Masonry wall quality acceptance method and system based on intelligent identification

Abstract

The invention relates to the technical field of intelligent identification, in particular to a masonry wall quality acceptance method and system based on intelligent identification. Comprising the steps of obtaining architectural drawings and floor wall component information; forming a wall brick arrangement drawing according to the obtained building drawing and the floor wall component information; building a BIM model based on the wall brick arrangement drawing; a robot is used for cruising to shoot a wall image, and a construction result is uploaded; performing intelligent comparison on uploaded construction result image data and the BIM model to obtain an inspection result; and forming a quality problem list from the inspection result and reporting. Through linkage of AI monitoring and intelligent hardware, quality inspection omission caused by human negligence can be effectively reduced, and the quality production management level is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent recognition, and particularly to a method and system for quality acceptance of masonry walls based on intelligent recognition. Background Art

[0002] At present, with the booming development of the construction industry, the scale of construction projects continues to expand, and the requirements for project quality are becoming increasingly stringent. As a key component of the building structure, the quality of masonry walls directly affects the overall safety and stability of buildings. The traditional method for quality acceptance of masonry walls mainly relies on manual operation. The acceptance personnel, relying on personal experience, use means such as visual inspection with the naked eye and simple tool measurement to check aspects such as mortar joints, fullness of top plugs, block layout, position of lintel beams, tie bars, grooving and pipe laying, reserved holes, and appearance quality of masonry walls.

[0003] However, this manual acceptance mode has many drawbacks. On the one hand, it is difficult to achieve a comprehensive and non - omissive inspection of large - area masonry walls through manual inspection. It is very easy to miss key quality problems due to negligence, resulting in the retention of quality hazards. On the other hand, the acceptance results are greatly affected by subjective factors such as the personal experience, responsibility, and working state of the acceptance personnel. There may be differences in the judgment criteria for the same quality problem among different acceptance personnel, resulting in the lack of consistency and accuracy of the acceptance results. Moreover, with the strengthening of the trend of refined and standardized management in the construction industry, the traditional manual acceptance is inefficient and cannot meet the quality control requirements during the rapid progress of large - scale projects, nor can it adapt to the requirements of information - based management of modern construction projects.

[0004] At the same time, with the rapid progress of technology, intelligent recognition technology has been widely applied in many fields and achieved remarkable results. Computer vision technology enables a computer to understand and interpret image information, and deep - learning algorithms endow the computer with powerful self - learning and pattern - recognition capabilities. Introducing these advanced intelligent recognition technologies into the quality acceptance work of masonry walls is expected to break through the bottleneck of traditional acceptance methods, achieve efficient, accurate, and comprehensive detection of the quality of masonry walls, improve the quality management level of construction projects, and promote the construction industry to move towards the direction of intelligence and digitization. Summary of the Invention

[0005] To solve the above-mentioned problems, the present invention provides an intelligent auction method and system. The present invention combines the intelligent recognition technology of masonry wall acceptance photos. By taking photos of the walls completed on-site and uploading them to the system, and using computer vision technology and deep learning algorithms, the system automatically identifies masonry quality problems (including checking mortar joints, fullness of top plugs, block layout quality, lintel position, identification of tie bars, grooving and pipe laying, position of reserved holes, appearance quality, etc.). After automatically identifying the problems, a quality problem list is immediately formed and reported to the management personnel for timely handling. Through this invention, full inspection of the masonry walls is achieved, and the masonry quality is improved by implementing monitoring.

[0006] In a first aspect, a method for intelligent recognition and masonry wall quality acceptance provided by the present invention adopts the following technical solutions: A method for intelligent recognition and masonry wall quality acceptance includes: Obtain architectural drawings and floor wall component information; Form a wall bricklaying plan based on the obtained architectural drawings and floor wall component information; Construct a BIM model based on the wall bricklaying plan; Use a robot to cruise and take wall images, and upload the construction results; Intelligently compare the uploaded construction result image data with the BIM model to obtain an inspection result; Report the inspection result by forming a quality problem list.

[0007] Further, the forming of the wall bricklaying plan according to the obtained architectural drawings and floor wall component information includes obtaining the block type and size according to the architectural drawings, starting from the bottom of the wall and performing bricklaying design upward through the greedy algorithm, calculating the number of blocks required for each row according to the wall length, block length, and mortar joint thickness, and at the same time calculating the number of block layers required according to the wall thickness. Finally, use graphic drawing software to draw the wall bricklaying plan according to the calculation results, showing the position of each block, mortar joint distribution, and treatment of special parts.

[0008] Further, the construction of the BIM model based on the wall bricklaying plan includes importing the drawn wall bricklaying plan into Revit in a format supported by BIM software such as DWG through Revit software. In Revit, according to the information in the bricklaying plan, use the relevant family libraries of walls and building blocks to create a wall model. For each building block, assign accurate size and material property information; utilize the parametric design function of Revit to establish parametric associations between the building blocks and the wall to ensure that when the wall size or bricklaying plan changes, the model can be automatically updated; at the same time, add lintel and tie bar components to the model. The lintel is placed according to the size and position required by the design, and the tie bars are arranged according to the spacing and length required by the specifications through the steel bar drawing tool of Revit.

[0009] Further, the use of the robot for cruising and photographing wall images includes constructing a two-dimensional network map based on the stair position and layout in the BIM model, dividing the entire floor space into grid cells of equal size, with each grid cell representing a node. Use the Manhattan method to estimate the coordinates of the target node, and the path from the starting point to the target node can be obtained by backtracking the parent node through the A* algorithm path search method. At the same time, use the Dijkstra algorithm for local search optimization to obtain the cruising path, and finally obtain the smoothed cruising path through Bezier curve fitting.

[0010] Further, the use of the robot for cruising and photographing wall images also includes calculating the optimal distance between the robot and the wall according to the height, width of the wall and the set shooting parameters, calculating the optimal shooting angle according to the shooting horizontal view angle. For wide walls, the method of taking multiple-angle photos and then stitching is used to obtain wall images. Among them, the floor height is obtained by using a laser range finder sensor with the ground distance as a reference, and the floor label is obtained by using an RFID sensor. The floor result is obtained by using a weighted method based on the floor height and the floor label.

[0011] Further, the method of taking multiple-angle photos and then stitching to obtain wall images includes performing wall image stitching based on the SIFT algorithm. Convolve the obtained image with the original image using a Gaussian convolution kernel to generate Gaussian pyramids of different scales. Subtract adjacent two-layer Gaussian images to obtain Gaussian difference images. Find local extreme points by comparing each pixel point with its adjacent 26 pixel points as feature points. In the neighborhood of the feature points, calculate the gradient amplitude and direction of each pixel point, and in each sub-region, after statistically analyzing the direction range of the gradient direction histogram, combine the 8-dimensional histogram vectors of the small regions to obtain a 128-dimensional feature descriptor vector.

[0012] Further, the intelligent comparison of the uploaded data with the BIM model to obtain the inspection result includes identifying the wall completed on-site based on the computer vision algorithm, and performing intelligent comparison on the recognition result and the BIM model through the deep learning algorithm. Among them, the Canny edge detection algorithm is used to extract the edges of the uploaded wall photo, the image is smoothed by Gaussian filtering, the gradient amplitude and direction of the image are calculated, and then non-maximum suppression and double-threshold detection are performed to finally obtain a clear wall edge contour; based on the template matching algorithm, the extracted wall edge contour is matched with the theoretical contour of the wall in the BIM model. Among them, the similarity between the two is calculated by using the normalized cross-correlation coefficient NCC, which is expressed as: NCC(A,B)=Σ[(A(i)-μA)(B(i)-μB)] / [sqrt(Σ(A(i)-μA)^2)×sqrt (Σ(B(i)-μB)^2)], where A and B respectively represent the image patch to be matched and the template image patch, i represents the image pixel point, μA and μB are the means of image patches A and B respectively, and it is judged whether the wall contour matches according to the NCC value. If the NCC value is greater than the set threshold, it is considered that the matching is successful.

[0013] Further, the use of the robot to cruise and photograph the wall image and upload the construction results also includes using the JPEG algorithm for image transmission. Among them, the image is quantized through the DCT transform coefficient, the quantized coefficients are scanned in a zigzag pattern, the two-dimensional coefficients are converted into a one-dimensional sequence, and Huffman coding is used to perform entropy coding on it to compress the data. When transmitting the data, an HTTP POST request is constructed, and the sorted image data and metadata are sent to the server as the request body, and the robot adds a Range field to the request header to handle resume data transfer.

[0014] Further, the intelligent comparison of the recognition result and the BIM model through the deep learning algorithm includes using the convolutional neural network ResNet model to extract features and classify the wall photo. Among them, for the calculation of the mortar joint fullness, the mortar joint area is segmented from the image through the image segmentation algorithm, and the ratio of the segmented mortar joint area to the theoretical mortar joint area is calculated to obtain the mortar joint fullness; for the evaluation of the block layout quality, the arrangement order and position deviation of the blocks in the image are detected and compared with the bricklaying plan in the BIM model. If the block position deviation exceeds the set threshold, it is considered that there is a problem with the layout quality.

[0015] In a second aspect, a system for the quality acceptance method of masonry walls based on intelligent recognition includes: A data acquisition module configured to acquire building drawings and floor wall component information; A composition module, configured to form a bricklaying plan for walls according to the obtained building drawings and floor wall component information; A model construction module, configured to construct a BIM model based on the bricklaying plan for walls; A data transmission module, configured to use a robot to cruise and capture wall images and upload construction results; An intelligent comparison module, configured to perform intelligent comparison between the uploaded construction result image data and the BIM model to obtain an inspection result; A reporting module, configured to form a list of quality problems from the inspection result for reporting.

[0016] Thirdly, the present invention provides a computer-readable storage medium, in which multiple instructions are stored, and the instructions are suitable for being loaded and executed by a processor of a terminal device for the method for intelligent identification-based quality acceptance of masonry walls.

[0017] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium. The processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor for the method for intelligent identification-based quality acceptance of masonry walls.

[0018] In summary, the present invention has the following beneficial technical effects: 1. The linkage between AI monitoring and intelligent hardware can effectively reduce the omission of quality inspections caused by human negligence and improve the quality production management level.

[0019] 2. The skills of operators vary during the construction process. It effectively improves the serious attitude of operators, improves their skills, and improves the project quality level.

[0020] 3. In the traditional inspection method, for a single wall, it is easy to miss quality problems. Through this system, all quality problems can be comprehensively identified and marked at each location without omission.

[0021] 4. For a wide operation surface, a large amount of masonry work, and scattered operations, the traditional management method cannot cover everything, resulting in losses in quality costs. This acceptance system manages to cover the entire operation surface. Through this acceptance system, the quality of operations at all locations is improved.

[0022] 5. After quality problems occur, traditionally, only the subcontractor responsible can be held accountable and then rectified. Through the system, it is convenient for later traceability. By linking with the settlement of workers, the implementation of quality production responsibilities is improved. Description of the Drawings

[0023] Figure 1 It is a schematic diagram of a method for intelligent identification-based quality acceptance of masonry walls according to Embodiment 1 of the present invention. Detailed implementation mode

[0024] The present invention will be further described in detail below with reference to the accompanying drawings.

[0025] Embodiment 1 Refer to Figure 1 , a method for intelligent recognition and quality acceptance of masonry walls in this embodiment includes: Obtain building drawings and floor wall component information; Form a wall bricklaying plan based on the obtained building drawings and floor wall component information; Construct a BIM model based on the wall bricklaying plan; Upload the image data of the on-site construction results of the workers; Intelligently compare the uploaded construction result image data with the BIM model to obtain inspection results; Form a list of quality problems with the inspection results and report them.

[0026] Specifically: I. Obtain building drawings and floor wall component information (1) Acquisition and preprocessing of building drawings Use a high-precision scanner to convert paper building drawings into electronic images with a resolution set to 600 dpi to ensure clear details of the drawings. Adopt OCR (Optical Character Recognition) technology and use a complex character feature extraction algorithm to recognize and extract the text information on the drawings. This algorithm is based on a deep learning convolutional neural network (CNN). Input the character image into the network, and through multiple convolutional and pooling operations, extract the feature vectors of the characters, and then perform classification recognition through a fully connected layer to accurately obtain key text information such as drawing numbers, project names, and design units.

[0027] For the graphic information in the drawings, use a vectorization algorithm for processing. Based on the contour tracking vectorization method, start from image edge detection and use the Canny edge detection algorithm. This algorithm smooths the image through Gaussian filtering to reduce noise interference, then calculates the gradient magnitude and direction of the image, performs non-maximum suppression to remove false edges, and finally determines the true edge contour through double-threshold detection. Convert the detected edge contour into a vector line segment representation for subsequent precise analysis of the graphics.

[0028] (2) Collection of floor wall component information If a BIM (Building Information Modeling) model has been established for the project, the detailed information of the floor wall components can be directly extracted from the model with the help of professional BIM software interface through specific API (Application Programming Interface) functions, including the precise location coordinates (X, Y, Z) of the wall, length, width, height dimensions, wall material, and the location and dimensions of door and window openings.

[0029] When there is no ready-made BIM model, use a total station for measurement. The total station uses the principle of triangulation to determine the location of the target point by transmitting and receiving electromagnetic waves. Suppose the coordinates of the total station are (X0, Y0, Z0), the horizontal angle of the target point is α, the vertical angle is β, and the slope distance is S. According to the trigonometric formula, the coordinates of the target point (X, Y, Z) are calculated as follows: X = X0 + S × cosβ × cosα Y = Y0 + S × cosβ × sinα Z = Z0 + S × sinβ Laser rangefinders are used to measure linear dimensions such as length and height of walls. Their working principle is to calculate the distance by measuring the time difference between laser emission and reception and combining it with the speed of light.

[0030] 2. Form a wall brick layout diagram based on the acquired architectural drawings and floor wall component information (1) Determination of block parameters From the material description section of the building drawing, carefully check the type of block, such as ordinary concrete block, shale brick or aerated concrete block, and obtain its standard size, which is recorded as length a, width b and height c respectively.

[0031] Considering the mortar joint factors during construction, the horizontal mortar joint thickness is usually between 8 and 12mm, and the average value is 10mm, recorded as δ1; the vertical mortar joint thickness is between 10 and 12mm, and the average value is 11mm, recorded as δ2.

[0032] (2) Brick arrangement algorithm design The wall brick arrangement design is based on the idea of ​​dynamic programming algorithm. Starting from the first row at the bottom of the wall, the number of bricks required for each row n1 is calculated based on the wall length L, the block length a, and the mortar joint thickness δ1: in, is a rounding function. In the calculation process, the symmetry and aesthetics of the block arrangement should be taken into consideration, and the mortar joints should be evenly distributed as much as possible. For the wall height H, calculate the required number of block layers n2: .

[0033] For special parts such as door and window openings, a lintel needs to be set above the door and window openings. The length of the lintel is generally determined by adding 250mm to each side of the opening width, that is, the lintel length , where is the width of the door and window opening. In the bricklaying plan, accurately mark the position and size of the lintel to ensure that the bricklaying plan matches the structural design.

[0034] Use professional architectural drawing software, such as AutoCAD, and according to the calculation results, write a LISP (List Processing) language program to automatically draw the bricklaying plan of the wall. During the drawing process, visually present the position of each block, the width of the mortar joint, and the treatment of special parts clearly.

[0035] III. Constructing a BIM model based on the bricklaying plan of the wall (1) BIM software operation and data import Select the widely used Revit software in the industry to construct a BIM model. Import the drawn bricklaying plan of the wall into Revit in DWG format, and use the import setting function of Revit to accurately match the coordinate system and unit of the drawing to ensure the accuracy of the imported data.

[0036] (2) Model creation and parameter setting In Revit, use family libraries such as "wall" and "block" to create a wall model. For each block, through parametric settings, assign accurate size (length a, width b, height c), material (such as properties such as density and strength grade of concrete material) information.

[0037] Utilize the powerful parametric design function of Revit to establish a parametric association between the block and the wall. By creating parametric formulas, such as the association formula between the wall length and the number of blocks: , when the wall size or the bricklaying plan changes, the model can automatically update according to the parametric formula to maintain the consistency and accuracy of the model. Add components such as lintels and tie bars to the model. The lintel is accurately placed according to the designed dimensions (length , width , height ). The tie bars are set at intervals of 500mm along the wall height according to the specification requirements, and the length extending into the wall is not less than 1000mm. Through the steel bar drawing tool of Revit, arrange them according to the set spacing and length, and set the connection parameters between the tie bars and the wall to ensure the stability of the structure. IV. Robot cruise route planning: When planning the cruising route of the robot, it is necessary to comprehensively consider the complex environment of the construction site, the distribution of walls, and the movement characteristics of the robot itself. In this embodiment, the advantages of the A* algorithm and the Dijkstra algorithm are combined, and the dynamic changes of obstacles are considered to achieve efficient and safe path planning.

[0038] 4.1.1 Environmental Modeling First, according to the positions and layouts of the walls on each floor in the BIM model and the information about possible obstacles at the construction site, a two-dimensional grid map is constructed. The entire floor space is divided into equal-sized grid cells, and each grid cell can be represented as a node. The state of the node is divided into passable and impassable. The impassable nodes represent the positions of obstacles, such as large equipment at the construction site, piled-up building materials, etc.

[0039] To more accurately reflect the environmental information, the dynamic changes of obstacles are considered. Sensors, lidars, and cameras are arranged at the construction site to monitor the positions and states of obstacles in real time. At regular intervals, the states of the nodes in the grid map are updated to ensure the real-time and accuracy of path planning.

[0040] 4.1.2 Initial Parameter Settings Determination of the starting point and the ending point: The starting point is the initial position of the robot, and the ending point is the position of the last wall to be inspected. According to the distribution of the walls, all the walls to be inspected are numbered in a certain order, and the robot visits the walls in the numbered order.

[0041] Selection of the heuristic function: The core of the A* algorithm is the heuristic function , where is the actual cost from the starting node to node n, is the estimated cost from node n to the target node. In this solution, can be defined as the number of grid cells passed from the starting node to node n multiplied by the side length of each grid cell, uses the Manhattan distance for estimation, that is , where is the coordinate of node n, is the coordinate of the target node.

[0042] 4.1.3 A* Algorithm Path Search Initialization of the open list and the closed list: The open list is used to store the nodes to be explored, and the closed list is used to store the nodes that have been explored. Initially, the open list only contains the starting node, and the closed list is empty.

[0043] Node expansion: Select from the open list Expand the node n with the smallest value. For the adjacent node m of node n, calculate its and values, and update . If node m is not in the open list and the closed list, add it to the open list and set node n as the parent node of node m. If node m is already in the open list and the new value is smaller, update and values, and set node n as the parent node of node m.

[0044] Target node judgment: If the expanded node n is the target node, the search ends, and the path from the starting point to the target node can be obtained by backtracking the parent nodes. If the open list is empty, it means that no feasible path is found.

[0045] 4.1.4 Dijkstra Algorithm Optimization In some cases, the A* algorithm may result in a non-optimal path due to the inaccuracy of the heuristic function. To further optimize the path, based on the path obtained by the A* algorithm, the Dijkstra algorithm is used for local optimization.

[0046] The Dijkstra algorithm is a breadth-first search algorithm. It starts from the starting point and gradually expands nodes until the target node is found. When expanding nodes, the Dijkstra algorithm only considers the actual distance between nodes and does not use the heuristic function. Take the path obtained by the A* algorithm as the initial path, use the nodes on the path as the starting points, and use the Dijkstra algorithm for local search to try to find a shorter path. If a shorter path is found, replace the original path.

[0047] 4.1.5 Path Smoothing To make the movement of the robot smoother, the planned path needs to be smoothed. Adopt the method of Bezier curve fitting, use the nodes on the path as control points, and generate a smooth curve. The formula of the Bezier curve is: , where n is the number of control points, is the coordinate of the i-th control point, and t is a parameter with a value range of [0,1]. By adjusting the value of t, different points on the curve can be obtained.

[0048] 4.2 Determination of Photographing Posture and Angle According to the height and width of the wall and the parameters of the photographing device (such as focal length, viewing angle, etc.), calculate the optimal photographing posture and angle.

[0049] 4.2.1 Height and Distance Calculation For higher walls, in order to ensure the integrity of shooting, it is necessary to adjust the height of the robot or the distance from the wall. Assume the height of the wall is H, the vertical viewing angle of the shooting device is , and the distance between the robot and the wall is d. Then we have: , and the optimal distance d between the robot and the wall can be calculated through the above formula. If the robot cannot reach this distance, it is necessary to adjust its own height to ensure that the top of the wall can be photographed.

[0050] 4.2.2 Angle Calculation In order to ensure that the captured wall image has no obvious distortion, it is necessary to adjust the shooting angle. Assume the width of the wall is W, and the horizontal viewing angle of the shooting device is , then the shooting angle is calculated through the following formula: , and the optimal shooting angle can be calculated through the above formula .

[0051] 4.2.3 Multi-Angle Shooting Strategy For wider walls, it may be necessary to use the method of shooting from multiple angles and then stitching. Divide the wall into several regions, and each region corresponds to a shooting angle. According to the width of the wall and the horizontal viewing angle of the shooting device, calculate the width of each region and the corresponding shooting angle. The robot shoots the images of each region in turn according to the calculated angles, and then uses an image stitching algorithm to stitch these images into a complete wall image.

[0052] 4.3 Implementation of Floor Judgment Function The robot is equipped with a variety of floor recognition sensors to improve the accuracy and reliability of floor judgment.

[0053] 4.3.1 Laser Rangefinder Sensor The laser rangefinder sensor judges the floor height by measuring the distance between the robot and the ceiling or the ground. Assume the height of each floor is h, and the distance measured by the robot is d. Then the floor n where it is located can be calculated through the following formula: , where is the floor function.

[0054] 4.3.2 RFID Sensor Set RFID tags at specific positions on each floor, and the floor number information is stored in the tags. The robot is equipped with an RFID reader. When the robot approaches the RFID tag, the reader reads the information in the tag to determine the current floor where it is located.

[0055] 4.3.3 Multi-Sensor Fusion To improve the accuracy of floor judgment, the data of the laser ranging sensor and the RFID sensor are fused. The weighted average method can be adopted, and different weights are set according to the reliability of the two sensors to calculate the final floor judgment result. For example, the weight of the laser ranging sensor is , and the weight of the RFID sensor is , and , then the final floor judgment result is: , where n1 is the floor judged by the laser ranging sensor and n2 is the floor judged by the RFID sensor.

[0056] V. The robot takes pictures of the wall The robot moves to the specified shooting point according to the planned cruise route and takes pictures of the wall according to the determined shooting posture and angle. The shooting device uses an 8-megapixel camera, and the resolution is set to more than 4000×3000 pixels to ensure image clarity. According to the on-site light environment, the white balance parameters are automatically adjusted to ensure accurate color reproduction of the captured images.

[0057] During the shooting process, the robot records in real time its own position information (through the built-in positioning system, such as GPS or indoor positioning system), shooting time, shooting angle and other metadata, and associates this metadata with the captured images.

[0058] For large-area walls, a multi-view shooting strategy is adopted, and then the SIFT (Scale-Invariant Feature Transform) algorithm based on feature point matching is used to stitch multiple images into a complete wall image. The SIFT algorithm detects feature points in the image, uses the scale invariance and rotation invariance of the feature points, calculates the Euclidean distance between the feature points as a similarity measure, and realizes the accurate stitching of the images.

[0059] Among them, 5.1 Multi - perspective shooting strategy For large - area walls, to ensure the integrity and accuracy of image information, a multi - perspective shooting strategy needs to be adopted. According to the actual size, shape of the wall and the parameters of the shooting equipment (such as the field of view, resolution, etc.), the shooting area is reasonably divided. The wall is divided into several overlapping sub - regions, and the overlapping part of each sub - region should be between 20% - 30% to ensure that enough matching feature points can be found for subsequent image stitching. If the wall is rectangular, it can be evenly divided into multiple small rectangular regions horizontally and vertically, and the robot moves to the shooting positions of each region in turn and takes pictures according to the pre - determined shooting postures and angles. During the shooting process, the consistency of shooting parameters, including aperture, shutter speed, ISO, etc., is strictly controlled to ensure that the multiple images taken are consistent in terms of color, brightness and contrast, laying a good foundation for subsequent image stitching work.

[0060] 5.2 Image stitching based on SIFT algorithm 5.2.1 Scale - space extreme value detection The core of the scale - space theory is to simulate the human eye's perception of images at different resolutions by smoothing the image at different scales. The Gaussian convolution kernel is used to convolve with the original image to generate Gaussian pyramids at different scales. The expression of the Gaussian convolution kernel is: where (x, y) are the coordinates of the image pixels, is the scale factor, which controls the size of the Gaussian kernel. Subtracting two adjacent layers of Gaussian images gives the Difference of Gaussians (DoG) image, and its expression is: where I(x, y) is the original image and k is the scale factor between adjacent scales. In the DoG image, by comparing each pixel with its 26 adjacent pixels (including 8 neighborhood pixels at the same scale and 9 pixels at the adjacent upper and lower scales), local extreme points are found. These extreme points are the possible feature points.

[0061] 5.2.2 Key - point localization To improve the localization accuracy of feature points, the exact position and scale of the scale - space extreme points need to be determined precisely. The Taylor expansion formula is used to fit the DoG function, and the exact position of the feature points is obtained by solving the extreme points of the fitted function.

[0062] Let the Taylor expansion formula of the DoG function at a certain extreme point be: where, is the offset vector. By taking the partial derivative of \(D(X)\) and setting it to zero, the solution for the offset \(X\) can be obtained: If the absolute value of the offset \(X\) is greater than 0.5, the position of the feature point needs to be updated to the offset position, and the scale - space extremum detection is performed again. At the same time, in order to remove feature points with low contrast and edge responses, threshold judgments need to be made on the contrast and principal curvature of the feature points. The contrast is measured by the value of \(D(X)\), and the principal curvature is calculated by the Hessian matrix of the DoG function. If the contrast of the feature point is lower than the set threshold (such as 0.03) or the principal curvature is higher than the set threshold (such as 10), then the feature point is removed.

[0063] 5.2.3 Orientation Assignment To make the feature points rotation - invariant, a main orientation needs to be assigned to each feature point. In the neighborhood of the feature point, calculate the gradient magnitude \(m(x,y)\) and direction \(\theta(x,y)\) of each pixel point, and their calculation formulas are: where \(L(x,y)\) is the Gaussian scale - space image. In the neighborhood of the feature point, with the feature point as the center, count the histogram of the gradient direction. The histogram has 36 bins, and each bin corresponds to a 10° direction range. The peak of the histogram is the main direction of the feature point. If there are other peaks that reach more than 80% of the main - direction peak, then these directions are also used as the secondary directions of the feature point.

[0064] 5.2.4 Feature Descriptor Generation Select a 16x16 neighborhood around the feature point. Divide this neighborhood into 4x4 small regions, and each small region is further divided into 4x4 sub - regions. In each sub - region, count the histogram of the gradient direction. The histogram has 8 bins, and each bin corresponds to a 45° direction range.

[0065] Combine the 8 - dimensional histogram vectors of each small region to obtain a 128 - dimensional feature descriptor vector. To improve the robustness of the feature descriptor, normalize the feature descriptor vector to make it scale - and illumination - invariant.

[0066] 5.2.5 Feature Point Matching Calculate the Euclidean distance between feature descriptors in different images as a similarity metric. For each feature point in an image, find the two feature points in another image with the closest Euclidean distances to it. If the ratio of the Euclidean distance between the nearest neighbor feature point and the second nearest neighbor feature point is less than a set threshold, these two feature points are considered to be matched. In this way, a large number of matched feature point pairs can be found in the two images. These matched feature point pairs will be used for subsequent image transformation and stitching.

[0067] 5.2.6 Image Transformation and Stitching Based on the matched feature point pairs, use the least squares method to solve the transformation matrix between the images. Common transformation models include affine transformation and perspective transformation. The expression of affine transformation is: where (x, y) are the pixel coordinates in the original image, and (x', y') are the pixel coordinates in the transformed image. The expression of perspective transformation is: By solving the transformation matrix, different images can be transformed to the same coordinate system. Then, the transformed images are stitched together. A simple average fusion of the overlapping regions or a more complex multi-band fusion algorithm can be used to eliminate the stitching seam and obtain a complete wall image.

[0068] VI. Robot Data Upload 6.1.1 Image Data Compression: To reduce the amount of data transmitted and improve the transmission efficiency, an image compression algorithm such as the JPEG algorithm is used. The JPEG algorithm is based on the discrete cosine transform (DCT), and its main steps are as follows: Block Division: Divide the image into 8×8 pixel blocks. Perform DCT transformation on each 8×8 pixel block to convert the image in the spatial domain to the frequency domain. The DCT transformation formula is: where f(x, y) are the pixel values in the spatial domain, F(u, v) are the frequency domain coefficients, and C(u) and C(v) are constants.

[0069] 6.1.2 Quantization: Quantize the coefficients after DCT transformation. Control the retention precision of different frequency coefficients through a quantization table to achieve data compression. The quantization formula is: where Q(u, v) is the quantization value at the corresponding position in the quantization table.

[0070] 6.1.3 Encoding: Perform a zigzag scan on the quantized coefficients to convert the two-dimensional coefficients into a one-dimensional sequence, and perform entropy coding on it using Huffman coding to further compress the data. Metadata integration: Integrate metadata such as shooting time, location (latitude and longitude or indoor positioning coordinates), shooting angle, and floor information according to a specific data structure. For example, create a JSON object containing various metadata to facilitate subsequent transmission and parsing. 6.2 Selection of input protocol To ensure the reliability and efficiency of data transmission, select the HTTP or HTTPS protocol in the TCP / IP protocol stack for data upload.

[0071] Data transmission process: Establish a connection: The robot, as the client, initiates a TCP connection request to the specified port of the quality acceptance system server. After receiving the request, the server establishes a TCP connection with the client.

[0072] Construct a request: The client constructs an HTTP POST request and sends the organized image data and metadata as the request body to the server. The request header contains information such as data type and length to enable the server to correctly parse the request. Set Content-Type to multipart / form-data for uploading multiple files and data.

[0073] 6.3 Optimization of the transmission process During the transmission process, problems such as network fluctuations and signal interference may occur, affecting the stability and speed of data upload. Therefore, the following optimization strategies are adopted: Resume interrupted transfer: When an interruption occurs during the transmission process, record the position of the uploaded data block. After reconnecting, continue uploading the remaining data from the breakpoint to avoid repeating the upload of data that has been successfully transmitted. To implement resume interrupted transfer, add a Range field to the request header to inform the server to start receiving data from the specified position.

[0074] Concurrent transmission: To improve the upload speed, divide the image data and metadata into multiple data blocks and adopt the method of concurrent transmission. The robot creates multiple HTTP requests to upload different data blocks simultaneously. However, pay attention to the number of concurrent requests to avoid network congestion caused by excessive requests. The number of concurrent requests can be dynamically adjusted according to the network bandwidth and server load. By measuring the network bandwidth and server response time, establish a mathematical model to calculate the optimal number of concurrent requests. Let the network bandwidth be B (bps), the size of each data block be S (bit), and the average server response time be T (s), then the optimal number of concurrent requests N can be estimated by the formula Estimate.

[0075] 6.4 Data verification To ensure the integrity and accuracy of the uploaded data, the received data is verified on the server side. Hash verification: Before uploading the data, the client calculates the hash values of the image data and metadata, such as MD5, SHA-1, or SHA-256. The hash value is sent to the server as part of the request header. After receiving the data, the server recalculates the hash value of the data and compares it with the hash value sent by the client. If the two are consistent, it indicates that the data has not been tampered with during transmission; if not, the client is required to re-upload the data. Data integrity check: The server checks whether the received data is complete based on the data length recorded in the request header. If the data length does not match, an error message is sent to the client, requesting re-upload.

[0076] VII. Obtain the inspection result by performing an intelligent comparison between the uploaded construction result image data and the BIM model (1)Wall recognition based on computer vision algorithms The edge detection algorithm is used to process the uploaded wall photos, and the Canny edge detection algorithm is a common choice. First, the image is subjected to Gaussian filtering using the two-dimensional Gaussian function: ) where σ is the standard deviation of the Gaussian kernel, and the filtering intensity is controlled by adjusting the value of σ. After Gaussian filtering, the gradient magnitude and direction of the image are calculated. The Sobel operator is used to calculate the gradients in the horizontal and vertical directions, and then the gradient magnitude and direction . Then non-maximum suppression is performed to remove false edges, and finally the true wall edge contour is determined through double-threshold detection. Based on the template matching algorithm, the extracted wall edge contour is matched with the theoretical contour of the wall in the BIM model. The normalized cross-correlation coefficient (NCC) is used as the matching metric: where A and B represent the image patch to be matched and the template image patch respectively, i represents the image pixel point, and μA and μB are the means of image patches A and B respectively. A matching threshold of 0.8 is set. When the NCC value is greater than this threshold, the contour matching is considered successful. Displacement sensors, pressure sensors and other devices are used to monitor the deformation of the wall during the masonry process in real time.

[0077] The displacement sensor converts the wall displacement into an electrical signal. Let the displacement be d. Through the sensitivity coefficient k of the sensor, the displacement is converted into a voltage signal V = k×d. The pressure sensor converts the pressure value P into an electrical signal. After analog-to-digital conversion (ADC) technology, the analog electrical signal is converted into a digital signal and stored in the electronic quality problem summary database. By comparing the uploaded on-site masonry photos with the information in the database, relevant parameters of the actual on-site masonry are obtained, such as the flatness and verticality of the wall, and then compared and analyzed with the BIM model. The flatness of the wall is measured by calculating the deviation between the wall edge in the image and the ideal straight line, and the verticality is determined using a gravity sensor or an image-based vertical angle calculation method.

[0078] (2) The comparison and analysis based on the deep learning algorithm uses an advanced convolutional neural network (CNN) model, such as ResNet (Residual Network), to extract features and classify the wall photos. First, the image data is preprocessed, including normalization processing, which adjusts the image pixel value range to between [0, 1]. The formula is: where x is the original pixel value, and are the minimum and maximum values of the image pixel values respectively. The preprocessed image is input into the ResNet model. The model extracts and classifies the image features through multiple convolutional layers, pooling layers, and fully connected layers, and identifies information such as wall joint, top plug fullness, block layout quality, lintel position, tie bar identification, grooving and pipe laying, reserved hole position, and appearance quality. For the calculation of joint fullness, the joint area is segmented from the image through an image segmentation algorithm. A threshold segmentation-based method is adopted. According to the color or gray-scale characteristics of the joint, a suitable threshold T is set to divide the image into joint and non-joint areas. Calculate the ratio of the actual area S1 of the segmented joint to the theoretical joint area S0 to obtain the joint fullness P: For the evaluation of block layout quality, the arrangement order and position deviation of the blocks in the image are detected. Using an object detection algorithm, such as the deep learning-based YOLO (You Only Look Once) algorithm, the position of each block is identified. The detected block positions are compared with the bricklaying plan in the BIM model. If the block position deviation exceeds the set threshold (such as 10 mm), it is determined that there is a problem with the layout quality.

[0079] VIII. Form a quality problem list with the inspection results and report it (1)Problem Classification and Detailed Record Based on the inspection results obtained through intelligent comparison, classify the quality problems in detail. The mortar joint problems include insufficient mortar filling in the joints (filling degree lower than 80%), inconsistent mortar joint thickness (thickness deviation exceeding ±2mm); the top plug problems are loose top plugs (judged by the size of the gap between the top plug and the wall through image analysis), and the top plugs are not set as required (no 15 - 20mm space is reserved at the top of the wall for the top plug); the block layout problems cover incorrect block arrangement (inconsistent with the bricklaying plan), and the staggered joints do not meet the requirements (the vertical mortar joint stagger is less than 1 / 3 of the block length); the lintel problems include lintel position deviation (exceeding the design position by ±20mm), and the lintel size does not meet the design requirements (length, width, and height deviation exceeding ±5mm); the tie bar problems are insufficient number of tie bars (less than the required number in the design), and the tie bar length is not enough (less than the required length of 1000mm in the design); the grooving and pipe laying problems involve inaccurate grooving positions (deviating from the design position by ±50mm), and the pipes are not firmly fixed (observing whether the pipes are loose through images); the reserved hole problems include reserved hole position deviation (exceeding the design position by ±30mm), and the hole size does not meet the requirements (greater or smaller than the design size by ±10mm); the appearance quality problems are wall surface cracks (crack width exceeding 0.3mm), missing corners, etc.

[0080] Make a detailed record of each type of quality problem, including the wall position where the problem occurs (accurate to the floor, room number, and wall number), problem description, and problem severity (minor, general, severe). For example, on a certain numbered wall in a certain room on a certain floor, the problem of insufficient mortar filling in the joints is found, described as "the calculated filling degree of some mortar joints is 70%, lower than the standard requirement", and the severity is determined to be general.

[0081] (2)Reporting Method and Tracking Management Through the report generation function of the quality acceptance system, using report generation tools (such as JasperReports), generate a report of the quality problem list in Excel or PDF format. With the help of the email sending protocol (such as SMTP) or the API interface of instant messaging software, push the report to relevant management personnel. At the same time, establish a problem tracking module in the quality acceptance system, assign a unique identification code to each quality problem, record the rectification status of the problem through the database, and track the whole process from problem discovery, issuing a rectification notice, during rectification to rectification completion until the problem is properly solved to ensure the closed-loop of quality management.

[0082] Example 2 This example provides a system for intelligent recognition and masonry wall quality acceptance method, including: A data acquisition module, configured to acquire building drawings and floor wall component information; A composition module, configured to form a wall bricklaying plan according to the obtained building drawings and floor wall component information; A model construction module, configured to construct a BIM model based on the wall bricklaying plan; A data transmission module, configured to use a robot to cruise and capture wall images and upload construction results; An intelligent comparison module, configured to perform an intelligent comparison between the uploaded construction result image data and the BIM model to obtain an inspection result; A reporting module, configured to form a list of quality problems with the inspection result and report it.

[0083] A computer-readable storage medium, in which multiple instructions are stored, and the instructions are suitable for being loaded and executed by a processor of a terminal device to perform the method described above.

[0084] A terminal device, including a processor and a computer-readable storage medium, where the processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor to perform the method described above.

[0085] The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A quality acceptance method for masonry walls based on intelligent recognition, characterized in that, Including: Obtain architectural drawings and floor wall component information; Form a wall bricklaying plan based on the obtained architectural drawings and floor wall component information; Construct a BIM model based on the wall bricklaying plan; Use a robot to cruise and capture wall images and upload the construction results; Perform intelligent comparison between the uploaded construction result image data and the BIM model to obtain an inspection result; Form a quality problem list from the inspection result and report it.

2. The method for quality acceptance of a masonry wall based on intelligent recognition according to claim 1, characterized in that, The forming of the wall bricklaying plan according to the obtained architectural drawings and floor wall component information includes obtaining the block type and size according to the architectural drawings, starting from the bottom of the wall and performing bricklaying design upward through the greedy algorithm, calculating the number of blocks required for each row according to the wall length, block length, and mortar joint thickness, and at the same time calculating the number of block layers required according to the wall thickness. Finally, use graphic drawing software to draw the wall bricklaying plan according to the calculation results, showing the position of each block, the mortar joint distribution, and the treatment of special parts.

3. The method for quality acceptance of a masonry wall based on intelligent recognition according to claim 2, characterized in that, The constructing of the BIM model based on the wall bricklaying plan includes, through Revit software, importing the drawn wall bricklaying plan into Revit in a format supported by BIM software such as DWG. In Revit, according to the information in the bricklaying plan, use the relevant family libraries of walls and blocks to create a wall model. For each block, assign accurate size and material property information; utilize the parametric design function of Revit to establish a parameter association between the block and the wall to ensure that when the wall size or bricklaying plan changes, the model can be automatically updated; at the same time, add lintel and tie bar components to the model. The lintel is placed according to the required size and position in the design, and the tie bars are arranged according to the required spacing and length in the specification through the steel bar drawing tool of Revit.

4. The method for quality acceptance of a masonry wall based on intelligent recognition according to claim 3, characterized in that, The using of the robot to cruise and capture wall images includes, according to the stair position and layout in the BIM model, constructing a two-dimensional network map, dividing the entire floor space into grid cells of equal size, with each grid cell representing a node, using the Manhattan method to estimate the coordinates of the target node, and obtaining the path from the starting point to the target node by backtracking the parent node through the A* algorithm path search method. At the same time, use the Dijkstra algorithm for local search optimization to obtain the cruise path, and finally obtain the smoothed cruise path through Bezier curve fitting.

5. The method for quality acceptance of a masonry wall based on intelligent recognition according to claim 4, characterized in that, The using of the robot to cruise and capture wall images also includes calculating the optimal distance between the robot and the wall according to the height, width, and shooting parameters of the wall, calculating the optimal shooting angle according to the shooting horizontal angle. For walls with a width greater than the set value, use the multi-angle shooting and stitching method to obtain wall images. Among them, obtain the floor height by using a laser distance sensor with the ground distance as a reference, obtain the floor label through an RFID sensor, and obtain the floor result by using a weighted method based on the floor height and floor label.

6. The method for intelligent recognition-based quality acceptance of masonry walls according to claim 5, characterized in that, The method of obtaining wall images by stitching images taken from multiple angles includes stitching wall images based on the SIFT algorithm. The obtained images are convolved with the original images using a Gaussian convolution kernel to generate Gaussian pyramids at different scales. The adjacent two layers of Gaussian images are subtracted to obtain Gaussian difference images. Local extreme points are found by comparing each pixel point with its 26 adjacent pixel points and used as feature points. In the neighborhood of the feature points, the gradient magnitude and direction of each pixel point are calculated. After statistically analyzing the direction range of the gradient direction histogram in each sub-region, the 8-dimensional histogram vectors of small regions are combined to obtain a 128-dimensional feature descriptor vector.

7. The method for quality acceptance of a masonry wall based on intelligent recognition according to claim 6, wherein The method of obtaining inspection results by intelligently comparing uploaded data with the BIM model includes identifying the wall completed on-site based on computer vision algorithms and intelligently comparing the recognition results with the BIM model using deep learning algorithms. Among them, the Canny edge detection algorithm is used to extract the edges of the uploaded wall photos. The image is smoothed by Gaussian filtering, the gradient magnitude and direction of the image are calculated, and then non-maximum suppression and double-threshold detection are performed to finally obtain a clear wall edge contour. Based on the template matching algorithm, the extracted wall edge contour is matched with the theoretical contour of the wall in the BIM model. Among them, the similarity between the two is calculated by using the normalized cross-correlation coefficient NCC, which is expressed as: NCC(A,B)=Σ[(A(i)-μA)(B(i)-μB)] / [sqrt(Σ(A(i)-μA)^2)×sqrt (Σ(B(i)-μB)^2)], where A and B represent the image patch to be matched and the template image patch respectively, i represents the image pixel point, μA and μB are the means of image patches A and B respectively. Whether the wall contour is matched is judged according to the NCC value. If the NCC value is greater than the set threshold, it is considered that the matching is successful.

8. A quality acceptance method for masonry walls based on intelligent recognition according to claim 7, characterized in that, The method of using a robot to cruise and shoot wall images and upload construction results also includes using the JPEG algorithm for image transmission. Among them, the image is quantized through DCT transform coefficients, the quantized coefficients are scanned in a zigzag pattern, the two-dimensional coefficients are converted into a one-dimensional sequence, and Huffman coding is used for entropy coding to compress the data. When transmitting the data, an HTTP POST request is constructed, and the organized image data and metadata are sent to the server as the request body. The robot adds a Range field to the request header to handle resume data transfer.

9. The method for quality acceptance of a masonry wall based on intelligent recognition according to claim 8, wherein, The intelligent comparison of the recognition result and the BIM model through the deep learning algorithm includes using the convolutional neural network ResNet model to extract features and classify the wall photos. Among them, for the calculation of the mortar joint fullness, the mortar joint area is segmented from the image through the image segmentation algorithm, and the ratio of the segmented mortar joint area to the theoretical mortar joint area is calculated to obtain the mortar joint fullness; for the evaluation of the block layout quality, by detecting the arrangement order and position deviation of the blocks in the image and comparing them with the bricklaying plan in the BIM model, if the block position deviation exceeds the set threshold, it is considered that there is a problem with the layout quality.

10. A system for the quality acceptance method of a masonry wall based on intelligent recognition, characterized in that, Including: A data acquisition module configured to acquire building drawings and floor wall component information; A composition module configured to form a wall bricklaying plan according to the acquired building drawings and floor wall component information; A model construction module configured to construct a BIM model based on the wall bricklaying plan; A data transmission module configured to use a robot to cruise and photograph wall images and upload the construction results; An intelligent comparison module configured to perform intelligent comparison between the uploaded construction result image data and the BIM model to obtain an inspection result; A reporting module configured to form a quality problem list from the inspection result and report it.

Images