Method for solving missing detection and repeated detection of road diseases based on computer vision

Abstract

The invention discloses a method for solving missing detection and repeated detection of road diseases based on computer vision, and the method comprises the steps: S1, video collection and stream pulling: carrying out the real-time video stream pulling of an image collection device, and carrying out the real-time frame extraction of the video stream after stream pulling; s2, dynamic frame extraction: performing dynamic frame extraction on the video stream based on different driving speeds of the AI road recognition terminal to obtain a disease image; s3, caching: caching all disease images obtained by dynamic frame extraction in a sliding window mode; and S4, disease point duplicate removal: putting the disease images, which record a plurality of states, of the same disease point described in the disease images obtained in the step S3 into a buffer area, judging whether the disease images are the same disease or not based on the disease features, and if the disease images are the same disease, removing duplicate images and reserving one image. According to the method, the technical problems of missing detection and repeated detection of road disease detection are solved, it is guaranteed that the disease image uploaded to the platform from the inspection box is free of missing and repeated detection, and uniqueness is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of road damage detection, and in particular relates to a solution to missed detection and repeated detection of road damage based on computer vision. Background Art

[0002] Traffic road pavement is prone to different types of defects such as cracks, deformations, and surface damage due to natural factors such as climate and temperature stress, or due to human factors such as vehicle overloading, improper construction, and excessive traffic flow, or due to material factors such as material quality and material ratio design. The emergence of these defects will not only affect driving safety, but will also affect the pavement base, destroy the entire road, and ultimately affect the service life of the road and road traffic. Therefore, it is necessary to regularly inspect the roads, detect pavement defect points, and promptly maintain and repair the discovered defects.

[0003] In the process of implementing the prior art, the inventor found that the traditional method of road disease inspection relies on manual inspection. When the inspector finds the disease, he will stop the car immediately to take pictures and record the disease information. There are problems such as unsafe parking and low efficiency. The traditional method has been replaced by intelligent inspection technology. Intelligent inspection technology usually places the camera equipment and computer vision recognition terminal on the vehicle. During the vehicle's movement, the recognition terminal will recognize the image collected by the camera equipment and record the disease information. However, the camera equipment of the prior art is usually 25 frames / second, 30 frames / second or 60 frames / second. If the recognition terminal extracts each frame of the image and recognizes it, a large number of repeated disease reports will appear. Therefore, the recognition terminal usually extracts the frame of the camera equipment, and the image after the frame is extracted is identified by the computer vision disease algorithm, and the disease point is recorded after it is identified. However, during the actual vehicle inspection, if the frame is extracted at a fixed frequency, when the vehicle speed is slow, the disease point is easy to be repeatedly detected and recorded, and a large number of repeated images are sent back to the system platform. When the vehicle speed is fast, this will cause the disease point to be easily missed. How to ensure that when a vehicle passes through a road surface, only one image of the road surface's diseased points is accurately uploaded to the platform without omission is an urgent problem to be solved.

[0004] Chinese patent document CN 116630911 A discloses a method for uniquely identifying road defects based on machine vision. Its structure includes a visual identification system and a laser identification system. The above two systems are used to identify and collect data on the defects on the road. The image pixel coordinates of the marking points and the corresponding physical coordinates of the motion positioning platform are uploaded to the computer as data, and the unique identification of the road defects is completed through the calibration algorithm program. The road defects are imaged by using a dual calibration method of vision and laser. The computer analyzes the uploaded road defect information through preset image data, which greatly reduces the omission and mislabeling, and can upload the defect data more accurately and timely, reducing the safety hazards of the road. Although the technical solution describes the phenomenon of omission of defect information, the description of the implementation process is not clear, and the technical solution only describes the phenomenon of omission, which cannot solve the problem of repeated reporting of defects.

[0005] Chinese patent document CN 115239969 A discloses a road disease detection method, device, electronic device and storage medium, the method comprising: obtaining target position information and target image corresponding to the road disease to be processed, extracting image features of the target image as target features. Based on the target features, target position information and the disease information stored in the pre-established road disease library, determining whether there are existing road diseases in the road disease library that are repeated with the road disease to be processed, wherein the disease information includes the correspondence between the pre-acquired road disease identifiers and the image features and position information. If there are, it is determined that the road disease to be processed is a repeated disease. If not, the identifier of the road disease to be processed is stored in the road disease library in correspondence with the target features and the target position information. This technical solution determines whether the disease point is compared with the disease information in the disease library by image features, however, the image features are easily affected by various external noise factors such as the distance, angle, light, etc. when the camera shoots the disease point when the vehicle moves, which has a great impact on the accuracy of the results and cannot solve the problems of missed detection and repeated detection.

[0006] Chinese patent document CN 118226457 A discloses a method for accurately counting the amount of road surface defects based on a single-point laser ranging radar, which relates to the field of road detection technology, including monitoring the number of wheel rotations based on a single-point laser ranging radar to obtain ranging data related to the wheel rotation frequency; judging whether the detection vehicle is in a parking state by analyzing the fluctuation characteristics of the laser radar ranging data; obtaining the time point corresponding to each fluctuation cycle based on the laser radar ranging data in the driving state, and then obtaining the image frame sequence of the complete road section from the road surface detection video; calculating the mileage pile number of the location of the road surface defect based on the number of wheel rotations monitored by the laser radar; identifying the image frame sequence to obtain the accurate statistical results of the amount of defects in the complete road section. Although this technical solution avoids repeated statistics and omissions, it can accurately evaluate and count the amount of defects in the detected road section. It is necessary to install a single-point laser ranging radar, which has high hardware costs. In addition, this technical solution depends on the number of wheel revolutions. The wheel size, tire pressure, and driving route will result in inconsistent numbers of wheel revolutions for the same distance traveled, which will affect the accuracy of the results and cannot effectively solve the problems of missed detection and repeated detection.

[0007] Therefore, it is necessary to provide a solution to missed and repeated detection of road defects based on computer vision, so as to solve the technical problems of missed and repeated detection of road defects, ensure that the defect images uploaded from the inspection box to the platform are not missed and there is no repeated detection, ensure uniqueness, achieve comprehensive data detection, avoid missed detection, and eliminate duplicate data to increase the accuracy of defect inspection. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide a solution to missed detection and repeated detection of road defects based on computer vision, so as to achieve comprehensive data detection, avoid missed detection, eliminate duplicate data, and increase the accuracy of defect inspection.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is: the solution to missed detection and repeated detection of road diseases based on computer vision specifically includes the following steps:

[0010] S1 video acquisition and streaming: the image acquisition device is used to stream the video in real time, and frames are extracted in real time from the stream after streaming;

[0011] S2 dynamic frame extraction: Based on the different driving speeds of the AI ​​road recognition terminal, the video stream is dynamically extracted to obtain the disease image;

[0012] S3 cache: caches all the disease images obtained by dynamic frame extraction in a sliding window manner;

[0013] S4: Deduplication of defect points: The defect images recording multiple states of the same defect point described in the defect image obtained in step S3 are placed in a buffer, and based on the defect features, it is determined whether they are the same defect. If they are the same defect, duplicate images are removed and one image is retained.

[0014] By adopting the above technical solution, missed detection is avoided by collecting and pulling streams and real-time dynamic frame extraction, and thus repeated defects occur. By judging whether there are repeated disease images and then removing the repeated disease images, the technical problems of missed detection and repeated detection of road defects are solved, ensuring that the disease images uploaded from the inspection box to the platform are not missed and there is no repeated detection, ensuring uniqueness, achieving comprehensive data detection, avoiding missed detection, and eliminating duplicate data, thereby increasing the accuracy of defect inspection.

[0015] Preferably, the specific steps of step S2 are:

[0016] S21 frame extraction strategy: the image acquisition device is placed in front of the AI ​​road recognition terminal, and the illumination range is 0 meters to 50 meters; due to the different pitch angles of the camera, the distance of the road surface illuminated is also different. Due to the resolution characteristics of the camera, irradiating too far will cause the image of the disease point to be blurred. Usually, the illumination range of 0 meters to 50 meters is the best state;

[0017] S22 frame extraction logic: During the driving process of the AI ​​road recognition terminal, one or more images are extracted within a coverage range of the image acquisition device. Since the frame extraction operation is performed within the camera's coverage range, at least one image is extracted within a camera's coverage range for defect recognition. The frame extraction frequency cannot be too low to avoid the extracted image within a camera's coverage range not being able to completely cover the road surface, resulting in missed detection of defect points. Of course, the same defect point may be framed and photographed multiple times during the driving process of the AI ​​road recognition terminal. We select the image with the best quality that is closest to the center of the image and record it, and discard the rest.

[0018] S23 frame calculation: The frame frequency changes accordingly according to the different driving speeds of the AI ​​road recognition terminal, and the frame interval is set within the irradiation range to obtain all disease images.

[0019] Preferably, the specific steps of step S23 are:

[0020] Assume that the illumination distance of the image acquisition device (binocular camera) is L 照射 , the AI ​​road recognition terminal travels at a speed of V 车速 (V 车速 The GPS is reported to the AI ​​recognition terminal in real time. Then, within the illumination range L of the image acquisition device (binocular camera), 照射The time required for the AI ​​road recognition terminal to travel is t 照射 It is also the maximum interval between two frame extractions; assuming that the frame rate of the image acquisition device (binocular camera) is a constant C 帧率 , then the irradiation range of the image acquisition device (binocular camera) is L 照射 The calculation formula for the maximum interval between two frames is:

[0021]

[0022] In an illumination range of the image acquisition device (binocular camera), assume that the number of frames to be drawn (e.g., 3 or 5 frames to be drawn in an illumination range) is a constant C. 照射抽帧 ;

[0023] From the above, it can be concluded that the frame extraction interval n within the irradiation range 照射抽帧 The formula is:

[0024]

[0025] To avoid missing frames, n 照射抽帧 It should be rounded down to get the frame extraction interval n within the irradiation range. 照射抽帧 .

[0026] Preferably, the specific steps of determining whether the diseases are the same based on the disease characteristics in step S4 are:

[0027] S41: First, determine whether the disease characteristic types are consistent. If they are consistent, go to step S42;

[0028] S42: retaining the defect image closest to the center of the image from among the multiple defect images determined to be the same defect point according to the variance of the defect point's deviation from the center of the image, and deleting the remaining defect images of the defect point.

[0029] Preferably, in step S41, the disease characteristics include disease type and disease size.

[0030] Preferably, the specific steps of step S41 are:

[0031] S411: First, determine whether the disease feature types are consistent. If they are consistent, go to step S412;

[0032] S412: Determine whether the sizes of the defects are consistent. If they are consistent, determine that they are the same defect point.

[0033] Preferably, the types of defects in step S411 include potholes, mesh cracks, damaged joints, bumps, subsidence, strip cracks, pockmarks, rutting and road surface damage; in step S412, the defect size includes length, width and area, and the errors of the length, width and area are set to 5-10%. If the errors of the length, width and area of ​​the defect points in the multiple defect images in the buffer zone are all within the range of 5-10%, and the defect size types are consistent, they are judged to be the same defect point.

[0034] Preferably, the specific steps of calculating the variance of the defect point deviating from the image center position in step S42 are:

[0035] S421: Assume that the coordinates of the lower left corner of the image are (0,0) and the coordinates of the upper right corner of the image are (x0,y0); the coordinates of the lower left corner of the defect point are (x1,y1) and the coordinates of the upper right corner are (x2,y2), and calculate the variance S of the defect point deviating from the center of the image 2 , the formula is:

[0036]

[0037] S422: Calculate the variance S of each defect point in the buffer zone from the center of the image one by one 2 ;

[0038] S423: retain the variance S of the defect points in the multiple defect images that deviate from the image center position 2 The smallest disease point is the diseased image, and the rest of the diseased images are discarded.

[0039] Preferably, in step S1, an AI road recognition terminal is used for inspection to realize video capture and streaming of the image acquisition device (binocular camera); and step S5 is also included: a road is selected for test comparison to verify the missed detection rate and deduplication rate of the AI ​​road recognition terminal.

[0040] Preferably, the inspection system for missed and repeated detection of road defects based on computer vision includes an AI road recognition terminal, an image acquisition device and a positioning device. The image acquisition device is arranged in front of the AI ​​road recognition terminal, and the positioning device is arranged in the AI ​​road recognition terminal and is connected to the image acquisition device.

[0041] Using the above technical solution, the image acquisition device is a vehicle-mounted binocular camera. The system collects road image data through the vehicle-mounted camera, uses the edge AI recognition terminal to perform image analysis, and detects road defects through disease recognition. GPS is used to locate the location of the disease point, and the recognition result is uploaded to the system platform through the 4G / 5G communication network. The inspectors can turn on the equipment and drive to patrol the road.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] (1) High flexibility: Applicable to various road scenes, the binocular camera can automatically calculate the distance in various scenes and maintain high accuracy in scenes with low light, thus improving work efficiency;

[0044] (2) Improved accuracy: Solve the technical problems of missed and repeated detection of road defects, ensure that the defect images uploaded from the inspection box to the platform are not missed and there is no repeated detection, ensure uniqueness, achieve comprehensive data detection, avoid missed detection, and eliminate duplicate data to increase the accuracy of defect inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of a solution to missed detection and repeated detection of road hazards based on computer vision of the present invention;

[0046] Figure 2 The schematic diagram of frame extraction logic of step S2 of the solution to missed detection and repeated detection of road defects based on computer vision of the present invention;

[0047] Figure 3 It is a schematic diagram (sample) of frame extraction of step S2 of the solution to missed detection and repeated detection of road defects based on computer vision of the present invention;

[0048] Figure 4 A schematic diagram of recording the disease state of the solution to missed detection and repeated detection of road diseases based on computer vision of the present invention;

[0049] Figure 5 Nine types of road diseases that are solutions to missed detection and repeated detection of road diseases based on computer vision of the present invention;

[0050] Figure 6 A schematic diagram of the size of a road defect according to the present invention's solution to missed detection and repeated detection of road defects based on computer vision;

[0051] Figure 7 A schematic diagram of the coordinates of road disease points of the solution to missed detection and repeated detection of road diseases based on computer vision of the present invention;

[0052] Figure 8 A GPS location distribution map of road diseases according to the present invention for solving the missed detection and repeated detection of road diseases based on computer vision;

[0053] Fig. 9 A detailed map of road damage obtained by solving the problem of missed detection and repeated detection of road damage based on computer vision of the present invention;

[0054] Fig.10 The system diagram of the present invention is a solution to missed detection and repeated detection of road defects based on computer vision. DETAILED DESCRIPTION

[0055] The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the protection scope of the present invention.

[0056] Example: Figure 1 As shown, the solution to missed detection and repeated detection of road defects based on computer vision specifically includes the following steps:

[0057] S1 video acquisition and streaming: the video stream of the image acquisition device is streamed in real time, and frames are extracted in real time for the video stream after the stream is streamed; in the step S1, an AI road recognition terminal is used for inspection to realize video acquisition and streaming of the image acquisition device (a binocular camera is used in this embodiment);

[0058] S2 dynamic frame extraction: based on the different driving speeds of the AI ​​road recognition terminal (the vehicle in this embodiment), the video stream is dynamically extracted to obtain the disease image;

[0059] The specific steps of step S2 are:

[0060] S21 frame extraction strategy: the image acquisition device (binocular camera) is placed in front of the AI ​​road recognition terminal, i.e., on the front hood of the vehicle, with an illumination range of 0 to 50 meters. Due to the different pitch angles of the camera, the distance of the road surface illuminated is also different. Due to the resolution characteristics of the camera, irradiating too far will cause the image of the diseased point to be blurred. Usually, the illumination range of 0 to 50 meters is the best state.

[0061] S22 frame extraction logic: During the driving process of the AI ​​road recognition terminal (the vehicle in this embodiment), one or more images are extracted within a coverage range illuminated by the image acquisition device (binocular camera); since the frame extraction operation is performed within the illumination range of the camera, at least one image is extracted within one illumination range of the camera for disease recognition, and the frame extraction frequency cannot be too low to avoid the extracted image within one illumination range of the camera failing to completely cover the road surface, resulting in missed detection of disease points; of course, the same disease point may be framed and photographed multiple times during the driving process of the AI ​​road recognition terminal (the vehicle in this embodiment). We select the image with the best quality that is closest to the center of the image of the disease point and make a record, and discard the rest; during the driving process of the AI ​​road recognition terminal (the vehicle in this embodiment), within the 1st, 2nd...m illumination ranges of the camera, each illumination range needs to be frame extracted. Schematic diagram, as shown in Figure 2 As shown:

[0062] S23 frame calculation: according to the different speeds of the AI ​​road recognition terminal (the vehicle in this embodiment), the frame frequency is changed accordingly, and the frame interval is set within the irradiation range to obtain all the disease images.

[0063] The specific steps of step S23 are:

[0064] Assume that the illumination distance of the image acquisition device (binocular camera) is L 照射 , the AI ​​road recognition terminal (the vehicle in this embodiment) travels at a speed of V 车速 (V 车速 The GPS is reported to the AI ​​recognition terminal in real time. Then, within the illumination range L of the image acquisition device (binocular camera), 照射 The time required for the AI ​​road recognition terminal (the vehicle in this embodiment) to travel is t 照射 It is also the maximum interval between two frame extractions; assuming that the frame rate of the image acquisition device (binocular camera) is a constant C 帧率 , then the irradiation range of the image acquisition device (binocular camera) is L 照射 The calculation formula for the maximum interval between two frames is:

[0065]

[0066] In an illumination range of the image acquisition device (binocular camera), assume that the number of frames to be drawn (e.g., 3 or 5 frames to be drawn in an illumination range) is a constant C. 照射抽帧 ;

[0067] From the above, it can be concluded that the frame extraction interval n within the irradiation range 照射抽帧 The formula is:

[0068]

[0069] To avoid missing frames, n 照射抽帧 It should be rounded down to get the frame extraction interval n within the irradiation range. 照射抽帧 ;

[0070] In this embodiment, it is specifically assumed that the binocular camera illumination distance is L illumination = 25 meters, the vehicle speed is V vehicle speed = 5 meters / second, C illumination frames = 4 images are drawn within an illumination range, and the frame rate of the camera is C frame rate = 25 frames / second;

[0071] The frame extraction interval within the binocular camera illumination range is:

[0072]

[0073] Therefore, in order to avoid missing frames, n irradiation frames are rounded down, and after rounding, n irradiation frames = 18 (frames); the schematic diagram of frame extraction is shown in Figure 3 As shown;

[0074] S3 cache: All disease images obtained by dynamic frame extraction are cached in a sliding window manner. The disease status record diagram is as follows Figure 4 As shown; From the above steps, it can be seen that the same irradiation range may be framed multiple times. In other words, the same defect point may be recorded by multiple frames of photos. Therefore, deduplication processing is required to avoid the same defect being reported multiple times, which will cause trouble to the subsequent system processing; From the characteristic analysis of the defect record, it can be seen that when the AI ​​road recognition terminal (the vehicle in this embodiment) is moving, the same defect point is recorded multiple times, as shown in the figure below, so it is necessary to put multiple images of the defect record into the buffer of the sliding window, select the best photo record according to the characteristics of the defect, and discard the rest of the photos;

[0075] S4: Deduplication of defect points: putting defect images recording multiple states of the same defect point described in the defect image obtained in step S3 into a buffer, and judging whether they are the same defect based on the defect features, if they are the same defect, removing duplicate images and retaining one image;

[0076] The specific steps of determining whether the disease is the same based on the disease characteristics in step S4 are:

[0077] S41: First, determine whether the disease feature types are consistent. If they are consistent, go to step S42; in step S41, the disease features include disease type and disease size;

[0078] The specific steps of step S41 are:

[0079] S411: First, determine whether the types of damage characteristics are consistent. If they are consistent, go to step S412. The types of damage in step S411 include potholes, reticular cracks, damaged joints, bumps, subsidence, strip cracks, pitting, rutting and road surface damage. The nine types of damage are as follows: Figure 5 As shown;

[0080] S412: Determine whether the sizes of the disease are consistent. If they are consistent, determine that they are the same disease point;

[0081] S42: retaining the defect image closest to the center of the image from among the multiple defect images determined to be the same defect point according to the variance of the defect point's deviation from the center of the image, and deleting the remaining defect images of the defect point;

[0082] like Figure 6As shown, in step S412, the defect size includes length, width and area, and the errors of length, width and area are all set to be 5-10%. If the errors of length, width and area of ​​the defect points in the multiple defect images in the buffer are all within the range of 5-10%, and the defect size type is consistent, then they are determined to be the same defect point;

[0083] like Figure 7 As shown, the specific steps of calculating the variance of the defect point deviating from the image center position in step S42 are:

[0084] S421: Assume that the coordinates of the lower left corner of the image are (0,0) and the coordinates of the upper right corner of the image are (x0,y0); the coordinates of the lower left corner of the defect point are (x1,y1) and the coordinates of the upper right corner are (x2,y2), and calculate the variance S of the defect point deviating from the center of the image 2 , the formula is:

[0085]

[0086] S422: Calculate the variance S of each defect point in the buffer zone from the center of the image one by one 2 ;

[0087] S423: retain the variance S of the defect points in the multiple defect images that deviate from the image center position 2 The smallest disease point disease image is used, and the rest of the disease images are discarded;

[0088] In this embodiment, it is assumed that there are three defect points in the buffer zone, namely a, b, and c, and the variance S^2 of each defect point is calculated. The data expression is as follows:

[0089]

[0090] S5 Actual measurement comparison: select a road for test comparison to verify the missed detection rate and deduplication rate of the AI ​​road recognition terminal. Specifically: select a road for test comparison, use the AI ​​road recognition terminal for inspection, and record the standard answer through manual investigation, test and check the effect. The road disease GPS location distribution map is as follows: Figure 8 The detailed map of road damage is shown in Fig. 9 As shown in the figure, after manual survey, there were 142 defect points in this road. After detection by the AI ​​road recognition terminal, 2 defect points were missed, with an missed reporting rate of 1.35%; 3 defect points were reported repeatedly, with a repeated reporting rate of 2.07%; the algorithm effect will be optimized through further iterations.

[0091] like Fig.10As shown, the inspection system for missed inspection and repeated inspection of road defects based on computer vision includes an AI road recognition terminal, an image acquisition device and a positioning device. The image acquisition device is arranged in front of the AI ​​road recognition terminal, and the positioning device is arranged in the AI ​​road recognition terminal and connected to the image acquisition device. The image acquisition device is a vehicle-mounted binocular camera. The system collects road image data through the vehicle-mounted camera, uses the edge AI recognition terminal for image analysis, and detects road defects by disease recognition. GPS is used to locate the location of the defect point, and the recognition result is uploaded to the system platform through the 4G / 5G communication network. The inspector can drive the car to inspect the road by turning on the equipment.

[0092] For ordinary technicians in this field, the specific embodiments are only illustrative descriptions of the present invention. It is obvious that the specific implementation of the present invention is not limited to the above-mentioned methods. As long as various non-substantial improvements are made using the method concepts and technical solutions of the present invention, or the concepts and technical solutions of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A solution to missed detection and repeated detection of road hazards based on computer vision, characterized in that: The specific steps include: S1 video acquisition and streaming: the image acquisition device is used to stream the video in real time, and frames are extracted in real time from the stream after streaming; S2 dynamic frame extraction: Based on the different driving speeds of the AI ​​road recognition terminal, the video stream is dynamically extracted to obtain the disease image; S3 cache: caches all the disease images obtained by dynamic frame extraction in a sliding window manner; S4: Deduplication of defect points: The defect images recording multiple states of the same defect point described in the defect image obtained in step S3 are placed in a buffer, and based on the defect features, it is determined whether they are the same defect. If they are the same defect, duplicate images are removed and one image is retained.

2. The method for solving missed detection and repeated detection of road defects based on computer vision according to claim 1 is characterized in that: The specific steps of step S2 are: S21 frame extraction strategy: the image acquisition device is placed in front of the AI ​​road recognition terminal, and the illumination range is 0 meters to 50 meters; S22 frame extraction logic: During the driving process of the AI ​​road recognition terminal, one or more images are extracted within the coverage area of ​​the image acquisition device; S23 frame calculation: The frame frequency changes accordingly according to the different driving speeds of the AI ​​road recognition terminal, and the frame interval is set within the irradiation range to obtain all disease images.

3. The method for solving missed detection and repeated detection of road defects based on computer vision according to claim 2 is characterized in that: The specific steps of step S23 are: Assume that the irradiation distance of the image acquisition device is L 照射 , the AI ​​road recognition terminal travels at a speed of V 车速 , then within the irradiation range L of the image acquisition device 照射 The time required for the AI ​​road recognition terminal to travel is t 照射 It is also the maximum interval between two frame extractions; assuming that the frame rate of the image acquisition device is a constant C 帧率 , then the irradiation range of the image acquisition device is L 照射 The calculation formula for the maximum interval between two frames is: In an illumination range of the image acquisition device, assume that the number of frames to be drawn is a constant C 照射抽帧 ; From the above, it can be concluded that the frame extraction interval n within the irradiation range 照射抽帧 The formula is: To avoid missing frames, n 照射抽帧 It should be rounded down to get the frame extraction interval n within the irradiation range. 照射抽帧 .

4. The method for solving missed detection and repeated detection of road defects based on computer vision according to claim 2 is characterized in that: The specific steps of determining whether the disease is the same based on the disease characteristics in step S4 are: S41: First, determine whether the disease characteristic types are consistent. If they are consistent, go to step S42; S42: retaining the defect image closest to the center of the image from among the multiple defect images determined to be the same defect point according to the variance of the defect point's deviation from the center of the image, and deleting the remaining defect images of the defect point.

5. The method for solving missed detection and repeated detection of road defects based on computer vision according to claim 4 is characterized in that: In step S41, the disease characteristics include the disease type and the disease size.

6. The method for solving missed detection and repeated detection of road defects based on computer vision according to claim 5 is characterized in that: The specific steps of step S41 are: S411: First, determine whether the disease feature types are consistent. If they are consistent, go to step S412; S412: Determine whether the sizes of the defects are consistent. If they are consistent, determine that they are the same defect point.

7. The method for solving missed detection and repeated detection of road defects based on computer vision according to claim 6 is characterized in that: The types of defects in step S411 include potholes, mesh cracks, damaged joints, bumps, subsidence, strip cracks, pockmarks, rutting and road surface damage; in step S412, the defect size includes length, width and area, and the errors of the length, width and area are set to be 5-10%. If the errors of the length, width and area of ​​the defect points in the multiple defect images in the buffer zone are all within the range of 5-10%, they are judged to be the same defect point.

8. The method for solving missed detection and repeated detection of road defects based on computer vision according to claim 6 is characterized in that: The specific steps of calculating the variance of the defect point deviating from the image center position in step S42 are: S421: Assume that the coordinates of the lower left corner of the image are (0,0) and the coordinates of the upper right corner of the image are (x0,y0); the coordinates of the lower left corner of the defect point are (x1,y1) and the coordinates of the upper right corner are (x2,y2), and calculate the variance S of the defect point deviating from the center of the image 2 , the formula is: variance S422: Calculate the variance S of each defect point in the buffer zone from the center of the image one by one 2 ; S423: retain the variance S of the defect points in the multiple defect images that deviate from the image center position 2 The smallest disease point is the diseased image, and the rest of the diseased images are discarded.

9. The method for solving missed detection and repeated detection of road defects based on computer vision according to claim 6, characterized in that: The step S1 uses an AI road recognition terminal to conduct inspections to achieve video capture and streaming of the image acquisition device; and also includes a step S5 actual measurement comparison: selecting a road for test comparison to verify the missed detection rate and deduplication rate of the AI ​​road recognition terminal.

10. The method for solving missed detection and repeated detection of road defects based on computer vision according to claim 6, characterized in that: The inspection system for missed and repeated detection of road defects based on computer vision includes an AI road recognition terminal, an image acquisition device and a positioning device. The image acquisition device is arranged in front of the AI ​​road recognition terminal, and the positioning device is arranged in the AI ​​road recognition terminal and connected to the image acquisition device.

Images