House apparent crack accurate positioning and detection method, medium and equipment

Abstract

This invention relates to the field of civil engineering monitoring technology, and particularly to a method, medium, and equipment for precise location and detection of apparent cracks in buildings. It includes: unfolding and cropping a scanned 3D point cloud model of the building wall surface into N local 3D planar point cloud models; driving a wall-climbing robot to capture high-definition local images of the wall surface; using the high-definition local images to repair the blurred local 2D wall images corresponding to the local 3D planar point cloud models to obtain repaired local wall images; and based on the repaired local wall images, mapping the crack information of the crack images sequentially to the local 3D planar point cloud models and the building wall surface 3D point cloud model to obtain the building wall surface 3D point cloud model. This invention automates the entire process of crack detection, from data acquisition, route planning, and image capture to crack identification and 3D location, and can intuitively and accurately present the 3D location and distribution characteristics of cracks on the building wall surface.

Description

Technical Field

[0001] This invention relates to the field of civil engineering monitoring technology, and in particular to a method, medium and equipment for accurately locating and detecting apparent cracks in buildings. Background Technology

[0002] Cracks in building walls are an important indicator of the health of concrete structures. Their appearance and development may indicate safety hazards such as decreased structural strength and insufficient stability. Therefore, accurate detection and location of cracks in building walls is a core aspect of building operation and maintenance.

[0003] Existing methods for detecting building cracks are mainly divided into two categories: manual inspection and semi-automated inspection. Manual inspection relies on inspectors carrying tools to conduct on-site surveys, observing the cracks visually and measuring their location, length, and width with a ruler. This method is not only inefficient, but the results are also easily affected by the inspectors' experience and subjective judgment, making it difficult to guarantee accuracy. Furthermore, in scenarios such as high-altitude walls and high-rise buildings, the personal safety of inspectors is at great risk. Semi-automated inspection methods mostly use single 3D laser scanning or wall-climbing robot photography. While pure 3D laser scanning can obtain a 3D model of the building's appearance, it lacks accuracy in identifying small cracks in point cloud data and is easily interfered with by debris on the wall (such as air conditioner outdoor units, window frames, pipelines, etc.), leading to data redundancy. Pure wall-climbing robot photography suffers from blind route planning, easily resulting in missed or repeated shots. Moreover, the captured images lack precise coordinate correlation with the 3D structure of the building, making it impossible to achieve accurate 3D crack positioning, associate crack information with the overall structure of the building, and intuitively present the overall health status of the building's concrete.

[0004] In addition, existing crack recognition models mostly use traditional U-net or U-net3+ models, which have limited accuracy in recognizing small, slender, and irregular targets such as wall cracks, and are prone to missed or false recognition. At the same time, during the fusion of 3D point clouds and 2D images, there are often coordinate mapping deviations, which cause the repaired image to not match the actual wall scene, affecting the accuracy of subsequent crack recognition.

[0005] Therefore, it is necessary to provide a new method for accurately locating and detecting apparent cracks in buildings to solve the above-mentioned technical problems. Summary of the Invention

[0006] The main objective of this invention is to provide a method, medium, and equipment for precise location and detection of apparent cracks in buildings, aiming to solve the problems that existing methods cannot achieve precise three-dimensional location of cracks and are difficult to associate crack information with the overall structure of the building.

[0007] To achieve the above objectives, this invention proposes a method for accurately locating and detecting apparent cracks in buildings, comprising the following steps: S1: Set up a 3D laser scanner at preset points around the house to be tested, and use the 3D laser scanner to scan the house to be tested from all directions to obtain a 3D point cloud model of the house wall surface. S2: Unfold the 3D point cloud model of the house wall into a 3D planar model of the wall, and trim it into N local 3D planar point cloud models of the wall; S3: Drive the wall-climbing robot to capture high-definition local images of the wall surface corresponding to the 3D planar point cloud model of each local wall surface; S4: Using high-resolution local wall images, the blurred local 2D wall image corresponding to the 3D planar point cloud model of the local wall is repaired to obtain the repaired local wall image; specifically including: S4.1. The perspective projection algorithm is used to project the three-dimensional planar point cloud model of each local wall surface onto a two-dimensional plane to obtain a blurred local two-dimensional wall surface image; S4.2. Using image registration and fusion algorithms, a blurred local two-dimensional wall image is repaired using a high-definition local wall image to obtain a repaired local wall image. S5: Use the semantic segmentation model of wall cracks to identify crack images of each high-definition local wall image and extract the corresponding geometric parameters. Based on the repaired local wall image, map the crack information of the crack image to the local wall 3D planar point cloud model and the house wall appearance 3D point cloud model in sequence to obtain the house wall appearance 3D point cloud model with accurate crack information, where: crack information includes location, shape and geometric parameters.

[0008] Optionally, S1 includes: S1.1 Confirm the location of the preset points around the house. At least one 3D laser scanner shall be set up at each preset point. When setting up the 3D laser scanner, the level of the 3D laser scanner shall be calibrated by a level and the absolute coordinates of the scanner shall be obtained by a GPS positioning module. The setting height and angle of the 3D laser scanner shall be adjusted according to the height of the house and the wall area. S1.2. Start the 3D laser scanner to scan the walls of the house from all angles to obtain the original point cloud model; S1.3. The statistical filtering algorithm is used to remove isolated noise points caused by environmental interference in the original point cloud model to obtain the three-dimensional point cloud model of the building wall surface.

[0009] Optionally, S2 includes: S2.1. The three-dimensional point cloud model of the building wall surface is unfolded into a three-dimensional planar point cloud model using a planar fitting algorithm. Specifically: Establish a local coordinate system for the wall surface, with one vertex of the wall surface as the origin, the horizontal direction of the wall surface as the X-axis, the vertical direction as the Y-axis, and the direction perpendicular to the wall surface as the Z-axis. Based on the local coordinate system of the wall, the three-dimensional coordinates of each point cloud in the three-dimensional point cloud model of the building wall are converted into two-dimensional coordinates (X,Y) in the local coordinate system of the wall. The Z coordinate is retained as the normal distance of the wall, realizing the expansion of the three-dimensional point cloud model into a three-dimensional planar point cloud model. S2.2. The wall debris data of the 3D planar point cloud model is removed by using the point cloud semantic segmentation model to obtain the wall 3D planar point cloud model after removal. S2.3. Based on the camera capture size of the wall-climbing robot, the discarded 3D planar point cloud model of the wall is cropped into N local 3D planar point cloud models; where: N is the number of local 3D planar point cloud models, and the specific formula is as follows: N = ceil(W / A) × ceil(H / B); Where: W is the total horizontal length of the wall, H is the total vertical height of the wall, ceil() is the round-up function, A is the length of the camera's field of view of the wall-climbing robot, and B is the length of the camera's field of view of the wall-climbing robot.

[0010] Optionally, S2.2 includes: S2.2.1 Construct a point cloud semantic segmentation model based on PointNet++. Input the 3D planar point cloud model into the point cloud semantic segmentation model based on PointNet++. Extract the local and global features of the point cloud through the MLP layer and max pooling layer of the PointNet++ network, and divide the point cloud in the 3D planar point cloud model into wall point cloud and debris point cloud. S2.2.2 Calculate the normal vector gradient of each point cloud in the three-dimensional planar point cloud model, and remove the clutter point clouds whose normal vector gradient exceeds the set gradient threshold to obtain the wall three-dimensional planar point cloud model after removal.

[0011] Optionally, S3 includes: S3.1 An improved ant colony algorithm is used to plan the scanning route of the wall-climbing robot, specifically including: S3.1.1 The ant colony algorithm is improved as follows to obtain the improved ant colony algorithm: The optimized heuristic function is obtained by using the Euclidean distance between the center points of the local wall 3D planar point cloud model as the core parameter of the heuristic function. The specific formula is as follows: η ij =1 / d ij ; Where: η ij d is the heuristic value for the ant moving from the i-th center point to the j-th center point; ij Let be the Euclidean distance between two center points i and j; i = 1, 2, 3, ..., N; j = 1, 2, 3, ..., N; i ≠ j; An adaptive pheromone evaporation coefficient is adopted, and the total number of iterations T of the improved ant colony algorithm is used as the dividing criterion. The iteration process is divided into an initial stage and a later stage, and evaporation coefficients are set for the initial stage and the later stage respectively. Specifically, the iteration number range of the initial stage is set to [1, 0.4T] to [1, 0.5T], and the pheromone evaporation coefficient τ1 in the initial stage is set to a value range of 0.08 to 0.12; the iteration number range of the later stage is set to (0.4T, T] to (0.5T, T], and the pheromone evaporation coefficient τ2 in the later stage is set to a value range of 0.28 to 0.32. A global optimal solution reward mechanism is introduced, which increases the pheromone concentration for ants that find the globally optimal route; the specific rules are as follows: ① Determining the Global Optimal Route: After each iteration of the improved ant colony algorithm, calculate the total length of the routes taken by all ants in that round. The route with the shortest total length, no repeated visits to nodes, and no obstacles is determined as the optimal route for the current round. Compare the current optimal routes of all iteration rounds and determine the route with the shortest total length as the global optimal route. The ant that finds the global optimal route is recorded as the optimal ant. ② Pheromone Concentration Increase Rule: After updating the pheromone evaporation of all paths according to the adaptive pheromone evaporation coefficient, an additional pheromone concentration is added to the path between all adjacent center points traversed by the optimal ant from the starting node to the ending node. The increase in pheromone concentration Δτ is calculated using the following formula: Δτ = α × τ; Where: α is the reward coefficient, with a value ranging from 1.5 to 2.5, and τ is the average pheromone concentration of all paths taken by ants in the current iteration; The blank areas left by the removed clutter point cloud are regarded as obstacles. When calculating the path between the center points, a collision detection algorithm is used to determine whether the path crosses the obstacle. If it does, the path is adjusted to avoid the obstacle. S3.1.2. Using the center point of all local wall surface 3D planar point cloud models as the nodes to be visited by the ant colony algorithm, the center point of the local model of the leftmost lowest point is set as the starting node S, and the improved ant colony algorithm is started to select the globally optimal route as the scanning route of the wall climbing robot. S3.2 Drive the wall-climbing robot to move along the planned scanning route. When it moves to the center point of any local wall surface 3D planar point cloud model, take a high-definition local image of the wall surface and store it in association with the corresponding center point coordinates.

[0012] Optionally, the wall-climbing robot includes a main frame, a vacuum adsorption module, a tracked mobile module, a servo drive module, a positioning module, and a high-definition camera module. The vacuum adsorption module includes a vacuum pump and multiple vacuum suction cups mounted on the main frame. Each vacuum suction cup is connected to the vacuum pump, which can extract air from the vacuum suction cups to create negative pressure, allowing the main frame to adhere to the wall surface via the vacuum suction cups. The tracked mobile module is mounted on the main frame and can move along the wall surface. The servo drive module is electrically connected to the tracked mobile module, can receive control commands corresponding to the scanning route, and drive the tracked mobile module to move according to the scanning route. The positioning module is mounted on the main frame and is used to obtain the current coordinates of the wall-climbing robot in real time. The high-definition camera module is located at the front end of the main frame, and the lens of the high-definition camera module is perpendicular to the wall surface, used to capture high-definition partial images of the wall surface.

[0013] Optionally, step S4.1 specifically involves: using the center point of the local wall surface three-dimensional planar point cloud model as the projection center and the XY plane of the local wall surface coordinate system as the projection plane, projecting the three-dimensional coordinates (X,Y,Z) of all points in the local wall surface three-dimensional planar point cloud model onto the XY plane to obtain a blurred local two-dimensional wall surface image, and obtaining the corresponding two-dimensional coordinates (X',Y'). S4.2 includes: Image registration specifically involves: extracting SIFT feature points from a high-resolution partial image of a wall and a blurred two-dimensional wall image; matching the SIFT feature points using the FLANN matching algorithm, eliminating incorrect matching points, and obtaining accurate feature point correspondences; and calculating the homography matrix between the high-resolution partial image of the wall and the blurred two-dimensional wall image based on the feature point correspondences. Image fusion, specifically, involves using the Poisson fusion algorithm to fuse the texture information of a high-definition local wall image into the two-dimensional wall image to be blurred, while retaining the coordinate information in the blurred two-dimensional wall image, resulting in a restored local wall image that has both clear texture and accurate coordinate information.

[0014] Optionally, S5 includes: S5.1 Local image crack recognition, specifically: input the high-definition local image of the wall taken by the wall-climbing robot into the wall crack semantic segmentation model, output the binary image of the crack, and extract the geometric parameters of the crack; The construction process of the semantic segmentation model for wall cracks is as follows: ① The improved U-net3+ model is obtained by making the following improvements to the traditional U-net3+ model: A ResNet50 residual network is introduced as a backbone in the encoder part of the traditional U-net3+ model; In the skip connection part of the encoder and decoder of the traditional U-net3+ model, a CBAM channel attention module is added to enhance the weight of the crack feature channel by weighting the feature maps of different channels. The weighted sum of the Dice loss function and the cross-entropy loss function is used as the loss function for the improved U-net3+ model; ② Construct a wall crack dataset containing several wall images labeled with cracks, and perform data augmentation on the wall crack dataset to obtain the augmented dataset; ③ Divide the enhanced dataset into a training set and a validation set, input the improved U-net3+ model and iteratively train it according to the set training parameters until the accuracy of the validation set reaches a stable level and no longer improves, then stop training to obtain the trained semantic segmentation model of wall cracks. S5.2 Local 3D mapping of crack information, specifically: based on the coordinate correspondence of the repaired local wall image and the homography matrix, mapping the crack pixel coordinates in the binary crack image to the corresponding local wall 3D planar point cloud model; specifically including: The homography matrix is ​​used to convert the two-dimensional pixel coordinates (u,v) of each crack pixel in the binary image of the crack output by the semantic segmentation model of the wall crack into the two-dimensional coordinates (X,Y) of the crack in the local wall image after repair. Based on the preset coordinate correspondence between the repaired local wall image and the local wall 3D planar point cloud model, the converted crack 2D coordinates (X,Y) are mapped to the corresponding local wall 3D planar point cloud model. S5.3 Global 3D Mapping: Specifically, the crack information in each local wall 3D planar point cloud model is integrated into the apparent 3D point cloud model of the house wall. Based on the coordinate position of each local model in the global 3D point cloud model, the crack information is finally mapped to the apparent 3D point cloud model of the house wall, resulting in an apparent 3D point cloud model of the house wall with accurate crack information.

[0015] In addition, the present invention also provides a readable storage medium storing computer program instructions, which, when executed by a processor, implement the method for accurately locating and detecting apparent cracks in a building as described above.

[0016] The present invention also provides an electronic device, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, wherein the computer program instructions are executed by the processor to perform the method for precise location and detection of apparent cracks in a building as described above.

[0017] This invention automates the entire process of identifying and mapping building cracks, from data acquisition, route planning, and image capture to crack recognition and 3D localization, through 3D laser scanning, automatic shooting by a wall-climbing robot, and automatic model recognition and mapping. This eliminates the need for manual intervention, effectively reducing labor intensity and avoiding the safety risks of working at heights. Simultaneously, through precise registration and fusion of 3D point clouds and 2D images, it achieves accurate mapping of crack information from 2D images to local 3D point clouds and then to global 3D point clouds. This allows for a direct and accurate presentation of the 3D location and distribution characteristics of cracks on building walls, providing precise data support for assessing the health of building concrete. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the arrangement of the three-dimensional laser scanner in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the three-dimensional point cloud model of the building wall surface in an embodiment of the present invention; Figure 3 This is a schematic diagram of the point cloud segmentation result of the point cloud semantic segmentation model based on PointNet++ in an embodiment of the present invention; Figure 4 This is a schematic diagram of the wall-climbing robot in an embodiment of the present invention; Figure 5 This is a high-definition partial image of the wall taken by the wall-climbing robot in an embodiment of the present invention; Figure 6 for Figure 5 The corresponding binary image of the crack.

[0020] Explanation of icon numbers: 1 House to be tested, 2 Wall-climbing robot, 2.1 Main frame, 2.2 Camera module, 2.3 Vacuum adsorption module, 2.4 Tracked movement module, 2.5 Positioning module, 3 3D laser scanner.

[0021] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0024] Furthermore, in this invention, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0025] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0026] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0027] This invention proposes a method, medium, and equipment for precise location and detection of apparent cracks in buildings, aiming to solve the problems that existing methods cannot achieve precise three-dimensional location of cracks and cannot correlate crack information with the overall structure of the building.

[0028] This embodiment provides a method for accurately locating and detecting apparent cracks in a building, including the following steps: S1: Set up a 3D laser scanner at preset points around the house to be tested, and use the 3D laser scanner to scan the house to be tested from all directions to obtain a 3D point cloud model of the house wall surface. S1 includes: S1.1 Confirm the location of the preset points around the house. At least one 3D laser scanner shall be set up at each preset point. When setting up the 3D laser scanner, the level of the 3D laser scanner shall be calibrated by a level and the absolute coordinates of the scanner shall be obtained by a GPS positioning module. The setting height and angle of the 3D laser scanner shall be adjusted according to the height of the house and the wall area. In this embodiment, see Figure 1 The 3D laser scanner employs a pulsed laser ranging principle, boasting a scanning accuracy of ≤0.1mm and a ranging range of ≥100m. Its scanning field of view is no less than 120°×90°, ensuring complete coverage of all walls to be inspected within the building. The number of preset points in this embodiment is determined based on actual needs to guarantee that the walls on all sides of the building can be scanned. During setup, the scanner's horizontal position is calibrated using a level, and its absolute coordinates are obtained via a GPS positioning module. The scanner's height and angle are adjusted according to the building's height and the area of ​​the walls to avoid obstructions (such as trees or billboards) between the scanner and the walls. S1.2. The 3D laser scanner is activated to perform a full-range scan of the building's walls to obtain the original point cloud model. Specifically, the 3D laser scanner emits laser pulses to the wall surface, receives the reflected pulses, calculates the laser propagation time, and combines this with the scanner's spatial attitude parameters to obtain the 3D coordinate information of a massive number of points on the wall surface. The collection of all 3D coordinate points forms the original 3D point cloud model of the building's wall appearance (see [link]). Figure 2 ).

[0029] S1.3. A statistical filtering algorithm is used to remove isolated noise points caused by environmental interference in the original point cloud model to obtain a three-dimensional point cloud model of the building wall surface. In this embodiment, in order to improve the quality of the point cloud model, the original point cloud model is preprocessed for denoising, and a statistical filtering algorithm is used to remove isolated noise points caused by environmental interference (such as dust and light reflection), thereby retaining the effective wall surface point cloud data.

[0030] S2: Unfold the 3D point cloud model of the house wall into a 3D planar model of the wall, and trim it into N local 3D planar point cloud models of the wall; S2 includes: S2.1. The three-dimensional point cloud model of the building wall surface is unfolded into a three-dimensional planar point cloud model using a planar fitting algorithm. Specifically: Establish a local coordinate system for the wall surface, with one vertex of the wall surface as the origin, the horizontal direction of the wall surface as the X-axis, the vertical direction as the Y-axis, and the direction perpendicular to the wall surface as the Z-axis. Based on the local coordinate system of the wall, the three-dimensional coordinates of each point cloud in the apparent three-dimensional point cloud model of the house wall are converted into two-dimensional coordinates (X,Y) under the local coordinate system of the wall. The Z coordinate is retained as the normal distance of the wall, realizing the expansion of the three-dimensional point cloud model into a three-dimensional planar point cloud model. In this embodiment, the expanded three-dimensional planar point cloud model can intuitively reflect the planar distribution characteristics of the wall.

[0031] S2.2. The wall debris data of the three-dimensional planar point cloud model is obtained by removing the wall debris data from the point cloud semantic segmentation model; in this embodiment, the wall debris data includes air conditioner outdoor units, window frames, pipelines, decorative lines, etc.

[0032] S2.2 includes: S2.2.1 Construct a point cloud semantic segmentation model based on PointNet++. Input a 3D planar point cloud model into the PointNet++-based point cloud semantic segmentation model. Extract local and global features of the point cloud through the MLP layer and max pooling layer of the PointNet++ network, and divide the point cloud in the 3D planar point cloud model into wall point cloud and debris point cloud; see [link to documentation]. Figure 4 Blue dot clouds represent walls, red dot clouds represent windows, and green dot clouds represent air conditioners.

[0033] This embodiment uses a 3D laser scanner to acquire over 500 wall point cloud models. Then, using CloudCompare software, these models are converted into labeled wall point cloud models containing both wall point clouds and debris point clouds. These 500 labeled point cloud data are then used to train a PointNet++ model, resulting in a PointNet++-based point cloud semantic segmentation model. The core of this model is the extraction of local and global features from the point cloud using the MLP and max-pooling layers of the PointNet++ network, thereby enabling the segmentation of the 3D wall point cloud model into wall point clouds and debris point clouds.

[0034] S2.2.2 Calculate the normal gradient of each point cloud in the 3D planar point cloud model, and remove clutter point clouds whose normal vector gradients exceed a set gradient threshold to obtain the clean 3D planar point cloud model of the wall. Since the normal vector gradient in cluttered areas (such as the edges of air conditioner outdoor units and window frames) can change abruptly, a gradient threshold is set. Points with abrupt gradient changes in the semantically segmented clutter point cloud region are verified, and clutter point clouds with abrupt gradient changes are removed. At the same time, wall point clouds with gentle gradients are retained to avoid the loss of effective wall point clouds due to semantic segmentation misjudgments, ultimately obtaining a clean 3D planar point cloud model of the wall.

[0035] S2.3. Based on the camera capture size of the wall-climbing robot, the discarded 3D planar point cloud model of the wall is cropped into N local 3D planar point cloud models; where: N is the number of local 3D planar point cloud models, and the specific formula is as follows: N = ceil(W / A) × ceil(H / B); Where: W is the total horizontal length of the wall, H is the total vertical height of the wall, ceil() is the round-up function, A is the length of the camera's field of view of the wall-climbing robot, and B is the length of the camera's field of view of the wall-climbing robot.

[0036] S2.3 Specifically: First, determine the camera's shooting size for the wall-climbing robot. The high-definition camera equipped on the wall-climbing robot is an industrial-grade CCD camera with a resolution of no less than 3000×4000 pixels and a lens focal length of 25mm. Its shooting field of view corresponds to the actual shooting range of the wall surface as A×B (e.g., 0.5m×0.5m). This shooting size is the cropping size of the local wall surface. Based on the cropping size, on the clean 3D planar point cloud model of the wall surface, a mesh is divided according to the principle of "uniform distribution of rows and columns." Calculate the number N of the cropped local wall surface 3D planar point cloud models. Each local wall surface 3D planar point cloud model corresponds to a cell in the mesh. Extract the point cloud data within each cell to form N local wall surface 3D planar point cloud models of the shooting size, and record the boundary coordinates of each local wall surface 3D planar point cloud model.

[0037] See Figure 4 The wall-climbing robot includes a main frame, a vacuum adsorption module, a tracked mobile module, a servo drive module, a positioning module, and a high-definition camera module. The vacuum adsorption module includes a vacuum pump and multiple vacuum suction cups mounted on the main frame. Each vacuum suction cup is connected to the vacuum pump, which can extract air from the vacuum suction cups to create negative pressure, allowing the main frame to adhere to the wall surface via the vacuum suction cups. The tracked mobile module is mounted on the main frame and can move along the wall surface. The servo drive module is electrically connected to the tracked mobile module, can receive control commands corresponding to the scanning route, and drive the tracked mobile module to move according to the scanning route. The positioning module is mounted on the main frame and is used to obtain the current coordinates of the wall-climbing robot in real time. The high-definition camera module is located at the front end of the main frame, and the lens of the high-definition camera module is perpendicular to the wall surface, used to capture high-definition partial images of the wall surface.

[0038] In this embodiment, the main frame is made of lightweight aluminum alloy, ensuring the robot is lightweight yet possesses sufficient structural strength. The vacuum adsorption module uses a vacuum pump to extract air from the suction cups, creating negative pressure to firmly adhere the robot to the wall surface, preventing it from falling. The tracked mobile module uses rubber tracks with anti-slip textures to improve the robot's stability on the wall. The servo drive module receives route planning instructions and drives the tracked mobile module to move, achieving precise robot movement. The positioning module uses IMU inertial navigation combined with local coordinate system positioning on the wall to obtain the robot's current coordinates in real time, with a positioning accuracy of ≤1mm. The high-definition camera module is fixed to the front end of the robot's main frame, with the lens perpendicular to the wall to ensure the shooting direction is perpendicular to the wall.

[0039] S3: Drive the wall-climbing robot to capture high-definition local images of the wall surface corresponding to the 3D planar point cloud model of each local wall surface; S3 includes: S3.1 An improved ant colony algorithm is used to plan the scanning route of the wall-climbing robot, specifically including: S3.1.1 The ant colony algorithm is improved as follows to obtain the improved ant colony algorithm: The optimized heuristic function is obtained by using the Euclidean distance between the center points of the local wall 3D planar point cloud model as the core parameter of the heuristic function. The specific formula is as follows: η ij =1 / d ij ; Where: η ij d is the heuristic value for the ant moving from the i-th center point to the j-th center point; ij Let be the Euclidean distance between two center points i and j; i = 1, 2, 3, ..., N; j = 1, 2, 3, ..., N; i ≠ j; An adaptive pheromone evaporation coefficient is adopted, and the iteration process is divided into an initial stage and a later stage based on the total number of iterations T of the improved ant colony algorithm. Evaporation coefficients are set for the initial and later stages respectively. Specifically: the iteration range for the initial stage is [1, 0.4T] to [1, 0.5T], and the pheromone evaporation coefficient τ1 is set to a value range of 0.08 to 0.12; the iteration range for the later stage is set to (0.4T, T] to (0.5T, T], and the pheromone evaporation coefficient τ2 is set to a value range of 0.28 to 0.32. In this embodiment, T=100, τ1=0.1, and τ2=0.3 are preferred. The initial stage uses a smaller pheromone evaporation coefficient to allow pheromones to accumulate quickly on the path through a lower evaporation rate, accelerating the algorithm's convergence speed. The later stage uses a larger pheromone evaporation coefficient to avoid excessive pheromone accumulation that could lead the algorithm into a local optimum, ensuring the algorithm's global search capability.

[0040] A global optimal solution reward mechanism is introduced, which increases the pheromone concentration for ants that find the globally optimal route; the specific rules are as follows: ① Determining the Global Optimal Route: After each iteration of the improved ant colony algorithm, calculate the total length of the routes taken by all ants in that round. The route with the shortest total length, no repeated visits to nodes, and no obstacles is determined as the optimal route for the current round. Compare the current optimal routes of all iteration rounds and determine the route with the shortest total length as the global optimal route. The ant that finds the global optimal route is recorded as the optimal ant. ② Pheromone Concentration Increase Rule: After updating the pheromone evaporation of all paths according to the adaptive pheromone evaporation coefficient, an additional pheromone concentration is added to the path between all adjacent center points traversed by the optimal ant from the starting node to the ending node. The increase in pheromone concentration Δτ is calculated using the following formula: Δτ = α × τ; Where: α is the reward coefficient, and the value of α ranges from 1.5 to 2.5; τ is the average pheromone concentration of all paths taken by all ants in the current iteration round; in this embodiment, after each iteration, the pheromone evaporation of all paths is updated according to the adaptive pheromone evaporation coefficient, and then the above-mentioned pheromone concentration increase operation is performed on the path of the best ant to ensure that the reward mechanism does not affect the adaptive adjustment of the evaporation coefficient.

[0041] The blank areas left by the removed clutter point cloud are regarded as obstacles. When calculating the path between the center points, a collision detection algorithm is used to determine whether the path crosses the obstacle. If it does, the path is adjusted to avoid the obstacle. S3.1.2. Using the center point of all local wall surface 3D planar point cloud models as the nodes to be visited by the ant colony algorithm, the center point of the local model of the leftmost lowest point is set as the starting node S, and the improved ant colony algorithm is started to select the globally optimal route as the scanning route of the wall-climbing robot. In this embodiment, before starting the improved ant colony algorithm, it is also necessary to initialize parameters such as the number of ants (e.g., 50 ants), the initial concentration of pheromones, and the maximum number of iterations (e.g., 100 times).

[0042] S3.2 Drive the wall-climbing robot to move along the planned scanning route. When it reaches the center point of any local wall surface 3D planar point cloud model, capture a high-resolution image of the local wall surface and store it in association with the corresponding center point coordinates. Specifically, S3.2 converts the planned scanning route into motion control commands for the robot and sends them to the servo drive module of the wall-climbing robot. The robot moves along the planned route on the wall surface, and the positioning module provides real-time feedback on the robot's current coordinates. When the robot reaches the center point of a local wall surface 3D planar point cloud model, the positioning module sends a trigger signal to control the high-resolution camera to capture a high-resolution image of the local wall surface. During shooting, the camera's exposure parameters (such as shutter speed and ISO) are automatically adjusted according to the wall surface light intensity to ensure a clear image with complete details. After shooting, the image data is transmitted in real-time to the backend processing system via a wireless transmission module and stored in association with the corresponding local model center point coordinates.

[0043] S4: Using high-resolution local wall images, the blurred local 2D wall image corresponding to the 3D planar point cloud model of the local wall is repaired to obtain the repaired local wall image; specifically including: S4.1. The perspective projection algorithm is used to project the three-dimensional planar point cloud model of each local wall surface onto a two-dimensional plane to obtain a blurred local two-dimensional wall surface image; Specifically, S4.1 involves using the center point of the local wall surface 3D planar point cloud model as the projection center and the XY plane of the local wall surface coordinate system as the projection plane. The 3D coordinates (X,Y,Z) of all points in the local wall surface 3D planar point cloud model are projected onto the XY plane to obtain a blurred local 2D wall surface image and the corresponding 2D coordinates (X',Y'). Since the texture information of the wall surface is lost during the projection process, the generated local 2D wall surface image is a blurred image. However, this blurred image is accompanied by the coordinate information of each pixel in the local wall surface coordinate system, thus realizing the preliminary coordinate association between the 2D image and the 3D point cloud.

[0044] S4.2 includes: Image registration specifically involves: extracting SIFT feature points from a high-resolution partial image of a wall and a blurred two-dimensional wall image; matching the SIFT feature points using the FLANN matching algorithm, eliminating incorrect matching points, and obtaining accurate feature point correspondences; and calculating the homography matrix between the high-resolution partial image of the wall and the blurred two-dimensional wall image based on the feature point correspondences. Image fusion, specifically, involves using the Poisson fusion algorithm to fuse the texture information of a high-definition local wall image into the two-dimensional wall image to be blurred, while retaining the coordinate information in the blurred two-dimensional wall image, resulting in a restored local wall image that has both clear texture and accurate coordinate information.

[0045] S4.2. Using image registration and fusion algorithms, a blurred local two-dimensional wall image is repaired using a high-definition local wall image to obtain a repaired local wall image. S5: Use the semantic segmentation model of wall cracks to identify crack images of each high-definition local wall image and extract the corresponding geometric parameters. Based on the repaired local wall image, map the crack information of the crack image to the local wall 3D planar point cloud model and the house wall appearance 3D point cloud model in sequence to obtain the house wall appearance 3D point cloud model with accurate crack information, where: crack information includes location, shape and geometric parameters.

[0046] S5 includes: S5.1 Local image crack recognition, specifically: inputting a high-resolution local image of the wall surface captured by the wall-climbing robot into the wall crack semantic segmentation model, and outputting a binary image of the crack (see...). Figure 6 (White pixels represent cracks, and black pixels represent the background), and extract the geometric parameters of the cracks (such as length, width, and area). The construction process of the semantic segmentation model for wall cracks is as follows: ① The improved U-net3+ model is obtained by making the following improvements to the traditional U-net3+ model: The ResNet50 residual network is introduced as a backbone in the encoder part of the traditional U-net3+ model; thereby solving the gradient vanishing problem in deep network training through residual connections and improving the model's feature extraction ability, especially for the extraction of features from small cracks.

[0047] In the skip connection part of the encoder and decoder of the traditional U-net3+ model, a CBAM channel attention module is added. By assigning weights to the feature maps of different channels, the weights of the crack feature channels are enhanced, while the weights of the background noise channels are suppressed, thereby improving the model's crack recognition accuracy.

[0048] The weighted sum of the Dice loss function and the cross-entropy loss function is used as the loss function to improve the U-net3+ model; this solves the class imbalance problem caused by the small proportion of crack pixels in wall crack images and improves the training effect of the model.

[0049] ② Construct several wall images with crack labels (see...) Figure 5This embodiment uses a wall crack dataset and performs data augmentation on it to obtain an augmented dataset. The dataset contains 1000 wall images labeled with cracks, covering different crack widths (0.1mm~5mm), crack shapes (straight, zigzag, mesh), lighting conditions (sunny, cloudy, indoor), and wall materials (concrete, brick). Data augmentation (such as rotation, flipping, scaling, and adding noise) is performed on the dataset to improve the model's generalization ability.

[0050] ③ Divide the enhanced dataset into training and validation sets, input the improved U-net3+ model, and iteratively train it according to the set training parameters until the validation set accuracy reaches a stable level and no longer improves. Then stop training to obtain the trained wall crack semantic segmentation model. In this embodiment, the enhanced dataset is divided into training and validation sets in an 8:2 ratio, and the improved U-net3+ model is input for training. The training parameters are set as follows: batch size of 8, learning rate of 0.001, optimizer of Adam, and 100 training epochs. During training, the model's recognition accuracy is monitored in real time using the validation set. When the validation set accuracy reaches a stable level and no longer improves, training is stopped, and the trained wall crack semantic segmentation model is obtained.

[0051] S5.2 Local 3D Mapping of Crack Information: Specifically, based on the coordinate correspondence of the repaired local wall image and the homography matrix, the crack pixel coordinates in the binary crack image are mapped to the corresponding local wall 3D planar point cloud model; this achieves the association between crack information and the local 3D point cloud. This includes: The homography matrix is ​​used to convert the two-dimensional pixel coordinates (u,v) of each crack pixel in the binary image of the crack output by the semantic segmentation model of the wall crack into the two-dimensional coordinates (X,Y) of the crack in the repaired local wall image. The obtained two-dimensional coordinates of the crack are completely consistent with the two-dimensional coordinates in the local coordinate system of the wall, and correspond to the XY plane coordinates of the three-dimensional planar point cloud model of the local wall.

[0052] Based on the preset coordinate correspondence between the repaired local wall image and the local wall 3D planar point cloud model, the converted crack 2D coordinates (X,Y) are mapped to the corresponding local wall 3D planar point cloud model; this achieves precise association between crack information and local 3D point cloud, ensuring the consistency between crack location and local wall 3D structure.

[0053] S5.3 Global 3D Mapping: Specifically, the crack information in each local wall 3D planar point cloud model is integrated into the apparent 3D point cloud model of the house wall. Based on the coordinate position of each local model in the global 3D point cloud model, the crack information is finally mapped to the apparent 3D point cloud model of the house wall, resulting in an apparent 3D point cloud model of the house wall with accurate crack information.

[0054] This embodiment also includes a readable storage medium storing computer program instructions, which, when executed by a processor, implement the method for accurately locating and detecting apparent cracks in a building as described above.

[0055] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0056] This embodiment also includes an electronic device, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, wherein the computer program instructions are executed by the processor to perform the method for precise location and detection of apparent cracks in a building as described above.

[0057] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0058] The electronic device can be a mobile phone, desktop computer, laptop, handheld computer, cloud server, or other computing device. The electronic device may include, but is not limited to, processors and memory. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0059] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0060] The memory can be used to store the computer program and / or modules. The processor implements the computer program by running or executing the computer program and / or modules stored in the memory, and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0061] If the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0062] The above description is only a preferred embodiment of the present invention and does not limit the scope of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the protection scope of the present invention.

Claims

1. A method for accurately locating and detecting apparent cracks in a building, characterized in that, Includes the following steps: S1: Set up a 3D laser scanner at preset points around the house to be tested, and use the 3D laser scanner to scan the house to be tested from all directions to obtain a 3D point cloud model of the house wall surface. S2: Unfold the 3D point cloud model of the house wall into a 3D planar model of the wall, and trim it into N local 3D planar point cloud models of the wall; S3: Drive the wall-climbing robot to capture high-definition local images of the wall surface corresponding to the 3D planar point cloud model of each local wall surface; S4: Using high-resolution local wall images, the blurred local 2D wall image corresponding to the 3D planar point cloud model of the local wall is repaired to obtain the repaired local wall image; specifically including: S4.

1. The perspective projection algorithm is used to project the three-dimensional planar point cloud model of each local wall surface onto a two-dimensional plane to obtain a blurred local two-dimensional wall surface image; S4.

2. Using image registration and fusion algorithms, a blurred local two-dimensional wall image is repaired using a high-definition local wall image to obtain a repaired local wall image. S5: Use the semantic segmentation model of wall cracks to identify crack images of each high-definition local wall image and extract the corresponding geometric parameters. Based on the repaired local wall image, map the crack information of the crack image to the local wall 3D planar point cloud model and the house wall appearance 3D point cloud model in sequence to obtain the house wall appearance 3D point cloud model with accurate crack information, where: crack information includes location, shape and geometric parameters.

2. The method for precise location and detection of apparent cracks in buildings according to claim 1, characterized in that, S1 includes: S1.1 Confirm the location of the preset points around the house. At least one 3D laser scanner shall be set up at each preset point. When setting up the 3D laser scanner, the level of the 3D laser scanner shall be calibrated by a level and the absolute coordinates of the scanner shall be obtained by a GPS positioning module. The setting height and angle of the 3D laser scanner shall be adjusted according to the height of the house and the wall area. S1.

2. Start the 3D laser scanner to scan the walls of the house from all angles to obtain the original point cloud model; S1.

3. The statistical filtering algorithm is used to remove isolated noise points caused by environmental interference in the original point cloud model to obtain the three-dimensional point cloud model of the building wall surface.

3. The method for precise location and detection of apparent cracks in buildings according to claim 2, characterized in that, S2 includes: S2.

1. The three-dimensional point cloud model of the building wall surface is unfolded into a three-dimensional planar point cloud model using a planar fitting algorithm. Specifically: Establish a local coordinate system for the wall surface, with one vertex of the wall surface as the origin, the horizontal direction of the wall surface as the X-axis, the vertical direction as the Y-axis, and the direction perpendicular to the wall surface as the Z-axis. Based on the local coordinate system of the wall, the three-dimensional coordinates of each point cloud in the three-dimensional point cloud model of the building wall are converted into two-dimensional coordinates (X,Y) in the local coordinate system of the wall. The Z coordinate is retained as the normal distance of the wall, realizing the expansion of the three-dimensional point cloud model into a three-dimensional planar point cloud model. S2.

2. The wall debris data of the 3D planar point cloud model is removed by using the point cloud semantic segmentation model to obtain the wall 3D planar point cloud model after removal. S2.

3. Based on the camera capture size of the wall-climbing robot, the discarded 3D planar point cloud model of the wall is cropped into N local 3D planar point cloud models; where: N is the number of local 3D planar point cloud models, and the specific formula is as follows: N = ceil(W / A) × ceil(H / B); Where: W is the total horizontal length of the wall, H is the total vertical height of the wall, ceil() is the round-up function, A is the length of the camera's field of view of the wall-climbing robot, and B is the length of the camera's field of view of the wall-climbing robot.

4. The method for precise location and detection of apparent cracks in buildings according to claim 3, characterized in that, S2.2 includes: S2.2.1 Construct a point cloud semantic segmentation model based on PointNet++. Input the 3D planar point cloud model into the point cloud semantic segmentation model based on PointNet++. Extract the local and global features of the point cloud through the MLP layer and max pooling layer of the PointNet++ network, and divide the point cloud in the 3D planar point cloud model into wall point cloud and debris point cloud. S2.2.2 Calculate the normal vector gradient of each point cloud in the three-dimensional planar point cloud model, and remove the clutter point clouds whose normal vector gradient exceeds the set gradient threshold to obtain the wall three-dimensional planar point cloud model after removal.

5. The method for precise location and detection of apparent cracks in buildings according to claim 4, characterized in that, S3 includes: S3.1 An improved ant colony algorithm is used to plan the scanning route of the wall-climbing robot, specifically including: S3.1.1 The ant colony algorithm is improved as follows to obtain the improved ant colony algorithm: The optimized heuristic function is obtained by using the Euclidean distance between the center points of the local wall 3D planar point cloud model as the core parameter of the heuristic function. The specific formula is as follows: or ij =1 / d ij ; Where: η ij d is the heuristic value for the ant moving from the i-th center point to the j-th center point; ij Let be the Euclidean distance between two center points i and j; i = 1, 2, 3, ..., N; j = 1, 2, 3, ..., N; i ≠ j; An adaptive pheromone evaporation coefficient is adopted, and the total number of iterations T of the improved ant colony algorithm is used as the dividing criterion. The iteration process is divided into an initial stage and a later stage, and evaporation coefficients are set for the initial stage and the later stage respectively. Specifically, the iteration number range of the initial stage is set to [1, 0.4T] to [1, 0.5T], and the pheromone evaporation coefficient τ1 in the initial stage is set to a value range of 0.08 to 0.12; the iteration number range of the later stage is set to (0.4T, T] to (0.5T, T], and the pheromone evaporation coefficient τ2 in the later stage is set to a value range of 0.28 to 0.

32. A global optimal solution reward mechanism is introduced, which increases the pheromone concentration for ants that find the globally optimal route; the specific rules are as follows: ① Determining the Global Optimal Route: After each iteration of the improved ant colony algorithm, calculate the total length of the routes taken by all ants in that round. The route with the shortest total length, no repeated visits to nodes, and no obstacles is determined as the optimal route for the current round. Compare the current optimal routes of all iteration rounds and determine the route with the shortest total length as the global optimal route. The ant that finds the global optimal route is recorded as the optimal ant. ② Pheromone Concentration Increase Rule: After updating the pheromone evaporation of all paths according to the adaptive pheromone evaporation coefficient, an additional pheromone concentration is added to the path between all adjacent center points traversed by the optimal ant from the starting node to the ending node. The increase in pheromone concentration Δτ is calculated using the following formula: Δτ = α × τ; Where: α is the reward coefficient, with a value ranging from 1.5 to 2.5, and τ is the average pheromone concentration of all paths taken by ants in the current iteration; The blank areas left by the removed clutter point cloud are regarded as obstacles. When calculating the path between the center points, a collision detection algorithm is used to determine whether the path crosses the obstacle. If it does, the path is adjusted to avoid the obstacle. S3.1.

2. Using the center point of all local wall surface 3D planar point cloud models as the nodes to be visited by the ant colony algorithm, the center point of the local model of the leftmost lowest point is set as the starting node S, and the improved ant colony algorithm is started to select the globally optimal route as the scanning route of the wall climbing robot. S3.2 Drive the wall-climbing robot to move along the planned scanning route. When it moves to the center point of any local wall surface 3D planar point cloud model, take a high-definition local image of the wall surface and store it in association with the corresponding center point coordinates.

6. The method for precise location and detection of apparent cracks in buildings according to claim 5, characterized in that, The wall-climbing robot includes a main frame, a vacuum adsorption module, a tracked mobile module, a servo drive module, a positioning module, and a high-definition camera module. The vacuum adsorption module includes a vacuum pump and multiple vacuum suction cups mounted on the main frame. Each vacuum suction cup is connected to the vacuum pump, which draws air from the suction cups to create negative pressure, allowing the main frame to adhere to the wall surface via the suction cups. The tracked mobile module is mounted on the main frame and can move along the wall surface. The servo drive module is electrically connected to the tracked mobile module, receives control commands corresponding to the scanning route, and drives the tracked mobile module to move according to the scanning route. The positioning module is mounted on the main frame and is used to obtain the current coordinates of the wall-climbing robot in real time. The high-definition camera module is located at the front end of the main frame, with its lens perpendicular to the wall surface, for capturing high-definition partial images of the wall surface.

7. The method for precise location and detection of apparent cracks in a building according to any one of claims 1-5, wherein S4.1 specifically comprises: taking the center point of the local wall surface three-dimensional planar point cloud model as the projection center, taking the XY plane of the local wall surface coordinate system as the projection plane, projecting the three-dimensional coordinates (X,Y,Z) of all points in the local wall surface three-dimensional planar point cloud model onto the XY plane to obtain a blurred local two-dimensional wall surface image, and obtaining the corresponding two-dimensional coordinates (X',Y'); S4.2 includes: Image registration specifically involves: extracting SIFT feature points from a high-resolution partial image of a wall and a blurred two-dimensional image of a wall; matching the SIFT feature points using the FLANN matching algorithm, eliminating incorrect matching points, and obtaining accurate feature point correspondences; Based on the correspondence of feature points, calculate the homography matrix between the high-definition local wall image and the blurred two-dimensional wall image; Image fusion, specifically, involves using the Poisson fusion algorithm to fuse the texture information of a high-definition local wall image into the two-dimensional wall image to be blurred, while retaining the coordinate information in the blurred two-dimensional wall image, resulting in a restored local wall image that has both clear texture and accurate coordinate information.

8. The method for precise location and detection of apparent cracks in buildings according to claim 7, wherein step S5 includes: S5.1 Local image crack recognition, specifically: input the high-definition local image of the wall taken by the wall-climbing robot into the wall crack semantic segmentation model, output the binary image of the crack, and extract the geometric parameters of the crack; The construction process of the semantic segmentation model for wall cracks is as follows: ① The improved U-net3+ model is obtained by making the following improvements to the traditional U-net3+ model: A ResNet50 residual network is introduced as a backbone in the encoder part of the traditional U-net3+ model; In the skip connection part of the encoder and decoder of the traditional U-net3+ model, a CBAM channel attention module is added to enhance the weight of the crack feature channel by weighting the feature maps of different channels. The weighted sum of the Dice loss function and the cross-entropy loss function is used as the loss function for the improved U-net3+ model; ② Construct a wall crack dataset containing several wall images labeled with cracks, and perform data augmentation on the wall crack dataset to obtain the augmented dataset; ③ Divide the enhanced dataset into a training set and a validation set, input the improved U-net3+ model and iteratively train it according to the set training parameters until the accuracy of the validation set reaches a stable level and no longer improves, then stop training to obtain the trained semantic segmentation model of wall cracks. S5.2 Local 3D mapping of crack information, specifically: based on the coordinate correspondence of the repaired local wall image and the homography matrix, mapping the crack pixel coordinates in the binary crack image to the corresponding local wall 3D planar point cloud model; specifically including: The homography matrix is ​​used to convert the two-dimensional pixel coordinates (u,v) of each crack pixel in the binary image of the crack output by the semantic segmentation model of the wall crack into the two-dimensional coordinates (X,Y) of the crack in the local wall image after repair. Based on the preset coordinate correspondence between the repaired local wall image and the local wall 3D planar point cloud model, the converted crack 2D coordinates (X,Y) are mapped to the corresponding local wall 3D planar point cloud model. S5.3 Global 3D Mapping: Specifically, the crack information in each local wall 3D planar point cloud model is integrated into the apparent 3D point cloud model of the house wall. Based on the coordinate position of each local model in the global 3D point cloud model, the crack information is finally mapped to the apparent 3D point cloud model of the house wall, resulting in an apparent 3D point cloud model of the house wall with accurate crack information.

9. A readable storage medium, characterized in that, It stores computer program instructions, which, when executed by a processor, implement the method for accurately locating and detecting apparent cracks in a house as described in any one of claims 1 to 8.

10. An electronic device, characterized in that, include: The method for accurately locating and detecting apparent cracks in a building as described in any one of claims 1 to 8 includes at least one processor, at least one memory, and computer program instructions stored in the memory, which are executed by the processor when the computer program instructions are executed.

Images