize the ground control point.

[0092] Relative angle requirement of ground control points: The angle α∈[0,30°] formed by the plane normal of the center point of the ground control point and the straight line from the camera to the point. The optimal situation is 0° to avoid large recognition errors, such as Figure 3 .

[0093] Ground control point style requirements: Ground control points should be clearly distinguishable from surrounding objects, such as black and white steel plates, to facilitate on-site identification. Figure 8 .

[0094] In step S1-4, estimating the relative pose of the camera based on the COLMAP framework and calculating the relative three-dimensional coordinates A(x1, y1, z1) of the GCP include the following steps:

[0095] Step S1-4-1, detect feature points for each image and perform feature point matching.

[0096] Step S1-4-2, obtain the internal and external parameters of the camera and estimate the relative position of the camera.

[0097] Step S1-4-3, based on the relative position of the camera and the GCP 2D point coordinates, use COLMAP's built-in triangulation to calculate the GCP 3D coordinates. The calculation process is as follows:

[0098] Let the coordinates of a point in the world coordinate system (hereinafter referred to as world coordinates) be X w =[x w ,y w ,z w ] T , the coordinates in the camera coordinate system (hereinafter referred to as camera coordinates) are X c =[x c ,y c,z c ] T , the coordinates of the camera image projection (hereinafter referred to as image coordinates) are X = [u, v] T .

[0099] The relationship between camera coordinates and image coordinates is expressed by the intrinsic parameter matrix N:

[0100]

[0101]

[0102] Among them, N is the internal parameter matrix, X c is the coordinate of the camera coordinate system, u and v are the coordinates of the camera image projection, z c represents the depth, f x and f y is the pixel ratio, c x and c y are the pixel coordinates of the center of the phase plane.

[0103] The external parameter matrix includes the translation matrix t and the rotation matrix R, which can be used to obtain the camera coordinates X c and world coordinate X w Changes:

[0104] Among them, the dimension of the rotation matrix R is 3×3, and the dimension of the translation matrix t is 3×1.

[0105] Combining and simplifying the two transformations and eliminating the camera coordinates yields:

[0106]

[0107] Thus the relative three-dimensional coordinates are calculated.

[0108] Step S1-6 includes the following steps:

[0109] Step S1-6-1, based on the GCP three-dimensional coordinates A (x1, y1, z1) and the real coordinates B (x2, y2, z2), perform least squares fitting according to the following relationship:

[0110] AT=B

[0111] Solve for the least squares solution:

[0112] T=(A T A) -1 A T B

[0113] Get the coordinate transformation rotation matrix T.

[0114] Step S1-6-2, using the orthogonality of the rotation matrix, change the camera pose, adjust the format to conform to the COLMAP format, realize three-dimensional point projection, and restore the true scale.

[0115] In this example, step S2 includes the following steps:

[0116] Step S2-1, preliminarily process the point cloud in the reconstructed site, use a voxel filter to downsample the point cloud, filter out sparse outliers, reduce the number of point clouds, and reduce noise interference.

[0117] Step S2-2: Use a height filter to separate the ground points in the model based on the large height difference between the ground object and the ground.

[0118] Step S2-3: Based on the characteristic that the point clouds are coherently distributed near the corresponding objects, the DBSCAN clustering algorithm is used to divide the site point clouds into different clusters, and relatively independent object blocks are clustered into one category.

[0119] Step S2-4: Adaptively optimize the height filter according to the clustering results.

[0120] Step S2-5: for different construction scenarios, according to the different characteristics of different main building heights, retrieve the database, set the height threshold h0, determine the building exceeding the height threshold as the main building, and extract the main building model.

[0121] Voxel downsampling of a point cloud using a voxel filter consists of the following steps:

[0122] Step S2-1-1, dividing the point cloud into voxels;

[0123] Step S2-1-2, calculate the centroid of the non-empty voxel to replace all the points in the voxel to achieve downsampling of the point cloud. The calculation method is as follows: Given a non-empty voxel V containing m points, the centroid P of the voxel cen (x cen ,y cen ,z cen )for:

[0124]

[0125] where x i ,y i and z i are the coordinates of the points inside the voxel, and m is the number of points inside the voxel.

[0126] Step S2-2 includes the following steps:

[0127] Step S2-2-1, initialize the filter and set the sliding window size D.

[0128] Step S2-2-2, setting the height difference threshold Δh max , when the height difference Δh between the highest point cloud and the lowest point cloud in the window is less than the height difference threshold Δh max , it is determined to be a ground point. Otherwise, it is determined to be a ground feature point.

[0129] Step S2-2-3, using a sliding window, traverse and filter the site model, separate ground points and object points, and remove ground points, such as Figure 5 .

[0130] Step S2-2-4, calculate the silhouette coefficient S and ARI index R of the clustering result of step S2-3:

[0131] For a sample set, its silhouette coefficient is the average of all sample silhouette coefficients. The value range of the silhouette coefficient is [-1,1]. The closer the distance between samples of the same category is, and the farther the distance between samples of different categories is, the higher the score is.

[0132] The silhouette coefficient s for a sample i =(ba) / Max(a,b), where a represents the average distance between a sample and other samples in its cluster, and b represents the average distance between a sample and samples in other clusters.

[0133] The overall silhouette coefficient of clustering Where N is the number of samples.

[0134] ARI is a clustering evaluation algorithm that measures the similarity between two clustering results by calculating the number of sample points in the same cluster and in different clusters. The calculation formula is as follows:

[0135]

[0136] Among them, R is the ARI index, e represents the number of point pairs that belong to the same cluster in both clustering results, f represents the number of point pairs that belong to the same cluster in clustering result one but not in clustering result two, c represents the number of point pairs that do not belong to the same cluster in clustering result one but belong to the same cluster in clustering result two, d represents the number of point pairs that do not belong to the same cluster in both clustering results, and the value range of ARI is [-1, 1]. The larger the value, the higher the similarity and the better the clustering stability.

[0137] Step S2-2-5, construct a new evaluation index ShRI through the weighting method to comprehensively measure the clustering effect; based on the genetic algorithm GA, obtain the window size and height difference threshold under the optimal ShRI.

[0138] Among them, "obtaining the window size and height difference threshold under the optimal ShRI based on the genetic algorithm GA" includes the following steps:

[0139] 1) Initialize the population, set the number of individuals n = 100, the maximum number of genetic iterations k = 100, the crossover probability P1 = 0.7, and the mutation probability P2 = 0.1.

[0140] 2) Set the input parameters sliding window size D and height difference threshold Δh max Binary conversion method, define the objective function ShRI calculation method:

[0141] ShRI=ω1S+ω2R

[0142] Among them, ω1 is the weight corresponding to the silhouette coefficient, S is the normalized silhouette coefficient, ω2 is the weight corresponding to the ARI indicator, when n=1, ω2=0, and R is the normalized ARI indicator.

[0143] It is worth noting that the comparison object of the ARI index in each traversal is the optimal individual in the current population, that is, the clustering result corresponding to the maximum ShRI.

[0144] 3) The number of individuals of the randomly crossed parents and the height difference threshold correspond to the binary conversion result, the binary code is mutated at a specific position, the offspring is obtained, and the offspring is added to the population. The ShRI of the population is reordered, and n = 100 elite individuals are retained to complete a genetic iteration. The iteration is repeated until the number of iterations k is reached, and the individual with the highest fitness, that is, the largest ShRI, and the optimal ShRI value are output.

[0145] In step S3, based on the point cloud distribution characteristics of the construction work surface, the main building height is counted, the work surface is identified, the work surface height is obtained, and the process of evaluating the construction progress includes the following steps:

[0146] Step S3-1, rough statistics (preliminary statistics): for different main buildings, different height intervals Δs are set, a point cloud quantity-height distribution histogram is drawn, and the original construction site point cloud elevation distribution is counted.

[0147] Step S3-2: According to the statistical results, select the point cloud density elevation range [h1, h2] as the preliminary working surface height range, segment the point cloud model, and obtain a preliminary working surface point cloud model, such as Figure 6 .

[0148] Step S3-3, according to the different types of working surface structures, if the working surface is a cast-in-place structure and the surface flatness of the working surface is low, the segmented point cloud model is Gaussian sampled and denoised; if the working surface is a steel structure, due to the high flatness of the steel structure working surface, Gaussian sampling is not required.

[0149] Step S3-4, Precise Statistics (Secondary Statistics): Further refine the height spacing, perform further height statistics on the obtained preliminary working surface point cloud model, and identify the histogram peak based on the AMPD (Amplitude Modulation Peak Detection) peak recognition algorithm, as Figure 7 。

[0150] When the target building involves multiple working surfaces, identify them in a man-machine interaction manner according to different situations, and make judgments according to the following methods: If the working surface of the target building is a combination of a steel frame and a core tube, based on the original statistical results, import the screening algorithm to perform peak recognition, and output two groups of working surface height ranges: [h1, h2] and [h1′, h'2] in sequence, where h1′ > h2, and repeat Step S3-1 for each of the two groups of working surface height ranges; If the working surface of the target building is a core tube and an access platform, according to the statistical results of Step S3-4, remove the peak range with a higher height and retain the peak range with a lower height as the core tube working surface height range.

[0151] Step S3-6, Plane Output: According to the statistical results of Step S3-5, based on the ransac (Random Sample Consensus) plane fitting algorithm, fit the local points in the peak region of the plane to improve the height recognition accuracy. Set the number of iterations n and the distance threshold s0. Randomly select three points from the local peak points to form a plane, construct a coordinate matrix, and solve the plane equation. Calculate the distance s of all points to the plane i ,and count the number of points (valid points) where s i < s0. After n iterations, select the plane with the largest number of valid points as the optimal plane to obtain the plane expression ax + by + cz = d, where (a, b, c) constitutes the normal vector of the plane, and d is the distance from the plane to the origin. Due to the flatness characteristics of the working surface, a and b approach 0, and c approaches 1. Therefore, the working surface height can be represented by d.

[0152] Step S3-7, Add plane visualization display, manually verify the visualization result, and output the precise height of the working surface;

[0153] Step S3-8, Import the output precise height of the working surface into the construction system database, and calculate the corresponding construction progress according to the following interpolation method based on the working surface height to evaluate the construction progress.

[0154] Two consecutive standard heights h in the database i 、h i+1 correspond to construction progress j i 、j i+1 , when the output precise height h of the working surface a is located in [h i , hi+1 ], corresponding to the construction progress j a The calculation method is:

[0155]

[0156] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0157] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0158] The processing unit performs the various methods and processes described above, such as methods S1 to S3. For example, in some embodiments, methods S1 to S3 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S3 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S3 in any other appropriate manner (e.g., by means of firmware).

[0159] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0160] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0161] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0162] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for identifying the progress of a work surface based on three-dimensional modeling of a construction scene, characterized in that: The method comprises the following steps: Step S1, obtaining an image set of the target building captured by a tower crane camera from multiple perspectives, arranging ground control points on site, and implementing three-dimensional reconstruction and scale restoration based on an improved COLMAP framework based on GCP to obtain a point cloud site model with a real scale; Step S2, constructing an adaptive filter, identifying the main building of the target building according to the point cloud site model obtained in step S1, and obtaining a main building model; Step S3, based on the point cloud distribution characteristics of the construction work surface, count the height of the main building, identify the work surface, obtain the height of the work surface, and realize the construction progress evaluation.

2. The method for identifying the progress of a work surface based on three-dimensional modeling of a construction scene according to claim 1, characterized in that: In step S1, the process of obtaining a point cloud site model with a real scale includes: Step S1-1, laying out a slide rail at the bottom of the tower crane boom, laying out a camera at the front end of the boom, and adjusting the relative position of the camera on the tower crane boom and the shooting angle of the camera; Step S1-2, photographing an image of a target building during the operation of the tower crane to obtain an image set of the target building; Step S1-3, arranging ground control points on site, measuring the real coordinates of the ground control points, and identifying the two-dimensional coordinates of the GCPs in the target building image set; Step S1-4, detecting and matching the feature points of each image in the image set, obtaining the internal and external parameters of the camera, and estimating the relative position and posture of the camera based on the COLMAP framework; Steps 1-5, based on the relative position of the camera and the two-dimensional coordinates of the GCP, use COLMAP's built-in triangulation to calculate the GCP relative spatial coordinates; Step S1-6, based on the least squares method, the coordinate transformation matrix between the relative space coordinates and the real coordinates is calculated, the camera posture is changed, and the format is adjusted to conform to the COLMAP format to obtain a point cloud model with a real scale.

3. The method for identifying the progress of a work surface based on three-dimensional modeling of a construction scene according to claim 2, characterized in that: The ground control points are arranged on site according to corresponding requirements, including: Ground control point distribution requirements: multiple ground control points are located at different heights and have set height differences; Ground control point size requirement: The pixel value of the ground control point in the actual image should be greater than the pixel threshold to ensure that the camera can recognize the ground control point; Relative angle requirements for ground control points: The angle between the plane normal of the center point of the ground control point and the straight line from the camera to the point is within the set range; Ground control point style requirements: The style of the ground control point should be different from the surrounding features.

4. The method for identifying the progress of a work surface based on three-dimensional modeling of a construction scene according to claim 1, characterized in that: The step S2 comprises: Step S2-1, using a voxel filter to downsample the point cloud of the point cloud site model, filtering out sparse outliers; Step S2-2, using a height filter to separate ground points in the point cloud site model; Step S2-3, using the DBSCAN clustering algorithm to divide the construction site point cloud into different clusters, and clustering relatively independent ground object blocks into one category; Step S2-4, adaptively optimizing the height filter according to the clustering result; Step S2-5, setting corresponding height thresholds for different construction scenarios, determining buildings exceeding the height thresholds as main buildings, and extracting the main building model.

5. The method for identifying the progress of a work surface based on three-dimensional modeling of a construction scene according to claim 4, characterized in that: In the step S2-1, downsampling the point cloud of the point cloud site model using a voxel filter includes: Step S2-1-1, dividing the point cloud into voxels; Step S2-1-2, calculate the centroid of the non-empty voxel to replace all the points in the voxel to achieve downsampling of the point cloud. The calculation method is as follows: Given a non-empty voxel V containing m points, the centroid P of the voxel cen (x cen ,y cen ,z cen )for: where x i ,y i and z i are the coordinates of the points inside the voxel, and m is the number of points inside the voxel.

6. The method for identifying the progress of a working surface based on three-dimensional modeling of a construction scene according to claim 4, characterized in that: In step S2-2, using a height filter to separate ground points in the point cloud site model includes the following steps: Step S2-2-1, initialize the filter and set the sliding window size; Step S2-2-2, setting a height difference threshold, when the height difference between the highest point cloud and the lowest point cloud in the window is less than the height difference threshold, it is determined to be a ground point; otherwise, it is determined to be a ground object point; Step S2-2-3, using a sliding window to traverse and filter the site model, separate ground points and object points, and remove ground points; Step S2-2-4, calculating the silhouette coefficient and ARI index of the clustering result of step S2-3; Step S2-2-5, construct a new evaluation index ShRI through the weighting method to comprehensively measure the clustering effect; based on the genetic algorithm GA, obtain the window size and height difference threshold under the optimal ShRI.

7. The method for identifying the progress of a work surface based on three-dimensional modeling of a construction scene according to claim 1, characterized in that: The step S3 comprises the following steps: Step S3-1, draw a point cloud quantity-height distribution histogram, and count the elevation distribution of the original construction site point cloud; Step S3-2, according to the statistical results of step S3-1, select the elevation range where the point cloud density is higher than the set density threshold as the preliminary working surface height range, segment the point cloud model, and obtain the preliminary working surface point cloud model; Step S3-3, processing the preliminary work surface point cloud model according to the work surface structure type; Step S3-4, further performing height statistics on the obtained preliminary work surface point cloud model, and identifying the histogram peak based on the AMPD peak recognition algorithm; Step S3-5, identifying the height range of the working surface according to the statistical results of step S3-4; Step S3-6, according to the result of step S3-5, based on the RANSAC plane fitting algorithm, the local point in the plane fitting peak area is used as the height of the working surface; Step S3-7, add plane visualization display, manually verify the visualization results, and output the precise height of the working surface; Step S3-8: Import the outputted precise height of the working surface into the construction system database, calculate the corresponding construction progress according to the working surface height by using the interpolation method, and realize the construction progress evaluation.

8. The method for identifying the progress of a working surface based on three-dimensional modeling of a construction scene according to claim 7, characterized in that: In the step S3-3, if the working surface is a cast-in-place structure, the segmented point cloud model is Gaussian sampled and denoised; if the working surface is a steel structure, Gaussian sampling is not required.

9. The method for identifying the progress of a working surface based on three-dimensional modeling of a construction scene according to claim 7, characterized in that: In step S3-5, identifying the height range of the working surface includes: When the target building involves multiple working surfaces, human-computer interaction is used for identification according to different situations, and judgment is made according to the following method: If the working surface of the target building is a combination of steel frame and core tube, according to the original statistical results, the screening algorithm is imported for peak recognition, and two groups of working surface height ranges are output in turn, and step S3-1 is repeated for the two groups of working surface height ranges respectively; if the working surface of the target building is a core tube and an access platform, according to the original statistical results, only the peak ranges of the two groups of working surface heights with relatively low heights are retained as the core tube working surface height range.

10. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 9 is implemented.

Images