High and large building foundation settlement monitoring system
By constructing a foundation settlement monitoring system based on convolutional neural networks and utilizing depth cameras and image processing technology, the problem of large measurement errors in traditional monitoring methods has been solved, achieving efficient and accurate settlement monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-29
- Publication Date
- 2026-04-03
AI Technical Summary
Traditional methods for monitoring foundation settlement are susceptible to environmental factors, leading to measurement errors. In particular, they are difficult to accurately monitor foundation settlement under long-term monitoring and complex terrain conditions.
A high-rise building foundation settlement monitoring system based on convolutional neural networks is adopted. Image data is collected by a depth camera, and the data is processed for sharpening and binarization. A convolutional neural network model is constructed to calculate the differences in pixel features and obtain settlement deformation monitoring data.
It enables efficient and accurate monitoring of foundation settlement under complex terrain and different lighting conditions, improves the efficiency of automated detection, and reduces measurement errors.
Smart Images
Figure CN121789102A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of foundation monitoring, and in particular relates to a foundation settlement monitoring system for tall buildings. Background Technology
[0002] The foundation, as one of the basic structural components of a building, has an undeniable impact on the quality of the structure. Settlement monitoring is necessary during foundation construction. Traditionally, the main method for monitoring foundation settlement relies on setting up stakes / poles, with workers using levels to observe and obtain settlement change parameters. Foundation settlement refers to the subsidence of the foundation surface caused by the compaction of the foundation soil layers under additional stress. Excessive settlement, especially uneven settlement, can cause buildings to tilt, crack, and become unusable.
[0003] Current ground monitoring methods primarily rely on total station measurements. However, the results of total station measurements can be affected by environmental conditions such as atmospheric pressure, temperature, and humidity. These factors can introduce measurement errors, which can accumulate, especially over long periods. Total stations typically need to be able to locate the measurement points; therefore, in areas with complex terrain or obstructions, it may be necessary to set up more measurement points. Summary of the Invention
[0004] The purpose of this invention is to provide a foundation settlement monitoring system for tall buildings to solve the problems existing in the prior art.
[0005] To achieve the above objectives, the present invention provides a foundation settlement monitoring system for tall buildings, comprising:
[0006] The data collection subsystem is used to collect foundation image data of tall buildings using a depth camera based on time series data.
[0007] A data processing subsystem, connected to the data collection subsystem, is used to sharpen the foundation image data of the tall building and then perform binarization processing to obtain processed image data.
[0008] A computing subsystem, connected to the data processing subsystem, is used to construct a convolutional neural network model, calculate the feature differences of pixels in the processed image data based on the convolutional neural network model, and calculate the feature differences of pixels based on the distance parameters of the depth camera to obtain the distance difference value of pixels at the same position.
[0009] A monitoring subsystem, connected to the computing subsystem, is used to calculate the distance difference value of pixels at the same location to obtain continuous foundation settlement and deformation monitoring data.
[0010] Preferably, the data collection subsystem includes:
[0011] The camera shooting module is used to connect the depth camera to the cloud storage module, and to enable the depth camera to shoot the tall building to obtain a building video stream;
[0012] The frame-by-frame extraction module is used to set a similarity threshold, and to divide the building video stream based on the similarity threshold and the time series to obtain the foundation image data of the tall building.
[0013] Preferably, the camera shooting module includes:
[0014] A camera position confirmation unit is used to select a depth camera and determine the relative position of the depth camera based on the position of the tall building.
[0015] A camera position adjustment unit is used to set the position and orientation of the depth camera based on the relative position;
[0016] The shooting unit is used to activate the depth camera to shoot the tall building and obtain the building video stream.
[0017] Preferably, the data processing subsystem includes:
[0018] The mapping module is used to map the pixels in the foundation image data of the tall building to a three-dimensional coordinate system to obtain a mapping dataset;
[0019] The partitioning module is used to partition the coordinate points in the mapped dataset based on the edge detection algorithm to obtain a partitioned dataset.
[0020] A fixing module is used to fix the regions of the partitioned dataset and then perform a sharpening process to obtain a sharpened dataset.
[0021] The binarization module is used to binarize the sharpened dataset to obtain the processed image data.
[0022] Preferably, the mapping module includes:
[0023] The camera coordinate system construction unit is used to map the pixels in the foundation image data of the tall building to the perspective projection model of the camera based on the internal and external parameters of the depth camera, so as to obtain the camera coordinate system of the image.
[0024] The 3D mapping unit is used to map points in the camera coordinate system of the image to a 3D coordinate system based on the camera's external parameters to perform coordinate transformation and obtain the mapping dataset.
[0025] Preferably, the computing subsystem includes:
[0026] The knowledge distillation module is used to construct the convolutional neural network model. It improves and trains the convolutional neural network model based on the knowledge distillation method to obtain a visual monitoring model.
[0027] The difference calculation module is used to input the processed image data into the visual monitoring model for calculation to obtain the feature differences of pixels.
[0028] The distance calculation module is used to calculate the feature differences of the pixels based on the distance parameters of the depth camera, and obtain the distance difference value of pixels at the same position.
[0029] Preferably, the knowledge distillation unit includes:
[0030] Network building blocks are used to construct convolutional neural network models and long short-term memory networks;
[0031] The first training unit is used to acquire a sample set of building foundation settlement, and to propagate the Long Short-Term Memory Network forward using the sample set of building foundation settlement to obtain the output of the Long Short-Term Memory Network.
[0032] The second training unit is used to train the convolutional neural network model by using the output of the long short-term memory network as the input of the convolutional neural network model, so as to obtain the student network model.
[0033] An improved unit is used for the attention mechanism module and the residual module, and the attention mechanism module and the residual module are introduced into the student network model to generate an improved convolutional neural network;
[0034] The third training unit is used to train the improved convolutional neural network model using a sample set of building foundation settlement data to obtain a visual monitoring model.
[0035] Preferably, the second training module includes:
[0036] The training label generation unit is used to construct hard labels for data based on the sample set of the building foundation settlement, and to use the output of the long short-term memory network as soft labels.
[0037] The precision training unit is used to input the hard and soft labels of the data into the convolutional neural network model for training, and simultaneously optimize the precision of the convolutional neural network model through a multi-scale supervision method to obtain the student network model.
[0038] Preferably, the improved module includes:
[0039] The construction unit is used to construct the attention mechanism module and the residual module;
[0040] The first introduction unit is used to introduce the attention mechanism module into the first three hidden layers of the student network model to obtain an attention classification model;
[0041] The second introducing unit is used to introduce the residual module into the attention classification model to generate the improved convolutional neural network.
[0042] Preferably, the monitoring subsystem includes:
[0043] The difference calculation module is used to compare the distance difference value of pixels at the same position with the initial pixel coordinate value to obtain the settlement offset value.
[0044] The merging module is used to merge the settlement offset values based on the time series to obtain the continuous foundation settlement deformation monitoring data;
[0045] The report generation module is used to generate a visual report and issue an early warning based on the continuous foundation settlement and deformation monitoring data.
[0046] The technical effects of this invention are as follows:
[0047] This invention relates to a system for monitoring foundation settlement based on convolutional neural networks, enabling automated detection and analysis of foundation settlement. Compared to traditional manual measurement methods, this automation significantly improves efficiency, especially for large monitoring areas or long-term monitoring tasks. Furthermore, convolutional neural networks can learn and adapt to characteristics under different geographical environments and conditions. This means that accurate foundation settlement monitoring can be performed even under complex terrain and varying lighting conditions. Attached Figure Description
[0048] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:
[0049] Figure 1 This is a schematic diagram of a tall building foundation settlement monitoring system in an embodiment of the present invention;
[0050] Figure 2 This is a schematic diagram illustrating the use of the output of a long short-time neural network and corresponding target labels to train a convolutional network model in an embodiment of the present invention.
[0051] Figure 3 This is a schematic diagram of a convolutional network in an embodiment of the present invention. Detailed Implementation
[0052] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0053] Example 1
[0054] like Figure 1 As shown, this embodiment provides a foundation settlement monitoring system for tall buildings, including:
[0055] The data collection subsystem is used to collect foundation image data of tall buildings using a depth camera based on time series data.
[0056] The data processing subsystem, connected to the data collection subsystem, is used to sharpen the foundation image data of tall buildings and then perform binarization to obtain processed image data.
[0057] The computing subsystem, connected to the data processing subsystem, is used to construct a convolutional neural network model, calculate the feature differences of pixels in the processed image data based on the convolutional neural network model, and calculate the feature differences of pixels based on the distance parameters of the depth camera to obtain the distance difference value of pixels at the same position.
[0058] The monitoring subsystem, connected to the computing subsystem, is used to calculate the distance difference between pixels at the same location to obtain continuous monitoring data on foundation settlement and deformation.
[0059] The data collection subsystem was further optimized and includes:
[0060] The camera shooting module is used to connect the depth camera to the cloud storage module, enabling the depth camera to shoot tall buildings and obtain building video streams;
[0061] The video stream can be obtained as follows:
[0062] In stereo vision, given two views (e.g., images taken by left and right cameras), feature points in one image can be projected onto the other image using a fundamental matrix, and the lines connecting the corresponding feature points in the two images lie on the same straight line, which is called the epipolar line.
[0063] The relationship between the fundamental matrix (F) and the feature points (x) and (x') can be expressed by the following formula:
[0064] [x'^TFx=0]
[0065] Here, (x) and (x') are the homogeneous coordinates of feature points on the two images, respectively. This equation represents the epipolar geometric relationship between the corresponding point (x') on the other image and the original point (x).
[0066] Substituting (x) into the above equation, we obtain the polar equation corresponding to the specific feature point (x):
[0067] [l'=Fx]
[0068] Here, (l') is the polar line corresponding to (x').
[0069] The frame-by-frame extraction module is used to set a similarity threshold and segment the building video stream based on the similarity threshold and time series data to obtain foundation image data of tall buildings. Since monitoring building foundations is a long-term series problem, it is necessary to remove overly similar images. Therefore, a similarity threshold is set, and the building video stream is segmented based on the similarity threshold and time series data to obtain foundation image data of tall buildings.
[0070] Further optimizations were made to the solution, including the camera shooting module, which includes:
[0071] The camera position confirmation unit is used to select a depth camera and determine the relative position of the depth camera based on the location of the tall building.
[0072] A camera position adjustment unit is used to set the position and orientation of the depth camera based on its relative position.
[0073] The shooting unit is used to activate the depth camera to shoot tall buildings and obtain building video streams.
[0074] The data processing subsystem, which has been further optimized, includes:
[0075] The mapping module is used to map the pixels in the foundation image data of tall buildings to a three-dimensional coordinate system to obtain a mapped dataset;
[0076] The partitioning module is used to partition the coordinate points in the mapped dataset based on the edge detection algorithm to obtain a partitioned dataset;
[0077] The fixing module is used to fix the regions of the partitioned dataset and then perform sharpening processing to obtain a sharpened dataset;
[0078] The binarization module is used to binarize the sharpened dataset to obtain the processed image data.
[0079] The solution has been further optimized, and the mapping module includes:
[0080] The camera coordinate system construction unit is used to map the pixels in the foundation image data of tall buildings to the perspective projection model of the camera based on the internal and external parameters of the depth camera, so as to obtain the camera coordinate system of the image.
[0081] The 3D mapping unit is used to map points in the camera coordinate system of an image to a 3D coordinate system based on the camera's extrinsic parameters to perform coordinate transformation and obtain a mapped dataset.
[0082] The camera coordinate system construction unit is used to obtain the camera's internal parameters, including focal length, principal point coordinates, radial distortion, and tangential distortion. These are obtained by photographing a calibration board with known geometry and then using a camera calibration algorithm (such as Zhang's calibration method).
[0083] The known spatial points and corresponding image coordinates on the calibration plate are used to estimate the camera's intrinsic parameters.
[0084] Camera positioning technology is used to obtain the position and orientation of the camera. This is achieved by using GPS, inertial measurement units (IMUs), or other sensors.
[0085] External parameters can also be calculated through feature point matching, for example using the StructurefromMotion (SfM) technique.
[0086] Once the camera's intrinsic and extrinsic parameters are known, pixels can be mapped to the camera coordinate system. This process involves the camera's projection model, typically perspective projection or fisheye projection.
[0087] The mathematical model of perspective projection can be expressed by the following formula: [s\begin{bmatrix}u\v\1\end{bmatrix}=\begin{bmatrix}f_x&0&c_x\0&f_y&c_y\0&0&1\end{bmatrix}\begin{bmatrix}X\Y\Z\end{bmatrix}] where ((u,v)) are pixel coordinates, ((X,Y,Z)) are camera coordinates, (f_x,f_y) are focal lengths, (c_x,c_y) are principal point coordinates, and (s) is the scaling factor.
[0088] Given the camera's position and orientation, points in the camera coordinate system can be mapped to the world coordinate system. This typically involves coordinate transformations and rotation matrices.
[0089] Further optimization of the scheme includes the following computing subsystem:
[0090] The knowledge distillation module is used to build convolutional neural network models. It improves and trains convolutional neural network models based on knowledge distillation methods to obtain visual monitoring models.
[0091] The difference calculation module is used to input the processed image data into the visual monitoring model for calculation to obtain the feature differences of pixels;
[0092] The distance calculation module is used to calculate the feature differences of pixels based on the distance parameters of the depth camera, and obtain the distance difference value of pixels at the same position.
[0093] The solution has been further optimized, and the knowledge distillation unit includes:
[0094] Network building blocks are used to construct convolutional neural network models and long short-term memory networks;
[0095] The first training unit is used to obtain a sample set of building foundation settlement. The sample set of building foundation settlement is used to propagate forward through the Long Short-Term Memory Network to obtain the output of the Long Short-Term Memory Network.
[0096] The second training unit is used to train the convolutional neural network model by using the output of the long short-term memory network as the input to obtain the student network model.
[0097] An improved unit is used for the attention mechanism module and the residual module, and the attention mechanism module and the residual module are introduced into the student network model to generate an improved convolutional neural network.
[0098] The third training unit is used to train the improved convolutional neural network model using a sample set of building foundation settlement data to obtain a visual monitoring model.
[0099] like Figure 2-3 As shown, firstly, an LSTM network is used as the teacher network for training. LSTM is a type of recurrent neural network suitable for sequential data, with the ability to capture long-term dependencies, making it suitable for processing temporal data or tasks with temporal correlation.
[0100] Design a Convolutional Neural Network (CNN) as a student network. CNNs are primarily used for image processing and are able to effectively capture spatial features in images. Student networks are typically designed to be more lightweight and better suited for operating in resource-constrained environments than teacher networks.
[0101] The teacher network makes predictions on the training data, and these predictions are used as targets, represented by soft labels (probability distributions), rather than the original hard labels. These soft labels contain information about the teacher network's relative confidence in each category.
[0102] The student network is trained using soft labels. The goal of the student network is to replicate the behavior of the teacher network as closely as possible, i.e., to maintain consistency with the teacher network in terms of prediction. In this case, a loss function combining the original hard labels and soft labels is typically used.
[0103] The student network is evaluated to see its performance on the test set. Due to the distillation process, the student network can often achieve performance similar to the teacher network with a relatively small model size.
[0104] The second training module further optimizes the scheme and includes:
[0105] The training label generation unit is used to construct hard labels for data based on the sample set of building foundation settlement, and the output of the long short-term memory network is used as soft labels.
[0106] The precision training unit is used to input hard and soft labels of data into the convolutional neural network model for training, and at the same time optimizes the precision of the convolutional neural network model through a multi-scale supervision method to obtain a student network model.
[0107] The hard and soft labels of the data are input into the convolutional neural network model for training. At the same time, the accuracy of the convolutional neural network model is optimized by a multi-scale supervision method to obtain the student network model.
[0108] The loss function is:
[0109]
[0110] L dice =1 - dice;
[0111] Where, dice is the DICE coefficient, X represents the first image data from the camera, Y represents the second image data from the camera, and L... dice This is the loss function.
[0112] Further optimization of the solution and improvements to the modules include:
[0113] Building units are used to construct attention mechanism modules and residual modules;
[0114] The first introductory unit is used to introduce the attention mechanism module into the first three hidden layers of the student network model to obtain an attention classification model.
[0115] The second introductory unit is used to introduce the residual module into the attention classification model to generate an improved convolutional neural network.
[0116] The solution has been further optimized, and the monitoring subsystem includes:
[0117] The difference calculation module is used to compare the distance difference value of pixels at the same location with the initial pixel coordinate value to obtain the settlement offset value;
[0118] The merging module is used to merge settlement offset values based on time series to obtain continuous foundation settlement deformation monitoring data; the initial coordinate parameter values are subtracted from the measured coordinate parameter values to obtain the settlement offset values.
[0119] The report generation module is used to generate visual reports and issue early warnings based on continuous foundation settlement and deformation monitoring data.
[0120] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A foundation settlement monitoring system for tall buildings, characterized in that, include: The data collection subsystem is used to collect foundation image data of tall buildings using a depth camera based on time series data. A data processing subsystem, connected to the data collection subsystem, is used to sharpen the foundation image data of the tall building and then perform binarization processing to obtain processed image data. A computing subsystem, connected to the data processing subsystem, is used to construct a convolutional neural network model, calculate the feature differences of pixels in the processed image data based on the convolutional neural network model, and calculate the feature differences of pixels based on the distance parameters of the depth camera to obtain the distance difference value of pixels at the same position. A monitoring subsystem, connected to the computing subsystem, is used to calculate the distance difference value of pixels at the same location to obtain continuous foundation settlement and deformation monitoring data.
2. The high-rise building foundation settlement monitoring system according to claim 1, characterized in that, The data collection subsystem includes: The camera shooting module is used to connect the depth camera to the cloud storage module, and to enable the depth camera to shoot the tall building to obtain a building video stream; The frame-by-frame extraction module is used to set a similarity threshold, and to divide the building video stream based on the similarity threshold and the time series to obtain the foundation image data of the tall building.
3. The tall building foundation settlement monitoring system according to claim 2, characterized in that, The camera capturing module includes: A camera position confirmation unit is used to select a depth camera and determine the relative position of the depth camera based on the position of the tall building. A camera position adjustment unit is used to set the position and orientation of the depth camera based on the relative position; The shooting unit is used to activate the depth camera to shoot the tall building and obtain the building video stream.
4. The tall building foundation settlement monitoring system according to claim 1, characterized in that, The data processing subsystem includes: The mapping module is used to map the pixels in the foundation image data of the tall building to a three-dimensional coordinate system to obtain a mapping dataset; The partitioning module is used to partition the coordinate points in the mapped dataset based on the edge detection algorithm to obtain a partitioned dataset. A fixing module is used to fix the regions of the partitioned dataset and then perform a sharpening process to obtain a sharpened dataset. The binarization module is used to binarize the sharpened dataset to obtain the processed image data.
5. The tall building foundation settlement monitoring system according to claim 4, characterized in that, The mapping module includes: The camera coordinate system construction unit is used to map the pixels in the foundation image data of the tall building to the perspective projection model of the camera based on the internal and external parameters of the depth camera, so as to obtain the camera coordinate system of the image. The 3D mapping unit is used to map points in the camera coordinate system of the image to a 3D coordinate system based on the camera's external parameters to perform coordinate transformation and obtain the mapping dataset.
6. The tall building foundation settlement monitoring system according to claim 1, characterized in that, The computing subsystem includes: The knowledge distillation module is used to construct the convolutional neural network model. It improves and trains the convolutional neural network model based on the knowledge distillation method to obtain a visual monitoring model. 。 The difference calculation module is used to input the processed image data into the visual monitoring model for calculation to obtain the feature differences of pixels. The distance calculation module is used to calculate the feature differences of the pixels based on the distance parameters of the depth camera, and obtain the distance difference value of pixels at the same position.
7. The tall building foundation settlement monitoring system according to claim 6, characterized in that, The knowledge distillation module includes: Network building blocks are used to construct convolutional neural network models and long short-term memory networks; The first training unit is used to acquire a sample set of building foundation settlement, and to propagate the Long Short-Term Memory Network forward using the sample set of building foundation settlement to obtain the output of the Long Short-Term Memory Network. The second training unit is used to train the convolutional neural network model by using the output of the long short-term memory network as the input of the convolutional neural network model, so as to obtain the student network model. An improved unit is used for the attention mechanism module and the residual module, and the attention mechanism module and the residual module are introduced into the student network model to generate an improved convolutional neural network; The third training unit is used to train the improved convolutional neural network model using a sample set of building foundation settlement data to obtain a visual monitoring model.
8. The tall building foundation settlement monitoring system according to claim 7, characterized in that, The second training module includes: The training label generation unit is used to construct hard labels for data based on the sample set of the building foundation settlement, and to use the output of the long short-term memory network as soft labels. The precision training unit is used to input the hard and soft labels of the data into the convolutional neural network model for training, and simultaneously optimize the precision of the convolutional neural network model through a multi-scale supervision method to obtain the student network model.
9. The tall building foundation settlement monitoring system according to claim 7, characterized in that, The improved module includes: The construction unit is used to construct the attention mechanism module and the residual module; The first introduction unit is used to introduce the attention mechanism module into the first three hidden layers of the student network model to obtain an attention classification model; The second introducing unit is used to introduce the residual module into the attention classification model to generate the improved convolutional neural network.
10. The high-rise building foundation settlement monitoring system according to claim 1, characterized in that, The monitoring subsystem includes: The difference calculation module is used to compare the distance difference value of pixels at the same position with the initial pixel coordinate value to obtain the settlement offset value. The merging module is used to merge the settlement offset values based on the time series to obtain the continuous foundation settlement deformation monitoring data; The report generation module is used to generate a visual report and issue an early warning based on the continuous foundation settlement and deformation monitoring data.