An Automatic Layout Method for Construction Sites Based on Generative Adversarial Networks

Abstract

The present invention discloses an automatic construction site layout method based on a generative adversarial network. The site contour and the contour of the proposed building are obtained from the top view of the actual construction site of the building, and combined with the actual engineering data, they are input into the generator of the generative adversarial network to generate a construction site layout plan for the building. The true construction site layout plan of the building is input into the discriminator of the generative adversarial network to train the generator. The automatic generation of the construction site layout plan for the proposed building is realized through the trained generative adversarial network. The method provided by the present invention uses machine learning technology to realize the automatic generation of the site layout plan, can quickly generate a large number of feasible plans, greatly reduce the human resources and time consumed in designing the construction site layout plan, and the safety and rationality of the generated plan are guaranteed to a certain extent, providing strong technical support for the intelligent development of building construction.

Description

Technical Field

[0001] The present invention belongs to the field of construction, and more specifically, relates to an automatic construction site layout method based on a generative adversarial network. Background Art

[0002] The management and civilized construction of the construction site are the main parts of work safety, an important symbol of modern construction, and an inevitable requirement for realizing the vision of an intelligent construction site in the future. Formulating a reasonable and comprehensive construction site layout plan has far-reaching significance for improving operation efficiency and ensuring project safety and quality.

[0003] Currently, the existing means of construction site layout are to layout and divide the construction site based on experience and intuitive feelings on the basis of referring to and complying with national and local regulations. Although such a site layout plan can meet the minimum safety requirements of the project, there are also several significant problems - first, the method of formulating the plan based on experience is difficult to take into account all possible situations that the project may face in advance and is not comprehensive; second, the plan formulated based on the minimum safety requirements is difficult to ensure its rationality during implementation, and the efficiency of project construction is not high; finally, the plan formulated in this way has too high randomness and is limited by the personal experience of engineers. While having greater risks, it also places high requirements on the professional level of engineers, consuming time and manpower. Summary of the Invention

[0004] In view of the above defects or improvement requirements of the prior art, the present invention provides an automatic construction site layout method based on a generative adversarial network, thereby solving technical problems such as the existing construction site layout method relying on manual experience and low efficiency.

[0005] To achieve the above object, according to the first aspect of the present invention, there is provided an automatic construction site layout method based on a generative adversarial network, the method comprising:

[0006] Training stage:

[0007] Input images including building outlines, corresponding construction site outlines, and building project information into the generator of the generative adversarial network. The generator encodes and decodes the images to generate a construction site layout plan of the building and inputs it into the discriminator, and trains the generator with the goal of maximizing the similarity between the construction site layout plan and the true construction site layout plan.

[0008] Application stage:

[0009] Input the proposed building outline, corresponding construction site outline, and building project information into the trained generative adversarial network model to obtain the construction site layout plan of the proposed building.

[0010] The construction site layout plan includes road node classification information and site area classification information.

[0011] Preferably, image processing is performed on the top view of the construction site of the building to obtain the building outline, construction site outline, and true construction site layout plan of the building.

[0012] Preferably, contour recognition is performed on the top view of the construction site of the building to obtain the building outline and construction site outline of the building.

[0013] Preferably, the construction site layout plan includes road node classification information and site area classification information.

[0014] Preferably, the road node classification information includes: four types of single intersections, four types of L-shaped intersections, four types of T-shaped intersections, and crossroads;

[0015] The site area classification information includes: construction area, office area, living area, greening area, education area, processing area, and material yard.

[0016] Preferably, the building engineering information includes the number of people at the peak labor period and the maximum stacking amounts of steel bars, lumber, concrete, and prefabricated components during the construction period.

[0017] According to the second aspect of the present invention, there is provided a construction site automatic layout system based on a generative adversarial network, including: a computer-readable storage medium and a processor;

[0018] The computer-readable storage medium is used to store executable instructions;

[0019] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the method as described in the first aspect.

[0020] Generally speaking, compared with the prior art by the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

[0021] The automatic construction site layout method based on the generative adversarial network provided by the present invention obtains the site contour and the contour of the proposed building from the top view of the actual construction site of the building, combines the actual engineering data, and inputs them into the generator of the generative adversarial network to generate the construction site layout plan of the building. The true construction site layout plan of the building is input into the discriminator of the generative adversarial network to train the generator. Through the trained generative adversarial network, the automatic generation of the construction site layout plan for the proposed building is realized. The method provided by the present invention uses machine learning technology to automatically generate the site layout, can quickly generate a large number of feasible plans, greatly reduce the human resources and time consumed in designing the construction site layout plan, and the safety and rationality of the generated plan are guaranteed to a certain extent, providing strong technical support for the intelligent development of building construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 FIG. is a schematic diagram of the training stage of the automatic construction site layout method based on the generative adversarial network provided by the embodiment of the present invention;

[0023] Figure 2 FIG. is a schematic diagram of the application stage of the automatic construction site layout method based on the generative adversarial network provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0025] The embodiment of the present invention provides an automatic construction site layout method based on the generative adversarial network, which is characterized in that the method includes:

[0026] Training stage:

[0027] An image including the building contour, the construction site contour corresponding to the building contour (i.e., the site layout data map of the building) and the building engineering information corresponding to the building contour is input into the generator of the generative adversarial network. The generator encodes and decodes the image to generate the construction site layout map of the building and inputs it into the discriminator, and trains the generator with the goal of maximizing the similarity between the construction site layout map and the true construction site layout map;

[0028] Application stage:

[0029] Input an image including the outline of a proposed building, the outline of the construction site corresponding to the outline of the proposed building (i.e., the layout data map of the proposed building), and the building engineering information corresponding to the outline of the proposed building into the trained generative adversarial network model to obtain the layout plan of the construction site of the proposed building.

[0030] Preferably, perform image processing on the top view of the construction site of the building to obtain the building outline, the construction site outline, and the actual layout plan of the construction site of the building.

[0031] Preferably, perform contour recognition on the top view of the construction site of the building based on the image contour recognition algorithm to obtain the building outline and the site outline of the building.

[0032] Preferably, the building outline includes the vertex coordinates of the building outline, and the construction site outline includes the vertex coordinates of the construction site outline.

[0033] Preferably, the layout plan of the construction site includes road node classification information and site area classification information.

[0034] Specifically, the training data is obtained from the top view images of actual construction projects, and can also be obtained by processing the top view of the construction site of the building (for example: aerial photos of the construction site of the building).

[0035] Before the training stage, it also includes a training data preprocessing stage: perform image processing on the top view of the construction site of the building to obtain the building outline, the site outline, and the actual layout plan of the construction site of the building; wherein, the actual layout plan of the construction site of the building includes site road node classification information and site area classification information.

[0036] It can be understood that when obtaining the training data, the building refers to a building that already has a qualified and verified construction site layout plan (i.e., the actual construction site layout plan), and the construction site of the building has been arranged according to the actual construction site layout plan. Therefore, by performing corresponding processing on the top view of the construction site of the building, the key information (i.e., road node classification information and site area classification information) in the actual construction site layout plan of the building can be obtained. That is, whether it is the layout plan of the construction site generated by the generator or the actual layout plan of the construction site, both include road node classification information and site area classification information.

[0037] Among them, perform processing on the top view of the construction site of the building based on the image contour recognition algorithm, extract the building outline and the site outline, and obtain the vertex coordinates of the polygon contour. That is, identify the outer contour of the site and the outer contour of the proposed building, and the recognition result is represented by the vertex coordinates of the contour polygon.

[0038] Process the top view of the construction site of the building through a classification algorithm to classify the road nodes in the site; perform pixel-by-pixel classification on the in-site area to obtain the classification result of the site area.

[0039] That is, the road nodes are obtained by performing pixel-by-pixel classification on the top view of the construction site of the building to obtain the classification result of the road nodes; the classification of the site area is obtained by performing pixel-by-pixel classification on the top view of the construction site of the building to obtain the classification result of the site area.

[0040] The classification results of the road nodes and the area classification together form the real site layout plan, which will be input into the discriminator; according to the vertex coordinates of the building outline and the site outline, establish a site layout data map including the building outline and the site outline and input it into the generator for generating the site layout plan;

[0041] Preferably, the classification information of the road nodes includes: four types of single intersections, four types of L-shaped intersections, four types of T-shaped intersections and crossroads.

[0042] Specifically, as shown in Table 1, the road nodes are divided into 13 categories in total.

[0043] Table 1 Road Node Classification Table

[0044]

[0045]

[0046] The classification information of the site area includes: construction area, office area, living area, greening area, education area, processing area, material yard.

[0047] Specifically, perform pixel-by-pixel classification on the in-site area, and divide it into 7 categories including construction area, office area, living area, greening area, education area, processing area, and material yard.

[0048] Preferably, the building engineering information includes the number of people at the peak labor period and the maximum stacking quantities of steel bars, timbers, concrete and prefabricated components during the construction period.

[0049] Specifically, the actual engineering information (i.e., the building engineering information) includes the number of people at the peak labor period, the maximum material stacking quantities of steel bars, timbers, concrete and prefabricated components during the construction period, and is summarized as a 5-dimensional vector as the input.

[0050] During the generative adversarial training phase, the input of the generator is the actual engineering data (vector) and the site layout data map including the vertex coordinates of the proposed building outline and the site outline, and the generated site layout plan (road layout and site zoning image) is output; the discriminator will input the real site layout plan and the generated site layout plan at the same time, judge their authenticity and output the results to feedback to the generator and the discriminator, and then perform this process again after each optimizes and adjusts; the optimization process will end when the generator and the discriminator reach the Nash equilibrium.

[0051] In the actual site layout application phase, during the process of the actual site layout application, the input end of the model should be the outer contour of the construction site to be designed, the outer contour of the proposed building, and the actual engineering information (the number of workers at the peak of employment, the maximum material stacking amounts of steel bars, lumber, concrete, and precast components during the construction period); the output is the road layout and site zoning image.

[0052] In order to input the constraint information required for site layout into the network at the input end of the network, the network input of the present invention is divided into three parts. One part is a 5D feature vector covering the actual engineering information, and the five dimensions are the number of workers at the peak of employment, the maximum material stacking amounts of steel bars, lumber, concrete, and precast components during the construction period; the remaining two parts are the contour point coordinates of the proposed building and the contour point coordinates of the entire site;

[0053] In the method provided by the embodiment of the present invention, the real site layout plan used is represented by integrating the road node classification and the regional pixel-by-pixel classification results - the entire site is divided according to the road nodes, and the average value of the pixel-by-pixel classification probabilities in each region is taken to return the category of the entire region;

[0054] In the method provided by the embodiment of the present invention, the generated site layout plan used is generated by three inputs (5D feature vector and two contour point coordinates). The generated site layout plan is in the form of the roads in the site and the various partitions divided by the roads.

[0055] During the training process, for the collected top views of the construction site and their actual engineering data. After all the data are processed, 20% is randomly selected as the test set, and the remaining 80% is used for the training of the generative adversarial network.

[0056] The site layout plan generated by the network is represented by a simple function as:

[0057] x g =G(z b ),z b is the input contour point coordinates and feature vector

[0058] The generator G actually consists of two parts, namely, an encoder G for extracting the feature of the site layout rule logic diagram from images including building outlines and corresponding construction site outlines ec and a decoder G for generating a site layout plan image based on the features dc , and the generated result is x g

[0059] In contrast, the real data of the site layout is expressed as:

[0060] x r ∈X train

[0061] Then, for the discriminator, its input is the data jointly composed of x g and x r .

[0062] Preferably, the discriminator in the generative adversarial network of the embodiment of the present invention adopts the discriminator in patch GAN that can analyze discrete blocks. When the function value returned by the discriminator D is larger, the probability that x comes from the training data set X train is greater; on the contrary, it is very likely to be the data generated by the generator G

[0063] The purpose of the generative adversarial network is to reach the Nash equilibrium after repeated iterations. In such an iterative process, the following requirements are imposed on the discriminator D:

[0064]

[0065] That is:

[0066]

[0067] The following requirements are imposed on the generator G:

[0068] x g D(G(z))→0

[0069] To sum up, the above optimization objectives are written as an expression:

[0070]

[0071] Compared with traditional machine learning that requires defining a specific loss function, defining an unsupervised adversarial learning loss by using the generative adversarial network avoids the problem of fuzzy output results caused by improper selection of the loss function

[0072] In the process of training the above network, the specific training and optimization process is as Figure 1As shown below. First, input the real site layout plan into the discriminator D, and mark that the value it returns should be true; then input a site layout design document into the generator G. After the generator generates a layout plan, input it into the discriminator D, and mark that the value it returns should be false; after completing the above steps, feedback the result output by the discriminator D to the generator G, and adjust G according to the gap between the current output result and true.

[0073] The above process is a cycle of the adversarial generative network training. After a predetermined number of cycles, check whether the generative adversarial network reaches the Nash equilibrium and check whether the error value it returns meets the requirements. After completing the above work, the trained GAN model (which can be called Site-GAN) can be saved.

[0074] In the application stage, as Figure 2 shown, input the site layout design document into the trained GAN model to output the site layout plan of the proposed building.

[0075] In summary, the method provided by the present invention generates a site layout plan with the adversarial generative network as the core. The input of this method is actual engineering data, the outline of the proposed building, and the outline of the site. The latter two are input after being reflected on the site layout document in the form of the vertex coordinates of their polygon outlines; the output is the generated site layout plan of the proposed building.

[0076] The formats of the site layout document and the construction site layout plan can both be jpg images.

[0077] Finally, in order to evaluate the use of the pre-prepared test dataset X test test the network - select appropriate evaluation parameters and indicators (scheme rationality, scheme diversity, compatibility with the original design document) to evaluate the rationality of the site layout plan output by the Site-GAN model.

[0078] The automatic construction site layout method based on the generative adversarial network provided by the present invention can automatically generate a large number of site layout plans according to the input site layout design data after training, and present the results in a visual way. The training data preprocessing process is used to extract site layout information and site layout design data from actual photos; the generative adversarial training process continuously optimizes between the generator and the discriminator to achieve the effect of generating site layout plans that meet the conditions; the actual site layout application can generate a large number of site layout plans through the trained generator. For the evaluation of the model's effect, it mainly evaluates the rationality, diversity, and compatibility with the original design data of the generated plans. This method starts from the site contour, the contour of the proposed building, and the actual engineering data, and uses machine learning technology to automatically generate the site layout, aiming to quickly generate a large number of feasible plans. Using this method can greatly reduce the human resources and time consumed in designing the construction site layout plan, and the safety and rationality of the generated plan are guaranteed to a certain extent. It provides strong technical support for the intelligent development of building construction.

[0079] An embodiment of the present invention provides an automatic construction site layout system based on a generative adversarial network, including: a computer-readable storage medium and a processor;

[0080] The computer-readable storage medium is used to store executable instructions;

[0081] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the method described in any of the above embodiments.

[0082] Those skilled in the art can easily understand that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An automatic construction site layout method based on a generative adversarial network, characterized in that, the method comprises: Training stage: Input images including building outlines, corresponding construction site outlines and corresponding construction project information into the generator of the generative adversarial network. The generator encodes and decodes the images to generate a construction site layout plan of the building and inputs it into the discriminator. The generator is trained with the goal of maximizing the similarity between the construction site layout plan and the true construction site layout plan of the building; Application stage: Input images including the outlines of the proposed building, the corresponding construction site outlines and the corresponding construction project information into the trained generative adversarial network model to obtain the construction site layout plan of the proposed building; wherein, the construction site layout plan includes road node classification information and site area classification information; the construction project information includes the number of workers during the peak labor period and the maximum stacking quantities of steel bars, lumber, concrete and precast components during the construction period; The generator includes an encoder for extracting the feature of the site layout rule logic diagram from the images of the building outline and the corresponding construction site outline, and a decoder for generating the construction site layout plan based on this feature.

2. The method according to claim 1, characterized in that, Image processing is performed on the top view of the construction site of the building to obtain the building outline, the construction site outline and the true construction site layout plan of the building.

3. The method according to claim 2, characterized in that, Contour recognition is performed on the top view of the construction site of the building to obtain the building outline and the construction site outline of the building.

4. The method according to any one of claims 1-3, characterized in that, The building outline includes the vertex coordinates of the building outline, and the construction site outline includes the vertex coordinates of the construction site outline.

5. The method according to claim 1, characterized in that, The road node classification information includes: four types of single intersections, four types of L-shaped intersections, four types of T-shaped intersections and crossroads; The site area classification information includes: construction area, office area, living area, greening area, education area, processing area, material yard.

6. An automatic construction site layout system based on a generative adversarial network, characterized in that, comprises: A computer-readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the method according to any one of claims 1-5.

Images