A method for automatically estimating construction progress based on multi-view matching

Abstract

The application discloses a kind of based on multi-view view matching's construction progress automatic estimation method, the method comprises: determining the three-dimensional model corresponding to construction site;The pre-designed building information model BIM is registered with the three-dimensional model, so that both are rotated and aligned;According to each construction equipment on the BIM, the corresponding position information of each construction equipment on the three-dimensional model is determined, and the three-dimensional model is individualized and split;Each equipment three-dimensional grid after the three-dimensional model individualization is reconstructed model surface denoising treatment by denoising method;Determine the perspective picture of each three-dimensional model, and determine the pixel matrix of the perspective picture;According to the pixel matrix, the construction progress of different equipment is determined.The scheme provided in the embodiment of the application can realize the automatic acquisition of the construction progress of the construction area, and reduce the interference of artificial subjectivity.

Description

Technical Field

[0001] This application relates to the field of machine vision technology, and in particular to a method for automatically estimating construction progress based on multi-view matching. Background Technology

[0002] Construction projects all require progress monitoring, and scientific monitoring methods are a prerequisite for ensuring the smooth implementation of construction. Traditional construction progress monitoring mainly involves manual on-site inspection and verification according to the construction schedule. This method is inefficient, labor-intensive, and has significant blind spots, and can no longer meet the needs of the fast-paced industry. With the development of drone technology, aerial photography using drones has emerged for construction progress monitoring. However, this method still has drawbacks, is prone to errors, and is labor-intensive.

[0003] In recent years, with the development of 3D reconstruction technology, methods have emerged that utilize drone aerial photography to reconstruct 3D mesh or point cloud models of actual scenes, and then compare these models with the actual scene models to obtain basic construction information. Currently, methods for estimating construction progress using reconstructed models can be broadly categorized as follows: 1) Obtaining point clouds of the construction site using drones and 3D reconstruction technology, breaking down the individual buildings in the construction area, obtaining the elevations of these buildings using construction drawings, and comparing the elevations to determine the corresponding construction progress; 2) Reconstructing point clouds of the construction site using drone oblique photography technology, voxelizing the point clouds obtained from the construction site, comparing changes in the construction site models at different stages, and monitoring construction progress; 3) Obtaining relevant scene images using cameras, analyzing the obtained point clouds, and registering the construction site building point clouds with the planned BIM point clouds for progress awareness.

[0004] At present, point cloud registration is often used for construction progress monitoring. However, when the construction scale is large, the on-site data is more complex, making it difficult to achieve accurate registration between the construction site point cloud and the BIM design model point cloud. Using point cloud to obtain basic construction information is prone to deviation and makes it difficult to measure construction progress. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention proposes a method for automatically estimating construction progress based on multi-view matching. This method enables the automatic acquisition of construction progress for large-scale construction areas, reducing subjective interference from human intervention. The specific solution is as follows:

[0006] This application provides a method for automatically estimating construction progress based on multi-view matching, the method comprising:

[0007] Determine the 3D model corresponding to the construction site;

[0008] Register the pre-designed Building Information Model (BIM) with the 3D model, and rotate and align the two.

[0009] Based on the various construction equipment in the BIM, determine the corresponding position information of each construction equipment on the three-dimensional model, and then decompose the three-dimensional model into individual units.

[0010] The method of denoising the three-dimensional mesh of each device after the three-dimensional model is individualized is used to achieve surface denoising of the reconstructed model;

[0011] Determine the images of the 3D model from various viewpoints, and determine the similarity matrix of the viewpoint images;

[0012] The construction progress of different equipment is determined based on the similarity numerical matrix.

[0013] Optionally, determining the three-dimensional model corresponding to the construction site includes:

[0014] Plan the flight path of the UAV, set up corresponding ground control points, acquire aerial images of the construction site taken by the UAV, and use the corresponding ground control points and images of related locations to realize the three-dimensional reconstruction of the construction site and generate a three-dimensional model of the construction site. The three-dimensional model contains the location information of different equipment at the construction site and the corresponding external construction information.

[0015] Optionally, registering the pre-designed Building Information Model (BIM) with the 3D model, and rotating and aligning them, includes:

[0016] Principal component analysis (PCA) is used to obtain the three principal directions of the pre-designed Building Information Model (BIM) and the 3D model.

[0017] Projecting each three-dimensional point of the three-dimensional model onto the three main directions yields the projections of each point.

[0018] Based on the extreme values ​​of the point projections, the eight corner points of the OBB bounding box of the three-dimensional model are obtained;

[0019] The BIM and the 3D model are aligned by rotation based on the eight corner points;

[0020] The BIM model and the bounding box of the 3D model are registered by scaling.

[0021] Optionally, the step of splitting the 3D model into individual units includes:

[0022] Each electrical and civil engineering device in the BIM is labeled individually, and the different devices in the BIM are presented as individual units;

[0023] Traverse each device in the BIM and determine the corresponding OBB bounding box of electrical and civil equipment to obtain the coordinate values ​​of 8 points;

[0024] Save the coordinates of the four points on the bottom surface of all devices that need to be individually separated, along with the name of the 3D model, to a txt file to obtain the polygon points that need to be segmented.

[0025] The corresponding 3ds Max script is invoked to perform individual segmentation of the 3D model based on the polygon points that need to be segmented.

[0026] Optionally, the step of calling the corresponding 3ds Max script to perform single-unit decomposition of the 3D model based on the polygon points that need to be segmented includes:

[0027] Read the txt file containing the 3D model and splitting module point set corresponding to the construction site that needs to be split;

[0028] Traverse all modules to be segmented in the file, obtain the set of segmentation points in the module to be segmented, and generate the polygonal face to be segmented from the set of points.

[0029] A 3D polygon is formed by looking upwards from the polygonal face to be split. The segmentation part is obtained by Boolean calculation. The points and triangular mesh faces that need to be split are selected to achieve automatic splitting function.

[0030] Optionally, determining the images of the 3D model from various perspectives includes:

[0031] Set the background color and size of the blender, the engine and camera to orthographic photography, and the corresponding camera's shooting range and height;

[0032] Iterate through the different equipment in the design BIM model and the different equipment on the construction site, and import them into the scene;

[0033] Obtain top view, front view, left view, and two 45-degree oblique views from different devices.

[0034] Optionally, determining the similarity matrix of the viewpoint images includes:

[0035] Input a unified color view of the original image I and the template image T;

[0036] The template T is slid in the image I. The sliding refers to moving the template image one pixel at a time and performing measurement calculations at different positions to determine the similarity value of a certain region of the original image in the current pixel, so as to obtain the similarity value corresponding to different pixel positions.

[0037] For different positions where template T covers I, the calculated metric values ​​are saved to the resulting image matrix R, which contains the matching metric value corresponding to each position;

[0038] Find the maximum and minimum values ​​in the result matrix R. Since the standard correlation matching criterion is used, the maximum value of the matching value can be regarded as the similarity value of the image matching.

[0039] Iterate through all the different perspectives of the equipment that need to measure the construction progress to obtain the corresponding similarity matrix s[i][j], where i represents the model number and j represents the view number.

[0040] Optionally, determining the construction progress of different devices based on the similarity value matrix includes:

[0041] The construction progress is estimated to be divided into three stages: construction completed, construction in progress, and construction not yet started.

[0042] When the similarity between the left view and the front view is greater than 60% and the overall average similarity is greater than 70%, the equipment construction is considered complete.

[0043] When the similarity between the left view and the front view is less than 30%, it is determined that the equipment is not yet in operation.

[0044] When the similarity between the left view and the front view is greater than 30%, the model is identified as a soil mound by training the corresponding soil mound image. If it is a soil mound, it is classified as equipment that has not yet been constructed; if it is not a soil mound, it is classified as equipment that is under construction.

[0045] This invention presents a method for automatically estimating construction progress based on multi-view matching. It obtains the construction progress of different equipment at the construction site by comparing and analyzing multi-view views between a 3D model reconstructed from oblique photogrammetry representing the construction site and a BIM design model. This invention improves matching accuracy through multiple processing of the two models, offering advantages such as high visualization and automation, while overcoming the drawbacks of existing construction progress estimation methods, such as high workload and error rate. Furthermore, this invention can be applied to drones with lower precision, reducing operating costs and demonstrating promising application prospects in construction scenarios. Attached Figure Description

[0046] Figure 1 This is the overall flowchart of the present invention;

[0047] Figure 2 This is a flowchart of the model registration module;

[0048] Figure 3 is a flowchart of the automatic module splitting process;

[0049] Figure 4 This is a flowchart of surface smoothing processing;

[0050] Figure 5 is a flowchart of the construction progress forecast;

[0051] Figure 6 shows a schematic diagram of the model after smoothing.

[0052] Figure 7. Schematic diagram of the equipment from multiple perspectives;

[0053] Figure 8 A numerical diagram illustrating the similarity of the completed construction sections;

[0054] Figure 9 Similarity values ​​for unfinished construction sections are shown in the schematic diagram. Detailed Implementation

[0055] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0056] Example 1

[0057] A method for automatically estimating construction progress based on multi-view matching includes the following steps:

[0058] Step 1: Plan the drone flight path, deploy corresponding ground control points (GCPs), acquire aerial images of the construction site, and use the GCPs and related images to perform 3D reconstruction of the construction site, generating a 3D mesh model. This model includes the location information of different equipment at the construction site and relevant external construction information.

[0059] Step 2: Register the planned BIM model with the on-site reconstruction model, and rotate and align the two.

[0060] Step 3: Based on the construction equipment on the corresponding planned BIM model, determine the corresponding location information, and use the 3ds Max script to split the site reconstruction model into individual units.

[0061] Step 4: After the on-site model is individualized, the relevant equipment is denoised using the L0 norm 3D mesh denoising method to reconstruct the model surface denoising.

[0062] Step 5: Use a Blender script to automatically acquire images from the five main viewpoints of the model;

[0063] Step 6: Obtain a similarity matrix of five perspectives for a device based on the modified image template matching algorithm;

[0064] Step 7: Determine the construction progress of different equipment based on relevant similarity thresholds;

[0065] Furthermore, the specific steps of the registration estimation method in step 2 are as follows:

[0066] Step 2.1: Use Principal Component Analysis (PCA) to obtain the three principal directions of the two models;

[0067] Step 2.2: Project the model points onto the three principal directions;

[0068] Step 2.3: Based on the extreme values ​​of the point projections, obtain the 8 corner points of the OBB bounding box of the model;

[0069] Step 2.4: Rotate and align the planned BIM model with the real-world reconstruction model;

[0070] Step 2.5: Register the bounding boxes of the two models by scaling.

[0071] Furthermore, step 3, which involves splitting the model into individual units, mainly requires calling the 3ds Max model splitting script. The specific steps are as follows:

[0072] Step 3.1: Label each electrical and civil equipment in the planned BIM model and realize the individual presentation of different equipment in the design model;

[0073] Step 3.2: Traverse each device in the BIM design model and determine the corresponding OBB bounding box of electrical and civil equipment, and obtain the coordinate values ​​of 8 points;

[0074] Step 3.3: Save the coordinates and model names of the four points on the bottom surface of all devices that need to be individually split into units to a txt file to obtain the polygon points that need to be segmented.

[0075] Step 3.4: Call the corresponding 3ds Max script to split the model;

[0076] Furthermore, the split script in 3ds Max is mainly divided into the following sub-steps;

[0077] Step 3.4.1: Read the txt file containing the construction site model to be split and the point set of the split modules;

[0078] Step 3.4.2: Traverse all modules to be segmented in the file, obtain the segmentation point set in the segmentation module, and generate the required polygonal faces from the point set;

[0079] Step 3.4.3: Form a 3D polygon from the polygonal face to be split upwards, obtain the segmented parts through Boolean calculation, select the points and triangular mesh faces that need to be split, and realize the automatic splitting function.

[0080] Furthermore, the specific steps for smoothing the reconstructed model in step 4 are as follows:

[0081] Step 4.1: Obtain the coordinates p of the corresponding vertex of the model that needs to be smoothed;

[0082] Step 4.2, set the optimized differential operator D(e) applicable to the mesh edges;

[0083] Step 4.3, introduce regularization terms

[0084] R(p) = (p1 – p2 + p3 – p4) 2 ;

[0085] Step 4.4, the optimization objective becomes

[0086] min p,δ |p–p*| 2 +α|R(p)| 2 +β|D(p)–δ| 2 +λ|δ|0;

[0087] Step 4.5, fix p and optimize δ, that is

[0088] min δ β|D(p)–δ| 2 +λ|δ|0, when δ i =0, otherwise δ i =D i ;

[0089] Step 4.6, fix δ and optimize p, i.e.

[0090] min p |p–p*| 2 +α|R(p)| 2 +β|D(p)–δ| 2 ;

[0091] Step 4.7: Iterate repeatedly until β ≥ 10. 3 .

[0092] Furthermore, the specific steps for automatically obtaining the view in step 5 are as follows:

[0093] Step 5.1: Set the background color and size of the blender, the engine and camera to orthographic photography, and the corresponding camera's shooting range and height;

[0094] Step 5.2: Iterate through the different equipment in the design BIM model and the different equipment on the construction site, and import them into the scene;

[0095] Step 5.3: Obtain the top view, front view, left view, and two 45-degree oblique views for different devices.

[0096] Furthermore, the specific steps of the algorithm for obtaining graph similarity values ​​through view matching in step 6 are as follows:

[0097] Step 6.1: Input a unified color view of the original image I and the template image T;

[0098] Step 6.2 involves sliding the template T across image I. Sliding here means moving the template image one pixel at a time, performing measurement calculations at different positions to determine the similarity value between the current pixel and a corresponding region in the original image. The measurement method here utilizes the standard relevance matching criterion (TM_CCOEFF_NORMED).

[0099]

[0100] R(x,y)=∑ x',y' (T'(x',y')×I'(x+x',y+y')),

[0101] This allows us to obtain similarity values ​​corresponding to different pixel locations;

[0102] Step 6.3: For different positions where template T covers I, save the calculated metric values ​​to the result image matrix R, where R contains the matching metric value corresponding to each position;

[0103] Step 6.4: Find the maximum and minimum values ​​in the result matrix R. Since the standard correlation matching criterion is used, the maximum value of the matching value can be regarded as the similarity value of the image matching.

[0104] Step 6.5: Using the above algorithm, traverse all the different perspectives of the equipment that need to measure the construction progress to obtain the corresponding similarity numerical matrix s[i][j], where i represents the model number and j represents the view number.

[0105] Furthermore, the specific steps for estimating the construction progress in step 7 are as follows:

[0106] Based on the construction progress, the estimated construction progress is divided into three stages: construction completed, construction in progress, and construction not yet started.

[0107] Prioritize determining whether construction equipment has completed its work, based on the following criteria:

[0108]

[0109] The equipment construction can only be considered complete when the similarity between the left view and the front view is greater than 60% and the overall average similarity is greater than 70%.

[0110] The criteria for distinguishing between equipment under construction and equipment not yet under construction are as follows:

[0111]

[0112] If the similarity between the left view and the front view is less than 30%, it indicates that the equipment is not yet in operation.

[0113] For cases where the similarity score is above 30%, the model is trained on the corresponding soil pile image to identify whether it is a soil pile. If it is a soil pile, it is classified as equipment that has not yet been constructed; if it is not a soil pile, it is classified as equipment that is under construction.

[0114] Example 2

[0115] The following provides a more detailed explanation of this application through a more specific approach.

[0116] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0117] This invention proposes a method for automatically estimating construction progress based on multi-view matching. Utilizing 3D reconstruction, model registration, automatic splitting, surface smoothing, and view matching, it achieves construction progress estimation by comparing a reconstructed construction site model with a planned BIM model. (Refer to...) Figure 1 The method for automatic estimation of construction progress based on multi-view matching of the present invention includes the following steps:

[0118] Step 1: Plan the drone flight path, deploy corresponding ground control points (GCPs), acquire aerial images of the construction site, and use the GCPs and related images to perform 3D reconstruction of the construction site, generating a 3D mesh model. This model includes the location information of different equipment at the construction site and relevant external construction information.

[0119] Step 2: Register the planned BIM model with the on-site reconstruction model, and rotate and align the two.

[0120] Step 3: Based on the construction equipment on the corresponding planned BIM model, determine the corresponding location information, and use the 3ds Max script to split the site reconstruction model into individual units.

[0121] Step 4: After the on-site model is individualized, the relevant equipment is denoised using the L0 norm 3D mesh denoising method to reconstruct the model surface denoising.

[0122] Step 5: Use a Blender script to automatically acquire images from the five main viewpoints of the model;

[0123] Step 6: Obtain a similarity matrix of five perspectives for a device based on the modified image template matching algorithm;

[0124] Step 7: Determine the construction progress of different equipment based on the relevant similarity threshold.

[0125] In the above embodiments, such as Figure 1 The diagram shows the overall framework of a method for automatic construction progress estimation based on multi-view matching. This method consists of four main modules: model registration, automatic segmentation, smoothing, and progress estimation. The model registration, automatic segmentation, and smoothing modules provide model data support for construction progress estimation, improving the accuracy of the final estimation. The model registration module primarily utilizes OOB (Out-of-Box) bounding boxes. Through position matching, it can accurately estimate the construction progress of equipment at corresponding locations, minimizing estimation errors. The automatic segmentation module automatically segments the equipment whose progress needs to be estimated at the construction site according to its location, facilitating subsequent comparison of individual equipment to obtain progress. The smoothing module smooths the surface of the reconstructed model. Due to surface roughness issues in the model reconstructed by UAV oblique photography, which affect progress estimation, surface smoothing is necessary. The progress estimation module uses a template matching algorithm to estimate the similarity matching values ​​of the model from different perspectives, and obtains the equipment construction status by evaluating the similarity values.

[0126] The specific process of the model registration module is as follows: Figure 2 As shown. The process for step 2 is as follows:

[0127] First, input the two models to be registered. Then, use Principal Component Analysis (PCA) to obtain the three principal directions of the two models: x, y, and z. Both models are triangular mesh models. Project the vertices of the two 3D models along the principal axes to obtain boundary values ​​and determine the eight corner points of the OBB bounding box. The two models will form two bounding boxes. Rotate and align the two bounding boxes. Since the height of the reconstructed model at the construction site will differ from the planned BIM model height, ground alignment (length and width registration) is used for registration. Finally, output the registered model for subsequent processing.

[0128] Figure 3 is a flowchart illustrating the model splitting module. The automatic model splitting module is mainly divided into two parts: the first part is obtaining the set of equipment points to be split in the planned BIM design model, and the second part is implementing the automatic splitting. The implementation steps for obtaining the set of equipment points to be split in the planned BIM design model in the first part are roughly as follows:

[0129] (1) Label the electrical and civil equipment in the planned BIM model one by one, and complete the individual presentation of different equipment in the design model;

[0130] (2) Traverse each device in the BIM design model and determine the corresponding electrical and civil equipment OBB bounding boxes to obtain the values ​​of 8 coordinate points;

[0131] (3) Save the coordinates and model names of the four points on the bottom surface of all devices that need to be split into individual units to a txt file to obtain the polygon points that need to be split.

[0132] The second part involves calling the splitting script in 3ds Max to achieve automatic splitting, which mainly consists of the following steps;

[0133] (1) Read the txt file containing the construction site model to be split and the point set of the split module;

[0134] (2) Traverse all modules to be segmented in the file, obtain the set of segmentation points in the module to be segmented, and generate the polygonal face to be segmented from the set of points;

[0135] (3) A three-dimensional polygon is formed from the polygon to be split upwards. The part to be split is calculated by Boolean, and the points and triangular mesh faces to be split are selected to achieve automatic splitting function.

[0136] The process of smoothing the reconstruction model is as follows Figure 4 As shown. This part mainly uses the L0 norm minimization-based triangular mesh denoising algorithm to achieve model smoothing. The main process of the algorithm is as follows:

[0137] (1) Obtain the corresponding vertex coordinates p of the model that needs to be smoothed;

[0138] (2) Set an optimized differential operator D(e) applicable to the mesh edges;

[0139] (3) Introducing regularization terms

[0140] R(p) = (p1 – p2 + p3 – p4) 2 ;

[0141] (4) The optimization objective becomes

[0142] min p,δ |p–p*| 2 +α|R(p)| 2 +β|D(p)–δ| 2 +λ|δ|0;

[0143] (5) Fix p and optimize δ, i.e.

[0144] min δ β|D(p)–δ| 2 +λ|δ|0, when δ i =0, otherwise δ i =D i ;

[0145] (6) Fix δ and optimize p, i.e.

[0146] min p |p–p*| 2 +α|R(p)| 2 +β|D(p)–δ| 2 ;

[0147] (7) Iterate repeatedly until β≥10 3 ;

[0148] (8) Complete the surface smoothing process and output the triangular mesh model p.

[0149] The basic flowchart of the construction progress estimation module is shown in Figure 5. The construction progress estimation module consists of three parts: automatic acquisition of multi-view images, view similarity matching, and construction progress estimation. The steps for the first part, automatic acquisition of multi-view images, are as follows:

[0150] (1) Set the background color and size of the blender, the engine and camera to orthographic photography, and the corresponding camera's shooting range and height;

[0151] (2) Iterate through the different equipment in the design BIM model and the different equipment on the construction site, and import them into the Blender scene set in (1);

[0152] (3) Call the functions in the blender library to automatically obtain the top view, front view, left view and two 45-degree oblique views of different devices, and save them as output.

[0153] The second part, the algorithm for obtaining image similarity values ​​through view matching, uses an image template matching algorithm. The specific process of this algorithm is as follows:

[0154] (1) Input a unified color view of the original image I and the template image T;

[0155] (2) The template T is slid across image I. Sliding here means moving the template image one pixel at a time, performing measurement calculations at different positions to determine the similarity value between the current pixel and a certain region in the original image. The measurement method here uses the standard relevance matching criterion (TM_CCOEFF_NORMED).

[0156]

[0157] R(x,y)=∑ x',y' (T'(x',y')×I'(x+x',y+y')),

[0158] This allows us to obtain similarity values ​​corresponding to different pixel locations;

[0159] (3) For different positions where template T covers I, the calculated metric values ​​are saved to the result image matrix R, which contains the matching metric value corresponding to each position;

[0160] (4) Find the maximum value in the result matrix R. Since the standard correlation matching criterion is used, the maximum value of the matching value can be regarded as the similarity value of the image matching.

[0161] (5) By using the above algorithm to traverse all the different perspectives of the equipment that need to measure the construction progress, the corresponding similarity matrix s[i][j] can be obtained, where i represents the model number and j represents the view number.

[0162] The third part involves achieving a final estimate of the construction schedule, matching the similarity score with the actual construction progress. Based on the construction situation, the construction schedule estimate can be divided into three stages: the completion stage, the ongoing construction stage, and the stage before construction begins. The specific process is as follows:

[0163] (1) Input the similarity matching matrix s

[0164] (2) Prioritize distinguishing whether construction equipment construction is completed, based on the following criteria:

[0165]

[0166] The equipment construction can only be considered complete when the similarity between the left view and the front view is greater than 60% and the overall average similarity is greater than 70%.

[0167] (3) The criteria for distinguishing between equipment under construction and equipment not yet under construction are as follows:

[0168]

[0169] If the similarity between the left view and the front view is less than 30%, it indicates that the equipment is not yet in operation.

[0170] (4) For cases where the similarity value is above 30%, the model is identified as a soil mound by training the corresponding soil mound image. If it is a soil mound, it is classified as equipment that has not yet been constructed. If it is not a soil mound, it is classified as equipment that is under construction.

[0171] (5) Based on the above, the equipment is displayed using visualization, and a corresponding construction progress monitoring report is generated.

[0172] The general effects of this invention in the embodiments are as follows:

[0173] (1) The surface smoothing of the model is shown in Figure 6. The rough three-dimensional model is input as shown in Figure 6(a). After processing by the model surface smoothing module, the smooth three-dimensional model in Figure 6(b) is finally obtained. The surface smoothing performance is good.

[0174] (2) The final effect of the automatic acquisition of the 3D model view is shown in Figure 7. The model is centered and its five perspective images are obtained. The generated images match the original model well, and the automation effect is good.

[0175] (3) Figure 8 and Figure 9 The basic effect of view matching in obtaining image similarity values ​​is roughly demonstrated. The matching effect is quite good for models with completed construction, with a similarity of around 80%. The similarity is generally low for models with incomplete construction. Therefore, by determining the threshold, the construction progress can be basically monitored.

[0176] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

Claims

1. A method for automatic construction progress estimation based on multi-view matching, characterized in that, The method comprises: determining the three-dimensional model corresponding to the construction site; aligning the pre-designed building information model BIM with the three-dimensional model, including: obtaining the three principal directions of the pre-designed building information model BIM and the three-dimensional model by principal component analysis PCA; projecting each three-dimensional point of the three-dimensional model in the three principal directions to obtain a point projection; obtaining eight corner points of the OBB bounding box of the three-dimensional model according to the extreme values of the point projection; rotating the BIM and the three-dimensional model based on the eight corner points; and realizing the length-width registration of the BIM and the three-dimensional model by scaling; determining the corresponding position information of each construction equipment on the three-dimensional model according to each construction equipment on the BIM, and performing monomerization splitting on the three-dimensional model; the monomerization splitting of the three-dimensional model comprises: labeling the electrical and soil construction equipment in the BIM one by one, and performing monomerization presentation on different equipment in the BIM; traversing each equipment in the BIM, and determining the OBB bounding box of the corresponding electrical and soil construction equipment to obtain eight coordinate point values; saving the coordinate values of the bottom surface direction of the four points of all equipment that need to be monomerized and split to a txt file to obtain polygon points that need to be segmented; calling a corresponding 3dmax script to monomerize and split the three-dimensional model based on the polygon points that need to be segmented; reconstructing the model surface denoising processing by the three-dimensional grid denoising method of each equipment after monomerization of the three-dimensional model; determining the perspective pictures of the three-dimensional model, and determining the similarity value matrix of the perspective pictures, comprising: inputting the uniform color view of the original image I and the template image T; sliding the template image T in the input original image I, which means moving the template picture one pixel at a time, calculating the similarity value of the corresponding area of the original image in the current pixel at different positions, so as to obtain the similarity value corresponding to different pixel positions; for different positions of the template image T covering on the input original image I, save the calculated metric value to the result image matrix R, which contains the matching metric value corresponding to each position; find the maximum value in the result image matrix R, since the standard correlation matching criterion is used to measure, the maximum value of the matching value can be identified as the similarity value of image matching; traverse all equipment with different perspectives that need to measure the construction progress to obtain the corresponding similarity value matrix s[i][j], wherein i represents the model number and j represents the view number; determining the construction progress of different equipment according to the similarity value matrix.

2. The method of claim 1, wherein, The method comprises: The application discloses a method for planning a flight route of a UAV, laying corresponding image control points, obtaining aerial images of a construction site shot by the UAV, realizing three-dimensional reconstruction of the construction site by using the corresponding image control points and images of related positions, generating a three-dimensional model corresponding to the construction site, and containing position information of different equipment in the construction site and corresponding external construction information in the three-dimensional model.

3. The method of claim 1, wherein, The corresponding 3dmax script is called to perform single-body split on the three-dimensional model based on polygon points that need to be split, including: reading a three-dimensional model corresponding to the construction site that needs to be split and a txt file of a split module point set; traversing all to-be-split modules in the file, obtaining a split point set in the to-be-split module, generating a polygon face that needs to be split from the point set; forming a three-dimensional polygon from the polygon face that needs to be split, obtaining a split part by Boolean calculation, selecting a point and a triangular mesh face that need to be split, and realizing an automatic split function.

4. The method of claim 1, wherein, The three-dimensional model is determined to obtain a picture of each view, including: setting a background color and size of blender, an engine and a camera to be orthogonal photography, and a photography range and height of the corresponding camera; traversing different equipment of a designed BIM model and different equipment of a construction site, and importing the equipment into a scene; obtaining a top view, a front view, a left view and two 45-degree oblique views of the different equipment.

5. The method of claim 1, wherein, The construction progress of the different equipment is determined according to the similarity value matrix, including: dividing construction progress estimation into three stages: construction completion, construction in progress and construction not yet started; when the similarity of the left view and the front view is greater than 60% and the overall similarity average is greater than 70%, it is determined that the equipment construction is completed; when the matching similarity of the left view and the front view is below 30%, it is determined that the equipment is not started yet; when the matching similarity of the left view and the front view is above 30%, whether the model is a soil pile is identified by training corresponding soil pile pictures, if the model is a soil pile, it is classified as equipment not yet started, and if the model is not a soil pile, it is classified as equipment under construction.

Images