Road quality automatic detection method and system based on object recognition

Abstract

The invention discloses a road quality automatic detection method and system based on object recognition, and the system comprises an image collection module which comprises a plurality of devices and is capable of adjusting parameters and collecting road images; the data transmission module is stable in transmission and supports wired and wireless modes; the image preprocessing module is used for denoising, enhancing contrast and correcting distortion; the object recognition module is used for recognizing a road problem by using a CNN model; the quality evaluation module is used for evaluating according to standards and formulas; the storage and display module is used for storing data and displaying results, and all the modules cooperate to guarantee detection. The road quality can be efficiently and accurately detected, manual errors are reduced, images can be rapidly collected and processed, various road diseases can be recognized, the quality can be automatically evaluated, data storage and display are convenient, safe and stable data transmission can be guaranteed, and road maintenance and management intelligentization is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of highway quality detection systems, and in particular to an automatic highway quality detection method and system based on object recognition. Background Art

[0002] With the booming development of the transportation industry, the highway network is becoming increasingly large and complex. The inspection and maintenance of highway quality has become a key link in ensuring traffic safety and road service life. Traditional highway quality inspection methods mainly rely on manual inspections and some simple measurement tools.

[0003] There are many significant disadvantages of manual inspection. On the one hand, inefficiency is its primary problem. Inspectors need to inspect section by section along the highway. Whether it is cracks, potholes on the road surface, or damage to ancillary facilities, they need to be observed and recorded manually by the naked eye. This method is time-consuming and labor-intensive. For long-distance highway inspections, it requires a lot of manpower and time costs, and it is difficult to meet the needs of rapid inspection of large-scale highway networks. On the other hand, the accuracy and objectivity of manual inspections are easily affected by the inspectors' own experience, skill level, and working status. Different inspectors may have different judgment criteria for highway diseases. For example, the definition of small cracks in the road surface and the assessment of pothole depth may vary due to personal subjective factors, resulting in inaccurate and unreliable inspection results.

[0004] In addition, traditional detection methods also have difficulties in data recording and analysis. The data recorded manually are often not standardized and comprehensive, and the subsequent data collation and analysis work is cumbersome, making it difficult to conduct effective statistics and trend analysis on the quality of highways. Moreover, some traditional detection tools, such as simple rulers and levels, can only measure limited quality indicators and cannot effectively detect complex highway conditions and hidden diseases (such as internal problems of the roadbed).

[0005] In recent years, although some detection methods based on sensor technology have emerged, these methods can only detect specific types of diseases or specific locations, and lack the ability to comprehensively evaluate the overall quality of the highway. At the same time, with the continuous increase in highway traffic volume, the timeliness of highway quality detection is also increasingly required. Traditional detection methods cannot meet the needs of quickly obtaining highway quality information to quickly take maintenance measures.

[0006] With the rapid development of information technology, computer vision and deep learning technologies have provided new ideas and methods for highway quality inspection. Using object recognition technology to realize automatic inspection of highway quality can effectively overcome the shortcomings of traditional inspection methods, improve inspection efficiency and accuracy, and provide strong support for highway maintenance and management. Summary of the invention

[0007] The present invention proposes a method and system for automatically detecting road quality based on object recognition to solve the problems mentioned in the above-mentioned prior art.

[0008] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a highway quality automatic detection system based on object recognition, comprising:

[0009] An image acquisition module is used to acquire images of the road surface, roadbed, and ancillary facilities. The image acquisition equipment includes but is not limited to high-definition cameras and camera devices carried by drones. The installation position and angle are adjustable, and the acquisition frequency is set according to the detection speed and accuracy requirements;

[0010] The data transmission module transmits the collected image data to the processing center in real time, supporting wired and wireless transmission methods. Wireless transmission includes but is not limited to 4G and 5G networks to ensure the stability and efficiency of data transmission;

[0011] An image preprocessing module performs preprocessing on the received image, including but not limited to denoising, contrast enhancement, and geometric distortion correction. An adaptive filtering algorithm is used to remove image noise, contrast enhancement is based on histogram equalization, and geometric distortion correction is performed through perspective transformation.

[0012] The object recognition module uses a trained deep learning model to identify detection objects in the image. The objects include road cracks, potholes, damage, sign wear, roadbed settlement, and guardrail damage. The model is built based on a convolutional neural network (CNN) and is trained using a large amount of labeled data;

[0013] The quality assessment module assesses the road quality based on the identified object information and predefined quality assessment standards. The assessment indicators include road surface smoothness P, damage rate D, and facility integrity F. The module outputs the road quality assessment results and a detailed report. The road surface smoothness calculation formula is: L i is the length of cracks or potholes on the road that exceed the threshold, n is the number of cracks or potholes, L total is the length of the detection section; the damage rate calculation formula is: S d is the road damage area, S total is the total road surface area; the calculation formula for facility integrity is: N g is the number of intact road ancillary facilities, N total is the total number of ancillary facilities;

[0014] The storage and display module stores the collected original images, processed images and evaluation results, provides users with a visual interface to display evaluation reports and images, and supports historical data query and comparative analysis.

[0015] Furthermore, the high-definition camera of the image acquisition module has optical image stabilization and autofocus functions, as follows:

[0016] The optical image stabilization function uses a suspended optical image stabilization lens group. The gyroscope sensor detects the camera's shaking direction and amplitude in real time, drives the image stabilization lens group to move in a plane perpendicular to the optical axis, compensates for the light deviation caused by shaking, effectively reduces image blur caused by factors such as road bumps and vehicle vibrations, and can ensure image clarity within a certain frequency and amplitude range;

[0017] The autofocus function is based on a combination of phase detection autofocus technology and contrast detection autofocus technology. Phase detection autofocus uses paired pixels on the image sensor to detect the phase difference of light and quickly determine the focus direction and distance. Contrast detection autofocus fine-tunes the focus by analyzing the contrast changes of the picture to ensure accurate focus under different lighting conditions and shooting distances. The adaptable light intensity range is from 0.1lux to 100000lux, and the shooting distance range is from 0.5m to 100m.

[0018] The camera device carried by the drone can automatically fly and collect images according to the preset route. The drone flight control system can achieve high-precision positioning and navigation based on the global positioning system (GPS), inertial navigation system (INS) and visual positioning system (VPS). The preset route can be planned and adjusted through the ground control station software. During the flight, the flight attitude and speed can be automatically adjusted according to real-time environmental information (such as wind speed, wind direction, obstacles, etc.), ensuring that the camera device can collect comprehensive and accurate images of the highway at the predetermined angle and height.

[0019] Furthermore, the data transmission module uses an encryption protocol during wireless transmission to ensure the security of data transmission, specifically:

[0020] The Advanced Encryption Standard (AES) algorithm is used to encrypt the transmitted data. The image data is encrypted block by block according to the key generated by the system. The key length is 256 bits. The plaintext data is encrypted into ciphertext at the data sending end and transmitted through the wireless channel. The receiving end uses the same key for decryption. At the same time, combined with the security mechanism in the wireless transmission protocol, such as the use of IPSec-based VPN technology in 4G networks, the native security architecture of 5G networks is used to enhance the confidentiality, integrity and authentication of data transmission;

[0021] Wired transmission uses optical fiber transmission to improve transmission bandwidth and anti-interference ability. The optical fiber uses single-mode optical fiber with a core diameter of 9μm, a cladding diameter of 125μm, and an operating wavelength of 1310nm or 1550nm. The optical transmitter converts the electrical signal into an optical signal for transmission in the optical fiber, and the optical receiver restores the optical signal to an electrical signal at the receiving end. The transmission process can effectively resist electromagnetic interference, and the supported transmission bandwidth can reach more than 10Gbps.

[0022] Furthermore, the image preprocessing module can automatically select a suitable filtering algorithm according to the image noise type during the denoising process, specifically:

[0023] When the image noise is Gaussian noise, the Wiener filtering algorithm is used to estimate the local mean and variance of the image and calculate the optimal filter coefficient for each pixel, effectively removing the Gaussian noise while maintaining the image details.

[0024] For salt and pepper noise, the median filtering algorithm is used to replace the gray value of each pixel in the image with the median gray value of the pixels in its neighborhood, which can eliminate the salt and pepper noise without destroying the edges and details of the image.

[0025] If mixed noise is detected in the image, an adaptive weighted median filtering algorithm is used to dynamically adjust the weights of the median filter and the mean filter according to the distribution of noise in the pixel neighborhood to effectively remove the mixed noise.

[0026] Geometric distortion correction can automatically calculate the correction parameters by identifying the feature points in the image. The specific process is as follows: first, the scale-invariant feature transform (SIFT) algorithm or the speeded up robust feature (SURF) algorithm is used to extract the feature points in the image, and the corresponding feature points in images of different perspectives are found through a feature matching algorithm (such as the nearest neighbor matching algorithm based on a descriptor). Then, based on the coordinate relationship of these feature point pairs, the perspective transformation model or the affine transformation model is used to calculate the parameters required for geometric distortion correction, thereby achieving accurate correction of image geometric distortion.

[0027] Furthermore, a method for automatically detecting road quality based on object recognition comprises the following steps:

[0028] Image acquisition step, using image acquisition equipment to obtain image data of the highway, equipment parameters and acquisition path are set according to the highway type and detection requirements;

[0029] Data transmission step: transmitting the collected image data to the processing center to ensure the stability of the transmission process and the integrity of the data;

[0030] Image preprocessing step: preprocess the image to improve the image quality for subsequent analysis;

[0031] In the object recognition step, the trained deep learning model is used to perform object recognition on the preprocessed images to obtain information about abnormal objects on the road;

[0032] Quality assessment step: quantitatively assess the quality of the highway based on the identification results and quality assessment standards. The quantitative assessment is based on the above-mentioned calculation formulas for road surface flatness, damage rate and facility integrity;

[0033] The result storage and display step stores the evaluation results and related images and displays them to the user for easy query and analysis.

[0034] Furthermore, for long-distance highway detection, multiple sets of image acquisition devices are used to work together, and the acquired images contain overlapping areas for subsequent splicing, specifically:

[0035] Multiple groups of image acquisition devices are distributed along the highway at a certain distance. The acquisition areas of two adjacent groups of devices overlap by 10%. The acquisition parameters of each group of devices (such as shooting angle, focal length, resolution, etc.) are pre-adjusted according to their location and local features of the highway to ensure that the acquired images can fully cover the highway pavement, roadbed and ancillary facilities.

[0036] During the acquisition process, a time synchronization mechanism is used to ensure the time consistency of images collected by different devices, which facilitates the subsequent accurate stitching and analysis of the images. The time synchronization accuracy can reach 10 milliseconds. At the same time, wireless communication technology is used to achieve collaborative work between devices. Status information and acquisition instructions can be transmitted between devices in real time to ensure the efficiency and reliability of the entire acquisition process.

[0037] Furthermore, in the data transmission step, when the transmission is interrupted, it has automatic reconnection and data caching functions to ensure that data is not lost, as follows:

[0038] The data transmission module has a built-in automatic reconnection mechanism. When a wireless or wired connection is detected to be disconnected, it will try to reconnect every 3-10 seconds according to the preset reconnection strategy, and will try to reestablish the connection automatically up to 5 times.

[0039] At the same time, it is equipped with a data cache module. When the transmission is interrupted, the data to be transmitted will be temporarily stored in the cache. The cache capacity is 10GB, and a first-in-first-out (FIFO) storage strategy is adopted to ensure that the data before the interruption can continue to be transmitted after the reconnection is successful, avoiding data loss.

[0040] Furthermore, for images with uneven illumination, a local contrast enhancement algorithm is used to improve image quality, specifically:

[0041] The algorithm based on Local Adaptive Histogram Equalization (CLAHE) is used to divide the image into multiple non-overlapping pixel blocks, and histogram equalization is performed on each pixel block to enhance the contrast within each small block.

[0042] In order to avoid artificial traces between blocks (such as discontinuity of grayscale values), the bilinear interpolation algorithm is used to smooth the boundaries between adjacent blocks, so that the entire image can maintain a natural visual effect while enhancing the contrast, improve the recognizability of object features in the image, and facilitate subsequent object recognition and quality assessment.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] In terms of detection efficiency, the present invention can quickly obtain a large amount of image information of the highway through the image acquisition module, whether it is through the fixed installation of high-definition cameras or the large-area acquisition of drone-mounted camera devices, which greatly shortens the detection time. Compared with traditional manual detection, it can complete the comprehensive detection of long-distance highways in a shorter time, discover quality problems in time, and reduce the possibility of disease deterioration due to untimely detection.

[0045] In terms of accuracy, the object recognition module of the deep learning model can accurately identify various road diseases such as road cracks, potholes, sign wear, roadbed settlement, etc., avoiding the subjective errors of manual detection. Based on predefined quality assessment standards and corresponding formulas, such as calculating indicators such as road surface flatness, damage rate, and facility integrity, the quality of the road is quantitatively evaluated, making the evaluation results more scientific and objective.

[0046] From the perspective of data processing, the data transmission module of the system ensures stable and efficient transmission of image data, and the image preprocessing module improves the image quality, which is conducive to subsequent analysis. The storage and display module facilitates the storage and query comparison of data, providing strong support for the long-term monitoring and analysis of highway quality. Through historical data comparison, the changing trend of highway quality can be understood and a more reasonable maintenance plan can be formulated.

[0047] In addition, the security and stability of the system are also guaranteed. The encryption protocol and optical fiber transmission method of the data transmission module ensure the security and stable transmission of data, and can avoid data leakage and transmission interruption even in complex environments. The multi-device coordination, automatic reconnection and caching functions in long-distance detection further improve the reliability of the system, ensure the smooth progress of the entire detection process, and effectively promote the development of highway quality detection and maintenance towards intelligence and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic block diagram of an automatic road quality detection system based on object recognition proposed by the present invention;

[0049] Figure 2 This is a schematic block diagram of an automatic road quality detection method based on object recognition proposed by the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0052] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, and it can be the internal connection of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0053] Reference Figure 1-2 :A highway quality automatic detection system based on object recognition, comprising:

[0054] Image acquisition module, used to collect images of highway pavement, roadbed, and ancillary facilities. Image acquisition equipment includes but is not limited to high-definition cameras and camera devices carried by drones. The installation position and angle can be adjusted, and the acquisition frequency is set according to the detection speed and accuracy requirements;

[0055] The data transmission module transmits the collected image data to the processing center in real time, supporting wired and wireless transmission methods. Wireless transmission includes but is not limited to 4G and 5G networks to ensure the stability and efficiency of data transmission;

[0056] An image preprocessing module performs preprocessing on the received image, including but not limited to denoising, contrast enhancement, and geometric distortion correction. An adaptive filtering algorithm is used to remove image noise, contrast enhancement is based on histogram equalization, and geometric distortion correction is performed through perspective transformation.

[0057] The object recognition module uses a trained deep learning model to identify detection objects in the image. The objects include road cracks, potholes, damage, sign wear, roadbed settlement, and guardrail damage. The model is built based on a convolutional neural network (CNN) and is trained using a large amount of labeled data.

[0058] The quality assessment module assesses the road quality based on the identified object information and predefined quality assessment standards. The assessment indicators include road surface smoothness P, damage rate D, and facility integrity F. The module outputs the road quality assessment results and a detailed report. The road surface smoothness calculation formula is: L i is the length of cracks or potholes on the road that exceed the threshold, n is the number of cracks or potholes, L total is the length of the detection section; the damage rate calculation formula is: S d is the road damage area, S total is the total road surface area; the calculation formula for facility integrity is: N g is the number of intact road ancillary facilities, N total is the total number of ancillary facilities;

[0059] The storage and display module stores the collected original images, processed images and evaluation results, provides users with a visual interface to display evaluation reports and images, and supports historical data query and comparative analysis.

[0060] In the present invention, the high-definition camera of the image acquisition module has an optical image stabilization function and an auto-focus function, as follows:

[0061] The optical image stabilization function uses a suspended optical image stabilization lens group. The gyroscope sensor detects the camera's shaking direction and amplitude in real time, drives the image stabilization lens group to move in a plane perpendicular to the optical axis, compensates for the light deviation caused by shaking, effectively reduces image blur caused by factors such as road bumps and vehicle vibrations, and can ensure image clarity within a certain frequency and amplitude range;

[0062] The autofocus function is based on a combination of phase detection autofocus technology and contrast detection autofocus technology. Phase detection autofocus uses paired pixels on the image sensor to detect the phase difference of light and quickly determine the focus direction and distance. Contrast detection autofocus fine-tunes the focus by analyzing the contrast changes of the picture to ensure accurate focus under different lighting conditions and shooting distances. The adaptable light intensity range is from 0.1lux to 100000lux, and the shooting distance range is from 0.5m to 100m.

[0063] The camera device carried by the drone can automatically fly and collect images according to the preset route. The drone flight control system can achieve high-precision positioning and navigation based on the global positioning system (GPS), inertial navigation system (INS) and visual positioning system (VPS). The preset route can be planned and adjusted through the ground control station software. During the flight, the flight attitude and speed can be automatically adjusted according to real-time environmental information (such as wind speed, wind direction, obstacles, etc.), ensuring that the camera device can collect comprehensive and accurate images of the highway at the predetermined angle and height.

[0064] In the present invention, the data transmission module adopts an encryption protocol during wireless transmission to ensure the security of data transmission, specifically:

[0065] The Advanced Encryption Standard (AES) algorithm is used to encrypt the transmitted data. The image data is encrypted block by block according to the key generated by the system. The key length is 256 bits. The plaintext data is encrypted into ciphertext at the data sending end and transmitted through the wireless channel. The receiving end uses the same key for decryption. At the same time, combined with the security mechanism in the wireless transmission protocol, such as the use of IPSec-based VPN technology in 4G networks, the native security architecture of 5G networks is used to enhance the confidentiality, integrity and authentication of data transmission;

[0066] Wired transmission uses optical fiber transmission to improve transmission bandwidth and anti-interference ability. The optical fiber uses single-mode optical fiber with a core diameter of 9μm, a cladding diameter of 125μm, and an operating wavelength of 1310nm or 1550nm. The optical transmitter converts the electrical signal into an optical signal for transmission in the optical fiber, and the optical receiver restores the optical signal to an electrical signal at the receiving end. The transmission process can effectively resist electromagnetic interference, and the supported transmission bandwidth can reach more than 10Gbps.

[0067] In the present invention, the image preprocessing module can automatically select a suitable filtering algorithm according to the image noise type during the denoising process, specifically:

[0068] When the image noise is Gaussian noise, the Wiener filtering algorithm is used to estimate the local mean and variance of the image and calculate the optimal filter coefficient for each pixel, effectively removing the Gaussian noise while maintaining the image details.

[0069] For salt and pepper noise, the median filtering algorithm is used to replace the gray value of each pixel in the image with the median gray value of the pixels in its neighborhood, which can eliminate the salt and pepper noise without destroying the edges and details of the image.

[0070] If mixed noise is detected in the image, an adaptive weighted median filtering algorithm is used to dynamically adjust the weights of the median filter and the mean filter according to the distribution of noise in the pixel neighborhood to effectively remove the mixed noise.

[0071] Geometric distortion correction can automatically calculate the correction parameters by identifying the feature points in the image. The specific process is as follows: first, the scale-invariant feature transform (SIFT) algorithm or the speeded up robust feature (SURF) algorithm is used to extract the feature points in the image, and the corresponding feature points in images of different perspectives are found through a feature matching algorithm (such as the nearest neighbor matching algorithm based on a descriptor). Then, based on the coordinate relationship of these feature point pairs, the perspective transformation model or the affine transformation model is used to calculate the parameters required for geometric distortion correction, thereby achieving accurate correction of image geometric distortion.

[0072] In the present invention, a method for automatically detecting road quality based on object recognition comprises the following steps:

[0073] Image acquisition step, using image acquisition equipment to obtain image data of the highway, equipment parameters and acquisition path are set according to the highway type and detection requirements;

[0074] Data transmission step: transmitting the collected image data to the processing center to ensure the stability of the transmission process and the integrity of the data;

[0075] Image preprocessing step: preprocess the image to improve the image quality for subsequent analysis;

[0076] In the object recognition step, the trained deep learning model is used to perform object recognition on the preprocessed images to obtain information about abnormal objects on the road;

[0077] Quality assessment step: quantitatively assess the quality of the highway based on the identification results and quality assessment standards. The quantitative assessment is based on the above-mentioned calculation formulas for road surface flatness, damage rate and facility integrity;

[0078] The result storage and display step stores the evaluation results and related images and displays them to the user for easy query and analysis.

[0079] In the present invention, for long-distance highway detection, multiple groups of image acquisition devices are used to work together, and the acquired images contain overlapping areas for subsequent splicing, specifically:

[0080] Multiple groups of image acquisition devices are distributed along the highway at a certain distance. The acquisition areas of two adjacent groups of devices overlap by 10%. The acquisition parameters of each group of devices (such as shooting angle, focal length, resolution, etc.) are pre-adjusted according to their location and local features of the highway to ensure that the acquired images can fully cover the highway pavement, roadbed and ancillary facilities.

[0081] During the acquisition process, a time synchronization mechanism is used to ensure the time consistency of images collected by different devices, which facilitates the subsequent accurate stitching and analysis of the images. The time synchronization accuracy can reach 10 milliseconds. At the same time, wireless communication technology is used to achieve collaborative work between devices. Status information and acquisition instructions can be transmitted between devices in real time to ensure the efficiency and reliability of the entire acquisition process.

[0082] The present invention has a powerful protection mechanism in the data transmission step to deal with the problem of transmission interruption and ensure the integrity of the data. Specifically, a sophisticated automatic reconnection mechanism is set inside the data transmission module. During the operation of the system, whether the wireless or wired connection is interrupted, the system can quickly detect it. At this time, take action according to the preset reconnection strategy. The reconnection time interval is carefully set within a reasonable range of 3-10 seconds, which is determined by comprehensively considering the common cycle of network fluctuations and avoiding excessive frequent requests from burdening the network. In addition, in order to avoid consuming too many resources by trying to reconnect without limit, the system will make up to 5 reconnection attempts to re-establish the connection so that data transmission can return to normal as soon as possible.

[0083] At the same time, in order to further ensure that data is not lost during transmission interruptions, a data cache module is specially equipped. When the transmission is interrupted, the data that has not been successfully transmitted will be quickly and temporarily stored in the cache. The cache has a capacity of up to 10GB, which is enough to accommodate the amount of data accumulated during a certain interruption time. The cache adopts a first-in-first-out (FIFO) storage strategy, which simulates the natural order of data transmission and ensures that after the reconnection is successful, the data can continue to be transmitted in the original order, as if there has never been an interruption, thereby effectively avoiding the problem of data loss caused by transmission interruptions and providing a solid guarantee for the data transmission stability of the entire system.

[0084] The present invention uses an efficient local contrast enhancement algorithm to improve image quality for images with uneven illumination. Specifically, an algorithm based on local adaptive histogram equalization (CLAHE) is used. During the implementation of the algorithm, the image is first divided into a plurality of pixel blocks of suitable size and non-overlapping. These pixel blocks serve as independent processing units, and each is subjected to histogram equalization processing. In this way, the contrast within each pixel block can be significantly enhanced, the details that were originally blurred due to uneven illumination can be clearly displayed, the distribution of different gray levels is more reasonable, and the local image information is richer.

[0085] However, this block processing may produce artificial traces between adjacent pixel blocks, such as discontinuity of grayscale values, which will affect the overall visual effect of the image. To solve this problem, the bilinear interpolation algorithm is introduced. This algorithm smoothes the boundaries between adjacent pixel blocks and adjusts the grayscale value by calculating the weighted average of the pixels around the boundary, so that the boundary transition is natural. In this way, the entire image can maintain a natural visual effect while enhancing the contrast, and the object features in the image are easier to identify. This is of great significance for the subsequent object recognition and quality assessment links, and can improve the accuracy of recognition and the reliability of quality assessment.

[0086] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A highway quality automatic detection system based on object recognition, characterized in that: include: An image acquisition module is used to acquire images of the road surface, roadbed, and ancillary facilities. The image acquisition equipment includes a high-definition camera and a camera device mounted on a drone. The installation position and angle are adjustable, and the acquisition frequency is set according to the detection speed and accuracy requirements; The data transmission module transmits the collected image data to the processing center in real time, supporting wired and wireless transmission methods. Wireless transmission includes 4G and 5G networks; The image preprocessing module preprocesses the received images, including denoising, contrast enhancement, and geometric distortion correction. It uses an adaptive filtering algorithm to remove image noise, enhances contrast based on histogram equalization, and corrects geometric distortion through perspective transformation. The object recognition module uses the trained deep learning model to identify the detection objects in the image. The objects include road cracks, potholes, damage, sign wear, roadbed settlement, and guardrail damage. The model is built based on the convolutional neural network (CNN) and trained with labeled data. The quality assessment module assesses the road quality based on the identified object information and predefined quality assessment standards. The assessment indicators include road surface smoothness P, damage rate D, and facility integrity F. The module outputs the road quality assessment results and a detailed report. The road surface smoothness calculation formula is: L i is the length of cracks or potholes on the road that exceed the threshold, n is the number of cracks or potholes, L total is the length of the detection section; the damage rate calculation formula is: S d is the road damage area, S total is the total road surface area; the calculation formula for facility integrity is: N g is the number of intact road ancillary facilities, N total is the total number of ancillary facilities; The storage and display module stores the collected original images, processed images and evaluation results, provides users with a visual interface to display evaluation reports and images, and supports historical data query and comparative analysis.

2. According to the object recognition-based automatic road quality detection system of claim 1, the high-definition camera of the image acquisition module has an optical image stabilization function and an automatic focus function, as follows: The optical image stabilization function uses a suspended optical image stabilization lens group. The gyroscope sensor detects the camera's shaking direction and amplitude in real time, drives the image stabilization lens group to move in a plane perpendicular to the optical axis, compensates for the light deviation caused by shaking, and effectively reduces image blur caused by factors such as road bumps and vehicle vibrations. The autofocus function is based on a combination of phase detection autofocus technology and contrast detection autofocus technology. Phase detection autofocus uses paired pixels on the image sensor to detect the phase difference of light, and contrast detection autofocus fine-tunes the focus by analyzing the contrast changes of the picture. The adaptive light intensity range is from 0.1lux to 100000lux, and the shooting distance range is from 0.5m to 100m. The camera device carried by the drone can automatically fly and collect images according to the preset route. The drone flight control system can achieve high-precision positioning and navigation based on the global positioning system GPS, inertial navigation system INS and visual positioning system VPS. The preset route can be planned and adjusted through the ground control station software. During the flight, the flight attitude and speed can be automatically adjusted according to real-time environmental information.

3. The road quality automatic detection system based on object recognition as claimed in claim 1, characterized in that: The data transmission module uses encryption protocols during wireless transmission to ensure the security of data transmission, specifically: The Advanced Encryption Standard AES algorithm is used to encrypt the transmitted data. The image data is encrypted block by block according to the key generated by the system. The key length is 256 bits. The plaintext data is encrypted into ciphertext at the data sending end and transmitted through the wireless channel. The receiving end uses the same key for decryption. Combined with the security mechanism in the wireless transmission protocol, IPSec-based VPN technology is used in 4G networks, and its native security architecture is used in 5G networks. Wired transmission uses optical fiber transmission to improve transmission bandwidth and anti-interference ability. The optical fiber uses single-mode optical fiber with a core diameter of 9μm, a cladding diameter of 125μm, and an operating wavelength of 1310nm or 1550nm. The optical transmitter converts the electrical signal into an optical signal for transmission in the optical fiber, and the optical receiver restores the optical signal to an electrical signal at the receiving end. The transmission process can effectively resist electromagnetic interference, and the supported transmission bandwidth can reach more than 10Gbps.

4. According to the object recognition-based highway quality automatic detection system of claim 1, the image preprocessing module automatically selects a suitable filtering algorithm according to the image noise type during the denoising process, specifically: When the image noise is Gaussian noise, the Wiener filtering algorithm is used to estimate the local mean and variance of the image, calculate the optimal filter coefficient for each pixel, remove the Gaussian noise, and maintain the image details. For salt and pepper noise, the median filtering algorithm is used to replace the gray value of each pixel in the image with the median gray value of the pixels in its neighborhood, thus eliminating the salt and pepper noise without destroying the edges and details of the image. If mixed noise is detected in the image, an adaptive weighted median filter algorithm is used to dynamically adjust the weights of the median filter and the mean filter according to the distribution of noise in the pixel neighborhood to effectively remove the mixed noise. Geometric distortion correction automatically calculates correction parameters by identifying feature points in the image. The specific process is as follows: first, the scale-invariant feature transform SIFT algorithm or the accelerated robust feature SURF algorithm is used to extract feature points in the image, and the corresponding feature points in images of different perspectives are found through the descriptor-based nearest neighbor matching algorithm. Then, based on the coordinate relationship of these feature point pairs, the perspective transformation model or the affine transformation model is used to calculate the parameters required for geometric distortion correction, thereby achieving accurate correction of image geometric distortion.

5. A method for applying the road quality automatic detection system based on object recognition according to any one of claims 1 to 4, characterized in that: The following steps are involved: Image acquisition step, using image acquisition equipment to obtain image data of the highway, equipment parameters and acquisition path are set according to the highway type and detection requirements; Data transmission step: transmitting the collected image data to the processing center to ensure the stability of the transmission process and the integrity of the data; Image preprocessing step: preprocess the image to improve the image quality for subsequent analysis; In the object recognition step, the trained deep learning model is used to perform object recognition on the preprocessed images to obtain information about abnormal objects on the road; Quality assessment step: quantitatively assess the quality of the highway based on the identification results and quality assessment standards. The quantitative assessment is based on the above-mentioned calculation formulas for road surface flatness, damage rate and facility integrity; The result storage and display step stores the evaluation results and related images and displays them to the user for easy query and analysis.

6. According to the method for automatic highway quality detection based on object recognition in claim 5, in the image acquisition step, for long-distance highway detection, multiple groups of image acquisition devices are used to work together, and the acquired images contain overlapping areas, specifically: Multiple groups of image acquisition devices are distributed along the highway at a certain distance. The acquisition areas of two adjacent groups of devices overlap by 10%. The acquisition parameters of each group of devices are pre-adjusted according to their location and local characteristics of the highway. During the acquisition process, the time synchronization mechanism is used to ensure the time consistency of images collected by different devices, and the images are accurately stitched and analyzed. The time synchronization accuracy can reach 10 milliseconds. At the same time, wireless communication technology is used to achieve collaborative work between devices, and status information and acquisition instructions can be transmitted in real time between devices.

7. The method for automatic detection of road quality based on object recognition according to claim 5, characterized in that: In the data transmission step, when the transmission is interrupted, it has automatic reconnection and data caching functions to ensure that data is not lost, as follows: The data transmission module has a built-in automatic reconnection mechanism. When a wireless or wired connection is detected to be disconnected, it will try to reconnect every 3-10 seconds according to the preset reconnection strategy, and will try to reestablish the connection automatically up to 5 times. At the same time, it is equipped with a data cache module. When the transmission is interrupted, the data to be transmitted will be temporarily stored in the cache. The cache capacity is 10GB, and it adopts a first-in-first-out FIFO storage strategy. After the reconnection is successful, the data before the interruption can continue to be transmitted.

8. According to the method for automatic road quality detection based on object recognition in claim 5, in the image preprocessing step, for images with uneven illumination, a local contrast enhancement algorithm is used to improve the image quality, specifically: The algorithm based on local adaptive histogram equalization (CLAHE) is used to divide the image into multiple non-overlapping pixel blocks, and histogram equalization is performed on each pixel block to enhance the contrast within each small block. The bilinear interpolation algorithm is used to smooth the boundaries between adjacent blocks, so that the entire image maintains a natural visual effect while enhancing the contrast and improving the recognizability of object features in the image.

Images