Intelligent foundation pit supporting inclined pile monitoring system based on computer vision
By combining image processing, machine learning, and deep learning methods, automatic, real-time, and accurate monitoring of the displacement of inclined piles supporting foundation pits and surrounding soil has been achieved. This solves the problems of low efficiency and weak anti-interference ability of traditional monitoring methods, and improves the accuracy of monitoring and engineering safety.
Patent Information
- Application Number
- CN202511456531.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-13
- Publication Date
- 2026-02-24
AI Technical Summary
Traditional foundation pit monitoring methods are inefficient, lack real-time performance, and have weak anti-interference capabilities, making it difficult to achieve automatic, real-time, and accurate monitoring of the displacement of the inclined piles supporting the foundation pit and the surrounding soil.
By combining image processing, machine learning, and deep learning, real-time image data is acquired through high-resolution camera equipment and image target surfaces. Feature recognition is performed using the YOLOv8 network and CBAM attention mechanism, and displacement monitoring and risk prediction are performed by combining multi-target tracking algorithms. A 3D model and monitoring report are generated to achieve automated early warning.
It enables simple, efficient, automatic, real-time, and accurate monitoring of the displacement of the inclined piles supporting the foundation pit and the surrounding soil, improving the accuracy and anti-interference ability of the monitoring and providing engineering safety assurance.
Smart Images
Figure CN121564632A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of pile foundation engineering, and in particular to an intelligent monitoring system for inclined piles in foundation pit support based on computer vision. Background Technology
[0002] With the development of underground engineering in my country, the problems of traditional foundation pit support, such as large land occupation, large amount of solid waste generation, and high cost, have become increasingly prominent. Therefore, the application of inclined piles as foundation pit support has the advantages of larger construction space, less solid waste generation, and low cost, and has good application prospects.
[0003] During foundation pit excavation, changes in soil pressure cause displacement and deformation of the inclined piles supporting the foundation pit, leading to potential safety hazards in the foundation pit support structure. Therefore, accurately monitoring the displacement and deformation of the inclined piles and the surrounding soil, and proposing corresponding measures, is crucial for ensuring the safety and reliability of engineering construction. Traditional methods for monitoring foundation pit and surrounding soil displacement mainly rely on theodolites, laser rangefinders, and mechanical displacement sensors. These methods suffer from problems such as complex layout, difficult maintenance, low efficiency, poor real-time performance, weak anti-interference capabilities, and difficult data processing. Therefore, to achieve automatic, real-time, and accurate monitoring of the displacement of inclined piles and surrounding soil, there is an urgent need to propose an intelligent foundation pit support inclined pile monitoring technology based on computer vision. Summary of the Invention
[0004] In view of the shortcomings of the prior art, the technical problem to be solved by the present invention is to provide an intelligent monitoring system for inclined piles of foundation pit support based on computer vision. The system adopts a combination of image processing, machine learning and deep learning to achieve simple, efficient, automatic, real-time, accurate, anti-interference and convenient data processing intelligent monitoring of the displacement of inclined piles of foundation pit support and surrounding soil.
[0005] This invention is accomplished using the following technical solution: an intelligent monitoring system for inclined piles in foundation pit support based on computer vision, characterized by comprising an image acquisition module, an image processing module, a computer vision analysis module, a result display module, and an early warning module, specifically including: S1. Image acquisition module, including high-resolution camera equipment and image target surface, to acquire real-time image data of the displacement of the inclined pile and the displacement of the soil around the inclined pile; S2. Image processing module, including image preprocessing and feature recognition, transmits the real-time image data obtained in S1 to image preprocessing, constructs a three-dimensional model based on the real-time image information of the displacement of the soil around the inclined pile after processing, inputs the preprocessed cloud map into the three-dimensional image semantic segmentation model of YOLOv8 network and attention mechanism, automatically identifies the target under multiple interference factors, and quantifies the multi-channel parameters of the target. S3. Computer vision analysis module, including displacement monitoring and pattern recognition. The displacement monitoring is based on a multi-target tracking algorithm to track the three-dimensional contours identified by the YOLOv8 model. The displacement of the tilted pile is calculated through time series image comparison and image registration technology. The pattern recognition uses convolutional neural network deep learning technology to classify the condition of the tilted pile and predict risks. S4. Results display module, including data visualization and report generation. The data collected by displacement monitoring and pattern recognition in S3 are displayed through charts or heat maps. The report generation module periodically generates monitoring reports for inclined pile 6. S5. Early warning module: The monitoring system monitors the status of the tilted pile in real time. When the system detects abnormal displacement or potential danger, the alarm system automatically triggers an alarm.
[0006] Furthermore, the image preprocessing steps in S2 include: 1) smoothing the image using a Gaussian smoothing filter; 2) selecting a suitable Laplacian operator template; 3) convolving the Laplacian template with the Gaussian-smoothed image to obtain the Laplacian response at that location; 4) copying edge pixel values during the convolution process and then convolving the edges; 5) combining the original image with the Laplacian response to enhance the image's edges and details; and 6) performing post-processing operations such as adjusting contrast or brightness on the image after the above processes.
[0007] Furthermore, the specific steps in S2 for constructing a 3D model based on the real-time image information of the soil displacement around the processed inclined pile 6 include: 1) simultaneously capturing two images of the same scene using an adjustable binocular camera 3; 2) determining the internal and external parameters of each camera; 3) performing denoising, grayscale conversion, and filtering operations on the images; 4) adjusting the images from the two cameras to align them on the same plane; 5) matching the same areas in the two images; 6) estimating the depth of each point in the scene using a disparity map and generating a depth map; 7) reconstructing the scene in 3D using the depth map and camera parameters.
[0008] Furthermore, the feature recognition steps in S2 specifically include: 1) creating a target 3D dataset; 2) improving the YOLOv8 network by changing the four-dimensional parameters of the YOLOv8 network's detection head to six-dimensional parameters X, Y, Z, W, H, and D, where X, Y, and Z are the coordinates of the center point of the 3D target bounding box, and W, H, and D are the width, height, and depth of the 3D target bounding box, respectively; 3) inputting the expanded labeled cloud map from the target 3D dataset into YOLOv8 for learning; 4) dividing the target 3D dataset into three parts: training, test, and valid; 5) adding the CBAM attention mechanism to the YOLOv8 network; 6) inputting the training, test, and valid parts from step 4) into the improved YOLOv8 network with the added CBAM attention mechanism for training to obtain the trained YOLOv8 network; 7) the trained YOLOv8 network recognizing the target positions in the cloud map after image preprocessing; 8) saving the recognized target parts separately to a new folder.
[0009] Furthermore, the specific steps for displacement monitoring in S3 include: 1) improving the multi-target tracking algorithm by replacing the Kalman filter in BoT-SORT with seven-dimensional parameters: X, Y, Z, W, H, D, and XYZ; 2) inputting the three-dimensional contour identified by the YOLOv8 model into the improved BoT-SORT for tracking; and 3) outputting the three-dimensional motion trajectory of the target.
[0010] Furthermore, the steps for creating the target 3D dataset include: 1) taking 50 cloud images with target depth information; 2) annotating the 50 cloud images with target depth information; 3) expanding the 50 original target images and 50 annotated images into 400 images by adding noise, rotating, flipping, etc.
[0011] Furthermore, the high-resolution camera device includes a fixed camera and an adjustable binocular camera. The high-resolution camera device communicates with a computer via a data cable or wireless signal. The fixed camera is placed in front of the tilted pile monitoring point, and the binocular camera is placed in front of the soil around the tilted pile. Multiple image targets 1 are respectively placed in the field of view of the high-resolution camera device at the tilted pile displacement monitoring point through target surface fixing mounts.
[0012] Furthermore, the image target 1 includes a target background plate, a target positioning image, a target positioning line, a positioning light, and a target adjustment rod. The target background plate and the target fixing mount are connected by the target adjustment rod. The target positioning image is mounted on the target background plate through the target positioning line. The positioning light is mounted on the target background plate and connected to an external power source through a power wire.
[0013] Furthermore, the internal parameters include focal length, the external parameters include camera pose and distance between two cameras, the parallax refers to the horizontal pixel difference of the same scene point in two images, and the depth map represents the distance from each point in the scene to the camera.
[0014] This invention employs a combination of image processing, machine learning, and deep learning to achieve intelligent monitoring of the displacement of inclined piles and surrounding soil in foundation pit support. This monitoring method is simple to deploy, efficient, automatic, real-time, accurate, highly resistant to interference, and convenient for data processing. It can improve the accuracy of inclined pile settlement monitoring, provide technical support for the maintenance of inclined piles and support structures, and has significant theoretical and engineering application value for ensuring engineering safety. Attached Figure Description
[0015] Figure 1 This is a schematic diagram of an intelligent foundation pit support tilt pile monitoring technology based on computer vision. Figure 2 A top view showing the overall layout of the monitoring points for multiple inclined piles; Figure 3 This is a schematic diagram of the image target surface of the image acquisition module; Figure 4 This is a flowchart of the data processing process. Detailed Implementation
[0016] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structure, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.
[0017] Reference Figure 1-4 As shown, the present invention provides an intelligent monitoring system for inclined piles in foundation pit support based on computer vision, including an image acquisition module, an image processing module, a computer vision analysis module, a result display module, and an early warning module.
[0018] S1. Graphics acquisition module, including high-resolution camera equipment and image target 1, to acquire real-time image data of the displacement of the inclined pile 6 and the displacement of the soil around the inclined pile 6.
[0019] Specifically, the high-resolution camera device includes a fixed camera 2 and an adjustable binocular camera 3. The high-resolution camera device communicates with a computer 5 via a data cable 4 or a wireless signal. The fixed camera 2 is positioned in front of the monitoring point of the inclined pile 6 to acquire real-time image data of the vertical and horizontal displacement of the pile top and the horizontal displacement of the pile body on the soil surface. The binocular camera is positioned in front of the soil surrounding the inclined pile 6 to acquire real-time image data of the displacement of the soil around the inclined pile 6.
[0020] Multiple image target surfaces 1 are respectively mounted on the displacement monitoring points of the inclined pile 6 within the field of view of the high-resolution camera equipment via target surface fixing mounts 16. Multiple image target surfaces 1 can be simultaneously mounted on different monitoring points. Specifically, the image target surface 1 includes a target surface background plate 11, a target surface positioning image 12, a target surface positioning line 13, a positioning light 14, and a target surface adjustment rod 15. The target surface background plate 11 and the target surface fixing mount 16 are connected via the target surface adjustment rod 15. The target surface positioning image 12 is mounted on the target surface background plate 11 via the target surface positioning line 13. The positioning light 14 is mounted on the target surface background plate 11 and connected to an external power source via a power cable 17. The target surface adjustment rod 15 can be freely adjusted in length and orientation of the target surface background plate. The positioning light 14 is on all day, adapting to various weather conditions, illuminating the target surface image at night, and also serving a positioning function.
[0021] S2. Image processing module, including image preprocessing and feature recognition.
[0022] Specifically, the real-time image data obtained in S1 is transmitted to image preprocessing. Based on the Laplacian operator algorithm, the noise, blur, and distortion problems generated in the monitoring images are solved to obtain processed real-time image data. Then, based on the processed real-time image information of the soil displacement around the inclined pile 6, the depth information of the scene is calculated, and a 3D model is constructed (the 3D model is displayed in the form of a cloud map). The preprocessed cloud map is input into a 3D image semantic segmentation model with a YOLOv8 network and attention mechanism to automatically identify the target under multiple interference factors and quantify the multi-channel parameters of the target.
[0023] The image preprocessing steps include: 1) smoothing the image using a Gaussian smoothing filter; 2) selecting a suitable Laplacian operator template; 3) convolving the Laplacian template with the Gaussian-smoothed image to obtain the Laplacian response at that location; 4) copying edge pixel values during the convolution process and then convolving the edges; 5) combining the original image with the Laplacian response to enhance the image's edges and details; and 6) performing post-processing operations such as adjusting contrast or brightness on the image after the above processes.
[0024] The specific steps for constructing a 3D model based on real-time image information of soil displacement around inclined pile 6 include: 1) simultaneously capturing two images of the same scene using an adjustable binocular camera 3; 2) determining the internal and external parameters of each camera; 3) performing denoising, grayscale conversion, and filtering operations on the images; 4) adjusting the images from the two cameras to align them on the same plane; 5) matching the same areas in the two images; 6) estimating the depth of each point in the scene using a disparity map and generating a depth map; 7) reconstructing the scene in 3D using the depth map and camera parameters. The internal parameters include focal length, the external parameters include camera pose and the distance between the two cameras, the disparity refers to the horizontal pixel difference of the same scene point in the two images, and the depth map represents the distance from each point in the scene to the camera.
[0025] The specific steps of feature recognition include: 1) Creating a target 3D dataset; 2) Improving the YOLOv8 network by changing the four-dimensional parameters of the YOLOv8 network's detection head to six-dimensional parameters: X, Y, Z, W, H, and D, where X, Y, and Z are the coordinates of the center point of the 3D target bounding box, and W, H, and D are the width, height, and depth of the 3D target bounding box, respectively; 3) Inputting the expanded labeled cloud map from the target 3D dataset into YOLOv8 for learning; 4) Dividing the target 3D dataset into three parts: train, test, and valid, with a ratio of 7:2:1; 5) Adding the CBAM attention mechanism to the YOLOv8 network; 6) Inputting the train, test, and valid parts from step 4) into the improved YOLOv8 network with the added CBAM attention mechanism for training to obtain the trained YOLOv8 network; 7) Recognizing the target positions in the cloud map after image preprocessing using the trained YOLOv8 network; 8) Saving the recognized target parts separately to a new folder.
[0026] The steps for creating the target 3D dataset include: 1) taking 50 cloud images with target depth information; 2) annotating the 50 cloud images with target depth information; 3) expanding the 50 original target images and 50 annotated images into 400 images by adding noise, rotating, flipping, etc.; 4) creating the output dataset.
[0027] S3. Computer vision analysis module, including displacement monitoring and pattern recognition.
[0028] Specifically, the displacement monitoring is based on a multi-target tracking algorithm to track the three-dimensional contour identified by the YOLOv8 model, and the displacement of the inclined pile 6 is calculated through time-series image comparison and image registration techniques. The specific steps of displacement monitoring include: 1) improving the multi-target tracking algorithm by replacing the Kalman filter in BoT-SORT with seven-dimensional parameters (X, Y, Z, W, H, D, XYZ); 2) inputting the three-dimensional contour identified by the YOLOv8 model in S2 into the improved BoT-SORT for tracking; and 3) outputting the three-dimensional motion trajectory of the target.
[0029] The pattern recognition uses convolutional neural network deep learning technology to classify and predict the risk of the inclined pile 6. Specifically, after obtaining the three-dimensional displacement of the inclined pile, the displacement is divided into multiple levels, with different displacement values corresponding to a specific risk, thus achieving the purpose of risk prediction.
[0030] S4. Results Display Module, including data visualization and report generation.
[0031] Specifically, the data collected by displacement monitoring and pattern recognition in S3 are visualized through charts or heatmaps to facilitate user analysis and decision-making. The report generation system regularly generates monitoring reports for the tilted piles, providing engineers and decision-makers with detailed monitoring data and analysis results.
[0032] S5. Early Warning Module: The monitoring system monitors the status of the tilted pile 6 in real time, providing automated, 24 / 7, full-process monitoring. When the system detects abnormal displacement or potential danger, the alarm system automatically triggers an alarm, notifying relevant personnel to take timely measures.
[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. An intelligent monitoring system for inclined piles in foundation pit support based on computer vision, characterized in that: It includes an image acquisition module, an image processing module, a computer vision analysis module, a results display module, and an early warning module, specifically including: S1. Image acquisition module, including high-resolution camera equipment and image target surface, to acquire real-time image data of the displacement of the inclined pile and the displacement of the soil around the inclined pile; S2. Image processing module, including image preprocessing and feature recognition, transmits the real-time image data obtained in S1 to image preprocessing, constructs a three-dimensional model based on the real-time image information of the displacement of the soil around the inclined pile after processing, inputs the preprocessed cloud map into the three-dimensional image semantic segmentation model of YOLOv8 network and attention mechanism, automatically identifies the target under multiple interference factors, and quantifies the multi-channel parameters of the target. S3. Computer vision analysis module, including displacement monitoring and pattern recognition. The displacement monitoring is based on a multi-target tracking algorithm to track the three-dimensional contours identified by the YOLOv8 model. The displacement of the tilted pile is calculated through time series image comparison and image registration technology. The pattern recognition uses convolutional neural network deep learning technology to classify the condition of the tilted pile and predict risks. S4. Results display module, including data visualization and report generation. The data collected by displacement monitoring and pattern recognition in S3 are displayed through charts or heat maps. The report generation module generates tilt pile monitoring reports periodically. S5. Early warning module: The monitoring system monitors the status of the tilted pile in real time. When the system detects abnormal displacement or potential danger, the alarm system automatically triggers an alarm.
2. The intelligent foundation pit support inclined pile monitoring system based on computer vision according to claim 1, characterized in that: S2 The steps of image preprocessing include: 1) smoothing the image using a Gaussian smoothing filter; 2) selecting a suitable Laplacian operator template; 3) convolving the Laplacian template with the Gaussian-smoothed image to obtain the Laplacian response at that location; 4) copying edge pixel values during the convolution process and then convolving the edges; 5) combining the original image with the Laplacian response to enhance the image's edges and details; and 6) performing post-processing operations such as adjusting contrast or brightness on the image after the above processes.
3. The intelligent foundation pit support inclined pile monitoring system based on computer vision according to claim 1 or 2, characterized in that: The specific steps for constructing a 3D model based on real-time image information of soil displacement around the processed inclined pile in S2 include: 1) simultaneously capturing two images of the same scene using an adjustable binocular camera; 2) determining the internal and external parameters of each camera; 3) performing denoising, grayscale conversion, and filtering operations on the images; 4) adjusting the images from the two cameras to align them on the same plane; 5) matching the same areas in the two images; 6) estimating the depth of each point in the scene using a disparity map and generating a depth map; 7) reconstructing the scene in 3D using the depth map and camera parameters.
4. The intelligent foundation pit support inclined pile monitoring system based on computer vision according to claim 3, characterized in that: in S2 The specific steps of feature recognition include: 1) Creating a target 3D dataset; 2) Improving the YOLOv8 network by changing the four-dimensional parameters of the YOLOv8 network's detection head to six-dimensional parameters: X, Y, Z, W, H, and D, where X, Y, and Z are the coordinates of the center point of the 3D target bounding box, and W, H, and D are the width, height, and depth of the 3D target bounding box, respectively; 3) Inputting the expanded labeled cloud map from the target 3D dataset into YOLOv8 for learning; 4) Dividing the target 3D dataset into three parts: training, test, and valid; 5) Adding the CBAM attention mechanism to the YOLOv8 network; 6) Inputting the training, test, and valid parts from step 4) into the improved YOLOv8 network with the added CBAM attention mechanism for training to obtain the trained YOLOv8 network; 7) Recognizing the target positions in the cloud map after image preprocessing; 8) Saving the recognized target parts separately to a new folder.
5. A computer vision-based intelligent foundation pit support inclined pile monitoring system according to claim 1, 2, or 4, characterized in that: The specific steps for displacement monitoring in S3 include: 1) improving the multi-target tracking algorithm by replacing the Kalman filter in BoT-SORT with seven-dimensional parameters: X, Y, Z, W, H, D, and XYZ; 2) inputting the three-dimensional contour identified by the YOLOv8 model into the improved BoT-SORT for tracking; and 3) outputting the three-dimensional motion trajectory of the target.
6. The intelligent foundation pit support inclined pile monitoring system based on computer vision according to claim 4, characterized in that: The steps to create a target 3D dataset include: 1) taking 50 cloud images with target depth information; 2) annotating the 50 cloud images with target depth information; 3) expanding the 50 original target images and 50 annotated images into 400 images by adding noise, rotating, flipping, etc.
7. The intelligent foundation pit support inclined pile monitoring system based on computer vision according to claim 1, characterized in that: The high-resolution camera device includes a fixed camera and an adjustable binocular camera. The high-resolution camera device communicates with a computer via a data cable or wireless signal. The fixed camera is placed in front of the tilted pile monitoring point, and the binocular camera is placed in front of the soil around the tilted pile. Multiple image targets are respectively placed at the tilted pile displacement monitoring point within the field of view of the high-resolution camera device through target fixing mounts.
8. The intelligent foundation pit support inclined pile monitoring system based on computer vision according to claim 1 or 7, characterized in that: The image target surface includes a target background plate, a target positioning image, a target positioning line, a positioning light, and a target adjustment rod. The target background plate and the target fixing mount are connected by the target adjustment rod. The target positioning image is mounted on the target background plate through the target positioning line. The positioning light is mounted on the target background plate and connected to an external power source through a power cord.
9. The intelligent foundation pit support inclined pile monitoring system based on computer vision according to claim 3, characterized in that: The internal parameters include focal length, the external parameters include camera pose and distance between two cameras, the parallax refers to the horizontal pixel difference of the same scene point in two images, and the depth map represents the distance from each point in the scene to the camera.