A subway foundation pit over-excavation monitoring method based on binocular camera and computer vision

Abstract

The present invention provides a method for monitoring over-excavation of subway foundation pits based on binocular cameras and computer vision, which belongs to the field of monitoring over-excavation of subway foundation pits. The method aims to solve the problems that over-excavation monitoring mostly relies on manpower and has poor real-time performance; homogeneous soil is difficult to distinguish multiple planes; a single measuring instrument has poor tolerance; the working surface is uneven and it is difficult to represent the average height; and it is impossible to monitor whether the steel supports are erected in time. The binocular camera is used to collect two-dimensional images and three-dimensional point clouds, and then the point clouds are merged, and semantic segmentation is performed through pre-training; after the ground plane is converted into a plane equation, the distance from the point to the plane equation in the soil point cloud is divided, and the three-dimensional space plane fitting is used to accurately segment different working planes to determine whether the working surface is over-excavated or the slope is too large; and whether the steel supports are over-excavated is determined based on the soil point cloud segmentation results and the excavator point cloud. It saves manpower and material resources and has high real-time performance; it can improve the monitoring accuracy; and it can dynamically monitor whether the working surface is over-excavated and whether the steel supports are not erected in time while the excavation work is still in progress.

Description

Technical Field

[0001] The present invention relates to the technical field of subway foundation pit over-excavation, and in particular to a subway foundation pit over-excavation monitoring method based on a binocular camera and computer vision. Background Art

[0002] In deep foundation pit projects, the working conditions of the excavation must be consistent with the working conditions of the scheme design. This is the prerequisite for ensuring the safety of the foundation pit. When the foundation pit is excavated to the support design elevation, grooves should be cut and steel supports or concrete supports should be installed in a timely manner. Only after the supports meet the design requirements can excavation continue, which requires following the principle of "support first, then dig." From the analysis of the time and space effects of foundation pit excavation, we know that (the time and space effect refers to the fact that after the foundation pit is excavated, the upper soil is excavated, which is equivalent to unloading the base soil, breaking the original load balance, causing stress release in the base soil, and causing deformation and uplift of the foundation soil). The deformation of the retaining structure is related to the size of the unsupported exposed area and the length of exposure time. Therefore, strictly following the working conditions of the foundation pit project scheme design, excavating first and then digging, and adding supports in time are the guarantee for controlling the deformation of the foundation pit wall and the corresponding ground displacement and settlement.

[0003] "Strictly prohibit over-excavation" is another important principle that must be followed in foundation pit excavation. Excessive excavation is the biggest enemy in foundation pit excavation. It can cause unnecessary losses at the least and serious accidents at the most. It should be prevented during construction. Its harm is mainly manifested in the following four aspects:

[0004] (1) Over-excavation increases the exposed area of the retaining structure and delays the support installation time, which will significantly increase the deformation of the retaining structure wall and the corresponding ground displacement and settlement.

[0005] (2) If the bottom of the foundation pit is over-excavated and the retaining wall is not buried deep enough, the bottom of the retaining wall will move and the strength will be damaged.

[0006] (3) Over-excavation of the pit bottom also increases the total amount of soil unloading, increases the amount of soil bulge at the pit bottom, and also increases the ground settlement around the pit; over-excavation of the pit bottom also disturbs the foundation soil, reducing the bearing capacity of the foundation soil.

[0007] (4) Over-excavation of the pit bottom also prevents the timely pouring of the bottom plate, leaving the pit bottom exposed for a long time. Due to the rheological properties of clay, the displacement of the soil in the passive pressure zone of the wall and the displacement of the soil outside the wall into the pit will increase, thereby increasing surface settlement, especially on rainy days.

[0008] The following difficulties exist in the existing over-excavation detection process:

[0009] First of all, subway foundation pit excavation often adopts layered and segmented excavation. At present, there is no effective automated monitoring method for over-excavation during the entire excavation process. Generally, monitoring still relies on manual operation of instruments. Since the excavation cycle is very long and the operation continues day and night without special circumstances, this monitoring method is difficult to cover the entire excavation process, lacks real-time performance, and is highly uncontrollable.

[0010] Secondly, since there are many and constantly changing layered and segmented excavation working surfaces, each working surface is mainly composed of soil. If automated monitoring is to be achieved, it is necessary to overcome the problem of distinguishing homogeneous materials.

[0011] In addition, the steel supports of the foundation pit are supported as they are excavated, and are very dense, which will cause great obstruction to various measuring instruments. Therefore, using a single instrument to cover the entire monitoring process requires the use of very expensive precision instruments. Precision instruments have poor tolerance to complex and harsh construction environments and are extremely susceptible to damage or failure, and cannot meet the real-time requirements for monitoring the excavation process.

[0012] Finally, the definition of overexcavation is complex. Statically, overexcavation occurs when the excavation depth of the current soil plane exceeds the design elevation. However, the soil plane at the construction site is inevitably not perfectly flat, so a method is needed to effectively represent the average height of the current soil plane. Dynamically, if steel supports are not erected in a timely manner while excavation continues, this is also a form of overexcavation. A comprehensive assessment of both forms is necessary to determine overexcavation. Summary of the Invention

[0013] The technical problems to be solved by the present invention are:

[0014] In order to solve the problems that the existing over-excavation monitoring mostly relies on manpower and lacks real-time performance; each working surface is mainly composed of soil, which is difficult to distinguish through automatic monitoring; a single measuring instrument has poor tolerance and is difficult to meet implementation requirements; the working surface is uneven and it is difficult to indicate the average height; and it is impossible to monitor over-excavation in the erection of steel supports.

[0015] The present invention is to solve the above technical problems using the following technical solutions:

[0016] The present invention provides a method for monitoring over-excavation of a subway foundation pit based on a binocular camera and computer vision, comprising the following steps:

[0017] Step 1: Place several binocular cameras along the direction of the subway foundation pit excavation, and use the binocular cameras to take images at different positions to obtain corresponding two-dimensional images and three-dimensional point clouds;

[0018] Step 2: Using the point cloud registration method, the point clouds obtained in step 1 are merged and duplicate points are removed to obtain a merged point cloud. The semantic segmentation model trained in the previous step is used to perform semantic segmentation on the 2D image to identify the soil, steel supports, and excavator. These are mapped to the merged point cloud, thereby segmenting the soil, steel supports, and excavator point clouds.

[0019] Step 3: Clean the soil and steel support point clouds obtained by semantic segmentation in step 2 to remove scattered points. After the soil is cleaned, the coordinates are converted into a quaternion array according to the previous point cloud merging coordinate conversion and merged into a complete soil body; the steel support point cloud is first merged and then cleaned to obtain a complete soil point cloud and a complete steel support point cloud;

[0020] Step 4: Use the Euler coordinate transformation based on the camera IMU parameters to convert the ground plane to the plane equation in the camera coordinate system; calculate the vertical distance from each point of the complete soil point cloud to the ground plane and arrange them from low to high; use the polyline fitting method to fit the trend, and preliminarily divide the point cloud according to the point cloud point numbers corresponding to the polyline intersection points, so as to roughly obtain the plane or slope to which each point in the point cloud belongs;

[0021] Step 5: Perform three-dimensional space plane fitting on the clusters of divided points to obtain the plane equation in the camera coordinate system. By using plane fitting, the equation of the plane to which the current point cloud belongs can be fitted;

[0022] Step 6: Find the intersection line of the fitted planes and project them onto the ground plane to obtain a segmentation plane perpendicular to the ground plane. Based on the position of the point cloud before and after the segmentation plane, accurately segment the soil point cloud, including multiple planes and slopes. Compare the segmentation results with the design elevation to determine whether there is over-excavation on the plane and whether the slope is too steep. If any of the above phenomena exist, an early warning will be issued, requiring on-site personnel to check.

[0023] Step 7. Based on the location of the excavator point cloud, determine the slope being excavated and the plane corresponding to the need to set up steel supports. Based on the dividing plane between the plane and the slope, check whether there are steel support point clouds within the standard interval of the steel supports. If there are, mark it to indicate that they have been set up in time. If not, it means that they have not been set up in time, there is a possibility of over-excavation, and on-site personnel need to check;

[0024] Step 8: Use motion detection method to detect whether the excavator is performing excavation operation. If it is, the height of the plane formed by the excavation is in a critical state, and the position of the steel support is in a critical state, then remind the staff to strengthen monitoring.

[0025] Furthermore, in step one, when the number of binocular cameras is one, the binocular camera can be placed on a walkable gantry crane; when the number of binocular cameras is greater than one, multiple binocular cameras can be placed above the horizontal plane of the first layer of steel support in the foundation pit.

[0026] Furthermore, the spacing is arranged so that the fields of view of the two cameras overlap by more than 1 / 3.

[0027] Furthermore, in step one, the foundation pit can be divided into three parts: left, middle, and right during image acquisition, and the left, middle, and right parts can be imaged.

[0028] Furthermore, in the point cloud segmentation process, the two-dimensional images and three-dimensional point clouds collected from the left, middle and right parts acquired by the binocular camera are subjected to coordinate transformation matrix and coincidence elimination to obtain the soil point cloud, steel support point cloud and excavator point cloud.

[0029] Furthermore, the soil part is statistically removed of outliers before performing coordinate transformation matrix and eliminating coincident points.

[0030] Furthermore, risk judgment is mainly divided into three aspects. When judging whether the excavation surface of the foundation pit is over-excavated, the plane angle between the fitting plane and the site plane is calculated, the working surface and the excavation surface are distinguished, and it is judged whether the plane angle exceeds the design value, and the over-steep excavation surface is marked on the soil point cloud; when judging whether there is a situation where the steel support is not erected in time and the excavation work is still in progress, the split plane corresponding to each plane and slope of the soil point cloud and the height of each point are calculated to exceed the probability of the design elevation, and whether the working surface exceeds the design elevation is obtained. The distance from the steel support to the split plane is calculated for the detection of the working status of the excavator, and the steel supports within the range and the steel supports outside the range are marked on the point cloud merger, and a risk warning is issued for the steel supports outside the range; when monitoring the load status of the foundation pit edge area, the target in the foundation pit edge area is detected, and the load status of the foundation pit edge area is judged based on the number and weight of the excavators and whether the excavation surface is over-excavated.

[0031] Furthermore, in step seven, whether there is over-excavation is determined based on whether there is a steel support point cloud within a range of 1-1.2 times the standard steel support interval.

[0032] Furthermore, the binocular camera is a D455 camera.

[0033] Furthermore, in step 2, preliminary training is performed using the DeepLabV3+ model.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention discloses a method for monitoring overexcavation of a subway foundation pit based on binocular cameras and computer vision. Multiple binocular cameras are used to detect overexcavation. Compared with using a single, precise instrument, binocular cameras are more suitable for use in complex and harsh construction environments. They are relatively resistant to damage and malfunction, and can meet the implementation requirements for monitoring the excavation process. Compared with precise instruments, monitoring costs can be saved. If a single binocular camera is damaged, it can be easily repaired and replaced, and the cost is low. Compared with manual monitoring, the method can cover the entire excavation process that operates day and night, has high real-time performance and controllability, and does not waste manpower.

[0036] The soil is segmented during the monitoring process to avoid the difficulty in distinguishing different planes and slopes of the same soil material. During the automated monitoring process, different areas can be divided and over-excavation at the corresponding locations can be monitored, resulting in more accurate monitoring results. In addition, if the soil surface is uneven during actual excavation, the average height of the soil surface can be obtained, further increasing the accuracy of the judgment.

[0037] A movable binocular camera or multiple binocular cameras can avoid image occlusion when collecting point clouds. Through semantic segmentation, steel supports and soil can be distinguished to realize over-excavation monitoring of soil. It can also dynamically monitor whether there is a situation where steel supports are not erected in time while excavation work is still in progress, avoiding another form of over-excavation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a method for monitoring over-excavation of a subway foundation pit based on a binocular camera and computer vision in an embodiment of the present invention;

[0039] Figure 2 Schematic diagram of the position of the binocular camera in an embodiment of the present invention;

[0040] Figure 3 This is a model diagram of a subway foundation pit overexcavation monitoring method based on binocular cameras and computer vision in an embodiment of the present invention;

[0041] Figure 4 This is a flowchart of point cloud segmentation in an embodiment of the present invention;

[0042] Figure 5 This is a flowchart of point cloud merging in an embodiment of the present invention;

[0043] Figure 6 Flowchart of soil segmentation in an embodiment of the present invention;

[0044] Figure 7 Flowchart of risk assessment in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In the description of the present invention, it should be noted that the terminology in each embodiment, such as "up", "down", "front", "back", "left", "right", etc., which indicate directions, are only for simplifying the description of the positional relationship based on the drawings in the specification, and do not mean that the referred elements and devices must be operated in accordance with the specific directions and defined operations and methods and structures in the specification. Such directional nouns do not constitute a limitation to the present invention.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0047] Specific implementation plan 1: Combined Figures 1 to 7 As shown, the present invention provides a method for monitoring over-excavation of a subway foundation pit based on a binocular camera and computer vision, comprising the following steps:

[0048] Step 1: Arrange several binocular cameras along the excavation direction of the subway foundation pit. When there is only one binocular camera, the binocular camera can be placed on a walkable gantry crane. When there are more than one binocular camera, multiple binocular cameras can be placed above the horizontal plane of the first layer of steel support in the foundation pit, with the spacing between the two cameras overlapping by more than 1 / 3 to form a binocular camera array. Use one or more binocular cameras to perform imaging at different positions to obtain corresponding two-dimensional images and three-dimensional point clouds. Specifically, the foundation pit is divided into three parts: left, middle, and right, and the left, middle, and right parts are imaged. The point cloud refers to a spatial point with three-dimensional coordinates in the coordinate system of the camera.

[0049] In actual engineering applications, the method described in this study should use a binocular camera with a detection range of more than 10m and an accuracy of less than 2%, and be equipped with an IMU module. The binocular cameras should be arranged along the edge of the construction site, and the detection ranges of adjacent cameras should overlap by more than 1 / 3 to meet the requirements of point cloud registration. Binocular cameras can also be arranged within the site or in layers as needed. Before use, this method requires pre-determining the camera height from the site plane and calibration results, the physical characteristics of the target safety risk, and the safety coupling conditions. It should also conduct targeted point cloud registration and verify the accuracy of the target detection, semantic segmentation, and motion detection models. During use, warnings raised by the method should be promptly investigated; attention should be paid to the protection of the binocular cameras, and the layout angle or position of the binocular cameras should be adjusted in a timely manner according to the construction progress.

[0050] Step 2: Using the point cloud registration method, the point clouds at the left, center, and right positions are merged and duplicate points are removed to obtain a merged point cloud. The merged point cloud here is the point cloud of the entire foundation pit. The semantic segmentation model trained in the early stage is used to perform semantic segmentation on the two-dimensional images at the left, center, and right positions, identify the soil, steel support, and excavator, and map them to the merged point cloud, thereby segmenting the soil, steel support, and excavator point clouds. Since the pixels of the images at the left, center, and right positions correspond one-to-one to the points of the point cloud, the point cloud can be segmented using the semantic segmentation results.

[0051] Step 3: Clean the soil and steel support point clouds obtained by semantic segmentation in step 2 to remove scattered points and ensure the reliability of the point cloud data. After the soil is cleaned, the coordinates are converted into a quaternion array according to the previous point cloud merging coordinate conversion and merged into a complete soil body; the steel support point cloud is first merged and then cleaned, thereby obtaining a complete soil point cloud and a complete steel support point cloud; the point cloud merging here refers to the merging of separate point clouds of the soil and steel support; the soil is cleaned first and then merged because the number of points in the soil point cloud is large, and the cleaning effect of each part is better without mutual interference; the steel support is because the number of points in the steel support point cloud is relatively small, and the steel supports are independent of each other in space. Cleaning them separately will have a relatively poor effect and will lose a large number of valid points. The cleaning effect is better after merging;

[0052] Step 4: Use the Euler coordinate transformation based on the camera IMU parameters to convert the ground plane to the plane equation in the camera coordinate system; calculate the vertical distance from each point of the complete soil point cloud to the ground plane and arrange them from low to high; use the polyline fitting method to fit the trend, and preliminarily divide the point cloud according to the point cloud point numbers corresponding to the polyline intersection points, so as to roughly obtain the plane or slope to which each point in the point cloud belongs;

[0053] Step 5: Perform 3D plane fitting on the clusters of divided points to obtain the plane equation in the camera coordinate system. Although there is still some occlusion due to the steel support, this problem is largely eliminated by registering the point clouds at the left, center, and right positions. Plane fitting can fit the equation of the plane to which the current point cloud belongs, representing the vast majority of the point cloud points and automatically eliminating the clutter caused by the coarse classification in the previous step.

[0054] Step 6: Find the intersection line of the fitted planes and project them onto the ground plane to obtain a segmentation plane perpendicular to the ground plane. Based on the position of the point cloud before and after the segmentation plane, accurately segment the soil point cloud, including multiple planes and slopes. Compare the segmentation results with the design elevation to determine whether there is over-excavation on the plane and whether the slope is too steep. If any of the above phenomena exist, an early warning will be issued, requiring on-site personnel to check.

[0055] Step 7. According to the location of the excavator point cloud, determine the slope being excavated and the plane corresponding to the need to set up steel supports. According to the dividing plane between the plane and the slope, check whether there is a steel support point cloud within the range of 1-1.2 times the standard interval of the steel support. The specific interval range can be adjusted according to the actual situation of the project. If there is, it will be marked to indicate that it has been set up in time. If not, it means that it has not been set up in time, there is a possibility of over-excavation, and on-site personnel need to check;

[0056] Among them, checking whether there is a steel support point cloud within 1.2 times the standard steel support interval is preferred;

[0057] Step 8: Use motion detection method to detect whether the excavator is performing excavation operation. If it is, the height of the plane formed by the excavation is in a critical state, and the position of the steel support is in a critical state, then remind the staff to strengthen monitoring.

[0058] Multiple binocular cameras are used to detect over-excavation of foundation pits. Compared with using a single and precise instrument, such as a 3D laser scanner, binocular cameras are more suitable for use in more complex and harsh construction environments. They are relatively resistant to damage and failure, and can meet the implementation requirements of excavation process monitoring. Compared with precision instruments, they can save monitoring costs. If a single binocular camera is damaged, it is easy to repair and replace, and the cost is low. Compared with manual monitoring, they can cover the entire excavation process that operates around the clock, have higher real-time performance and controllability, and do not waste manpower.

[0059] The soil is segmented during the monitoring process to avoid the difficulty in distinguishing different planes and slopes of the same soil material. During the automated monitoring process, different areas can be divided and over-excavation at the corresponding locations can be monitored, resulting in more accurate monitoring results. In addition, if the soil surface is uneven during actual excavation, the average height of the soil surface can be obtained, further increasing the accuracy of the judgment.

[0060] A movable binocular camera or multiple binocular cameras can avoid image occlusion when collecting point clouds. Through semantic segmentation, steel supports and soil can be distinguished to realize over-excavation monitoring of soil. It can also dynamically monitor whether there is a situation where steel supports are not erected in time while excavation work is still in progress, avoiding another form of over-excavation.

[0061] Preferably, in step 1, the binocular camera is a D455 camera.

[0062] Preferably, in step 2, preliminary training is performed using the DeepLabV3+ model.

[0063] Preferably, during the point cloud segmentation process, the two-dimensional images and three-dimensional point clouds collected from the left, middle and right parts obtained by the binocular camera can be subjected to coordinate transformation matrix and coincidence elimination. The soil part can be statistically removed of outliers before and after the coordinate transformation matrix and coincidence elimination, and finally the soil point cloud, steel support point cloud and excavator point cloud are obtained.

[0064] Specific implementation plan 2: Combined Figures 1 to 7 As shown, risk assessment is mainly divided into three aspects. When judging whether the excavation surface of the foundation pit is over-excavated, it is necessary to calculate the plane angle between the fitting plane and the site plane, distinguish the working surface and the excavation surface, and determine whether the plane angle exceeds the design value. The over-steep excavation surface is marked on the soil point cloud. When judging whether the steel support is not erected in time while the excavation work is still in progress, the probability of exceeding the design elevation is calculated by using the planes and slopes of the soil point cloud, namely the AE area, the corresponding split planes and the height of each point. Then, whether the working surface exceeds the design elevation is determined and the corresponding position is marked on the soil point cloud. By detecting the working status of the excavator, the distance from the steel support to the split plane is calculated, and the steel supports within and outside the range are marked on the point cloud merge, thereby issuing a risk warning. When monitoring the load status of the foundation pit edge area, the load status of the foundation pit edge area is determined by detecting the target in the foundation pit edge area, combining the number and weight of the excavators and whether the excavation surface is over-excavated. The working surface is a plane, the excavation surface is a slope, and the working surface and the excavation surface are set at intervals. The other combinations and connection relationships of this embodiment are the same as those of the specific embodiment one.

[0065] Although the present invention is disclosed as above, the scope of protection disclosed by the present invention is not limited thereto. Those skilled in the art of the present invention may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A method for monitoring over-excavation of subway foundation pits based on binocular cameras and computer vision, characterized in that: The following steps are involved: Step 1: Place several binocular cameras along the direction of the subway foundation pit excavation, and use the binocular cameras to take images at different positions to obtain corresponding two-dimensional images and three-dimensional point clouds; Step 2: Using the point cloud registration method, the point clouds obtained in step 1 are merged and duplicate points are removed to obtain a merged point cloud. The semantic segmentation model trained in the previous step is used to perform semantic segmentation on the 2D image to identify the soil, steel supports, and excavator. These are mapped to the merged point cloud, thereby segmenting the soil, steel supports, and excavator point clouds. Step 3: Clean the soil and steel support point clouds obtained by semantic segmentation in step 2 to remove scattered points. After the soil is cleaned, the coordinates are converted into a quaternion array according to the previous point cloud merging coordinate conversion and merged into a complete soil body; the steel support point cloud is first merged and then cleaned to obtain a complete soil point cloud and a complete steel support point cloud; Step 4: Use the Euler coordinate transformation based on the camera IMU parameters to transform the ground plane into the plane equation in the camera coordinate system; Calculate the vertical distance from each point of the complete soil point cloud to the ground plane and arrange them from low to high; use the polyline fitting method to fit the trend, and perform a preliminary division of the point cloud according to the number of the point cloud points corresponding to the polyline intersection points, and roughly obtain the plane or slope to which each point in the point cloud belongs; Step 5: Perform three-dimensional space plane fitting on the clusters of divided points to obtain the plane equation in the camera coordinate system. By using plane fitting, the equation of the plane to which the current point cloud belongs can be fitted; Step 6: Find the intersection line of the fitted planes and project them onto the ground plane to obtain a segmentation plane perpendicular to the ground plane. Based on the position of the point cloud before and after the segmentation plane, accurately segment the soil point cloud, including multiple planes and slopes. Compare the segmentation results with the design elevation to determine whether there is over-excavation on the plane and whether the slope is too steep. If any of the above phenomena exist, an early warning will be issued, requiring on-site personnel to check. Step 7. Based on the location of the excavator point cloud, determine the slope being excavated and the plane corresponding to the need to set up steel supports. Based on the dividing plane between the plane and the slope, check whether there are steel support point clouds within the standard interval of the steel supports. If there are, mark it to indicate that they have been set up in time. If not, it means that they have not been set up in time, there is a possibility of over-excavation, and on-site personnel need to check; Step 8: Use motion detection method to detect whether the excavator is performing excavation operation. If it is, the height of the plane formed by the excavation is in a critical state, and the position of the steel support is in a critical state, then remind the staff to strengthen monitoring.

2. The method for monitoring overexcavation of a subway foundation pit based on a binocular camera and computer vision according to claim 1, characterized in that: In step 1, when the number of binocular cameras is one, the binocular camera can be placed on a walkable gantry crane; when the number of binocular cameras is greater than one, multiple binocular cameras can be placed above the horizontal surface of the first layer of steel support in the foundation pit.

3. The method for monitoring overexcavation of a subway foundation pit based on a binocular camera and computer vision according to claim 2, characterized in that: The layout interval is such that the fields of view of the two cameras overlap by more than 1 / 3.

4. The method for monitoring overexcavation of a subway foundation pit based on a binocular camera and computer vision according to claim 3, characterized in that: In step 1, the foundation pit can be divided into three parts: left, middle and right when collecting images, and the left, middle and right parts can be imaged.

5. The method for monitoring overexcavation of a subway foundation pit based on a binocular camera and computer vision according to claim 4, characterized in that: In the point cloud segmentation process, the two-dimensional images and three-dimensional point clouds collected from the left, middle and right parts of the binocular camera are subjected to coordinate transformation matrix and coincidence elimination to obtain soil point cloud, steel support point cloud and excavator point cloud.

6. The method for monitoring overexcavation of a subway foundation pit based on a binocular camera and computer vision according to claim 5, characterized in that: The soil part is statistically removed of outliers before performing coordinate transformation matrix and eliminating coincident points.

7. The method for monitoring overexcavation of a subway foundation pit based on a binocular camera and computer vision according to claim 5, characterized in that: Risk assessment is mainly divided into three aspects. When judging whether the excavation surface of the foundation pit is over-excavated, the plane angle between the fitting plane and the site plane is calculated, the working surface and the excavation surface are distinguished, and it is judged whether the plane angle exceeds the design value. The steep excavation surface is marked on the soil point cloud. When judging whether the steel support is not erected in time and the excavation work is still in progress, the probability of exceeding the design elevation is calculated for the split planes and heights of each point corresponding to each plane and slope of the soil point cloud. Whether the working surface exceeds the design elevation is determined, the distance from the steel support to the split plane is calculated for the detection of the excavator's working status, and the steel supports within the range and those outside the range are marked on the point cloud merging, and risk warnings are issued for steel supports outside the range. When monitoring the load status of the foundation pit edge area, the target of the foundation pit edge area is detected, and the load status of the foundation pit edge area is judged in combination with the number and weight of the excavators and whether the excavation surface is over-excavated.

8. The method for monitoring overexcavation of a subway foundation pit based on a binocular camera and computer vision according to claim 7, characterized in that: In step seven, whether there is over-excavation is determined based on whether there is a steel support point cloud within the range of 1-1.2 times the standard steel support interval.

9. The method for monitoring overexcavation of a subway foundation pit based on a binocular camera and computer vision according to claim 8, characterized in that: The binocular camera is a D455 camera.

10. The method for monitoring overexcavation of a subway foundation pit based on a binocular camera and computer vision according to claim 9, characterized in that: In step 2, preliminary training is performed using the DeepLabV3+ model.

Images