Intelligent building measurement system and method based on artificial intelligence

Abstract

The present invention relates to the field of building measurement, and in particular to an intelligent building measurement system and method based on artificial intelligence. First, multi-angle point cloud data of all measurement points of the building to be measured is collected, the multi-angle point cloud data is mapped to a depth image, and the implicit representation of the nodes in the depth image is extracted; then, the depth image is adaptively disparity optimized, and the point cloud data is processed using a graph convolutional network to obtain a new feature representation of each point, and the quality of the feature is further improved using a self-attention mechanism; finally, a measurement point recognition neural network model is constructed to identify the points to be measured corresponding to the feature information of the point cloud image, and obtain the measurement data of the actual building points to be measured. The problem that the prior art does not efficiently process and optimize the measurement data, resulting in noise, occlusion, artifacts, etc. in the data, as well as the problem that the feature extraction of the point cloud data is not accurate enough and there are difficulties in automatically identifying the measurement points, requiring manual intervention, is solved.

Description

Technical Field

[0001] The present invention relates to the field of building measurement, and in particular to an intelligent building measurement system and method based on artificial intelligence. Background Art

[0002] With the acceleration of urbanization and the rapid development of the construction industry, the demand for accurate measurement of buildings is increasing. Traditional building measurement methods often rely on manual operation, which is not only inefficient but also prone to errors, making it difficult to meet the high-precision and high-efficiency requirements of modern construction projects. In addition, with the rapid development of big data, cloud computing and artificial intelligence technologies, how to apply these advanced technologies to building measurement and improve the accuracy and efficiency of measurement has become the focus of the industry.

[0003] In recent years, point cloud data and depth image technology have been widely used in building measurement. They can provide rich 3D information to help engineers better understand and analyze building structures. However, how to extract useful features from a large amount of point cloud data and how to effectively apply these features to building measurement remains a challenge.

[0004] my country's patent application number: CN202210500351.3, publication date: 2022.06.10, discloses a building measurement method based on an intelligent robot. By installing a portable 3D laser scanning real-scene measurement device on an intelligent robot platform with autonomous navigation capabilities, based on the building CAD plan, it automatically divides the area and measurement task points, generates navigation and scanning paths, and after the robot moves autonomously to the target point, it starts autonomous measurement work, automatically completes the three-dimensional data collection of each room in the building and the output of the building quality report, realizes the autonomous movement and actual measurement of the indoor space of the building, and improves the efficiency of building measurement.

[0005] However, the above technology has at least the following technical problems: the existing technology does not efficiently process and optimize the measurement data, resulting in noise, occlusion, motion, artifacts and other problems in the data, affecting the accuracy of the measurement results; the feature extraction of point cloud data is not precise enough, resulting in instability and inaccuracy in the measurement results; there are difficulties in automatically identifying measurement points, and manual intervention is required. Summary of the invention

[0006] The embodiments of the present application provide an intelligent building measurement system and method based on artificial intelligence, which solves the problems that the existing technology does not efficiently process and optimize measurement data, resulting in noise, occlusion, motion, artifacts, etc. in the data, affecting the accuracy of the measurement results; the feature extraction of point cloud data is not accurate enough, resulting in instability and inaccuracy of the measurement results; there are difficulties in automatically identifying measurement points, requiring manual intervention. The intelligent building measurement system based on artificial intelligence is realized, and the accuracy, efficiency and adaptability of building measurement are improved through advanced data processing, feature extraction and deep learning technologies.

[0007] This application provides an intelligent building measurement system and method based on artificial intelligence, which specifically includes the following technical solutions:

[0008] The intelligent building measurement system based on artificial intelligence includes the following parts:

[0009] A data acquisition module, a conversion module, a first feature extraction module, a parallax optimization module, a second feature extraction module, a measurement point recognition module and a measurement output module;

[0010] The conversion module is used to map the multi-angle point cloud data to the depth image, and to superimpose more than one depth image to form the density of the depth image. Each point on the point cloud data is mapped to a valid pixel of the depth image. The conversion module is connected to the parallax optimization module by data transmission;

[0011] The first feature extraction module is used to encode the point cloud data using a variational graph autoencoder to extract a latent variable representation of each point as a feature, and the first feature extraction module is connected to the disparity optimization module by means of data transmission;

[0012] The disparity optimization module is used to perform disparity optimization on the depth image, remove noise points, including occlusion, motion, and artifacts, and perform adaptive disparity optimization. The disparity optimization module is connected to the second feature extraction module by means of data transmission;

[0013] The second feature extraction module is used to further process the point cloud data using a graph convolutional network to obtain a new feature representation of each point, and use a self-attention mechanism to further improve the quality of the feature, and the second feature extraction module is connected to the measurement point identification module by means of data transmission;

[0014] The measuring point recognition module is used to construct a measuring point recognition neural network model, and extracts feature information from a point cloud image of a point to be measured taken by a laser radar as a training sample. The training sample is input into the measuring point recognition neural network model, and after intelligent learning, the measuring point ID corresponding to the training sample is output. Through deep training and optimization of the measuring point recognition neural network model, the measuring point recognition neural network can recognize the point to be measured corresponding to the feature information of the point cloud image within a preset accuracy range. The measuring point recognition module is connected to the measurement output module by means of data transmission.

[0015] The intelligent building measurement method based on artificial intelligence includes the following steps:

[0016] S100: Collect multi-angle point cloud data of all measurement points of the building to be measured, map the multi-angle point cloud data to the depth image, and extract the implicit representation of the nodes in the depth image through the variational graph autoencoder;

[0017] S200: Adaptive parallax optimization of depth images;

[0018] S300: Use graph convolutional networks to further process point cloud data to obtain new feature representations for each point, and use self-attention mechanisms to further improve feature quality;

[0019] S400: Constructing a neural network model for measuring point recognition, identifying the points to be measured corresponding to the feature information of the point cloud image, and obtaining measurement data of the points to be measured of the actual building.

[0020] Preferably, the S100 specifically includes:

[0021] The multi-angle point cloud data is mapped to a depth image, and more than one depth images are superimposed to form the density of the depth image. The density of the depth image is the same as the number of multi-angle point cloud images, and each point on the point cloud data is mapped to an effective pixel of the depth image. A variational graph autoencoder is introduced to extract and match features, encode the point cloud data, obtain the latent variable representation of each point, and then use the latent variable representation for feature matching.

[0022] Preferably, the S100 further includes:

[0023] An implicit representation is generated for each node based on the structure and node features of the graph, and the adjacency matrix of the graph is reconstructed based on the implicit representation. The goal of the variational graph autoencoder is to maximize the lower bound of evidence and maximize the similarity between the reconstructed adjacency matrix and the true adjacency matrix.

[0024] Preferably, the S200 specifically includes:

[0025] An adaptive disparity optimization algorithm is used, which first calculates the depth value of each pixel point, then calculates local and global disparity gradients, dynamically adjusts disparity optimization parameters, and uses the adjusted disparity optimization parameters to perform disparity optimization.

[0026] Preferably, the S200 further includes:

[0027] A depth value is calculated for each pixel, where the depth value is based on the distance from the pixel to the center of the radar; the depth change rate of each pixel is quantified, the disparity gradient of each pixel is calculated, the edge of the depth change is identified, and the disparity gradient is calculated; and the global disparity gradient is further calculated.

[0028] Preferably, the S200 further includes:

[0029] According to the global disparity gradient, disparity optimization parameters are dynamically adjusted, and the adjusted parameters are used to perform disparity optimization; the disparity optimization optimizes and cleans the initially extracted features.

[0030] Preferably, the S300 specifically includes:

[0031] A graph convolutional network is used to process point cloud data to obtain a new feature representation for each point, and a weight is generated for each point using the new feature representation.

[0032] Preferably, the S300 further includes:

[0033] Use the self-attention mechanism to process the point cloud depth image data and obtain the weight of each point.

[0034] Preferably, the S400 specifically includes:

[0035] A measurement point recognition neural network model is constructed, and the extracted high-quality depth image features are used as training samples. The training samples are input into the measurement point recognition neural network model, and the measurement point ID corresponding to the training samples is output. Through deep training and optimization of the measurement point recognition neural network model, the measurement point recognition neural network can identify the points to be measured corresponding to the feature information of the point cloud image within a preset accuracy range.

[0036] Beneficial effects:

[0037] The multiple technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0038] 1. By using LiDAR to collect multi-angle point cloud data of the building to be measured, combined with depth image technology, high-precision measurement of the building can be achieved; multi-angle point cloud data of all measurement points of the building to be measured is obtained, and the multi-angle point cloud data is mapped to the depth image. Multiple depth images are superimposed to form the density of the depth image, which improves the accuracy of the collected data, enriches the comprehensive acquisition of building data, and provides a data basis for the accuracy of subsequent measurements;

[0039] 2. Through the variational graph autoencoder and adaptive disparity optimization algorithm, the quality and accuracy of the data can be effectively improved, thereby ensuring the reliability of the measurement results; combined with the graph convolutional network and self-attention mechanism, meaningful features can be efficiently extracted from the point cloud data, providing strong support for the subsequent measurement point identification; by constructing a measurement point identification neural network model, it can automatically identify the points to be measured corresponding to the feature information of the point cloud image, greatly improving the measurement efficiency; by dynamically adjusting the disparity optimization parameters, it can adapt to different measurement environments and conditions to ensure that the best measurement results can be obtained in various situations.

[0040] 3. The technical solution of this application can effectively solve the problems that the existing technology does not efficiently process and optimize the measurement data, resulting in noise, occlusion, motion, artifacts and other problems in the data, affecting the accuracy of the measurement results; the feature extraction of point cloud data is not accurate enough, resulting in instability and inaccuracy of the measurement results; there are difficulties in automatically identifying the measurement points, requiring manual intervention. It can realize an intelligent building measurement system based on artificial intelligence, and improve the accuracy, efficiency and adaptability of building measurement through advanced data processing, feature extraction and deep learning technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a structural diagram of the intelligent building measurement system based on artificial intelligence described in this application;

[0042] Figure 2 This is a flow chart of the intelligent building measurement method based on artificial intelligence described in this application; DETAILED DESCRIPTION

[0043] The embodiments of the present application provide an intelligent building measurement system and method based on artificial intelligence, which solves the problems that the prior art does not efficiently process and optimize measurement data, resulting in noise, occlusion, motion, artifacts and other problems in the data, affecting the accuracy of the measurement results; the feature extraction of point cloud data is not precise enough, resulting in instability and inaccuracy of the measurement results; there are difficulties in automatically identifying measurement points, which requires manual intervention.

[0044] The technical solution in the embodiment of the present application is to solve the above problems, and the overall idea is as follows:

[0045] By using LiDAR to collect multi-angle point cloud data of the building to be measured and combining it with depth image technology, high-precision measurement of the building can be achieved; multi-angle point cloud data of all measurement points of the building to be measured is obtained, and the multi-angle point cloud data is mapped to the depth image. Multiple depth images are superimposed to form the density of the depth image, which improves the accuracy of the collected data, enriches the comprehensive acquisition of building data, and provides a data basis for the accuracy of subsequent measurements; through the variational graph autoencoder and adaptive disparity optimization algorithm, the quality and accuracy of the data are effectively improved, thereby ensuring the reliability of the measurement results; combined with the graph convolutional network and self-attention mechanism, meaningful features are efficiently extracted from the point cloud data, providing strong support for the subsequent measurement point identification; by constructing a measurement point identification neural network model, the points to be measured corresponding to the feature information of the point cloud image can be automatically identified, greatly improving the efficiency of measurement; by dynamically adjusting the disparity optimization parameters, it can adapt to different measurement environments and conditions to ensure that the best measurement results can be obtained in various situations.

[0046] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0047] See attached Figure 1 The intelligent building measurement system based on artificial intelligence described in this application includes the following parts:

[0048] A data acquisition module 10, a conversion module 20, a first feature extraction module 30, a parallax optimization module 40, a second feature extraction module 50, a measurement point identification module 60 and a measurement output module 70;

[0049] The data acquisition module 10 is used to collect multi-angle point cloud data of all measurement points of the building to be measured by using a laser radar. The data acquisition module 10 is connected to the conversion module 20 by means of data transmission;

[0050] The conversion module 20 is used to map the multi-angle point cloud data to the depth image, and multiple depth images are superimposed to form the density of the depth image. Each point on the point cloud data is mapped to an effective pixel of the depth image. The conversion module 20 is connected to the parallax optimization module 40 by data transmission;

[0051] The first feature extraction module 30 is used to encode the point cloud data using a variational graph autoencoder to extract the latent variable representation of each point as a feature. The first feature extraction module 30 is connected to the disparity optimization module 40 by means of data transmission;

[0052] The disparity optimization module 40 is used to perform disparity optimization on the depth image, remove noise points such as occlusion, motion, artifacts, etc., and perform adaptive disparity optimization. The disparity optimization module 40 is connected to the second feature extraction module 50 by data transmission;

[0053] The second feature extraction module 50 is used to further process the point cloud data using a graph convolutional network to obtain a new feature representation of each point, and use a self-attention mechanism to further improve the quality of the feature. The second feature extraction module 50 is connected to the measurement point identification module 60 by means of data transmission;

[0054] The measuring point recognition module 60 is used to construct a measuring point recognition neural network model, and extracts feature information from the point cloud image of the to-be-measured point taken by the laser radar as a training sample, and inputs the training sample into the measuring point recognition neural network model. After intelligent learning, the measuring point ID corresponding to the training sample is output. Through deep training and optimization of the measuring point recognition neural network model, the measuring point recognition neural network can recognize the to-be-measured point corresponding to the feature information of the point cloud image within a preset accuracy range. The measuring point recognition module 60 is connected to the measurement output module 70 by means of data transmission;

[0055] The measurement output module 70 is used to output building measurement data.

[0056] See attached Figure 2 The artificial intelligence-based intelligent building measurement method described in this application comprises the following steps:

[0057] S100: Collect multi-angle point cloud data of all measurement points of the building to be measured, map the multi-angle point cloud data to the depth image, and extract the implicit representation of the nodes in the depth image through the variational graph autoencoder;

[0058] The data acquisition module 10 uses a laser radar to collect multi-angle point cloud data of all measurement points of the building to be measured, where the measurement points are the target points to be measured in the building. The building simulation space is created based on the point cloud data, and the current frame point cloud data is selected and represented as , t represents the tth moment, Represents the nth point in the point cloud data at time t.

[0059] The conversion module 20 converts the multi-angle point cloud data Mapped to the depth image, multiple depth images are superimposed to form the density of the depth image, the density of the depth image is the same as the number of multi-angle point cloud images, and each point on the point cloud data is mapped to an effective pixel of the depth image.

[0060] In order to further improve the quality and accuracy of the data, the first feature extraction module 30 introduces a variational graph autoencoder to extract and match features. The variational graph autoencoder encodes the point cloud data to obtain a latent variable representation of each point, and then uses these representations for feature matching.

[0061] Specifically, the variational graph autoencoder is a generative model for graph data that can extract implicit representations of nodes in the graph. The variational graph autoencoder is used to extract implicit representations of each point in the point cloud data and use these representations for feature matching. The encoding process of the variational graph autoencoder generates an implicit representation for each node based on the structure of the graph and the node features. The specific formula is:

[0062] ,

[0063] in, represents the output distribution of the encoder, usually a Gaussian distribution, Z is the latent variable, X is the input feature, A is the adjacency matrix, is the i-th input feature, It is the implicit representation generated for the i-th node based on the input features X and the adjacency matrix A.

[0064] The decoding process of the variational graph autoencoder is to reconstruct the adjacency matrix of the graph based on the implicit representation. The specific formula is:

[0065] ,

[0066] in, is the conditional probability distribution, It is the sigmoid function, which is used to map values ​​to between (0,1).

[0067] The goal of the variational graph autoencoder is to maximize the following evidence lower bound:

[0068] ,

[0069] in, represents the next term of evidence, KL is the KL divergence, is the expected value of the reconstruction error, is a prior distribution, usually selected as a standard normal distribution. Maximize the similarity between the reconstructed adjacency matrix A and the true adjacency matrix A, ensuring that the distribution of the implicit representation Z is close to the prior distribution. The use of variational graph autoencoders to encode point cloud data is to obtain the latent variable representation of each point, which can effectively extract the implicit features of each point in the point cloud data, which is a preliminary feature extraction step.

[0070] S200: Adaptive parallax optimization of depth images;

[0071] After the preliminary feature extraction, in order to further improve the quality of the data, the depth image is optimized for disparity. The disparity optimization module 40 preprocesses the superimposed depth image to remove noise points formed by occlusion, motion, artifacts, etc. In order to achieve disparity optimization, an adaptive disparity optimization algorithm is used. The algorithm first calculates the depth value of each pixel, then calculates the local and global disparity gradients, dynamically adjusts the disparity optimization parameters, and uses the adjusted parameters for disparity optimization. The specific disparity optimization process is as follows:

[0072] Calculate a depth value for each pixel , the depth value is based on the distance from the pixel to the center of the radar. The specific formula is:

[0073] ,

[0074] in, , , Respectively represent the distance from the i-th point in the point cloud data to the center of the radar in the x, y, and z directions. Quantify the depth change rate of each pixel, calculate the disparity gradient of each pixel, identify the edge of the depth change, and calculate the disparity gradient :

[0075] ,

[0076] In order to obtain an overall disparity change index, the global disparity gradient is further calculated. The global disparity gradient is the average of the local disparity gradients of all pixels. For the entire image, the global disparity gradient G is calculated:

[0077] ,

[0078] in, is the total number of pixels in the entire depth image, and p is a constant greater than 1, which is used to enhance the influence of the gradient value, so as to pay more attention to those areas with large depth changes. Dynamically adjust the disparity optimization parameters according to the global disparity gradient and , the specific formula is:

[0079] ,

[0080] ,

[0081] in, , is the original disparity optimization parameter, and q is a constant greater than 1, which is used to adjust the sensitivity of the parameter. The disparity optimization parameter is dynamically adjusted according to the current disparity gradient, so that the optimization is more adapted to the current data. The adjusted parameters are used for disparity optimization.

[0082] ,

[0083] in, is the depth value of the optimized pixel, is a weight that represents the influence of pixel j on pixel i. It can be defined as:

[0084] ,

[0085] in, is a constant that represents the spatial scale of the weight. The definition of this weight is based on a Gaussian function, which ensures that pixels closer to pixel i have a greater influence on it.

[0086] Parallax optimization optimizes and cleans the initially extracted features, laying a better foundation for subsequent advanced feature extraction.

[0087] S300: Use graph convolutional networks to further process point cloud data to obtain new feature representations for each point, and use self-attention mechanisms to further improve feature quality;

[0088] Based on the high-quality data obtained in the disparity optimization step, the second feature extraction module 50 uses a graph convolutional network to process the point cloud data, obtains a new feature representation for each point, and uses these new feature representations to generate weights for each point. The update rule of the graph convolutional network is:

[0089] ,

[0090] Among them, H is the node feature, D is the degree matrix, W is the weight matrix, is the number of layers. Based on the node features of the current layer And the adjacency matrix A to calculate the node features of the next layer The new feature representation contains not only the information of each point, but also the information of its neighbors, providing richer data for subsequent disparity optimization and anomaly detection.

[0091] The goal of a graph convolutional network is to minimize the following loss function, which is usually chosen as the cross entropy loss:

[0092] ,

[0093] in, represents the loss function, is the adjacency matrix An element in Represents transpose. The difference between the reconstructed adjacency matrix A and the true adjacency matrix A is minimized by the above formula.

[0094] The self-attention mechanism is a method for calculating the weighted sum of each element in the input sequence, where the weight is calculated by query, key and value. The second feature extraction module 50 uses the self-attention mechanism to process the point cloud data to obtain the weight of each point, further improving the quality of the feature and improving the accuracy of the measurement system.

[0095] The weight calculation of self-attention is:

[0096] ,

[0097] Where Q and K are the query matrix and key matrix. Based on the query Q and key K, a weight is calculated for the i-th element .

[0098] The output calculation of self-attention is:

[0099] ,

[0100] in, is the output of the ith element, Represents the attention weight of the ith element to the jth element, and V is the value matrix. The self-attention mechanism is used to process the point cloud data to obtain the weight of each point, and finally obtain high-quality deep image features. Combining the latest graph neural network and self-attention mechanism technology, it can perform building measurement more accurately.

[0101] S400: Constructing a neural network model for measuring point recognition, identifying the points to be measured corresponding to the feature information of the point cloud image, and obtaining measurement data of the points to be measured of the actual building.

[0102] The measurement point recognition module 60 constructs a measurement point recognition neural network model, uses the extracted high-quality depth image features as training samples, inputs the training samples into the measurement point recognition neural network model, and outputs the measurement point ID corresponding to the training samples after intelligent learning. Through deep training and optimization of the measurement point recognition neural network model, the measurement point recognition neural network can recognize the points to be measured corresponding to the feature information of the point cloud image within a preset accuracy range.

[0103] The neural network model for measuring point recognition includes an input layer, a mapping layer, a first hidden layer, a second hidden layer, and an output layer. The deep image feature information is represented as , Indicates the total number of feature information. For any feature information, express.

[0104] The training samples are input into the measurement point recognition neural network model from the input layer. The input layer passes the data to the mapping layer. The mapping layer maps the image feature information to the neural network calculation space to improve the calculation accuracy of the data in the neural network model and reduce the calculation complexity. First, the information entropy of the feature information is calculated, and then all the feature information is classified. The specific formula is:

[0105] ,

[0106] in, The information entropy represents the characteristic information. The characteristic information is classified according to the information entropy, and the basis for classification is:

[0107] ,

[0108] in, Indicates the basis for division, and To adjust the parameters, Indicates transposition. Divide the feature information from greater than Extract the feature information corresponding to the lowest overall information entropy from the feature information of The feature information corresponding to the highest overall information entropy is extracted from the feature information. The specific formula is:

[0109] ,

[0110] ,

[0111] in, and is the extracted representative feature, and Respectively represent the highest and lowest information entropy. The representative feature mean is used as the feature reference value to calculate the mapping feature information:

[0112] ,

[0113] in, Represents mapping feature information, which is the feature information of image features mapped to the neural network calculation space.

[0114] The mapped feature information is input into the first hidden layer, and the processing process of the first hidden layer is:

[0115] ,

[0116] in, represents the first hidden layer, represents an activation function, and the activation function can be a commonly used activation function. represents the weight of the neuron, Represents bias. The first hidden layer enters the second hidden layer after transformation, and the processing process of the second hidden layer is:

[0117] ,

[0118] in, represents the second hidden layer, represents the mean of any two neurons in the first hidden layer, and each neuron corresponds to an input of the first hidden layer. Indicates feature information, , The second hidden layer is transformed to obtain the model output, and the calculation of the output layer is:

[0119] ,

[0120] in, represents the output of the output layer, represents the weight of the output layer, Represents the bias of the output layer. The error between the output of the measurement point recognition neural network model and the actual measurement point ID is calculated, and the model parameters are optimized according to the gradient descent method until the final output error reaches the expected level, thus completing the training of the measurement point recognition neural network model.

[0121] The point cloud depth image features of the points to be measured are input into the trained measurement point recognition neural network model to obtain the outline and position of the measurement points. According to the size ratio between the image and the actual size, the measurement data of the actual building points to be measured are obtained, and the building measurement data is output by the measurement output module 70.

[0122] In summary, the artificial intelligence-based intelligent building measurement system and method described in this application are completed.

[0123] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0124] 1. By using LiDAR to collect multi-angle point cloud data of the building to be measured, combined with depth image technology, high-precision measurement of the building can be achieved; multi-angle point cloud data of all measurement points of the building to be measured is obtained, and the multi-angle point cloud data is mapped to the depth image. Multiple depth images are superimposed to form the density of the depth image, which improves the accuracy of the collected data, enriches the comprehensive acquisition of building data, and provides a data basis for the accuracy of subsequent measurements;

[0125] 2. Through the variational graph autoencoder and adaptive disparity optimization algorithm, the quality and accuracy of the data can be effectively improved, thereby ensuring the reliability of the measurement results; combined with the graph convolutional network and self-attention mechanism, meaningful features can be efficiently extracted from the point cloud data, providing strong support for the subsequent measurement point identification; by constructing a measurement point identification neural network model, it can automatically identify the points to be measured corresponding to the feature information of the point cloud image, greatly improving the measurement efficiency; by dynamically adjusting the disparity optimization parameters, it can adapt to different measurement environments and conditions to ensure that the best measurement results can be obtained in various situations.

[0126] Effect research:

[0127] The technical solution of this application can effectively solve the problems that the existing technology does not efficiently process and optimize the measurement data, resulting in noise, occlusion, motion, artifacts and other problems in the data, affecting the accuracy of the measurement results; the feature extraction of point cloud data is not accurate enough, resulting in instability and inaccuracy of the measurement results; there are difficulties in automatically identifying measurement points, requiring manual intervention. In addition, the above system or method has undergone a series of effect surveys and verifications, and can ultimately realize an intelligent building measurement system based on artificial intelligence, which improves the accuracy, efficiency and adaptability of building measurement through advanced data processing, feature extraction and deep learning technologies.

[0128] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0130] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0131] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An intelligent building measurement method based on artificial intelligence, characterized in that: The following steps are involved: S100: Collect multi-angle point cloud data of all measurement points of the building to be measured, map the multi-angle point cloud data to the depth image, and extract the implicit representation of the nodes in the depth image through the variational graph autoencoder; S200: Adopting an adaptive disparity optimization algorithm to perform adaptive disparity optimization on the depth image, wherein the adaptive disparity optimization algorithm first calculates the depth value of each pixel point, then calculates local and global disparity gradients, dynamically adjusts disparity optimization parameters, and uses the adjusted disparity optimization parameters to perform disparity optimization; In the implementation process of the adaptive disparity optimization algorithm, a depth value is calculated for each pixel, and the depth value is based on the distance from the pixel to the center of the radar; the depth change rate of each pixel is quantified, the disparity gradient of each pixel is calculated, the edge of the depth change is identified, and the disparity gradient is calculated; and the global disparity gradient is further calculated; According to the global disparity gradient, dynamically adjusting disparity optimization parameters, and using the adjusted parameters to perform disparity optimization; the disparity optimization optimizes and cleans the initially extracted features; S300: Use graph convolutional networks to further process point cloud data to obtain new feature representations for each point, and use self-attention mechanisms to further improve feature quality; S400: Constructing a neural network model for measuring point recognition, identifying the points to be measured corresponding to the feature information of the point cloud image, and obtaining measurement data of the points to be measured of the actual building.

2. The intelligent building measurement method based on artificial intelligence according to claim 1 is characterized in that: The S100 specifically includes: The multi-angle point cloud data is mapped to a depth image, and more than one depth images are superimposed to form the density of the depth image. The density of the depth image is the same as the number of multi-angle point cloud images, and each point on the point cloud data is mapped to an effective pixel of the depth image. A variational graph autoencoder is introduced to extract and match features, encode the point cloud data, obtain the latent variable representation of each point, and then use the latent variable representation for feature matching.

3. The intelligent building measurement method based on artificial intelligence according to claim 2 is characterized in that: The S100 further includes: An implicit representation is generated for each node based on the structure and node features of the graph, and the adjacency matrix of the graph is reconstructed based on the implicit representation. The goal of the variational graph autoencoder is to maximize the lower bound of evidence and maximize the similarity between the reconstructed adjacency matrix and the true adjacency matrix.

4. The intelligent building measurement method based on artificial intelligence according to claim 1 is characterized in that: The S300 specifically includes: A graph convolutional network is used to process point cloud data to obtain a new feature representation for each point, and a weight is generated for each point using the new feature representation.

5. The intelligent building measurement method based on artificial intelligence according to claim 1 is characterized in that: The S300 further includes: Use the self-attention mechanism to process the point cloud depth image data and obtain the weight of each point.

6. The intelligent building measurement method based on artificial intelligence according to claim 1 is characterized in that: The S400 specifically includes: A measurement point recognition neural network model is constructed, and the extracted high-quality depth image features are used as training samples. The training samples are input into the measurement point recognition neural network model, and the measurement point ID corresponding to the training samples is output. Through deep training and optimization of the measurement point recognition neural network model, the measurement point recognition neural network can identify the points to be measured corresponding to the feature information of the point cloud image within a preset accuracy range.

7. An intelligent building measurement system based on artificial intelligence, applied to the intelligent building measurement method based on artificial intelligence as claimed in claim 1, characterized in that: Includes the following parts: A data acquisition module, a conversion module, a first feature extraction module, a parallax optimization module, a second feature extraction module, a measurement point recognition module and a measurement output module; The conversion module is used to map the multi-angle point cloud data to the depth image, superimpose more than one depth image to form the density of the depth image, and each point on the point cloud data is mapped to an effective pixel of the depth image. The conversion module is connected to the first feature extraction module by data transmission; The first feature extraction module is used to encode the point cloud data in the mapped and superimposed depth image using a variational graph autoencoder, and extract the latent variable representation of each point as a feature, and the first feature extraction module is connected to the disparity optimization module by means of data transmission; The disparity optimization module is used to perform disparity optimization on the depth image, remove noise points, including occlusion, motion, and artifacts, and perform adaptive disparity optimization. The disparity optimization module is connected to the second feature extraction module by means of data transmission; The second feature extraction module is used to further process the point cloud data using a graph convolutional network to obtain a new feature representation of each point, and use a self-attention mechanism to further improve the quality of the feature, and the second feature extraction module is connected to the measurement point identification module by means of data transmission; The measuring point recognition module is used to construct a measuring point recognition neural network model, and extracts feature information from a point cloud image of a point to be measured taken by a laser radar as a training sample. The training sample is input into the measuring point recognition neural network model, and after intelligent learning, the measuring point ID corresponding to the training sample is output. Through deep training and optimization of the measuring point recognition neural network model, the measuring point recognition neural network can recognize the point to be measured corresponding to the feature information of the point cloud image within a preset accuracy range. The measuring point recognition module is connected to the measurement output module by means of data transmission.

Images