A method for measuring the flatness of prefabricated pavement panels based on machine vision

Abstract

The present application relates to the field of road construction, and discloses a method for measuring the flatness of prefabricated road panels based on machine vision, including the following steps: positioning and marking the assembled prefabricated panel area; generating point cloud data through three-dimensional scanning, and performing denoising and feature extraction; calculating the elevation of each point on the surface of the prefabricated panel based on the feature data, identifying the elevation change and generating flatness evaluation data; performing grayscale detection on the three-dimensional point cloud data, and further analyzing the defective area in combination with the elevation data; using the Gaussian process regression model to predict the elevation of the unscanned area or the boundary area, and finally outputting the flatness evaluation result. Through the scheme of the present invention, efficient and accurate detection of the surface flatness of the prefabricated panel can be achieved, feedback and adjustment suggestions can be provided for construction, and the detection accuracy can be effectively improved and the construction cost can be reduced.

Description

Technical Field

[0001] The invention relates to the technical field of road construction, and in particular to a method for measuring the flatness of an assembled pavement prefabricated slab based on machine vision. Background Art

[0002] In the construction of prefabricated pavement panels, flatness is one of the key factors that determine the quality of road use. A flat road surface can not only improve the comfort of vehicle driving, but also enhance the durability and safety of the road. Therefore, in the process of laying prefabricated panels, it is very important to ensure that the surface flatness of each prefabricated panel meets the design requirements. At present, common flatness detection methods include mechanical leveling, laser leveling, and vibration leveling, but these technologies still have many shortcomings in practical applications.

[0003] First, although the existing mechanical and laser leveling technologies can improve construction efficiency to a certain extent, they have poor adaptability to complex terrain or narrow construction sites, and it is difficult to accurately ensure the flatness of the area where multiple prefabricated panels are connected. In addition, these technologies rely on high-precision mechanical equipment, which requires a large initial investment and has high requirements for equipment maintenance and environmental conditions. Especially in strong light or bad weather, the accuracy of laser leveling is easily affected, resulting in reduced data reliability. Existing detection methods also mainly measure the elevation of a single prefabricated panel, making it difficult to quickly and effectively evaluate the overall flatness of the joint area of ​​multiple prefabricated panels.

[0004] In addition, currently commonly used manual inspection methods (such as rulers or levels) are inconvenient to operate during long-distance inspections, and the accuracy of manual visual inspections is affected by the subjective judgment of the observer, making it difficult to provide high-precision flatness data. The precise inspection technology for the joints of multiple prefabricated panels is still imperfect, with problems such as low accuracy, low efficiency, and poor environmental adaptability, which has become a bottleneck to be solved in prefabricated pavement construction projects. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for measuring the flatness of prefabricated pavement panels based on machine vision, which can accurately and efficiently detect the flatness of the surface of prefabricated pavement panels.

[0006] To achieve the above purpose, the present invention is implemented by the following technical scheme: a method for measuring the flatness of prefabricated pavement panels based on machine vision, comprising the following steps:

[0007] Locate, mark and count the assembled prefabricated panels, and record the location information of each prefabricated panel;

[0008] Select a location where the prefabricated panel area can be fully observed to install the detection machine, and use a level to calibrate the horizontal state of the detection machine;

[0009] Start the detection machine, use the detection machine to perform three-dimensional scanning on the assembled prefabricated panels, and generate three-dimensional point cloud data;

[0010] De-noising the three-dimensional point cloud data to extract features of the surface of the prefabricated panel;

[0011] Calculate the elevation of each point on the precast panel surface based on the characteristic data, identify elevation changes and generate flatness assessment data;

[0012] Perform grayscale detection on 3D point cloud data, mark surface defect areas, and perform defect analysis in combination with elevation data;

[0013] Based on the elevation assessment results and defect analysis of the precast panels, the flatness assessment results are output.

[0014] Preferably, when the three-dimensional point cloud data is subjected to denoising, a non-local mean algorithm is used to perform weighted averaging on similar pixels in the point cloud data to remove noise and maintain clarity of surface features.

[0015] Preferably, the step of calculating the elevation of each point on the surface of the prefabricated board according to the characteristic data, identifying the elevation change and generating the flatness evaluation data comprises the following steps:

[0016] Obtain three-dimensional point cloud data of the prefabricated panel surface, the point cloud data containing the position coordinates of multiple points on the prefabricated panel surface ,in Represents the horizontal coordinate of a point, Indicates the elevation of a point;

[0017] Calculate the elevation value of each point on the surface of the precast panel based on the point cloud data , and calculate the surface curvature using curvature analysis methods;

[0018] By comparing the curvature changes and elevation differences, the local convex or concave areas on the surface of the precast panel are identified, and these data are used to determine the elevation changes of the surface;

[0019] Elevation data for each precast panel Processing is performed to calculate the flatness of the surface by minimizing the sum of the squares of the surface height differences and to generate flatness evaluation data.

[0020] Preferably, the formula for calculating the flatness of the surface by minimizing the sum of the squares of the surface height differences is as follows:

[0021]

[0022] in, is the expected reference height, Indicates the actual elevation of the surface point of the precast panel.

[0023] Preferably, the method further comprises predicting the elevation of the prefabricated panels in the unscanned area or the boundary area by using a Gaussian process regression model, specifically comprising:

[0024] Using scanned 3D point cloud data As training data;

[0025] Define the covariance function for the Gaussian process regression model , used to measure the correlation between points, the covariance function is:

[0026]

[0027] in, and are the positions of any two points on the surface of the precast panel, is the signal amplitude, is a scale parameter used to adjust the spatial correlation between different location points. is the square of the Euclidean distance, indicating and distance in input space;

[0028] The elevation of unscanned areas or boundary areas is predicted by Gaussian process regression model, using known point cloud data Fit and predict unmeasured points Elevation and its uncertainty, the mean and variance of the predicted elevation;

[0029] Using the predicted elevation values ​​and uncertainties, the elevation distribution of unscanned areas or boundary areas is generated, and the flatness of the entire precast panel area is evaluated in combination with the elevation data;

[0030] Based on the predicted elevation distribution, potential areas of height deviation or irregular boundaries are identified.

[0031] Preferably, the mean and variance of the predicted elevation are calculated by the following formula:

[0032] Predicted mean :

[0033]

[0034] Prediction Variance :

[0035]

[0036] in, is the location of the prediction point; is the position matrix of the known points, containing the spatial coordinates of all known points; is the elevation matrix of known points; Forecast point The covariance vector between the known points; is the transpose symbol; is the covariance matrix between known points; is the noise variance; is the identity matrix; It is the inverse matrix of the covariance matrix plus the noise variance; Forecast point Its own covariance.

[0037] Preferably, the step of performing grayscale detection on the three-dimensional point cloud data, marking the surface defect area, and performing defect analysis in combination with the elevation data comprises the following steps:

[0038] Convert the color information of each point in the acquired three-dimensional point cloud data into a grayscale value;

[0039] Based on the generated grayscale image, a threshold is set, and the grayscale value is compared with the threshold to divide the defect area and the normal area;

[0040] According to the marked defective areas, a comprehensive analysis is carried out in combination with the elevation data of the precast panel surface. By analyzing the elevation and grayscale changes, different types of surface defects can be identified.

[0041] Preferably, the defect analysis includes further analyzing the defect area in the grayscale model using a convolutional neural network to identify depressions, protrusions and cracks on the surface of the prefabricated board.

[0042] Preferably, the step of outputting the flatness evaluation result according to the elevation evaluation result and defect analysis of the prefabricated board comprises the following steps:

[0043] According to the elevation information in the 3D point cloud data and reference height of precast panels , calculate the elevation deviation at each precast panel location;

[0044] Perform a flatness analysis on the entire precast panel area, and calculate the sum of squares of elevation deviations of all points to obtain the overall flatness error of the entire precast panel;

[0045] Based on the defect analysis data, each marked defect area is independently evaluated for flatness, with evaluation indicators including protrusion height, depression depth and crack width;

[0046] Combined with the specific evaluation results of flatness error and defective areas, a complete flatness evaluation report is generated, which includes the overall flatness deviation value of each precast panel, the location and size of the raised and recessed areas, and the width and length of the cracks;

[0047] Based on the flatness assessment report, the overall quality of the precast panels is graded and corresponding construction adjustment suggestions are given, including whether local areas need to be polished, filled or the precast panels need to be replaced.

[0048] The present invention also provides a device for measuring the flatness of prefabricated pavement panels based on machine vision, comprising:

[0049] 3D scanner, used to collect 3D point cloud data of prefabricated panels;

[0050] A level gauge is installed on the top of the testing machine to ensure that the testing machine is installed horizontally;

[0051] A processor, used for performing denoising, feature extraction and elevation calculation on the three-dimensional point cloud data;

[0052] Display, used to show elevation and defect data for post-inspection verification and flatness assessment.

[0053] The present invention provides a method for measuring the flatness of prefabricated pavement panels based on machine vision. It has the following beneficial effects:

[0054] 1. The present invention uses high-precision three-dimensional scanning and convolutional neural networks combined with elevation data to conduct a detailed analysis of the surface of the prefabricated panels. It can accurately identify surface defects such as subtle depressions, protrusions, and cracks, effectively improving the accuracy of flatness detection and ensuring that the installation quality of the prefabricated panels meets the design requirements.

[0055] 2. By introducing the Gaussian process regression model, the present invention can accurately predict the elevation of unscanned areas or boundary areas, supplementing the parts that are difficult to cover in traditional measurements, thereby ensuring the integrity assessment of the overall flatness and avoiding possible height deviations in boundary areas.

[0056] 3. The present invention adopts convolutional neural network (CNN) to analyze the grayscale image of the surface of the prefabricated board, automatically identifies and classifies surface defects such as dents, protrusions and cracks, reduces the subjective error of manual judgment, and improves the accuracy and efficiency of defect identification.

[0057] 4. By generating a detailed flatness assessment report, the present invention can provide construction personnel with accurate repair or adjustment suggestions and promptly handle problems during the installation of prefabricated panels, thereby reducing post-construction adjustment and maintenance costs and improving overall construction efficiency.

[0058] 5. The detection method of the present invention has strong adaptability. It is not only suitable for conventional flat pavement precast slab detection, but can also handle the flatness evaluation of complex surface materials, joint areas and boundary areas, ensuring the detection accuracy under different construction environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 One of the schematic flow charts of the method of the present invention;

[0060] Figure 2 This is the second schematic flow chart of the method of the present invention. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] Please see attached Figure 1 -Attached Figure 2 The embodiment of the present invention provides a method for measuring the flatness of a prefabricated pavement slab based on machine vision, comprising the following steps:

[0063] S1. Locate, mark and count the assembled prefabricated panels, and record the location information of each prefabricated panel.

[0064] First, it is necessary to locate, mark and count the assembled prefabricated panels. This step is the basis for subsequent flatness detection. By accurately recording the position and geometric information of each prefabricated panel, it is ensured that the spatial position of each prefabricated panel can be correctly associated during the 3D scanning process. The accuracy of positioning and marking will directly affect the correctness of the elevation data, and then affect the overall flatness assessment. Therefore, in this step, in addition to recording the spatial coordinates of each prefabricated panel, it is also necessary to mark the number of each prefabricated panel to provide a unified reference standard for subsequent detection and analysis.

[0065] In this embodiment, the assembled prefabricated panels are first accurately positioned. The spatial coordinates of each prefabricated panel are obtained by using high-precision measurement tools such as laser rangefinders or total stations. ,in and represents the horizontal coordinate of the precast panel, is the initial height value of the precast panel. These data are the important basis for subsequent 3D scanning and elevation calculation.

[0066] In this embodiment, after obtaining the geometric coordinates of the prefabricated panels, each prefabricated panel is numbered. The purpose of the numbering is to ensure that the spatial position information of each prefabricated panel can be correctly matched during the subsequent three-dimensional point cloud data processing, and to ensure that the specific position of each prefabricated panel can be accurately tracked in the subsequent flatness analysis. The numbering system can use digital or letter coding to number the prefabricated panels in the order of installation.

[0067] When identifying the prefabricated panels, in this embodiment, a prefabricated panel position information database is established, which records the geometric coordinates, numbering information of each prefabricated panel and the connection relationship between adjacent prefabricated panels. The position information in the database will be used for subsequent image processing and calibration of three-dimensional point cloud data.

[0068] In the positioning process, especially for complex or large-scale prefabricated panel areas, this embodiment combines multi-point measurement with data fusion technology to ensure that the position information of all prefabricated panels has high accuracy. In actual operation, the four corners of each prefabricated panel are measured multiple times using a total station, and the measurement results are fused and corrected to ensure that the final data error is within the allowable range (usually controlled at the millimeter level).

[0069] In addition, the installation angle of the prefabricated panels is also recorded in this embodiment to ensure that the angle information of each prefabricated panel during installation is consistent with the design requirements. Through these positioning and marking operations, an accurate geometric basis is provided for subsequent three-dimensional scanning and elevation calculation.

[0070] S2. Select a location where the prefabricated panel area can be fully observed to install the detection machine, and use a level to calibrate the horizontal state of the detection machine

[0071] After the prefabricated panel area is located, marked and counted, the next step is to select a suitable location to install the inspection machine to ensure that the prefabricated panel area can be fully covered during the subsequent 3D scanning process. The key to this step is to ensure that the inspection machine can observe all the prefabricated panel surfaces to be tested from an optimal angle, and to ensure that the equipment is level during installation to avoid data errors caused by tilt.

[0072] In this embodiment, first, according to the layout of the prefabricated panel area, a suitable location that can fully cover the surface of the prefabricated panel is selected to install the detection machine. The installation location of the detection machine should ensure that its field of view can cover the entire prefabricated panel area, and avoid any obstruction or environmental interference (such as direct strong light or uneven ground) to ensure the smooth progress of the 3D scanning process. In order to ensure that the scanning equipment can best observe all surfaces of the prefabricated panel, the height and angle of the selected location need to be adjusted according to the specific conditions of the construction site.

[0073] When installing the inspection machine, a level is used in this embodiment to calibrate the equipment horizontally. The level is used to ensure that the inspection machine remains horizontal during installation, because the tilt of the equipment will directly cause systematic deviations in the scan data, thereby affecting the subsequent flatness assessment. The specific process of horizontal calibration is to place a level at multiple locations on the equipment to ensure that the horizontal deviation in each direction is minimized.

[0074] S3. Start the detection machine and use it to perform 3D scanning on the prefabricated panels to generate 3D point cloud data.

[0075] After the inspection machine is installed and calibrated, start the inspection machine to perform a three-dimensional scan of the prefabricated panel and generate three-dimensional point cloud data of the prefabricated panel. The three-dimensional point cloud data can accurately record the spatial coordinate information of the surface of the prefabricated panel and is the key data basis for flatness measurement. In this step, the machine vision system will use scanning technology to carefully sample the surface of the prefabricated panel and obtain the height information of each sampling point. With these height data, the subsequent flatness evaluation and defect identification can be accurately analyzed.

[0076] In this embodiment, the surface of the prefabricated board is first scanned in three dimensions by a laser scanner or a structured light scanner. During the scanning process, the detection machine will gradually cover the entire prefabricated board area according to the pre-set scanning path. The scanning system will collect the height of each point on the surface of the prefabricated board and record its spatial coordinates. ,in and Represents the horizontal coordinate of each point on the precast panel, In this way, the inspection machine is able to generate a dense 3D point cloud dataset that describes the geometry of the precast panel surface in detail.

[0077] In this embodiment, the scanning is divided into multiple scanning levels to ensure that the entire prefabricated panel surface is collected in all directions and at multiple angles. The point data collected by each scan is integrated into a complete three-dimensional point cloud through automatic stitching technology. For each scan point, its three-dimensional coordinates will be stored in the built-in memory of the device, and the data can be transmitted in real time or analyzed later as needed.

[0078] S4. De-noising the three-dimensional point cloud data to extract the features of the prefabricated board surface

[0079] After generating the 3D point cloud data, the data needs to be denoised and feature extracted to ensure that the subsequent elevation calculation and defect identification can be analyzed based on clear and accurate data. Since the scanning process may be affected by environmental noise (such as light reflection, equipment jitter, etc.) or surface impurities, the original point cloud data often contains some errors or noise data, so it needs to be optimized through denoising. After removing invalid data, feature extraction is also required to analyze the height changes, protrusions, depressions and other features of the prefabricated board surface to provide necessary information for subsequent flatness evaluation.

[0080] In this embodiment, the three-dimensional point cloud data is first denoised. The denoising process uses the Non-local Means Algorithm (NLM), which reduces noise by weighted averaging similar points in the point cloud data while retaining important surface details. The formula of the NLM algorithm is as follows:

[0081]

[0082] in, Represents a point on the surface of a precast panel The height value at Indicate point and The similarity weight between Indicates traversing point cloud data All points , Represents the similarity weight of all points Normalize.

[0083] Through this algorithm, the noise data in the point cloud can be effectively removed without significantly affecting the detailed features of the precast panel surface.

[0084] The denoising process in this embodiment takes into account the material differences and surface characteristics of different regions. There may be some local material changes on the surface of the precast panel (such as steel bars or other embedded objects), so the denoising process not only considers the height value, but also combines data such as reflectivity to ensure that the denoising process can be optimized for the surfaces of different materials. At the same time, multi-scale denoising technology is used to adjust the denoising intensity according to the surface complexity of different regions. In relatively flat areas, the denoising intensity is greater, while in areas with obvious bumps and depressions, the denoising process will retain more details.

[0085] After the denoising process is completed, this embodiment performs feature extraction on the point cloud data. The purpose of feature extraction is to identify significant features on the surface of the prefabricated board, such as height changes, edges, and contours. By analyzing the neighborhood information of each point in the three-dimensional point cloud, the system can extract the key feature points of the prefabricated board surface. The specific feature extraction process includes the following steps:

[0086] 1. Height gradient calculation: By calculating the height gradient of each point, the areas with significant height changes on the surface are identified. The height gradient formula is:

[0087]

[0088] in, For point The height value of Indicates the rate of change of height at that point. Areas with large height gradients are usually convex or concave areas on the surface of the precast panel.

[0089] 2. Curvature calculation: After identifying significant changes in the surface, the local curvature of each point is further calculated to determine the smoothness and protrusion of the precast panel surface. Gaussian curvature and mean curvature are used to describe the geometry of the surface:

[0090] Gaussian curvature Calculated by the following formula:

[0091]

[0092] in, and are the principal curvatures on the surface of the precast panel, and is the corresponding radius of curvature.

[0093] Mean curvature Then:

[0094]

[0095] Gaussian curvature is used to identify bumps and depressions on a surface, while mean curvature measures the smoothness of a surface.

[0096] 3. Edge detection: In this embodiment, edge detection is also performed on the point cloud data to extract the contour features of the prefabricated panel surface. Edge detection identifies the obvious change areas on the surface by analyzing the height difference between adjacent points. By calculating the height difference between each point and the surrounding points, obvious edges or raised areas can be identified.

[0097] Through the above-mentioned denoising process and feature extraction, this embodiment can eventually generate a complete data set including the height variation, curvature information and edge features of the prefabricated board surface.

[0098] S5. Calculate the elevation of each point on the surface of the precast panel based on the characteristic data, identify elevation changes and generate flatness assessment data

[0099] After denoising and feature extraction, the elevation of each point on the prefabricated panel surface needs to be calculated based on the 3D point cloud data, and the flatness of the prefabricated panel is evaluated by analyzing the elevation changes. The core of this step is to identify the possible convexities or concavities on the surface of the prefabricated panel through accurate analysis of the point cloud data, and to evaluate the flatness based on these elevation data. This analysis not only relies on height data, but also combines curvature information to ensure that the surface geometry can be fully and accurately reflected.

[0100] In this embodiment, the position coordinates of each point on the surface of the prefabricated panel are first obtained through the three-dimensional point cloud data generated in the previous stage. For each point of the precast panel, the point cloud data describes in detail the spatial position and height information of the point.

[0101] Next, based on these point cloud data, this embodiment calculates the elevation value of each point on the surface of the prefabricated board. It is The value is obtained, and the local elevation change on the surface is identified by analyzing the height difference between each point and the adjacent points. In order to more accurately analyze the surface characteristics, this embodiment combines the curvature analysis method to further describe the surface shape. The curvature information helps to determine the local convexities and concavities of the surface.

[0102] The Gaussian curvature used in this embodiment and mean curvature It is an important indicator to characterize the surface shape. The average curvature is used to evaluate the smoothness of the precast board surface. The average curvature can be used to determine whether the surface is smooth or has obvious deformation.

[0103] Through comprehensive analysis of elevation data and curvature changes, this embodiment can identify highly abnormal areas on the surface of the prefabricated board. These abnormal areas usually appear as local convexities or concavities, and these anomalies are used to determine the elevation changes on the surface of the prefabricated board. The identification of elevation changes is crucial for subsequent flatness assessment and can reveal whether the surface is flat or uneven.

[0104] After identifying the elevation change, this embodiment further processes the elevation data of each precast panel. To evaluate the flatness of the entire surface, this embodiment quantifies the flatness error by minimizing the sum of squares of the surface height difference. The calculation formula is:

[0105]

[0106] in, Indicates the first The height value of the location, By means of this formula, the present embodiment can accurately calculate the overall flatness deviation of each prefabricated board, and the smaller the value, the better the surface flatness.

[0107] Finally, this embodiment generates flatness evaluation data, including the overall flatness deviation of each prefabricated board and the specific location and severity of local protrusions or depressions. These evaluation data provide a detailed reference for subsequent construction adjustments and defect repairs, ensuring that the surface of the prefabricated board meets the design requirements and providing data support for possible repairs.

[0108] In some embodiments, in order to more accurately evaluate the overall flatness of the prefabricated panel surface, especially for the elevation of the prefabricated panels in unscanned areas or boundary areas, a Gaussian Process Regression (GPR) model is used for prediction. The model can infer the elevation values ​​of unmeasured areas or joints based on the three-dimensional point cloud data of the scanned area. Boundary areas and joints are key points in flatness assessment because these areas often have height differences due to splicing or construction errors. Through the Gaussian process regression model, effective elevation predictions can be made for these areas to ensure that the flatness of the entire prefabricated panel area is fully evaluated.

[0109] In this embodiment, the scanned 3D point cloud data is first used as training data. These point cloud data contain the known spatial coordinates of the prefabricated panel surface. By analyzing the elevation data of the scanned area, a data basis can be provided for the unmeasured area. The location of each known point and elevation values All of them are included in the training set for subsequent regression fitting and prediction.

[0110] In this embodiment, the covariance function of the Gaussian process regression model is defined, which is used to measure the spatial correlation between different points. The form of the covariance function is:

[0111]

[0112] in:

[0113] and are the positions of any two points on the surface of the precast panel,

[0114] is the signal amplitude, which represents the size of the natural variation in the data,

[0115] is the square of the Euclidean distance, indicating and The distance in the input space,

[0116] is a scale parameter used to adjust the spatial correlation between different location points. Controls the model's sensitivity to spatial distance. Larger values Indicates that the correlation between adjacent points is strong and is suitable for smoother surfaces, while smaller It is suitable for surfaces with more local variations.

[0117] After the Gaussian process regression model is established, the model is fitted with known point cloud data and used to predict the elevation of unscanned areas or boundary areas. Through the covariance function of the model, the system can infer the elevation of each unmeasured point. Elevation and its uncertainty. The mean of the predicted elevation and variance Calculated by the following formulas:

[0118] Predicted mean :

[0119]

[0120] Prediction Variance :

[0121]

[0122] is the location of the prediction point; is the position matrix of the known points, containing the spatial coordinates of all known points; is the elevation matrix of known points; Forecast point The covariance vector between the known points; is the transpose symbol; is the covariance matrix between known points; is the noise variance; is the identity matrix; It is the inverse matrix of the covariance matrix plus the noise variance; Forecast point Its own covariance.

[0123] In this embodiment, the mean value is predicted Get the elevation prediction value of each unmeasured point and combine it with the prediction variance , and evaluate the elevation distribution of unscanned areas or boundary areas. In particular, the joints of prefabricated panels usually have certain height deviations due to splicing or material differences during construction. Through the elevation prediction of the Gaussian process regression model, this embodiment can identify potential height anomalies in these areas, thereby helping construction personnel to focus on the joints.

[0124] After generating the elevation prediction value, this embodiment combines the elevation data to comprehensively evaluate the flatness of the entire prefabricated panel area. By comparing the elevation prediction value of the unscanned area or the boundary area with the actual measurement value, the system can detect areas with large height deviations and provide guidance for subsequent flatness adjustments.

[0125] Based on the predicted elevation distribution, especially for the joint area of ​​the precast panel, potential height deviations or uneven areas at the boundary can be identified. In this way, local concave-convex problems at the boundary or joint can be discovered in time to ensure that the entire precast panel surface meets the design requirements, thus providing an effective reference for subsequent construction or repair.

[0126] S6. Perform grayscale detection on 3D point cloud data, mark surface defect areas, and perform defect analysis in combination with elevation data

[0127] After the elevation data and feature extraction are completed, the next step is to perform grayscale detection on the 3D point cloud data to mark the defective areas on the surface of the prefabricated board. The purpose of this step is to analyze the local unevenness, cracks or depressions on the surface of the prefabricated board through image processing technology, and further analyze the specific characteristics of these defects in combination with the aforementioned elevation data. Grayscale detection can convert complex 3D point cloud information into a more intuitive image form, thereby facilitating subsequent defect identification and analysis.

[0128] In this embodiment, the generated three-dimensional point cloud data is first converted into grayscale with respect to color information. Contains position coordinates and color information, and the color information usually includes red, green, and blue (RGB) channels. In order to simplify the data and highlight the surface features, this embodiment converts the color information of each point into a grayscale value. , the conversion formula is:

[0129]

[0130] in, , , Respectively represent the red, green, and blue channel values ​​of the point. Through this formula, each point in the point cloud data is converted into a grayscale value, which reflects the brightness information of the point, thereby better highlighting the surface details.

[0131] Next, this embodiment sets a grayscale threshold , which is used to distinguish normal areas from defective areas. With threshold By comparison, the defective area on the surface can be marked. The grayscale detection formula is as follows:

[0132]

[0133] in, Indicates that the point belongs to the defect area. Indicates that the point is a normal area. The threshold setting can be dynamically adjusted according to the site conditions and the material characteristics of the prefabricated panels to ensure adaptability to different surface materials.

[0134] After marking the defective area, this embodiment combines the elevation data Further defect analysis is performed on the defective area. By analyzing the elevation data, the specific features of the defective area are identified, including depressions, protrusions and cracks. The specific analysis steps are as follows:

[0135] 1. Depth calculation: For the points marked as defect areas, the system calculates the height difference between them and the surrounding points. For the concave area, the depth Calculated by the following formula:

[0136]

[0137] in, is the reference height of the normal area, is the lowest point of the concave area. Through this calculation, the depth of the concave area can be identified.

[0138] 2. Calculation of raised height: For the raised area, its height Calculated by the following formula:

[0139]

[0140] in, is the highest point of the raised area, By calculating the height of the protrusions, the local protruding area on the surface can be determined.

[0141] 3. Crack width and length: This embodiment also identifies cracks on the surface by analyzing the geometric shape of the defect area. The width and length of the crack can be determined by calculating the geometric boundary of the defect area. The length of the crack is calculated by the horizontal distance between the two end points. is the circumference of the crack boundary.

[0142] Through these calculations, the present embodiment can accurately describe the specific shape and size of the defect area and generate a detailed analysis report based on the type and severity of the defect.

[0143] Finally, this embodiment combines the elevation information and grayscale detection results in the 3D point cloud data to generate a prefabricated board surface defect analysis report, which includes the location, type, depth or height of each defect area, and the length and width of the crack. These analysis results provide a strong basis for subsequent repairs and provide construction personnel with a comprehensive assessment of the surface quality of the prefabricated board.

[0144] In order to further accurately analyze these defects, the present invention introduces a convolutional neural network (CNN) for a deeper analysis of the defective areas in the grayscale model. Through the image processing capabilities of CNN, different types of defects on the surface of prefabricated panels, including dents, bulges, and cracks, can be more accurately identified and classified.

[0145] In this embodiment, the grayscale area marked as a defect is first Extracted to form independent grayscale image blocks. Each image block corresponds to a local defect area on the surface of a prefabricated board. These image blocks serve as input data for the convolutional neural network to analyze.

[0146] Structural design of convolutional neural network

[0147] The convolutional neural network structure of this embodiment includes multiple convolutional layers, pooling layers and fully connected layers, which are used to extract high-level features from the input grayscale image and then classify the defect type. The specific CNN structure and processing steps are as follows:

[0148] 1. Convolutional layer: The function of the convolutional layer is to extract features from the input grayscale image. Each convolutional layer uses several convolutional kernels (filters) to perform convolution operations on the input image to generate feature maps. Through the convolutional layer, the model can extract important features such as the edge and shape of the defect area.

[0149] 2. Pooling layer: The pooling layer is used to downsample the feature map extracted by the convolution layer, reduce the data dimension, and retain key features. The pooling layer uses the maximum pooling method. The pooling operation can retain the significant features in the feature map and enhance the translation invariance of the model, thereby improving the robustness of defect recognition.

[0150] 3. Fully connected layer: After multiple convolution and pooling operations, the output of the convolutional neural network is input to the fully connected layer for further classification processing. The fully connected layer compresses all extracted feature maps into a one-dimensional vector and outputs the probability of each defect through the Softmax function. In this embodiment, the defect types on the surface of the prefabricated board are divided into three categories: dents, bulges and cracks.

[0151] Through the processing of the convolutional layer, pooling layer and fully connected layer of the CNN model, this embodiment can identify the defect type of each defect area. The model outputs the probability that each defect area belongs to a depression, a bulge or a crack. Based on the size of the probability, the model can determine the type of each defect and further calculate its location, size and shape characteristics.

[0152] Depression identification: For areas identified as depressions, the depth and range of the depressions are further calculated in combination with elevation data. The specific method is to analyze the difference between the lowest point of the depression area and the reference height of the normal surface. .

[0153] Bump recognition: For bump areas, the height of the bump is calculated by combining the CNN classification results and elevation data. and its spatial distribution.

[0154] Crack identification: For areas identified as cracks, the CNN model is used to extract the boundary information of the cracks and further analyze the length of the cracks. and width , and calculate the depth of the crack in combination with the elevation data.

[0155] Finally, this embodiment generates a detailed defect analysis report based on the classification results of the convolutional neural network and the elevation data. The report includes:

[0156] The type of each defect area (depression, protrusion or crack),

[0157] The location and shape of the defect (including area, length, width, etc.),

[0158] The depth of the depressions and the height of the protrusions,

[0159] The length, width and depth of the crack.

[0160] These analysis results provide a reliable reference for subsequent construction and repair, ensuring that construction workers can promptly address quality issues on the surface of precast panels.

[0161] S7. Output the flatness assessment result based on the elevation assessment result and defect analysis of the precast slab.

[0162] After completing the calculation of the surface elevation of the precast slab and the identification and analysis of defects, the final flatness assessment result needs to be output.

[0163] In this embodiment, firstly, the flatness of the prefabricated board is evaluated using the overall flatness error calculation formula based on the elevation data generated in the previous stage. , these deviation values ​​are used to calculate the flatness error of the entire precast panel surface The specific formula is as follows:

[0164]

[0165] By minimizing the error , the overall flatness of the precast panel surface can be evaluated. The smaller the error value, the flatter the surface.

[0166] This embodiment also combines defect analysis data of local areas, which are generated by convolutional neural networks and elevation analysis, including depression depth, protrusion height, crack width and length, etc. For each identified defect area, the system further calculates its impact on the overall flatness according to the severity of the depression or protrusion, and integrates these local data into the final flatness evaluation.

[0167] Depression area assessment: For the identified depression area, this embodiment calculates its depth using the above formula. Then, the system classifies it according to its depth and area to determine whether it needs to be repaired.

[0168] Raised Area Assessment: For raised areas, the system calculates the height using the formula above. Raised areas are also graded and assessed to determine if grinding or adjustment is required.

[0169] Crack area assessment: For crack areas, the system provides a detailed analysis report based on the length and width of the crack. If the width of the crack exceeds the set threshold, the system will recommend repair or replacement of the precast panel.

[0170] By evaluating all defective areas, this embodiment combines the effects of elevation deviation and local defects to generate a detailed flatness evaluation report. The report includes:

[0171] Overall flatness error;

[0172] Detailed data of each defect area (deepness of depression, height of protrusion, length and width of crack, etc.);

[0173] Severity classification of each defect area (based on depth, width and area);

[0174] Recommendations for adjustments to the construction, including the need to repair or replace precast panels.

[0175] This embodiment also provides clear feedback for construction based on the evaluation results. For example, if the flatness error of a certain area exceeds the allowable standard range, the system will automatically generate suggestions to prompt the construction personnel to polish or reinstall the area; if there are large cracks in certain areas, the system will recommend filling materials for repair.

[0176] The final evaluation results not only provide key feedback data for construction, but can also be used to judge the compliance of prefabricated panel installation during quality acceptance. Through the evaluation report, construction personnel can make on-site adjustments based on the suggestions provided by the system to ensure that the surface flatness of the entire prefabricated panel meets the design requirements.

[0177] The flatness assessment method of this embodiment combines elevation data, defect analysis and machine learning models to comprehensively and accurately assess the surface flatness of prefabricated panels, provide precise feedback information, and help improve construction quality and reduce subsequent maintenance costs.

[0178] In general, the present invention conducts a comprehensive and accurate flatness assessment of the surface of the prefabricated board through three-dimensional scanning, denoising, elevation calculation, defect analysis, and convolutional neural network. The method can accurately identify defects such as depressions, protrusions, and cracks on the surface of the prefabricated board, and combines the Gaussian process regression model to predict the elevation of the unscanned area to generate a complete flatness assessment report. Through the scheme of the present invention, the detection accuracy and efficiency can be significantly improved, and accurate feedback and adjustment suggestions can be provided for construction, thereby ensuring the installation quality of the prefabricated board and reducing subsequent maintenance costs.

[0179] The device for measuring the flatness of prefabricated pavement panels based on machine vision described below and the method for measuring the flatness of prefabricated pavement panels based on machine vision described above can be referred to each other.

[0180] The present invention also provides a device for measuring the flatness of prefabricated pavement panels based on machine vision, comprising:

[0181] 3D scanner, used to collect 3D point cloud data of prefabricated panels;

[0182] A level gauge is installed on the top of the testing machine to ensure that the testing machine is installed horizontally;

[0183] A processor, used for performing denoising, feature extraction and elevation calculation on the three-dimensional point cloud data;

[0184] Display, used to show elevation and defect data for post-inspection verification and flatness assessment.

[0185] The device of this embodiment can be used to execute the above method embodiment, and its principles and technical effects are similar, which will not be repeated here.

[0186] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for measuring the flatness of prefabricated pavement panels based on machine vision, characterized in that: The following steps are involved: Locate, mark and count the assembled prefabricated panels; Select a location where the inspection machine can be installed to fully observe the prefabricated panel area; Use a detection machine to perform 3D scanning on the prefabricated panels to generate 3D point cloud data; De-noising the three-dimensional point cloud data to extract features of the surface of the prefabricated panel; Calculate the elevation of each point on the precast panel surface based on the characteristic data, identify elevation changes and generate flatness assessment data; Perform grayscale detection on 3D point cloud data, mark surface defect areas, and perform defect analysis in combination with elevation data; Output the flatness assessment results based on the elevation assessment results and defect analysis of the precast panels; The step of calculating the elevation of each point on the surface of the prefabricated board according to the characteristic data, identifying the elevation change and generating the flatness evaluation data comprises the following steps: Obtain three-dimensional point cloud data of the prefabricated panel surface, the point cloud data containing the position coordinates of multiple points on the prefabricated panel surface ,in Represents the horizontal coordinate of a point, Indicates the elevation of a point; Calculate the elevation value of each point on the surface of the precast panel based on the point cloud data , and calculate the surface curvature using curvature analysis methods; By comparing the curvature changes and elevation differences, the local convex or concave areas on the surface of the precast panel are identified, and these data are used to determine the elevation changes of the surface; Elevation data for each precast panel Processing is performed to calculate the flatness of the surface by minimizing the sum of the squares of the surface height differences and to generate flatness evaluation data.

2. The method for measuring the flatness of prefabricated pavement slabs based on machine vision according to claim 1 is characterized in that: When the three-dimensional point cloud data is subjected to denoising, a non-local mean algorithm is used to perform weighted averaging on similar pixels in the point cloud data to remove noise and maintain clarity of surface features.

3. The method for measuring the flatness of prefabricated pavement slabs based on machine vision according to claim 1 is characterized in that: The formula for calculating the flatness of a surface by minimizing the sum of the squares of the surface height differences is as follows: in, is the expected reference height, Indicates the actual elevation of the surface point of the precast panel.

4. The method for measuring the flatness of prefabricated pavement slabs based on machine vision according to claim 1 is characterized in that: The method further includes predicting the elevation of the prefabricated panels in the unscanned area or the boundary area by using a Gaussian process regression model, specifically including: Using scanned 3D point cloud data As training data; Define the covariance function for the Gaussian process regression model , used to measure the correlation between points, the covariance function is: in, and are the positions of any two points on the surface of the precast panel, is the signal amplitude, is a scale parameter used to adjust the spatial correlation between different location points. is the square of the Euclidean distance, indicating and distance in input space; The elevation of unscanned areas or boundary areas is predicted by Gaussian process regression model, using known point cloud data Fit and predict unmeasured points Elevation and its uncertainty, the mean and variance of the predicted elevation; Using the predicted elevation values ​​and uncertainties, the elevation distribution of unscanned areas or boundary areas is generated, and the flatness of the entire precast panel area is evaluated in combination with the elevation data; Based on the predicted elevation distribution, potential areas of height deviation or irregular boundaries are identified.

5. The method for measuring the flatness of prefabricated pavement slabs based on machine vision according to claim 4 is characterized in that: The mean and variance of the predicted elevation are calculated using the following formula: Predicted mean : Prediction Variance : in, is the location of the prediction point; is the position matrix of the known points, containing the spatial coordinates of all known points; is the elevation matrix of known points; Forecast point The covariance vector between the known points; is the transpose symbol; is the covariance matrix between known points; is the noise variance; is the identity matrix; It is the inverse matrix of the covariance matrix plus the noise variance; Forecast point Its own covariance.

6. The method for measuring the flatness of prefabricated pavement slabs based on machine vision according to claim 1 is characterized in that: The steps of performing grayscale detection on the three-dimensional point cloud data, marking the surface defect area, and performing defect analysis in combination with the elevation data include the following steps: Convert the color information of each point in the acquired three-dimensional point cloud data into a grayscale value; Based on the generated grayscale image, a threshold is set, and the grayscale value is compared with the threshold to divide the defect area and the normal area; According to the marked defective areas, a comprehensive analysis is carried out in combination with the elevation data of the precast panel surface. By analyzing the elevation and grayscale changes, different types of surface defects can be identified.

7. The method for measuring the flatness of prefabricated pavement slabs based on machine vision according to claim 6 is characterized in that: The defect analysis includes further analyzing the defect area in the grayscale model using a convolutional neural network to identify depressions, protrusions and cracks on the surface of the prefabricated board.

8. The method for measuring the flatness of prefabricated pavement slabs based on machine vision according to claim 1 is characterized in that: The step of outputting the flatness evaluation result according to the elevation evaluation result and defect analysis of the prefabricated slab comprises the following steps: Calculate the elevation deviation at each precast panel location based on the elevation information in the 3D point cloud data and the reference height of the precast panel; Perform a flatness analysis on the entire precast panel area, and calculate the sum of squares of elevation deviations of all points to obtain the overall flatness error of the entire precast panel; Based on the defect analysis data, each marked defect area is independently evaluated for flatness, with evaluation indicators including protrusion height, depression depth and crack width; Combined with the specific evaluation results of flatness error and defective areas, a complete flatness evaluation report is generated, which includes the overall flatness deviation value of each precast panel, the location and size of the raised and recessed areas, and the width and length of the cracks; Based on the flatness assessment report, the overall quality of the precast panels is graded and corresponding construction adjustment suggestions are given, including whether local areas need to be polished, filled or the precast panels need to be replaced.

Images