A method and robot for automatically identifying cracks in concrete

Abstract

The application discloses a kind of concrete crack automatic identification method and robot, belong to civil engineering structure health monitoring technical field.The present application includes obtaining the gray image of the surface of concrete to be detected, and the gray median and gray standard deviation of the gray image are calculated;According to the gray median and gray standard deviation, determine an adaptive threshold;With fixed interval, the pixel of gray image is sampled, and the sampling point with the gray value lower than the adaptive threshold is marked as candidate crack pixel point;With candidate crack pixel point as seed point, region growth is carried out;When the pixel number of any growth formed connected region exceeds the preset upper limit, the growth of the region is terminated;The length-width ratio of the minimum circumscribed rectangle of each connected region is calculated, and the connected region with the length-width ratio greater than the set shape threshold is determined as crack region, and the position information of the crack region is output.The present application can realize identification efficiency and accuracy on the embedded microcontroller with limited resources.

Description

Technical Field

[0001] This invention relates to the field of civil engineering structural health monitoring technology, specifically to an automatic identification method and robot for concrete cracks. Background Technology

[0002] In the field of structural health monitoring in civil engineering, crack detection on the surface of concrete structures is a crucial step in assessing their safety, durability, and remaining lifespan. Automatic crack identification technology for concrete structures is the core of the intelligent development of structural health monitoring in civil engineering. Currently, this technology mainly falls into two categories: one is an offline processing mode based on a backend server. Most current automated inspection systems, such as drones and basic wall-climbing robots, only collect image or video data and transmit massive amounts of raw data wirelessly to a remote server or ground station, where a powerful computer runs complex identification algorithms for processing. The drawbacks of this mode are poor real-time performance. Limited by wireless bandwidth and data volume, there is a significant delay from data acquisition to obtaining identification results, making real-time on-site analysis impossible. Furthermore, in complex environments with weak or no signals, such as under bridges or deep in tunnels, data transmission is unstable or even interrupted. Moreover, the massive computational load of such algorithms cannot be embedded in mobile platforms. Another approach involves researchers deploying simple, traditional image processing algorithms on airborne microcontrollers to adapt to mobile platforms. However, due to the complex texture of concrete surfaces, uneven lighting, and interference from stains and shadows, these traditional algorithms suffer from low recognition rates under varying on-site conditions. Therefore, this invention proposes a novel automatic concrete crack recognition method and robot. Summary of the Invention

[0003] The purpose of this invention is to overcome the problems in the prior art and provide an automatic method and robot for identifying concrete cracks.

[0004] This invention provides an automatic method for identifying concrete cracks, comprising: A grayscale image of the concrete surface to be inspected is acquired, and the median and standard deviation of the grayscale image are calculated. Based on the median and standard deviation, an adaptive threshold is determined, and the adaptive threshold decreases as the standard deviation increases. The grayscale image is sampled at fixed intervals, and sampling points with grayscale values ​​lower than the adaptive threshold are marked as candidate crack pixels. The candidate crack pixels are used as seed points for region growth. For pixels to be incorporated into the current region, the pixel grayscale value must be lower than the threshold and its gradient direction must be consistent with the gradient direction of the seed point. When the number of pixels in any connected region formed by growth exceeds a preset upper limit, the growth of that region is terminated. Calculate the aspect ratio of the minimum bounding rectangle of each connected region, and determine the connected regions with aspect ratios greater than a set shape threshold as crack regions, and output the location information of the crack regions.

[0005] Preferably, during the region growth process, for the current seed point pixel, pixels that meet the conditions are searched in its eight connected neighborhoods for growth. A pixel that meets the conditions is one whose gray value of the neighboring pixel is lower than a threshold and whose absolute value of the difference between its gradient direction and the gradient direction of the seed point is less than a preset angle threshold.

[0006] Preferably, the angle threshold is 30 degrees, and the preset upper limit of pixels that meet the conditions is 500 pixels.

[0007] Preferably, if the set shape threshold is 2, then connected regions with an aspect ratio greater than or equal to 2 are determined to be cracks; otherwise, they are discarded.

[0008] Preferably, when outputting the location information of the crack region, the contour of the crack region is first mapped to the original image coordinate system, and a result image containing the contour annotation is generated.

[0009] Preferably, the gradient direction is obtained by convolving the grayscale image using an edge detection operator.

[0010] The present invention also provides a concrete crack automatic identification robot employing the above-described concrete crack automatic identification method, the robot comprising: Mobile platform for moving on concrete surfaces; A visual sensing device, installed on the mobile platform, is used to acquire image information of the concrete surface; A microcontroller board is mounted on the mobile platform and electrically connected to the visual sensing device; The microcontroller board has an embedded processing unit configured to perform any of the above-described automatic concrete crack identification methods.

[0011] Preferably, the bottom of the mobile platform is equipped with a rotor thruster adsorption device to generate a thrust perpendicular to the concrete surface, so that the robot can be adsorbed onto the concrete surface.

[0012] Preferably, the rotor propeller adsorption device includes a pipe housing and a brushless motor-driven fan installed inside it, and the microcontroller board is electrically connected to the brushless motor for adjusting the speed of the fan.

[0013] Preferably, it also includes a wireless remote controller for receiving manual operation instructions; the microcontroller board is communicatively connected to the wireless remote controller and is configured to control the movement of the mobile platform and / or the adsorption force of the rotor propeller adsorption device according to the instructions from the wireless remote controller.

[0014] Compared with existing technologies, the beneficial effects of this invention are as follows: the adaptive threshold of this invention effectively overcomes the interference of uneven lighting and texture changes on concrete surfaces, ensuring the environmental robustness of the algorithm; the skip sampling and early termination mechanism directly reduces the initial computation by about 75% and avoids invalid traversal, greatly improving processing efficiency; the gradient direction constraint accurately suppresses the growth of non-crack noise from the edge geometry characteristics; the aspect ratio screening realizes the final discrimination of the narrow shape of cracks; the above-mentioned algorithm of this invention can achieve a leapfrog improvement in unit image processing time from about 120ms in traditional methods to about 35ms on resource-constrained embedded microcontrollers, while maintaining high recognition accuracy. Attached Figure Description

[0015] Figure 1 This is a flowchart of the algorithm in an embodiment of the present invention.

[0016] Figure 2 This is a plan view of the robot in an embodiment of the present invention.

[0017] Figure 3 This is a diagram of a crack skeleton with linear extension characteristics according to an embodiment of the present invention.

[0018] Figure 4 This is the binarized image that is finally output in the embodiment of the present invention.

[0019] Figure 5 This is a comparison chart of single-frame processing time in an embodiment of the present invention.

[0020] Explanation of reference numerals in the attached figures: 1. Mobile platform; 11. Wheels; 12. DC motor; 2. Rotor propulsion adsorption system; 21. Pipe housing; 22. Brushless electronic speed controller; 3. Vision sensing system; 4. Microcontroller board; 5. Wireless communication module. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms “first,” “second,” and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as “comprising” or “including” indicate that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. Terms such as “connected” or “linked” are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as “upper,” “lower,” “left,” and “right” are used only to indicate relative positional relationships; when the absolute position of the described objects changes, the relative positional relationship may also change accordingly.

[0023] The present invention provides an automatic concrete crack identification method, comprising: A grayscale image of the concrete surface to be inspected is acquired, and the median and standard deviation of the grayscale image are calculated. Based on the median and standard deviation, an adaptive threshold is determined, and the adaptive threshold decreases as the standard deviation increases. The grayscale image is sampled at fixed intervals, and sampling points with grayscale values ​​lower than the adaptive threshold are marked as candidate crack pixels. The candidate crack pixels are used as seed points for region growth. For pixels to be incorporated into the current region, the pixel grayscale value must be lower than the threshold and its gradient direction must be consistent with the gradient direction of the seed point. When the number of pixels in any connected region formed by growth exceeds a preset upper limit, the growth of that region is terminated. Calculate the aspect ratio of the minimum bounding rectangle of each connected region, and determine the connected regions with aspect ratios greater than a set shape threshold as crack regions, and output the location information of the crack regions.

[0024] The adaptive threshold of this invention effectively overcomes the interference of uneven lighting and texture changes on the concrete surface, ensuring the environmental robustness of the algorithm; the skip sampling and early termination mechanism directly reduces the initial computation by about 75% and avoids invalid traversal, significantly improving processing efficiency; gradient direction constraints accurately suppress the growth of non-crack noise from the edge geometry; aspect ratio filtering achieves the final discrimination of the narrow shape of cracks; such as Figure 5 As shown, the algorithm of the present invention can achieve a leapfrog improvement in image processing time on resource-constrained embedded microcontrollers, reducing it from approximately 120ms to approximately 35ms in traditional methods, while maintaining high recognition accuracy.

[0025] like Figure 1As shown, this algorithm abandons the time-consuming full-image pixel traversal and complex window statistics of traditional methods, and introduces adaptive thresholding, skip sampling, gradient consistency constraints, growth termination mechanism, and shape selection. The specific steps are as follows: (1) Image preprocessing and adaptive threshold calculation In this embodiment, a visual sensing device is used to acquire on-site images. The acquired RGB images are first converted into grayscale images. To overcome the effects of uneven illumination and surface color difference, this embodiment employs an adaptive threshold based on image statistical features, rather than a fixed offset. (Calculation) gray median and standard deviation Threshold It is dynamically determined by the following formula:

[0026] (1); Among them, the adaptive coefficient The range of values ​​is For images with complex background textures and low contrast, the grayscale standard deviation σ is usually large. In this case, the algorithm automatically calculates a lower threshold T using the formula to retain more possible details of dark cracks. Conversely, for images with uniform backgrounds, σ is small, and the algorithm will calculate a higher threshold T, thereby effectively suppressing false detections caused by noise in uniform backgrounds.

[0027] In a preferred embodiment of the present invention, the adaptive coefficient k is set to 1.0. The image is an 8-bit grayscale image with pixel values ​​ranging from 0 to 255. The grayscale median B and standard deviation σ are obtained by statistically calculating all pixels of the input image Ig.

[0028] To illustrate its adaptive mechanism, this embodiment also provides the following two typical scenario calculation examples: Scenario 1: Crack Identification in Complex Backgrounds Assuming the acquired concrete surface image has complex texture due to oil stains and shadows, the calculated median grayscale value is B1 = 110, and the standard deviation is σ1 = 35. The adaptive threshold is calculated using the formula T = B − kσ.

[0029] T1 = 110 − 1.0 × 35 = 75; the calculated threshold T1 = 75 is significantly lower than the medium brightness level of the image (110). This lower threshold means that more dark pixels with gray values ​​below 75 will be marked as candidate points. In this scenario, this helps ensure that even if the crack has low contrast due to the overall dark environment—for example, a crack with a gray value of 80 is obvious against a bright background but not against a dark background—its pixels can still be effectively captured because they are below the threshold, thus avoiding missing real cracks in low-contrast complex backgrounds.

[0030] Scenario 2: Noise Suppression Against a Uniform Background Assuming the collected concrete surface is smooth and clean with uniform lighting, the calculated median grayscale value B² = 180 and the standard deviation σ² = 8. Calculate the adaptive threshold:

[0031] T2 = 180 − 1.0 × 8 = 172; the calculated threshold T2 = 172 is very close to the bright background level (180) of the image. This higher threshold will rigorously filter candidate pixels. In this scene, the gray values ​​of most concrete body pixels are concentrated around 180, while the gray values ​​of dark pixels caused by sensor noise or minor surface defects may only be around 175. Since T2 = 172, these slightly darkened noise pixels (gray value > 172) will not be misclassified as candidate crack points, thus significantly suppressing false alarms (false alarms) that misidentify noise as cracks in a uniform background. This step replaces the step in the original method that required manual experience to set the constant C, significantly improving environmental adaptability.

[0032] (2) Skip sampling and rapid candidate point selection To significantly reduce the number of pixels that need to be processed, the algorithm does not traverse every single pixel of the image, but instead uses a skip sampling method. The sampling interval is set. That is, sampling every single pixel. Only these sampling points are sampled. Perform threshold judgment: If If the crack is cleared, the point is marked as a candidate crack pixel; otherwise, it is ignored. This step reduces the initial judgment operation by about 75%, and since cracks are usually continuous, skip sampling has little impact on its ability to capture the overall shape.

[0033] (3) Gradient-guided region growth and early termination In the region growing stage, the algorithm introduces gradient information as a constraint to improve growing accuracy and efficiency. First, the Sobel operator is used to compute the grayscale image. Gradient magnitude plot G and gradient direction plot Subsequently, using each candidate pixel obtained in (2) as a seed point, region growth is initiated. The growth rule is: for the 8 neighboring pixels of the current pixel, it is included in the current connected region only if both of the following conditions are met simultaneously:

[0034] Gray value condition: gray value of neighboring pixels ; Gradient direction consistency condition: the gradient direction of neighboring pixels gradient direction of seed pixel The absolute value of the difference ,recommend This condition leverages the characteristic that the pixel gradient direction at the crack edge is perpendicular to the crack's direction, effectively preventing growth from spreading to non-crack dark areas with chaotic orientations. Furthermore, to avoid meaningless traversal of large stains or shadows, for example, an upper limit on the number of pixels in the growth region can be set. Once the number of pixels in the current connected region exceeds this limit, the growth process for that region is immediately terminated and it is discarded entirely. This directly skips all subsequent calculations for that large region.

[0035] (4) Shape filtering based on aspect ratio integer operations Traditional seepage algorithms use circularity for shape filtering, requiring calculation of the region's area and perimeter, which involves floating-point operations and incurs significant overhead on microcontrollers. This invention proposes using the minimum aspect ratio of the bounding rectangle. As a lightweight alternative indicator. For each connected region that survives after (3). Calculate its minimum bounding rectangle, and obtain the rectangle's length L and width W. Calculate the aspect ratio:

[0036] (2); Cracks are typically elongated (with a larger radius R), while pores, bubbles, and other noise features are approximately circular (with a larger radius R). Set threshold. .like Then determine This is the crack area; if If the impurity is round, it is identified as a circular impurity and removed. The entire judgment process only requires integer addition, subtraction, and comparison operations, making it extremely suitable for embedded platforms.

[0037] (5) Results fusion and output All connected components retained after filtering (4) The contour is mapped back to the original image coordinates.

[0038] The microcontroller board can also send these contour point sets or bounding rectangle coordinates to the host computer via the wireless communication module 5, or directly highlight the crack area in the locally generated binarized image. At this point, the lightweight seepage identification process is complete.

[0039] Data transmission and feedback: Processing results and image data are transmitted to a remote control terminal via a wireless communication module, supporting real-time monitoring and subsequent analysis. The robot can adjust its movement path based on the recognition results to achieve accurate detection.

[0040] Detection effect This robot can move on concrete surfaces with sufficient friction. Its rotor fan has a maximum speed of approximately 30,000 rpm, generating 2900g of thrust. A brushless motor is connected to an electronic speed controller, which provides three-phase control signals to the brushless motor. The system is powered by a 12V, 44Ah lithium battery, providing approximately 30 minutes of continuous operation on a single charge. The robot has a four-wheel structure, with each wheel independently driven by a micro DC motor. The fan speed can be adjusted according to the load, and the speed of the hub DC motors directly determines the climbing speed. The robot weighs approximately 800g and has dimensions of 20cm × 15cm × 4cm. It is equipped with four DC motors. The body is made of transparent acrylic sheet, and the entire robot is controlled via a wireless analog remote control.

[0041] This robot uses a single rotor thruster to achieve its adhesion function. The principle is that the reaction force generated by the downward jet of air from a fan presses the robot against the wall. The system has an open structure, making it highly adaptable to varying wall smoothness and roughness. The entire robot consists of only core components such as a mobile platform, rotor thruster, camera, and microcontroller. The rotor thruster is located at the center of the platform's bottom, with a compact structure that helps maintain overall stability. The four-wheel independent drive mechanism enables the robot to turn in place and provides greater flexibility during forward, backward, and turning movements, making it particularly suitable for precise positioning tasks on vertical surfaces.

[0042] like Figure 3 and 4 As shown, the crack recognition algorithm proposed in this embodiment performs exceptionally well in suppressing false detections in interfering areas such as oil stains and shadows. This is achieved by introducing a maximum pixel count constraint during the region growing stage. The algorithm terminates prematurely when the number of scan points exceeds 500, effectively avoiding redundant calculations for large-area non-crack targets and significantly reducing the false detection rate. Simultaneously, the skip sampling strategy significantly reduces the number of initial scan points by approximately 70%, and in conjunction with the early termination mechanism, reduces the overall computation time by approximately 60%–80% compared to the traditional full-pixel traversal method. Comparative results show that although the algorithm introduces sparsity processing during the sampling phase, thanks to crack continuity constraints and subsequent skeleton extraction and screening steps, the main crack structures can be identified intact, without obvious breaks or omissions. Furthermore, performance evaluation indicates that the average processing time per unit image is reduced from 120ms in the traditional method to 35ms, significantly improving real-time performance and providing a more efficient technical approach for this automated detection robot.

[0043] Traditional methods often use circularity to distinguish noise, but this is extremely unstable when there are few pixels. This embodiment uses aspect ratio as a filtering criterion, completely eliminating interference from nearly circular features such as micropores and pinholes commonly found on concrete surfaces. Figure 3 As shown, only the crack skeleton with linear extension characteristics is retained.

[0044] like Figure 4 As shown, by setting an upper limit on the number of pixels in the growth region and combining it with an early termination mechanism, the algorithm will quickly stop growth and discard the region when it detects oil stains, water stains, or large areas of shadow. In the final output binarized image, there are no large non-crack dark areas, which significantly reduces the false detection rate.

[0045] The robot body structure in this embodiment is as follows: Figure 2 As shown, the mobile platform of the concrete crack detection robot is a frame made of lightweight materials such as carbon fiber or aluminum alloy. As an assembly platform for other components, it has nine pre-drilled grooves for fitting with other parts, with the central location for the rotor thruster. The four wheels use high-friction rubber tires, and each wheel is independently driven by a DC motor, enabling flexible motion control.

[0046] The rotor propulsion adsorption system is fixed at the center of the chassis bottom. Its duct housing effectively gathers airflow, improving fan efficiency. The internal brushless motor-driven fan's speed is controlled by a brushless electronic speed controller. The microcontroller board dynamically adjusts the fan speed based on the wall roughness and the robot's total weight to provide just the right adsorption force. The vision sensing system uses a high-definition wireless camera mounted at the front of the vehicle to capture images of the concrete surface. The captured video stream is transmitted to a remote computer via wireless communication module 5, while the transmitted image data is processed locally by the microcontroller board to identify and outline crack locations in real time. The main control system uses an embedded microprocessor such as an STM32 as its core, responsible for receiving remote control commands, controlling the motors and fans, and processing measurement data.

[0047] In this embodiment, the robot is started via a remote control terminal. The microcontroller board controls the operation of the rotor propeller adsorption device, and the fan generates a downward thrust, enabling the robot to be stably adsorbed onto the concrete surface.

[0048] Motion and Image Acquisition: The microcontroller board controls the movement of the mobile platform on the concrete surface, and the vision sensing device continuously captures images or videos of the concrete surface and transmits them to a remote computer via the wireless communication module 5. In this embodiment, the mobile platform, the rotor propeller adsorption device, and the visual sensing device are all coordinated and controlled by the microcontroller board. Its operation includes two modes:

[0049] Manual remote control mode: In this mode, the operator sends commands via a wireless remote control. The wireless transceiver system adopts a three-channel design. After the commands are decoded by the microcontroller board, they are mapped into drive signals for the four DC motors and control signals for the brushless electronic speed controller, thereby enabling real-time manual adjustment of forward, backward, turning, and suction force.

[0050] Autonomous Inspection Mode: In this mode, the microcontroller board automatically generates a motion control sequence based on preset inspection tasks, controlling the mobile platform to move along the planned path. Simultaneously, it controls the vision sensor to continuously capture images of the concrete surface. The microcontroller board then uses its built-in crack recognition algorithm to process and analyze the images in real time.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for automatic identification of concrete cracks, characterized in that, include: Acquire a grayscale image of the concrete surface to be inspected, and calculate the median and standard deviation of the grayscale image. An adaptive threshold is determined based on the median gray level and the standard deviation of gray level, and the adaptive threshold decreases as the standard deviation increases. The grayscale image is sampled at fixed intervals, and sampling points with grayscale values ​​lower than the adaptive threshold are marked as candidate crack pixels. The candidate crack pixels are used as seed points for region growth. For pixels to be incorporated into the current region, the pixel grayscale value must be lower than the threshold and its gradient direction must be consistent with the gradient direction of the seed point. When the number of pixels in any connected region formed by growth exceeds a preset upper limit, the growth of that region is terminated. Calculate the aspect ratio of the minimum bounding rectangle of each connected region, and determine the connected regions with aspect ratios greater than a set shape threshold as crack regions, and output the location information of the crack regions.

2. The automatic concrete crack identification method as described in claim 1, characterized in that, During the region growth process, for the current seed point pixel, pixels that meet the conditions are searched in its eight connected neighborhoods for growth. A pixel that meets the conditions is one whose gray value of the neighboring pixel is lower than a threshold and whose absolute value of the difference between its gradient direction and the gradient direction of the seed point is less than a preset angle threshold.

3. The automatic concrete crack identification method as described in claim 2, characterized in that, The angle threshold is 30 degrees, and the preset upper limit for the number of pixels that meet the conditions is 500 pixels.

4. The automatic concrete crack identification method as described in claim 1, characterized in that, If the set shape threshold is 2, then connected regions with an aspect ratio greater than or equal to 2 are identified as cracks; otherwise, they are discarded.

5. The automatic concrete crack identification method as described in claim 1, characterized in that, When outputting the location information of the crack region, the contour of the crack region is first mapped to the original image coordinate system, and a result image containing the contour annotation is generated.

6. The automatic concrete crack identification method as described in claim 1, characterized in that, The gradient direction is obtained by convolving the grayscale image using an edge detection operator.

7. A concrete crack automatic identification robot employing the concrete crack automatic identification method according to any one of claims 1-6, characterized in that, include: Mobile platform for moving on concrete surfaces; A visual sensing device, installed on the mobile platform, is used to acquire image information of the concrete surface; A microcontroller board is mounted on the mobile platform and electrically connected to the visual sensing device; The microcontroller board has an embedded processing unit configured to perform the automatic concrete crack identification method as described in any one of claims 1 to 6.

8. A robot as described in claim 7, characterized in that, The bottom of the mobile platform is equipped with a rotor thruster adsorption device, which generates a thrust perpendicular to the concrete surface, allowing the robot to adhere to the concrete surface.

9. A robot as described in claim 8, characterized in that, The rotor propulsion adsorption device includes a pipe housing and a brushless motor-driven fan installed inside it. The microcontroller board is electrically connected to the brushless motor to adjust the fan speed.

10. A robot as described in claim 7, characterized in that, It also includes a wireless remote controller for receiving manual operation commands; the microcontroller board is communicatively connected to the wireless remote controller and is configured to control the movement of the mobile platform and / or the adsorption force of the rotor propeller adsorption device according to the commands from the wireless remote controller.

Images