Pavement disease detection system and method capable of adjusting detection picture in real time

Abstract

The invention belongs to the technical field of pavement disease detection, and particularly relates to a pavement disease detection system capable of adjusting a detection picture in real time, which comprises a focusable camera, a front-end controller, a front-end controller, a front-end controller and a rear-end controller, the power supply module is used for supplying power to the focus-adjustable camera and the front-end controller; the camera bracket is used for fixing the focus-adjustable camera on a car roof; and the data acquisition module is used for acquiring image data acquired by the focus-adjustable camera. According to the system, the focus-adjustable camera is connected with the front-end controller, the power supply module is used for stably supplying power, and the camera bracket is fixedly mounted, so that the system can still continuously acquire clear road surface images in dynamic scenes such as vehicle running, and a high-quality data source is provided for subsequent disease detection.

Description

Technical Field

[0001] This invention belongs to the field of pavement defect detection technology, and in particular relates to a pavement defect detection system and method that can adjust the detection screen in real time. Background Technology

[0002] In the field of road maintenance and management, timely and accurate detection of pavement defects (such as cracks, potholes, and repair areas) is a key link in ensuring road traffic safety and extending the service life of roads. With the development of intelligent detection technology, pavement defect detection systems based on machine vision are gradually replacing traditional manual inspections and becoming the mainstream detection method. Its core logic is to automatically identify and analyze defects by collecting pavement images through cameras and combining them with algorithm models.

[0003] However, existing road surface defect detection systems still face many technical bottlenecks in practical applications, making it difficult to meet the needs of complex inspection scenarios. Specific problems are as follows: The fixed detection area and insufficient targeting of existing systems mean that most adopt a "full-image detection" mode, where the detection area is the entire image captured by the camera, and the detection range cannot be dynamically adjusted according to the distribution of road defects. When only a local area of ​​the road surface has defects (such as dense damage on the right side of the road and intact on the left side) or when there is interference from non-target areas (such as road shoulders, green belts, and pedestrians and vehicles on the roadside), the system still needs to process the entire image data. This not only causes a large amount of redundant computing power consumption, but also makes it easy for defects to be falsely detected or missed due to pixel interference from non-target areas. This is especially true for small defects (such as tiny cracks less than 1 mm wide), which significantly affects the recognition accuracy.

[0004] In outdoor inspection scenarios, where environmental adaptability is poor and image quality is unstable, complex and variable lighting conditions (such as overexposure due to direct sunlight and occlusion caused by trees / tunnel entrances / exits) make existing systems prone to image preprocessing. This results in issues like lost detail and insufficient contrast in acquired images, making it difficult for subsequent algorithm models to extract clear disease features. While some systems incorporate infrared or polarization sensors to assist in optimizing image quality, the additional hardware not only increases system cost and weight but also raises the difficulty of equipment installation and maintenance, making them unsuitable for lightweight vehicle-mounted inspection scenarios.

[0005] The existing systems have limited adaptability and weak detection capabilities for small-sized defects. Most of them are based on traditional YOLO series networks to build detection models, and have not been optimized for the characteristics of "large size differences" of road defects. For large-sized defects such as large potholes, the model has a high recognition accuracy, but for small-sized defects such as fine cracks and edge peeling, the model is prone to missed detection due to insufficient ability to extract small-scale features. At the same time, the models are mostly trained in fixed scenarios and cannot be adapted to the special defect types in different regions (such as freeze-thaw cracks in the north and water-damaged potholes in the south). Users need to re-label a large amount of data to train new models, which has a very high technical threshold and time cost.

[0006] Insufficient interactivity and real-time performance result in low operational efficiency. The detection parameters of the existing system (such as detection range and focal length) are mostly preset during initialization. Users cannot adjust them in real time through the terminal during the inspection process. If the camera installation angle is off or the road surface deviates (such as a bend), the system needs to stop and adjust the hardware position, which seriously affects the inspection efficiency. In addition, the detection results are mostly exported in batches after the inspection is completed. Users cannot view the location and type of defects in real time, and cannot promptly identify key defect areas and formulate emergency maintenance plans, resulting in the continuous expansion of potential defects.

[0007] The incomplete location information of road defects limits their value for maintenance guidance. Existing systems can only output the image coordinates and category of defects, lacking the ability to accurately locate their geographical location and actual distance, and cannot associate defect information with the actual mileage or specific road segment. Maintenance personnel need to manually match defect locations using maps, which is not only time-consuming and labor-intensive, but also prone to errors and omissions in maintenance operations due to positioning inaccuracies, making it difficult to meet the needs of refined road management.

[0008] In summary, there is an urgent need for a road surface defect detection system that can dynamically adjust the detection area, adapt to complex environments, accurately identify defects of multiple sizes, and has real-time interactive and complete location positioning functions. This system would address the shortcomings of existing technologies, improve the efficiency and accuracy of road inspections, and provide more reliable technical support for road maintenance. Summary of the Invention

[0009] The purpose of this invention is to address the aforementioned technical problems by providing a road surface defect detection system that allows for real-time adjustment of the detection image.

[0010] In view of this, the present invention provides a road surface defect detection system that can adjust the detection screen in real time, comprising: An adjustable-focus camera is used for image acquisition and is connected to a front-end controller; The power supply module is used to power the adjustable focus camera and the front-end controller. Camera bracket, used to fix an adjustable-focus camera to the roof of a vehicle; The data acquisition module is used to acquire image data obtained by the adjustable focus camera; The data processing module is used to perform adaptive exposure and shadow suppression preprocessing on the acquired images, encode the preprocessed original images and transmit them to the external receiving terminal for real-time display via the RTSP protocol, and simultaneously crop and adjust the resolution of the original images to provide the algorithm detection module with images that meet the size requirements. The algorithm detection module, deployed on the front-end controller, is used to call the pre-trained pavement defect detection model, perform inference on the images output by the data processing module, and obtain information on the location, size, and category of pavement defects. The front-end controller is used to run the data acquisition module, data processing module, and algorithm detection module. It connects to an external receiving terminal via WiFi to transmit image data and interactive information, and is also used to measure the distance to the defect location.

[0011] Preferably, the process of constructing the pavement distress detection model specifically includes: Collect photos of road surface defects and manually label them to identify the different categories of road surface defects; To address the size differences in road surface defects, this paper introduces deformable convolution and CBAM attention mechanisms on the basis of the YOLOv8 network architecture, and adds a small-scale detector to construct an improved road defect target detection algorithm. The improved road defect detection algorithm was trained using a dataset of labeled road defect photos to obtain a road defect detection model.

[0012] Preferably, the external receiving terminal is used to display the real-time image transmitted by the data processing module. The user can drag and adjust the size and position of the detection window in the real-time screen of the external receiving terminal to form the detection area position information, and transmit the information to the data processing module of the front-end controller in real time.

[0013] Preferably, after receiving the detection area location information transmitted by the external receiving terminal, the data processing module performs copy-free cropping of the original image data based on the position and size of the detection area location information in the original image, and performs resolution self-adaptation processing according to the input requirements of the road surface defect detection model.

[0014] Preferably, the process of the algorithm detection module performing inference on the image specifically includes: The input image is normalized and features are extracted, and a multi-scale feature pyramid is constructed using an improved YOLOv8 network; Candidate box generation and regression calculation are performed on feature maps at various scales to obtain the location boxes and category confidence of candidate disease targets; Non-maximum suppression is used to filter overlapping candidate defect target locations, and the final pavement defect detection results are output.

[0015] Preferably, after receiving the road surface defect detection results output by the algorithm detection module, the front-end controller overlays the results onto the real-time video, encodes them using the RTSP protocol, and transmits them to an external receiving terminal for display.

[0016] Preferably, when constructing the pavement defect detection model, the user can import custom defect model information, which is obtained by AI training from a large number of defect images in different scenarios.

[0017] Preferably, when transmitting image information to an external receiving terminal, the front-end controller first compresses the image information.

[0018] Preferably, when the user adjusts the detection window on the external receiving terminal, the detection window is aligned with the road surface or an area with severe road damage. The external receiving terminal then transmits the image position information corresponding to the adjusted detection window to the data processing module of the front-end controller.

[0019] A method for detecting pavement defects with real-time adjustable detection images, applied to a pavement defect detection system with real-time adjustable detection images, includes the following steps: S1: Collect photos of road surface defects and manually label them to identify different categories of road surface defects; to address the differences in the size of road surface defects, introduce deformable convolution and CBAM attention mechanism on the basis of YOLOv8 network architecture, and add a small-scale detector to build an improved road defect target detection algorithm; train the algorithm using the labeled dataset to obtain the road surface defect detection model, and deploy it on the front-end controller. S2: Fix the adjustable focus camera to the roof of the vehicle using the camera bracket, connect the adjustable focus camera to the front controller via USB data cable, connect the external receiving terminal to the front controller via WiFi, and turn on the power supply module to supply power to the adjustable focus camera and the front controller. S3: The system is running, and the data acquisition module begins to collect road video image data acquired by the adjustable focus camera; S4: The data processing module performs adaptive exposure and shadow suppression preprocessing on the acquired image, encodes the preprocessed raw image, and transmits it to the external receiving terminal for real-time display via the RTSP protocol. S5: The user drags and adjusts the size and position of the detection window in the real-time screen of the external receiving terminal to form the detection area position information, and transmits the information to the data processing module of the front-end controller in real time. S6: After receiving the detection area location information, the data processing module performs copy-free cropping on the original image data based on the position and size of the detection area location information in the original image, and performs resolution self-adaptation processing according to the input requirements of the road surface defect detection model, and transmits the processed image to the algorithm detection module. S7: The algorithm detection module calls the road surface defect detection model to infer the location, size and category information of the road surface defects from the image output by the data processing module. S8: The front-end controller overlays the road defect detection results onto the real-time video, encodes them using the RTSP protocol, and transmits them to an external receiving terminal for display.

[0020] The beneficial effects of this invention are: This invention connects an adjustable-focus camera to a front-end controller, combines a stable power supply module with a fixed camera bracket, and ensures that the system can continuously acquire clear road images in dynamic scenarios such as vehicle movement, providing a high-quality data source for subsequent defect detection. The adaptive exposure and shadow suppression preprocessing of the data processing module can effectively counteract the interference of complex environments such as strong light and shadow on image quality and improve the system's environmental adaptability. Images are transmitted to an external receiving terminal in real time via the RTSP protocol. Combined with WiFi communication between the front-end controller and the terminal, real-time interaction between the user and the system is achieved. At the same time, cropping and resolution adjustment of the original image can reduce redundant data transmission and processing, thereby reducing the computational power consumption for subsequent algorithm detection. The front-end controller has the function of measuring the distance to the defect location. Combined with the defect category and size information output by the algorithm detection module, it can realize the output of comprehensive detection results of road defects, providing accurate defect information support for road maintenance. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the system composition of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0023] It should be noted that all directional and positional terms used in this invention, such as "up," "down," "left," "right," "front," "back," "vertical," "horizontal," "inner," "outer," "top," "lower," "lateral," "longitudinal," and "center," are only used to explain the relative positional relationships and connections between components in a specific state (as shown in the accompanying drawings). They are merely for the convenience of describing the invention and do not require the invention to be constructed and operated in a specific orientation; therefore, they should not be construed as limitations on the invention. Furthermore, 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.

[0024] In the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0025] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0026] like Figure 1 As shown, a road surface defect detection system capable of real-time adjustment of the detection screen includes: An adjustable-focus camera, used for image acquisition, connects to the front-end controller. The camera uses a 1 / 2.8-inch CMOS sensor with ≥2 million effective pixels, supports manual / automatic focal length adjustment from 8mm to 25mm, and a frame rate of up to 30fps@1080P. It features automatic white balance and backlight compensation. The connection to the front-end controller uses a USB 3.0 data cable with a transmission rate of ≥5Gbps, ensuring delay-free image acquisition of road surfaces in vehicle-moving scenarios (vehicle speed ≤60km / h). The camera lens employs an anti-glare coating to further reduce glare interference in strong light environments. Combined with the shadow suppression algorithm in the data processing module, this forms a dual anti-interference guarantee of both hardware and software. The power supply module powers the adjustable-focus camera and front-end controller. It uses a 12V / 5A vehicle power adapter with an input voltage range of 9V-36V, compatible with the power outputs of different vehicle models. It also has a built-in 10000mAh lithium battery, supporting continuous power supply for ≥2 hours after a power outage, preventing system interruptions due to vehicle shutdown or wiring faults and ensuring no loss of inspection data. The power adapter connects to the front-end controller and camera using waterproof connectors, with an IP65 protection rating, suitable for rainy days, flooded roads, and other damp environments. Camera bracket, used to fix an adjustable-focus camera to the roof of a vehicle; The data acquisition module is used to acquire image data obtained by the adjustable focus camera; The data processing module is used to perform adaptive exposure and shadow suppression preprocessing on the acquired images, encode the preprocessed original images and transmit them to the external receiving terminal for real-time display via the RTSP protocol, and simultaneously crop and adjust the resolution of the original images to provide the algorithm detection module with images that meet the size requirements. The external receiving terminal is specifically an Android tablet (system version ≥ Android 10.0, screen size ≥ 10.1 inches, resolution 1920×1200), which includes an operation interface, a real-time image interface, a detection result display interface, and a data communication module. The cropping in the data processing module is implemented based on memory mapping (mmap) technology: the original image data is stored in the shared memory area of ​​the front-end controller. After the data processing module receives the detection area location information (such as the upper left corner coordinates (x1, y1) and the lower right corner coordinates (x2, y2)) transmitted by the external receiving terminal, it directly points to the data of the corresponding area in the shared memory through the memory pointer. There is no need to copy the data to a new memory address. The cropping time is ≤10ms. Compared with the traditional "data copy + cropping" method, the memory usage is reduced by 60% and the computing power consumption is reduced by 40%, which effectively ensures the lightweight operation of the system.

[0027] Resolution adjustment follows the principle of "prioritizing the input size of the detection model": if the resolution of the image after cropping the detection area is 1280×720 pixels, while the input size of the road surface defect detection model is 640×640 pixels, the data processing module uses bilinear interpolation to scale the image to 640×640 pixels while keeping the aspect ratio unchanged (first cropping to 640×360 pixels, then padding with black pixels at the top and bottom to 640×640 pixels); if the resolution of the cropped image is smaller than the model input size (e.g., 320×320 pixels), the nearest neighbor interpolation algorithm is used to enlarge it to 640×640 pixels, ensuring that the input image size is fully matched with the model requirements and avoiding inference errors caused by resolution incompatibility.

[0028] When the front-end controller transmits image information to external receiving terminals, it adopts the H.265 video compression standard (compression ratio 10:1-20:1). Compared with the H.264 standard, this reduces the data volume by 50% while maintaining the same image quality, and lowers the WiFi transmission bandwidth requirement from 4Mbps to 2Mbps. This avoids image stuttering and frame drops in outdoor environments with weak network coverage (such as suburbs and mountainous areas). Simultaneously, the compression process enables ROI encoding, using high-quality encoding (quantization parameter QP=20) for image areas within the detection window and low-quality encoding (quantization parameter QP=35) for areas outside the detection window. This controls the data volume while ensuring clear image details in the affected areas. The algorithm detection module, deployed on the front-end controller, is used to call the pre-trained pavement defect detection model, perform inference on the images output by the data processing module, and obtain information on the location, size, and category of pavement defects. The front-end controller operates the data acquisition module, data processing module, and algorithm detection module. It connects to an external receiving terminal via WiFi to transmit image data and interactive information. It also measures the distance to defect locations. The front-end controller uses an industrial-grade embedded motherboard, equipped with a quad-core ARM Cortex-A53 processor (≥1.5GHz), 2GB DDR4 memory, and 16GB eMMC storage. It supports a wide operating temperature range of -20℃ to 60℃, adapting to the complex temperature conditions of outdoor inspections. Simultaneously, the front-end controller has a built-in GPS module (supporting BeiDou + GPS dual-mode positioning, with a positioning accuracy ≤1 meter) and a GPS data acquisition module. It can collect vehicle longitude, latitude, and altitude information in real time. Combined with the defect image coordinates output by the algorithm detection module, it calculates the horizontal distance between the defect and the vehicle using a perspective projection formula (given that the camera is 2.5 meters above the ground, has a 30° downward angle, and a 15mm focal length), achieving a three-dimensional information output of "defect location - geographic coordinates - distance".

[0029] This invention connects an adjustable-focus camera to a front-end controller, combines a stable power supply module with a fixed camera bracket, and ensures that the system can continuously acquire clear road images in dynamic scenarios such as vehicle movement, providing a high-quality data source for subsequent defect detection. The adaptive exposure and shadow suppression preprocessing of the data processing module can effectively counteract the interference of complex environments such as strong light and shadow on image quality and improve the system's environmental adaptability. Images are transmitted to an external receiving terminal in real time via the RTSP protocol. Combined with WiFi communication between the front-end controller and the terminal, real-time interaction between the user and the system is achieved. At the same time, cropping and resolution adjustment of the original image can reduce redundant data transmission and processing, thereby reducing the computational power consumption for subsequent algorithm detection. The front-end controller has the function of measuring the distance to the defect location. Combined with the defect category and size information output by the algorithm detection module, it can realize the output of comprehensive detection results of road defects, providing accurate defect information support for road maintenance.

[0030] As a preferred example of this application, the construction process of the pavement distress detection model specifically includes: Collect photos of road surface defects and manually label them to identify the different categories of road surface defects; To address the size differences in road surface defects, this paper introduces deformable convolution and CBAM attention mechanisms on the basis of the YOLOv8 network architecture, and adds a small-scale detector to construct an improved road defect target detection algorithm. The improved road defect detection algorithm was trained using a dataset of labeled road defect photos to obtain a road defect detection model. This application addresses the significant size differences in pavement distresses (such as microcracks and large potholes) by introducing deformable convolution and CBAM attention mechanisms on the basis of the YOLOv8 architecture. This enhances the model's ability to extract features of distresses at different scales. The addition of a small-scale detector effectively avoids the problem of traditional models missing small-sized distresses (such as microcracks) and covers more types and sizes of pavement distress detection scenarios. By training and improving the algorithm using manually labeled disease datasets, the model can accurately learn the characteristic patterns of different types of diseases, reducing model misjudgments caused by fuzzy data labeling. The final road disease detection model provides a high-performance inference foundation for the algorithm detection module, ensuring the accuracy and reliability of subsequent detection results.

[0031] As a preferred example of this application, the external receiving terminal is used to display the real-time image transmitted by the data processing module. The user can drag and adjust the size and position of the detection window in the real-time screen of the external receiving terminal to form the detection area position information, and transmit the information to the data processing module of the front-end controller in real time. In the example of this application, the user can drag and adjust the detection window in the real-time screen of the external receiving terminal, breaking the limitation of the traditional detection system's "fixed detection area". The detection window can be focused on the suspected area of ​​road defects (such as densely damaged sections) according to actual needs, reducing the interference of non-target areas (such as road shoulders and green belts) on the detection process and improving the detection targeting.

[0032] The detection area location information can be transmitted to the data processing module of the front-end controller in real time, without waiting for the system to process data in batches. This ensures that the adjusted detection area can be quickly applied to the subsequent image cropping and algorithm detection stages, avoiding detection delays caused by interaction latency and ensuring the real-time performance of the system in dynamic inspection scenarios.

[0033] As a preferred example of this application, after receiving the detection area location information transmitted by the external receiving terminal, the data processing module performs copy-free cropping of the original image data based on the position and size of the detection area location information in the original image, and performs resolution self-adaptation processing according to the input requirements of the road surface defect detection model. In the example of this application, the data processing module performs "copy-free cropping" on the original image based on the location information of the detection area, which can directly remove redundant data in non-detection areas and avoid the invalid computing power occupation in the traditional "full image processing" mode; at the same time, it performs resolution self-adaptation processing according to the input requirements of the detection model to ensure that the input image size matches the model requirements without the need for additional data format conversion, further reducing the computing power burden of the front-end controller and ensuring the lightweight operation of the system; The cropping process retains only the image of the detection area, avoiding interference from non-target area pixels on the model's feature extraction; resolution self-adaptation processing ensures that the image of the detection area remains clear and detailed after scaling, especially for small-sized lesions, providing high-quality input data for the algorithm detection module and reducing false detections and missed detections caused by image blurring.

[0034] As a preferred example of this application, the process of the algorithm detection module inferring from the image specifically includes: The input image is normalized and features are extracted, and a multi-scale feature pyramid is constructed using an improved YOLOv8 network; Candidate box generation and regression calculation are performed on feature maps at various scales to obtain the location boxes and category confidence of candidate disease targets; Non-maximum suppression is used to filter overlapping candidate defect target locations, and the final pavement defect detection results are output. This application constructs a multi-scale feature pyramid through an improved YOLOv8 network, which can extract feature information of pavement distress from different levels (such as edge features of small cracks and overall contour features of large potholes), avoids the omission of distress features caused by single-scale feature extraction, and improves the model's ability to identify complex distress. Candidate box generation and regression calculation can initially locate the target position of the defect. Combined with non-maximum suppression to screen overlapping candidate boxes, it can effectively eliminate duplicate and misjudged candidate boxes (such as candidate boxes that misjudge road stains as defects), ensuring that the final output of road defect detection results is more accurate and reducing the interference of invalid detection information on user decision-making.

[0035] As a preferred example of this application, after receiving the road defect detection result output by the algorithm detection module, the front-end controller superimposes the result onto the real-time video, encodes it using the RTSP protocol, and transmits it to an external receiving terminal for display. This application overlays the disease detection results onto real-time video, which can intuitively display the actual location, size and type of the disease on the road surface. Compared with simple text or data results, it is easier for users to quickly understand the distribution of the disease and reduce the cost of interpreting the results. The video overlay results are encoded and transmitted via the RTSP protocol, ensuring that the detection results can be quickly fed back to the external receiving terminal. Users can view the disease detection status in real time during the inspection process without waiting for the data to be exported in batches after the inspection ends. This facilitates the timely identification of key disease areas and the development of subsequent maintenance plans.

[0036] As a preferred example of this application, when constructing the pavement defect detection model, the user can import custom defect model information, which is obtained by AI training from a large number of defect images in different scenarios; In the example of this application, users are allowed to import custom defect model information, which enables the system to adapt to the needs of road surface defect detection in different regions and scenarios (such as freeze-thaw cracks in the north and water damage potholes in the south), avoids the decrease in detection accuracy of a single model in special scenarios, and improves the system's scenario adaptability. Custom defect models are generated by AI training from a large number of defect images in different scenarios. Users do not need to label data and train models from scratch. They only need to import the pre-trained model to expand the detection function, reducing the technical threshold and time cost for users and improving the ease of use of the system.

[0037] As a preferred example of this application, the front-end controller first compresses the image information when transmitting it to an external receiving terminal; In the example of this application, the front-end controller compresses the image information before transmitting it, which can significantly reduce the amount of data during WiFi communication, avoid image stuttering and delay caused by data transmission congestion, and ensure that the external receiving terminal can smoothly display real-time images and detection results. This is especially suitable for outdoor inspection scenarios with unstable network environments. Compressed image data has lower requirements for communication bandwidth, and can achieve stable transmission without relying on high-bandwidth networks, reducing communication costs caused by high bandwidth requirements. At the same time, it reduces the performance requirements of the communication modules between the front-end controller and external terminals, indirectly reducing the system hardware cost.

[0038] As a preferred example of this application, when the user adjusts the detection window on the external receiving terminal, the detection window is aligned with the road surface or an area with severe road damage, and the external receiving terminal transmits the image position information corresponding to the adjusted detection window to the data processing module of the front-end controller. In the example of this application, the user can further narrow the detection range by pointing the detection window at the road surface or the area with more serious damage, reduce invalid detection of non-road areas (such as sidewalks and roadside vegetation) or road surfaces without defects, and make the computing power of the algorithm detection module more concentrated on the core detection area, thereby improving the defect detection efficiency per unit time. The external receiving terminal transmits the adjusted detection window position information, which ensures that the data processing module can accurately locate the core detection area in the original image, avoids cropping errors caused by position information deviation (such as cropping out some disease areas), and ensures the accuracy of the input data for subsequent algorithm detection.

[0039] A method for detecting pavement defects with real-time adjustable detection images, applied to a pavement defect detection system with real-time adjustable detection images, includes the following steps: S1: Collect photos of road surface defects and manually label them to identify different categories of road surface defects; to address the differences in the size of road surface defects, introduce deformable convolution and CBAM attention mechanism on the basis of YOLOv8 network architecture, and add a small-scale detector to build an improved road defect target detection algorithm; train the algorithm using the labeled dataset to obtain the road surface defect detection model, and deploy it on the front-end controller. S2: Fix the adjustable focus camera to the roof of the vehicle using the camera bracket, connect the adjustable focus camera to the front controller via USB data cable, connect the external receiving terminal to the front controller via WiFi, and turn on the power supply module to supply power to the adjustable focus camera and the front controller. S3: The system is running, and the data acquisition module begins to collect road video image data acquired by the adjustable focus camera; S4: The data processing module performs adaptive exposure and shadow suppression preprocessing on the acquired image, encodes the preprocessed raw image, and transmits it to the external receiving terminal for real-time display via the RTSP protocol. S5: The user drags and adjusts the size and position of the detection window in the real-time screen of the external receiving terminal to form the detection area position information, and transmits the information to the data processing module of the front-end controller in real time. S6: After receiving the detection area location information, the data processing module performs copy-free cropping on the original image data based on the position and size of the detection area location information in the original image, and performs resolution self-adaptation processing according to the input requirements of the road surface defect detection model, and transmits the processed image to the algorithm detection module. S7: The algorithm detection module calls the road surface defect detection model to infer the location, size and category information of the road surface defects from the image output by the data processing module. S8: The front-end controller overlays the road defect detection results onto the real-time video, encodes them using the RTSP protocol, and transmits them to an external receiving terminal for display.

[0040] This method clarifies the entire process of "model training and deployment - hardware connection - data acquisition - image processing - region adjustment - algorithm inference - result display", making the system operation process more standardized, reducing system failures caused by chaotic operation sequence, and ensuring system stability during the inspection process. Among them, image preprocessing in step S4 improves data quality, detection region adjustment and cropping in steps S5-S6 reduce computing power consumption, algorithm inference in step S7 ensures detection accuracy, and result overlay and transmission in step S8 improves readability and timeliness. The synergistic effect of each step enables the system to achieve optimal results in terms of environmental adaptability, detection accuracy, and real-time performance, meeting the actual needs of outdoor dynamic road inspection.

[0041] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A road surface defect detection system with real-time adjustable detection screen, characterized in that: include: An adjustable-focus camera is used for image acquisition and is connected to a front-end controller; The power supply module is used to power the adjustable focus camera and the front-end controller. Camera bracket, used to fix an adjustable-focus camera to the roof of a vehicle; The data acquisition module is used to acquire image data obtained by the adjustable focus camera; The data processing module is used to perform adaptive exposure and shadow suppression preprocessing on the acquired images, encode the preprocessed original images and transmit them to the external receiving terminal for real-time display via the RTSP protocol, and simultaneously crop and adjust the resolution of the original images to provide the algorithm detection module with images that meet the size requirements. The algorithm detection module, deployed on the front-end controller, is used to call the pre-trained pavement defect detection model, perform inference on the images output by the data processing module, and obtain information on the location, size, and category of pavement defects. The front-end controller is used to run the data acquisition module, data processing module, and algorithm detection module. It connects to an external receiving terminal via WiFi to transmit image data and interactive information, and is also used to measure the distance to the defect location.

2. The road surface defect detection system with real-time adjustable detection screen according to claim 1, characterized in that, The construction process of the road surface distress detection model specifically includes: Collect photos of road surface defects and manually label them to identify the different categories of road surface defects; To address the size differences in road surface defects, this paper introduces deformable convolution and CBAM attention mechanisms on the basis of the YOLOv8 network architecture, and adds a small-scale detector to construct an improved road defect target detection algorithm. The improved road defect detection algorithm was trained using a dataset of labeled road defect photos to obtain a road defect detection model.

3. The road surface defect detection system with real-time adjustable detection screen according to claim 1, characterized in that, The external receiving terminal is used to display the real-time image transmitted by the data processing module. Users can drag and adjust the size and position of the detection window in the real-time screen of the external receiving terminal to form the detection area position information, and transmit this information to the data processing module of the front-end controller in real time.

4. The road surface defect detection system with real-time adjustable detection screen according to claim 3, characterized in that, After receiving the detection area location information transmitted by the external receiving terminal, the data processing module performs copy-free cropping of the original image data based on the position and size of the detection area location information in the original image, and performs resolution self-adaptation processing according to the input requirements of the road surface defect detection model.

5. The road surface defect detection system with real-time adjustable detection screen according to claim 1, characterized in that, The reasoning process of the algorithm detection module on the image specifically includes: The input image is normalized and features are extracted, and a multi-scale feature pyramid is constructed using an improved YOLOv8 network; Candidate box generation and regression calculation are performed on feature maps at various scales to obtain the location boxes and category confidence of candidate disease targets; Non-maximum suppression is used to filter overlapping candidate defect target locations, and the final pavement defect detection results are output.

6. The road surface defect detection system with real-time adjustable detection screen according to claim 1, characterized in that, After receiving the road defect detection results output by the algorithm detection module, the front-end controller overlays the results onto the real-time video, encodes them using the RTSP protocol, and transmits them to an external receiving terminal for display.

7. The road surface defect detection system with real-time adjustable detection screen according to claim 2, characterized in that, When constructing the pavement defect detection model, users can import custom defect model information, which is obtained by AI training from a large number of defect images in different scenarios.

8. The road surface defect detection system with real-time adjustable detection screen according to claim 1, characterized in that, When transmitting image information to an external receiving terminal, the front-end controller first compresses the image information.

9. The road surface defect detection system with real-time adjustable detection screen according to claim 3, characterized in that, When the user adjusts the detection window on the external receiving terminal, aligning the detection window with the road surface or an area with severe road damage, the external receiving terminal transmits the image position information corresponding to the adjusted detection window to the data processing module of the front-end controller.

10. A method for detecting pavement defects with a real-time adjustable detection screen, applied to the pavement defect detection system with a real-time adjustable detection screen as described in any one of claims 1-9, characterized in that, Includes the following steps: S1: Collect photos of road surface defects and manually label them to identify different categories of road surface defects; to address the differences in the size of road surface defects, introduce deformable convolution and CBAM attention mechanism on the basis of YOLOv8 network architecture, and add a small-scale detector to build an improved road defect target detection algorithm; train the algorithm using the labeled dataset to obtain the road surface defect detection model, and deploy it on the front-end controller. S2: Fix the adjustable focus camera to the roof of the vehicle using the camera bracket, connect the adjustable focus camera to the front controller via USB data cable, connect the external receiving terminal to the front controller via WiFi, and turn on the power supply module to supply power to the adjustable focus camera and the front controller. S3: The system is running, and the data acquisition module begins to collect road video image data acquired by the adjustable focus camera; S4: The data processing module performs adaptive exposure and shadow suppression preprocessing on the acquired image, encodes the preprocessed raw image, and transmits it to the external receiving terminal for real-time display via the RTSP protocol. S5: The user drags and adjusts the size and position of the detection window in the real-time screen of the external receiving terminal to form the detection area position information, and transmits the information to the data processing module of the front-end controller in real time. S6: After receiving the detection area location information, the data processing module performs copy-free cropping on the original image data based on the position and size of the detection area location information in the original image, and performs resolution self-adaptation processing according to the input requirements of the road surface defect detection model, and transmits the processed image to the algorithm detection module. S7: The algorithm detection module calls the road surface defect detection model to infer the location, size and category information of the road surface defects from the image output by the data processing module. S8: The front-end controller overlays the road defect detection results onto the real-time video, encodes them using the RTSP protocol, and transmits them to an external receiving terminal for display.

Images