A road surface disease de-duplication method based on high-precision positioning and deep neural network

Abstract

The present application relates to a kind of road surface disease deduplication method based on high-precision positioning and deep neural network, comprising the following steps: step 1, vehicle-mounted intelligent patrol equipment is equipped with positioning module, vehicle-mounted camera, central industrial computer for acquisition processing and upload, vehicle-mounted intelligent patrol equipment is started before the top-down transformation parameter calibration of vehicle-mounted camera;Step 2, start equipment to collect image and GPS data, and utilize neural network algorithm to identify road surface disease;Step 3, in combination with the GPS of vehicle-mounted camera calibration parameter and vehicle-mounted camera, the pixel coordinates of road disease in image recognition are combined, and the GPS coordinate of road disease is calculated;Step 4, utilize the GPS coordinate of disease and disease type and database data are checked, if repeat, then determine as repeated disease, if not repeat, then determine as new disease, and write into database.It can realize the high-precision positioning and label information of road disease quickly and accurately, and be used for the same disease judgment of routine patrol.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of road engineering, in particular to a road surface disease deduplication method based on high-precision positioning and deep neural networks. BACKGROUND

[0002] After any road is used for a period of time, various damages, deformations and other defects will appear one after another, which are collectively referred to as road surface diseases. The appearance of road surface diseases will affect driving comfort. If road surface diseases are not discovered and disposed of in time, the road surface diseases will continue to evolve under the action of multiple factors such as environment and vehicle load, and may even evolve into serious road surface diseases, affecting driving safety.

[0003] In the prior art, a vehicle-mounted intelligent patrol device is usually used to detect road surface diseases. The vehicle-mounted intelligent patrol device can replace manual work through a GNSS antenna to obtain GPS and a vehicle camera plus a target detection neural network to quickly and frequently detect road surface diseases, timely discover and notify maintenance personnel for repair, reduce the number of road surface diseases, and improve the comfort and safety of road driving.

[0004] At present, intelligent patrol devices are frequently patrolled, and among the road diseases discovered every day, there are the same diseases as those discovered in previous patrols, which are generally referred to as repeated diseases. A large number of repeated diseases are generated at the same time as high-frequency patrols, and a large number of repeated diseases are submitted to maintenance personnel every day. For repeated diseases, manual judgment is required for deduplication, otherwise it will affect the construction plan and construction progress of maintenance personnel and reduce the operation efficiency of the intelligent management and maintenance system. SUMMARY

[0005] In view of the deficiencies in the prior art, the present application aims to provide a road surface disease deduplication method based on high-precision positioning and deep neural networks, which uses high-precision positioning and deep neural networks to establish a unique information for each road disease. The disease discovered each time is matched with the existing information first, and if it is not matched, it is a new disease that needs to be notified to maintenance personnel for repair work, otherwise it is a repeated disease that does not need to be notified to maintenance personnel, greatly improving the operation efficiency of the intelligent management and maintenance system.

[0006] The above application object of the present application is realized by the following technical scheme:

[0007] A road surface disease deduplication method based on high-precision positioning and deep neural networks, comprising the following steps:

[0008] Step 1: The vehicle-mounted intelligent patrol device is equipped with a positioning module, a vehicle camera, and a central industrial computer for collecting, processing and uploading. Before the vehicle-mounted intelligent patrol device is started, the vehicle camera is subjected to a top-down transformation of external parameter calibration.

[0009] Step 2, start the device to collect images and GPS data, and use neural network algorithm to identify road surface diseases;

[0010] Step 3, combine the vehicle-mounted camera calibration parameters and the GPS of the vehicle-mounted camera to calculate the GPS coordinates of the road disease pixel coordinates in the image recognition;

[0011] Step 4, use the GPS coordinates of the disease and the disease type to query the database data, if repeated, it is determined as repeated disease, if not repeated, it is determined as new disease, and written into the database.

[0012] In a preferred example, the application can be further configured as: in step 1, first, the external parameters of the vehicle-mounted camera are calibrated by using a chessboard of known size before the patrol, the chessboard should be located in the middle part of the view angle of the vehicle-mounted camera, and the lower boundary is parallel to the horizontal view angle of the vehicle-mounted camera;

[0013] The pixel coordinates of the corner points of the chessboard and the size of the chessboard are used to calculate the homography of the vehicle-mounted camera, and the pixel position matrix and the external parameter matrix are multiplied to obtain the pixel position coordinates of the pixel point in the bird's eye view angle;

[0014] At the same time, the latitude and longitude coordinates of the current vehicle-mounted camera are known as [lon, lat], the pixel width w and the pixel height h of the collected picture can be obtained, so the pixel coordinates of the four corner points of the picture can be obtained as a = [0, 0], b = [0, h], c = [w, 0], d = [w, h], the ab line constitutes the left boundary of the picture, and the cd line constitutes the right boundary of the picture, and the a1, b1, c1 and d1 are obtained by multiplying the homography and abcd;

[0015] At this time, the intersection of a1b1 and c1d1 can be obtained by using linear equation, which is recorded as cam_pt(xcp,ycp), which is the bird's eye pixel coordinate position of the GNSS antenna;

[0016] Similarly, the bird's eye pixel coordinates cam_mid_pt(xcmp,ycmp) are obtained by doing bird's eye transformation on the lower boundary midpoint of the calibration image, and the bird's eye pixel coordinates cam_sel_pt(xcsp,ycsp) are obtained by translating cam_mid_pt to the right by d meters, because the relationship between the pixel point and the actual distance can be specified when doing external parameter calibration, so the pixel point is specified as equal to the actual 1mm, so ycmp=ycsp, xcsp=ycmp-d*1000 at this time;

[0017] At this time, the homography, cam_pt, cam_mid_pt and cam_sel_pt are the calibration parameters required.

[0018] In a preferred example, the application can be further configured to: in step 2, the pixel rectangular area of the road disease in the picture is recognized by the neural network algorithm, recorded as bbox[xmin, ymin, xmax, ymax], wherein (xmin, ymin) is the upper left pixel coordinate, (xmax, ymax) is the lower right pixel coordinate, and the road disease category name label of the detected target is also obtained.

[0019] In a preferred example, the application can be further configured to: in step 4, the disease deduplication includes the following steps:

[0020] S1, for the road disease, the center pixel coordinates (u, v) of the road disease on the image can be obtained according to the bbox:

[0021]

[0022] S2, the longitude and latitude and azimuth angle information of the picture can be obtained by the GPS module when the intelligent patrol device collects, respectively lon, lat, angle;

[0023] S3, the longitude and latitude coordinates are converted into the geodetic coordinate system, and the geodetic coordinates of the vehicle-mounted camera are obtained as cam_utm(cam_utm_x, cam_utm_y);

[0024] S4, the distance cam_dist is obtained by calculating the distance between the pixel coordinates of the vehicle-mounted camera and the pixel center point in the image, the unit is a pixel point, that is, the actual distance is 1mm, and then the azimuth angle and cam_dist, cam_utm_x, cam_utm_y are used to calculate the geodetic coordinates of the image center point cam_mid_utm(cam_mid_utm_x, cam_mid_utm_y);

[0025]

[0026] The geodetic coordinates cam_sel_utm(cam_sel_utm_x, cam_sel_utm_y) of the image center point right moving d meters of pixel points can be calculated from d, cam_utm_x, cam_utm_y;

[0027]

[0028] S5, at this time, the pixel coordinates have cam_pt, cam_mid_pt and cam_sel_pt, the geodetic coordinates have cam_utm, cam_mid_utm and cam_sel_utm, and the geodetic coordinates and the pixel coordinates have a large difference in order of magnitude, so the mapping matrix M_pt_to_utm of the pixel coordinates and the geodetic coordinates can be solved by taking cam_utm as a reference point (100, 100) to correct the geodetic coordinates, wherein 100 is a constant, and other values can be used;

[0029]

[0030] S6, finally, the pixel coordinates are obtained by using the disease center point (u, v) and the homography to do point multiplication, and then the geodetic coordinates bbox_utm (bbox_utm_x, bbox_utm_y) are obtained by using the mapping matrix M_pt_to_utm to do point multiplication to remove the order of magnitude, and then the geodetic coordinates are restored to obtain dst_utm (dst_utm_x, dst_utm_y); 100 in the formula is the same as 100 in 4.5.

[0031]

[0032] Finally, the geodetic coordinates are converted into longitude and latitude coordinates to output the accurate bbox_gps (dst_lng, dst_lat) of the disease.

[0033] S7, at this time, a high-precision positioning disease GPS position bbox_gps and a disease name label are obtained, and the data is uploaded to a cloud database through a collection device, and the bbox_gps and the label in the database are matched when the road disease is collected next time, if there is no matching, it is considered that the disease is a new disease, and the maintenance personnel can be notified, if the disease is matched, it is considered that the disease has been found before, and it is a repeated disease, and the maintenance personnel does not need to be notified again.

[0034] In a preferred example, the positioning module can be further configured as a GPS device or a GNSS positioning antenna.

[0035] In a preferred example, the central industrial computer is connected with the positioning module and the vehicle-mounted camera respectively, the central industrial computer is powered by the vehicle, and the central industrial computer sends an image collection instruction and completes a calculation step.

[0036] In summary, the present application has at least one of the following beneficial technical effects:

[0037] The method mentioned in the application can realize rapid and accurate high-precision positioning and label information of road diseases by using image analysis and target detection neural network algorithm based on a vehicle-mounted pavement intelligent patrol device, and can be used for daily patrol of the same disease.

[0038] In terms of execution efficiency, the method has fast calculation speed and low calculation cost, greatly reduces the workload of artificial work compared with the existing artificial determination, and more accurately and comprehensively feeds back the intelligent patrol result to road maintenance personnel, thereby improving the work efficiency of the maintenance personnel.

[0039] In terms of economic benefits, the method mentioned in the application greatly improves the uniqueness and accuracy of road disease data without increasing the existing cost.

[0040] In terms of social benefits, the method mentioned in the application can enable the maintenance personnel to quickly and efficiently perform maintenance work, and avoid the situation that the workload of the maintenance personnel is increased due to high-frequency patrol. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 A structural block diagram of a vehicle-mounted intelligent patrol device is shown.

[0042] Figure 2 A schematic diagram of vehicle-mounted camera calibration is shown.

[0043] Figure 3 A flowchart of repeated disease determination is shown.

[0044] Figure 4 A schematic diagram of the working of the vehicle-mounted intelligent patrol device is shown.

[0045] Reference signs: 1, central industrial computer; 2, vehicle-mounted camera; 3, positioning module. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application; obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments of the application, and all other embodiments obtained by those skilled in the art without creative labor on the basis of the embodiments in the application belong to the scope of protection of the application.

[0047] Embodiment one:

[0048] Reference Figures 3-4 A road disease deduplication method based on high-precision positioning and deep neural network is disclosed, comprising the following steps:

[0049] Step 1, the vehicle-mounted intelligent patrol device is provided with a positioning module 3, a vehicle-mounted camera 2, and a central industrial computer 1 for image acquisition, processing, and uploading. Before starting the vehicle-mounted intelligent patrol device, the vehicle-mounted camera 2 is subjected to top-down transformation and parameter calibration.

[0050] Step 2, the device is started to collect images and GPS data, and a neural network algorithm is used to identify road surface diseases.

[0051] Step 3, the road disease pixel coordinates in the image recognition are calculated by combining the calibration parameters of the vehicle-mounted camera 2 and the GPS of the vehicle-mounted camera 2.

[0052] Step 4, the GPS coordinates and disease types of the disease are used to check the database data. If it is repeated, it is determined as a repeated disease. If it is not repeated, it is determined as a new disease, and is written into the database.

[0053] In this embodiment, the positioning module 3 is a GPS device or a GNSS positioning antenna. Referring to Figure 1 , the central industrial computer 1 is connected with the positioning module 3 and the vehicle-mounted camera 2, respectively. The central industrial computer 1 is powered by the vehicle-mounted power supply, and the image acquisition instruction is sent by the central industrial computer 1.

[0054] Embodiment two:

[0055] A road surface disease de-duplication method based on high-precision positioning and deep neural networks, the specific implementation steps are:

[0056] 1. Hardware preparation:

[0057] The vehicle-mounted intelligent patrol device is provided with a high-precision positioning module 3, a camera, and an industrial computer for image acquisition, processing, and uploading. When the positioning module 3 is a GNSS positioning antenna, the high-precision GNSS positioning antenna is installed together with the camera.

[0058] 2. Camera calibration:

[0059] Referring to Figure 2 , first, the extrinsic parameters of the vehicle-mounted camera 2 are calibrated by using a chessboard with known size before the patrol. The chessboard should be located in the middle part of the view angle of the vehicle-mounted camera 2, and the lower boundary should be parallel to the horizontal view angle of the vehicle-mounted camera 2.

[0060] The pixel coordinates of the chessboard corner points and the size of the chessboard are used to calculate the top-down transformation extrinsic parameter matrix homography of the vehicle-mounted camera 2. The pixel position matrix and the extrinsic parameter matrix are multiplied to obtain the pixel position coordinates of the pixel points in the top-down angle.

[0061] At the same time, the current vehicle-mounted camera 2 is known as the latitude and longitude coordinates [lon, lat], the pixel width w and the pixel height h of the collected picture can be obtained, so the pixel coordinates of the four corners of the picture can be obtained at this time a = [0, 0], b = [0, h], c = [w, 0], d = [w, h], the ab line constitutes the left boundary of the picture, and the cd line constitutes the right boundary of the picture, at this time, the a1, b1, c1, d1 after the bird's eye view transformation are obtained by homography and abcd dot multiplication;

[0062] At this time, the intersection of a1b1 and c1d1 can be obtained by using a linear equation, which is recorded as cam_pt(xcp,ycp), which is the bird's eye pixel coordinate position of the GNSS antenna;

[0063] Similarly, the bird's eye view transformation of the midpoint of the lower boundary of the calibration image is obtained, and the bird's eye pixel coordinates cam_mid_pt(xcmp,ycmp) are obtained, and cam_mid_pt is translated to the right by d meters to obtain the bird's eye pixel coordinates cam_sel_pt(xcsp,ycsp), because the relationship between the pixel point and the actual distance can be specified when the external parameter calibration is performed, so 1 pixel point is equal to 1 mm in actual distance, at this time ycmp=ycsp, xcsp=ycmp-d*1000;

[0064] At this time, the homography, cam_pt, cam_mid_pt, and cam_sel_pt are the calibration parameters required.

[0065] 3. Disease identification:

[0066] In intelligent patrol, the target detection neural network algorithm is used, the recognition accuracy is > 90%, the road disease can be effectively identified, and the pixel rectangular area of the road disease in the picture can be obtained through algorithm recognition, which is recorded as bbox[xmin,ymin,xmax,ymax], wherein (xmin,ymin) is the left upper corner pixel coordinate, (xmax,ymax) is the right lower corner pixel coordinate, and the road disease category name label of the detected target is also obtained.

[0067] 4. Disease deduplication:

[0068] S1, for road diseases, the center pixel coordinates (u, v) of the road disease on the image can be obtained according to bbox:

[0069]

[0070] S2, the intelligent patrol equipment can obtain the latitude, longitude and azimuth information of the picture through the GPS module when collecting, which are lon, lat and angle respectively;

[0071] S3, convert the latitude and longitude coordinates into the geodetic coordinate system to obtain the geodetic coordinates of the vehicle-mounted camera 2 as cam_utm(cam_utm_x, cam_utm_y);

[0072] S4, calculate the distance cam_dist by the pixel coordinates of the vehicle-mounted camera 2 and the pixel midpoint of the image midpoint, the unit is a pixel point, that is, the actual distance 1 mm, and then calculate the geodetic coordinates cam_mid_utm(cam_mid_utm_x, cam_mid_utm_y) of the image midpoint by using the azimuth angle and cam_dist, cam_utm_x, cam_utm_y;

[0073]

[0074] The geodetic coordinates cam_sel_utm(cam_sel_utm_x, cam_sel_utm_y) of the image midpoint right-shifted by d meters of pixel points can be calculated by d, cam_utm_x, and cam_utm_y;

[0075]

[0076] S5, at this time, the pixel coordinates are cam_pt, cam_mid_pt, and cam_sel_pt, and the geodetic coordinates are cam_utm, cam_mid_utm, and cam_sel_utm, which are one-to-one corresponding, and since the order of magnitude of the geodetic coordinates and the pixel coordinates is quite different, the mapping matrix M_pt_to_utm of the pixel coordinates and the geodetic coordinates can be solved by taking cam_utm as the reference point (100, 100) to modify the geodetic coordinates, and 100 here is a constant, and other values can be used;

[0077]

[0078] S6, finally, the disease center point (u, v) and the homography are multiplied to obtain the pixel coordinates in the overhead view, and then the mapping matrix M_pt_to_utm is multiplied to obtain the geodetic coordinates bbox_utm(bbox_utm_x, bbox_utm_y) after the order of magnitude is removed, and then the geodetic coordinates are restored to obtain dst_utm(dst_utm_x, dst_utm_y); the 100 in the formula is the same as the 100 in 4.5;

[0079]

[0080] Finally, the geodetic coordinates are converted into latitude and longitude coordinates to output the accurate bbox_gps(dst_lng, dst_lat) of the disease;

[0081] S7, at this time, a high-precision positioning disease GPS position bbox_gps and disease name label are obtained, the data is uploaded to the cloud database through the collection device, and the bbox_gps and label in the database are matched when the road disease is collected next time. If no match is found, the disease is considered to be a new disease, and the maintenance personnel can be notified. If a match is found, the disease is considered to have been found before, and it is a repeated disease, so the maintenance personnel does not need to be notified again.

[0082] The implementation principle of the present application is that the method mentioned in the present application is based on a vehicle-mounted pavement intelligent patrol device, and image analysis and target detection neural network algorithm are used to realize rapid and accurate high-precision positioning and label information of road diseases, and the same disease judgment in daily patrol. In terms of execution efficiency, the method has fast calculation speed and low calculation cost, greatly reduces the workload of manual work compared with the existing manual determination, and more accurately and comprehensively feeds back the intelligent patrol result to the road maintenance personnel. The working efficiency of the maintenance personnel is improved. In terms of economic benefits, the method mentioned in the present application greatly improves the uniqueness and accuracy of road disease data without increasing the existing cost. In terms of social benefits, the method mentioned in the present application can enable the maintenance personnel to quickly and efficiently perform maintenance work, and avoid the situation that the workload of the maintenance personnel is increased due to high-frequency patrol.

[0083] The embodiments of the specific implementation are preferred embodiments of the present application, and do not limit the protection scope of the present application, so that: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A road disease de-duplication method based on high-precision positioning and deep neural network, characterized by: Comprise the following steps: Step 1, the vehicle-mounted intelligent patrol equipment carries the positioning module (3), the vehicle-mounted camera (2), the central industrial computer (1) for collecting, processing and uploading, the vehicle-mounted intelligent patrol equipment is started before the vehicle-mounted camera (2) is carried out the top view transformation parameter calibration; Step 2, start the device to collect image and GPS data, and utilize neural network algorithm to identify road surface disease; Step 3, in combination with the calibration parameters of the vehicle-mounted camera (2), the GPS coordinates of the vehicle-mounted camera (2) and the road disease pixel coordinates in image recognition, the GPS coordinates of the road disease are calculated; Step 4, the GPS coordinates of the disease and the disease type are used to check the database data, if repeated, it is determined as repeated disease, if not repeated, it is determined as new disease, and written into the database; In step 1, first, the known size of the checkerboard is used to calibrate the external parameter of the vehicle-mounted camera (2) before the patrol, the checkerboard should be located in the middle part of the view angle of the vehicle-mounted camera (2), and the lower boundary is parallel to the horizontal view angle of the vehicle-mounted camera (2); The pixel coordinates of the checkerboard corner points and the checkerboard size are used to calculate the vehicle-mounted camera (2) top view transformation homography matrix, and the pixel position matrix and the external parameter matrix are multiplied to obtain the pixel coordinates of the GNSS antenna in the top view; At the same time, the current longitude and latitude coordinates of the vehicle-mounted camera (2) are known as [lon, lat], the pixel width w and the pixel height h of the collected picture can be obtained, so the pixel coordinates of the four corners of the picture can be obtained as a = [0, 0], b = [0, h], c = [w, 0], d = [w, h], the ab line constitutes the left boundary of the picture, and the cd line constitutes the right boundary of the picture, and the a1, b1, c1 and d1 are obtained by multiplying the homography and abcd; At this time, the intersection of a1b1 and c1d1 can be obtained by using linear equation, which is recorded as cam_pt(xcp,ycp), which is the top view pixel coordinate position of the GNSS antenna; Similarly, the top view pixel coordinates cam_mid_pt(xcmp,ycmp) of the lower boundary midpoint of the calibration image are obtained, and the top view pixel coordinates cam_sel_pt(xcsp,ycsp) of cam_mid_pt are obtained by shifting right by d meters, because the relationship between the pixel point and the actual distance can be specified when the external parameter calibration is carried out, so 1 pixel point is equal to 1 mm in actual, so ycmp=ycsp, xcsp=ycmp-dx1000; At this time, the homography, cam_pt, cam_mid_pt and cam_sel_pt are the calibration parameters required.

2. The pavement disease de-duplication method based on high-precision positioning and deep neural network according to claim 1, characterized in that: In step 2, the pixel rectangular area of the road disease in the picture is obtained by neural network algorithm, which is recorded as bbox[xmin,ymin,xmax,ymax], wherein (xmin,ymin) is the left upper corner pixel coordinate, (xmax,ymax) is the right lower corner pixel coordinate, and the road disease category name label is also obtained.

3. The pavement disease de-duplication method based on high-precision positioning and deep neural network according to claim 2, characterized in that: In step 4, the disease deduplication includes the following steps: S1, the road disease center pixel coordinates (u, v) of the road disease on the image are obtained according to the bbox: u = (xmin+xmax) / 2; v = (ymin+yamx) / 2; (xmin, ymin) is the upper left pixel coordinate, (xmax, ymax) is the lower right pixel coordinate; S2, the longitude, latitude and azimuth angle information of the picture can be obtained by the GPS module when the intelligent patrol device collects, respectively lon, lat, angle; S3, the longitude and latitude coordinates are converted into the geodetic coordinate system, and the geodetic coordinates of the vehicle-mounted camera (2) are obtained as cam_utm (cam_utm_x, cam_utm_y); S4, the distance cam_dist is obtained by calculating the distance between the pixel coordinates of the vehicle-mounted camera (2) and the pixel midpoint of the image point, the unit is a pixel point, that is, the actual distance is 1mm, then the azimuth angle and cam_dist, cam_utm_x, cam_utm_y can be used to calculate the geodetic coordinates of the image point cam_mid_utm (cam_mid_utm_x, cam_mid_utm_y); cam_mid_utm_x = cam_utm_x + sin(angle)*cam_dist; cam_mid_utm_y = cam_utm_y + cos(angle)*cam_dist; The geodetic coordinates cam_sel_utm (cam_sel_utm_x, cam_sel_utm_y) of the image point right moving d meters of pixel points can be calculated from d, cam_utm_x, cam_utm_y; cam_sel_utm_x = cam_mid_utm_x + cos(angle)*d; cam_sel_utm_y = cam_mid_utm_y - sin(angle)*d; S5, at this time, the pixel coordinates are cam_pt, cam_mid_pt, cam_sel_pt; the geodetic coordinates are cam_utm, cam_mid_utm, cam_sel_utm three groups of coordinates and one-to-one correspondence, because the order of magnitude of the geodetic coordinates and the pixel coordinates is different, the mapping matrix M_pt_to_utm of the pixel coordinates and the geodetic coordinates can be solved after the geodetic coordinates are modified with cam_utm as the reference point (100, 100), here 100 is a constant; S6, finally, the pixel coordinates are obtained by multiplying the disease center point (u, v) and homography, and then the geodetic coordinates bbox_utm (bbox_utm_x, bbox_utm_y) after deduplication are obtained by multiplying the pixel coordinates and the mapping matrix M_pt_to_utm, and then the geodetic coordinates dst_utm (dst_utm_x, dst_utm_y) are obtained by restoring the order of magnitude; ​ dst_utm_x = bbox_utm_x + cam_utm_x - 100; dst_utm_y = - bbox_utm_y + cam_utm_y + 100; Finally, the geodetic coordinates are converted into longitude and latitude coordinates to output the accurate GPS position of the disease bbox_gps(dst_lng, dst_lat); S7, At this time, a high-precision positioning disease GPS position bbox_gps and disease name label are obtained, and the data is uploaded to the cloud database through the collection device. When the next road disease is collected, the bbox_gps and label in the database are matched. If there is no match, it is considered that the disease is a new disease, and the maintenance personnel can be notified. If it is matched, it is considered that the disease has been found before, which is a repeated disease, and the maintenance personnel does not need to be notified again.

4. The pavement disease de-duplication method based on high-precision positioning and deep neural network according to claim 1, characterized in that: The positioning module (3) is a GPS device or a GNSS positioning antenna.

5. The pavement disease de-duplication method based on high-precision positioning and deep neural network according to claim 1, characterized in that: The central industrial computer (1) is connected with the positioning module (3) and the vehicle-mounted camera (2) respectively, the central industrial computer (1) is powered by vehicle-mounted power supply, and the central industrial computer (1) sends an image acquisition instruction and completes a calculation step.

Images