A computer vision-based high template support system multi-point displacement real-time monitoring method

Abstract

The application provides a high and large template support system multi-point displacement real-time monitoring method based on computer vision, considers adverse conditions of insufficient light and low image resolution of a construction site, introduces an image resolution improvement technology, further improves measurement accuracy compared with a traditional visual recognition method, has more stable recognition effect in the case of partial obstruction of a marker, solves problems of inconvenient operation, low efficiency, inability to long-time monitoring and single-point monitoring of a traditional monitoring method, and solves problems of insufficient accuracy and low robustness of the traditional visual recognition method when facing adverse scenes such as dim light, partial obstruction of a marker and image blur.

Description

Technical Field

[0001] This invention belongs to the field of building construction technology, and in particular relates to a method for real-time monitoring of multi-point displacement of tall formwork support systems based on computer vision. Background Technology

[0002] Tall formwork support systems are temporary support structures erected during building construction to support concrete components such as beams and slabs. They typically consist of uprights, horizontal bars, vertical diagonal braces, and interlocking joints. The Ministry of Housing and Urban-Rural Development stipulates that concrete formwork support systems must be erected at a height of 8m or more, or with a span of 18m or more, or a total construction load of 15kN / m². 2 Projects with concentrated line loads of 20kN / m or above are considered high-risk sub-projects exceeding a certain scale, requiring the preparation of specialized construction plans and expert review.

[0003] In tall formwork support systems, the vertical uprights, due to their large height and small cross-section, have a high slenderness ratio. During or for a period after concrete pouring, the load on them suddenly increases, easily leading to settlement and lateral deformation. This could cause the structure to lose overall stability and endanger the entire construction project. To prevent accidents, monitoring of the support system is necessary. The displacement of key components of the support system is an important indicator reflecting the structural working state. Horizontal displacement can be used to understand the deformation trend of the frame in the horizontal direction and prevent overall instability. Settlement can be used to understand the deformation trend of the uprights in the vertical direction and is an important factor in predicting whether the frame will collapse.

[0004] Traditional displacement monitoring typically uses equipment such as levels, theodolites, and total stations to measure structural displacement. The measurement work requires manual operation and can only measure one point at a time, which has disadvantages such as inconvenience, low efficiency, and inability to monitor for a long time. Traditional displacement sensors are deployed near the support for automatic measurement, but due to the close distance, the measurement accuracy is easily affected by construction vibrations, and they can only measure a single point.

[0005] Computer vision is a simulation of biological vision. Collected images or videos are analyzed using specific algorithms to derive specific information needed by the user. Using computers to identify targets instead of the human eye is both fast and accurate. Applying computer vision recognition technology to construction monitoring in building engineering can quickly achieve non-contact monitoring of buildings and structures, ensuring construction safety.

[0006] Currently, commonly used computer vision algorithms include:

[0007] (1) Template matching method. This is a method for searching for known targets in a large image. This algorithm requires two images: a predefined template and a region of interest (ROI). The predefined template is moved pixel-by-pixel across the ROI, and a normalized correlation coefficient is used to describe the degree of matching between the template and the region. The location with the largest coefficient is the target's location. This method has a simple formula and high computational efficiency, but it is limited by its inability to adapt to situations where the target rotates, tilts, or changes in size.

[0008] (2) Corner detection method. A corner is the intersection of two sides. This algorithm has good robustness in using corners as tracking targets, and the recognition effect is not affected by the rotation, tilt, or size change of the target. However, its drawback is that it is very easy to make false recognitions and it is difficult to accurately identify the unique corner in the image.

[0009] (3) Optical flow method. The optical flow method achieves continuous tracking of target points between adjacent frames by solving the basic equation of optical flow. It eliminates the need to manually establish the connection between corresponding points in different frames, making it simpler and more efficient. However, one of the basic assumptions of the basic equation of optical flow is that the motion of the tracked target is small. The recognition accuracy depends on the vibration frequency and amplitude of the target structure and the sampling frequency of the camera, which limits its applicability.

[0010] In the monitoring of tall formwork support systems, the frame is tall and the members are dense, and there are many key parts that need to be monitored. The single-point measurement mode of traditional monitoring equipment is difficult to meet the needs of multi-point monitoring. In addition, the complex construction site environment has unfavorable conditions such as dim lighting, concrete seepage obscuring some markers, and low resolution of monitoring cameras. The accuracy and applicability of traditional visual recognition methods are limited. Therefore, all of the above methods have certain limitations. Summary of the Invention

[0011] The purpose of this invention is to solve the problems of traditional monitoring methods being inconvenient to operate, inefficient, unable to monitor for long periods of time, and only capable of single-point monitoring, as well as the problems of traditional visual recognition methods being inaccurate and lacking robustness in adverse scenarios such as dim lighting, partial occlusion of markers, and blurred images. The invention provides a real-time multi-point displacement monitoring method for tall template support systems based on computer vision.

[0012] Therefore, the above-mentioned objective of the present invention is achieved through the following technical solution:

[0013] A method for real-time multi-point displacement monitoring of a tall template support system based on computer vision, characterized in that: the displacement monitoring method includes:

[0014] S1. Design and produce a black and white marker for target detection in advance. The marker has a single prominent corner point in the center, and the rest of the part is composed of smooth lines. The overall pattern has high recognizability.

[0015] S2. Take hundreds of images of the landmarks from different angles, directions, distances, and lighting conditions. Train the YOLOX algorithm to obtain network weights. Select the clearer and higher resolution landmark images to train the SRGAN network to obtain network weights.

[0016] S3. Based on the sensor size S, focal length f, and field of view (FOV) of the surveillance camera, calculate the monitoring distance D that meets the monitoring accuracy requirements:

[0017]

[0018]

[0019] In the formula: Res is the image resolution; n r The required measurement accuracy in the real world, measured in mm; n p The measurement accuracy achievable by the identification algorithm is expressed in pixels.

[0020] S4. Select parts among the uprights, horizontal bars and other components of the template support frame that can reflect the actual deformation and change trend of the support frame, and attach markers as monitoring points; arrange monitoring cameras around the template support frame so that they are within the appropriate working distance D calculated in step S3, adjust the viewing angle so that they can capture all the markers and fix the viewing angle, and record the total number of markers in the camera's field of view as n.

[0021] S5. For any monitoring point, select two points nearby and connect them with a line that lies in a plane perpendicular to the camera's optical axis. Measure the actual length L of each line. i and the pixel length l in the image i Calculate the measurement point ratio coefficient s i :

[0022]

[0023] S6. Before officially starting monitoring, capture an image (image0) of the template support frame in its initial state. For the i-th monitoring point in the image, define a rectangular region (ROI) containing the possible future movement range of the marker at that monitoring point. i Ensure that each ROI contains only one marker and that no two ROIs overlap, i = 1, 2, ..., n;

[0024] S7. Solve for the local super-resolution coordinates of each measurement point in the image0 under the initial state;

[0025] S8. Set the sampling interval Δt and the horizontal displacement alarm value X. limit and vertical displacement alarm value Y limitMonitoring is initiated before critical processes such as rebar tying and concrete pouring, which increase the load on the support frame. An image is captured from the monitoring camera every Δt in real time, and the image at time t is recorded as image_t. t ;

[0026] S9. Solve for the image at time t. t Local super-resolution coordinates of each measuring point:

[0027] S10. Based on the proportional coefficient s of each monitoring point i Resolution improvement factor k i Using the local super-resolution coordinates of the central corner point, calculate the actual displacement of each measuring point at time t:

[0028]

[0029]

[0030]

[0031] In the formula: Let A be the actual displacement of each measuring point at time t; t Both A0 and A1 are coefficient matrices;

[0032] S11. Determine whether the actual displacement of each measuring point exceeds the limit. For any measuring point i (i = 1, 2, ..., n), if |X i-t |>X limit or |Y i-t |>Y limit The monitoring system issued an alarm to alert the staff;

[0033] S12. Repeat steps S9 to S11 to plot the displacement time history curves of each measuring point and update them in real time until the monitoring task ends.

[0034] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0035] As a preferred embodiment of the present invention, step S7 specifically includes the following steps:

[0036] S701. For measurement point i, within the rectangular region ROI... i The trained SRGAN network is used to upscale the image, resulting in a super-resolution image U. i-0 Super-resolution image U i-0 Relative to the rectangular region ROI i The horizontal or vertical pixel multiple is denoted as k. i-0 ;

[0037] S702, in super-resolution image U i-0The rectangular region where the marker is located is identified using the YOLOX algorithm within the area, denoted as B. i-0 ;

[0038] S703, in rectangular region B i-0 The Shi-Tomasi corner detection algorithm is used within the area to identify the coordinates of the center corners of the landmarks, and these coordinates are then transformed to the ROI. i Using a coordinate system, we obtain the local super-resolution coordinates (x, y) of measurement point i. i-0 y i-0 );

[0039] S704. Repeat steps S701 to S703 until the local super-resolution coordinates of all n measurement points are obtained.

[0040] As a preferred embodiment of the present invention, step S9 specifically includes the following steps:

[0041] S901. For measurement point i, within the rectangular region ROI... i The trained SRGAN network is used to upscale the image, resulting in a super-resolution image U. i-t Super-resolution image U i-t Relative to the rectangular region ROI i The horizontal or vertical pixel multiple is denoted as k. i-t ;

[0042] S902, in super-resolution image U i-t The rectangular region where the marker is located is identified using the YOLOX algorithm within the area, denoted as B. i-t ;

[0043] S903, in rectangular region B i-t The Shi-Tomasi corner detection algorithm is used within the area to identify the coordinates of the center corners of the landmarks, and these coordinates are then transformed to the ROI. i Using a coordinate system, we obtain the local super-resolution coordinates (x, y) of measurement point i. i-t y i-t );

[0044] S904. Repeat steps S901 to S903 until the local super-resolution coordinates of all n measurement points are obtained.

[0045] This invention provides a real-time multi-point displacement monitoring method for tall formwork support systems based on computer vision. Compared with traditional displacement monitoring methods for tall formwork support systems, this invention has the following advantages:

[0046] (1) Compared with traditional manual measurement methods, it overcomes the shortcomings of inconvenient operation, low efficiency and inability to monitor for a long time when using level, theodolite and total station measurement, and has the advantages of all-time, real-time and automated measurement.

[0047] (2) It overcomes the shortcomings of traditional displacement sensors, such as limited distance, single measurement point, and inconvenient installation. The long-distance non-contact measurement mode has a wider monitoring range and stronger environmental adaptability.

[0048] (3) High efficiency, convenience and low cost, meeting the needs of flexible layout on construction site, multi-point synchronous measurement and real-time displacement output;

[0049] (4) Considering the disadvantages of insufficient light and low image resolution at the construction site, image resolution enhancement technology is introduced, which further improves the measurement accuracy compared with traditional visual recognition methods, and has a more stable recognition effect when the marker is partially obscured. Attached Figure Description

[0050] Figure 1 This is a field test schematic diagram of the real-time multi-point displacement monitoring method for tall template support systems based on computer vision provided by the present invention.

[0051] In the diagram: ① is the tall formwork support system; ② is the marker affixed to the pole; ③ is the field of view of the monitoring camera; ④ is the monitoring camera; ⑤ is the data analysis center and results display screen.

[0052] Figure 2 This is an example of using an SRGAN network to upscale an image.

[0053] Figure 3 Examples of YOLOX algorithm for recognizing complete and partially occluded landmarks. Detailed Implementation

[0054] The present invention will be described in further detail with reference to the flowchart and accompanying drawings.

[0055] Figure 1 This is a field test schematic diagram of the real-time multi-point displacement monitoring method for a tall template support system based on computer vision provided by the present invention. The real-time multi-point displacement monitoring method for a tall template support system based on computer vision includes the following steps:

[0056] S1. Design and manufacture black and white markers for target detection in advance, such as... Figure 1 As shown in ②, the logo has a single prominent corner point at its center, and the rest of the parts are composed of smooth lines, giving the overall design a high degree of recognizability.

[0057] S2. Take 500 images of the landmark under different angles, directions, distances, and lighting conditions. Train the YOLOX algorithm to obtain network weights. Select the clearer and higher resolution landmark images and train the SRGAN network to obtain network weights.

[0058] S3, such as Figure 1 As shown, firstly, the maximum field of view (FOV) of the camera is calculated based on the camera resolution and accuracy requirements. Then, the monitoring distance D that meets the monitoring accuracy requirements is calculated based on the sensor size S, focal length f, and FOV of the monitoring camera.

[0059] S4, such as Figure 1 Attach markers ② to the uprights of the template support frame ① as monitoring points; arrange monitoring cameras ④ around the template support frame so that they are within the working distance D, adjust the viewing angle so that they can capture all the markers and fix the viewing angle, and record the total number of markers in the camera's field of view as n (there are 6 measuring points in the figure).

[0060] S5. For any monitoring point, select two points nearby and connect them with a line (this line should be in a plane perpendicular to the camera's optical axis), and measure its actual length L. i and the pixel length l in the image i Calculate the measurement point ratio coefficient s i :

[0061] S6. Before officially starting monitoring, capture an image (image0) of the template support frame in its initial state. For the i-th monitoring point in the image, define a rectangular region (ROI) containing the possible future movement range of the marker at that monitoring point. i ;

[0062] S7. Solve for the local super-resolution coordinates of each measurement point in the initial state image0:

[0063] S701. For measurement point i, in the ROI... i The trained SRGAN network is used internally to enhance image resolution, such as... Figure 2 As shown, the super-resolution image U is obtained. i-0 U i-0 Compared to ROI i The horizontal or vertical pixel multiple is denoted as k. i-0 ;

[0064] S702, in U i-0 The rectangular region where the marker is located is identified using the YOLOX algorithm within the area, denoted as B. i-0 , Figure 3 A schematic diagram illustrating the YOLOX algorithm for identifying landmarks;

[0065] S703, in B i-0 The Shi-Tomasi corner detection algorithm is used within the area to identify the coordinates of the center corners of the landmarks, and these coordinates are then transformed to the ROI. i Using a coordinate system, we obtain the local super-resolution coordinates (x, y) of measurement point i. i-0 y i-0 );

[0066] S704. Repeat steps S701 to S703 until the local super-resolution coordinates of all n measurement points are obtained.

[0067] S8. Set the sampling interval Δt and the horizontal displacement alarm value X. limit and vertical displacement alarm value Y limit Monitoring is initiated before the start of important construction procedures, and an image from the monitoring camera is captured in real time every Δt. The image at time t is recorded as image. t ;

[0068] S9. Following steps S701 to S704, solve for the image at time t. t Local super-resolution coordinates (x) of each measuring point i-t y i-t );

[0069] S10. Based on the proportional coefficient s of each monitoring point i Resolution improvement factor k i Using the local super-resolution coordinates of the central corner point, calculate the actual displacement of each measuring point at time t.

[0070] S11. Determine whether the actual displacement of each measuring point exceeds the limit. For any measuring point i (i = 1, 2, ..., n), if |X i-t |>X limit or |Y i-t |>Y limit The monitoring system issued an alarm to alert the staff;

[0071] S12. Repeat steps S9 to S11 to plot the displacement time history curves of each measuring point and update them in real time until the monitoring task ends. Data processing and result display are completed through step ⑤.

[0072] The above specific embodiments are used to explain and illustrate the present invention, and are only preferred embodiments of the present invention, not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for real-time monitoring of multi-point displacement in a tall template support system based on computer vision, characterized in that: The real-time displacement monitoring method includes: S1. Design and manufacture a black and white marker for target detection in advance. The marker has a single prominent corner point in the center, and the rest of the part is composed of smooth lines. S2. Take hundreds of images of the landmarks from different angles, directions, distances, and lighting conditions. Train the YOLOX algorithm to obtain network weights. Select the clear and high-resolution landmark images to train the SRGAN network to obtain network weights. S3. Based on the sensor size of the surveillance camera S ,focal length f and field of vision FOV Calculate the monitoring distance that meets the monitoring accuracy requirements. D : （1） （2） In the formula: Res Image resolution; n r The measurement accuracy is the same as that achieved in the real world, and the unit is mm; n p The measurement accuracy achieved by the identification algorithm is expressed in pixels. S4. Select locations on the uprights and horizontal bars of the formwork support frame that reflect the actual deformation and trend of the support, and affix markers as monitoring points; arrange monitoring cameras around the formwork support frame, placing them at the appropriate working distance calculated in step S3. D Within this range, adjust the viewing angle to capture all landmarks and then fix the angle. The total number of landmarks within the camera's field of view is recorded as follows: n ; S5. For any monitoring point, select two points nearby and connect them with a line that lies in a plane perpendicular to the camera's optical axis. Measure the actual lengths of these two lines. L i and pixel length in the image l i Calculate the measurement point ratio coefficient s i : （3） S6. Before officially starting monitoring, capture image0 of the template support frame in its initial state, and analyze the first... i Each monitoring point is defined as a rectangular region of interest (ROI) encompassing the future movement range of the marker at that monitoring point. i Ensure that each ROI contains only one marker and that no two ROIs overlap. i =1, 2, ..., n ; S7. Solve for the local super-resolution coordinates of each measurement point in the image0 under the initial state; S8. Set the sampling interval Δ t Horizontal displacement alarm value X limit and vertical displacement alarm value Y limit Monitoring is initiated before the start of critical procedures such as rebar tying and concrete pouring, which increase the load on the support frame. Monitoring is performed every Δ... t Capture an image from a surveillance camera in real time. t The time-lapse image is denoted as image. t ; S9, Solve t Moment Image t Local super-resolution coordinates of each measuring point: S10. Based on the proportional coefficient of each monitoring point s i Resolution improvement factor k i And the local super-resolution coordinates of the central corner point, calculate the coordinates of each measuring point in t Actual displacement at time: （4） （5） （6） In the formula: for t Actual displacement of each measuring point at any given time; A t and A 0 are all coefficient matrices; S11. Determine whether the actual displacement of each measuring point exceeds the limit. For any measuring point... i ( i =1, 2, ..., n If | X i-t | > X limit or | Y i-t |> Y limit The monitoring system issued an alarm to alert the staff; S12. Repeat steps S9 to S11 to plot the displacement time history curves of each measuring point and update them in real time until the monitoring task ends.

2. The method for real-time multi-point displacement monitoring of a tall template support system based on computer vision according to claim 1, characterized in that: Step S7 specifically includes the following steps: S701, For measuring points i In the rectangular region ROI i The trained SRGAN network is used to upscale the image, resulting in a super-resolution image. U i-0 Super-resolution images U i-0 Relative to the rectangular region ROI i The horizontal or vertical pixel multiple is denoted as k i-0 ; S702, in super-resolution images U i-0 The rectangular region where the marker is located is identified using the YOLOX algorithm within the area, denoted as . B i-0 ; S703, in a rectangular area B i-0 The Shi-Tomasi corner detection algorithm is used within the area to identify the coordinates of the center corners of the landmarks, and these coordinates are then transformed to the ROI. i Coordinate system, to obtain the measuring points i Local super-resolution coordinates ( x i-0 , y i-0 ); S704. Repeat steps S701 to S703 until all are obtained. n Local super-resolution coordinates of each measuring point.

3. The method for real-time multi-point displacement monitoring of a tall template support system based on computer vision according to claim 1, characterized in that: Step S9 specifically includes the following steps: S901, For measuring points i In the rectangular region ROI i The trained SRGAN network is used to upscale the image, resulting in a super-resolution image. U i-t Super-resolution images U i-t Relative to the rectangular region ROI i The horizontal or vertical pixel multiple is denoted as k i-t ; S902, in super-resolution images U i-t The rectangular region where the marker is located is identified using the YOLOX algorithm within the area, denoted as . B i-t ; S903, in the rectangular area B i-t The Shi-Tomasi corner detection algorithm is used within the area to identify the coordinates of the center corners of the landmarks, and these coordinates are then transformed to the ROI. i Coordinate system, to obtain the measuring points i Local super-resolution coordinates ( x i-t , y i-t ); S904. Repeat steps S901 to S903 until all are obtained. n Local super-resolution coordinates of each measuring point.

Images