A method for locating the mileage pile number of a road inspection target

Abstract

The present invention relates to a method for positioning the mileage pile number of a road inspection target. The method collects data through a self-developed road inspection system (RIS); uses YOLOv5 and PaddleOCR to detect mileage piles and extract their text information; uses semi-global block matching (SGBM) and an improved U-V disparity algorithm to generate a depth map; and calculates preliminary mileage pile number positioning information of the inspection data collection point based on automatic identification of mileage piles, mileage pile text information, binocular depth of field images, and the distance between the mileage piles and the collection points. After supplementary correction, the precise mileage positioning information of the inspection data collection point trajectory is output. The method integrates binocular stereo vision, satellite positioning technology, target detection algorithm, and text recognition algorithm into road mileage positioning; and can realize automated and intelligent inspection target mileage number positioning.

Description

Technical Field

[0001] The invention relates to the technical field of road traffic detection, and in particular to a method for positioning the mileage pile number of a road inspection target highway. Background Art

[0002] Road management departments need an accurate data list containing the road mileage locations of abnormal objects in order to carry out appropriate processing and ensure traffic safety and transportation infrastructure are in normal condition. At present, the parallax calculation of stereo matching in complex road traffic environments is seriously affected by noise, making it difficult to count UV parallax, resulting in difficulty in extracting the parallax value of road mileage piles. Summary of the invention

[0003] In view of this, the purpose of the present invention is to provide a method for locating the mileage post number of a highway target for road inspection, which collects multi-source road data through a road detection system; adopts the YOLOv5 deep learning algorithm to train a mileage post recognition model; based on the deep learning recognition results, Paddle OCR is used to extract text information on the mileage post; semi-global block matching (SGBM) is used for binocular stereo matching, and the UV disparity algorithm is used to calculate the disparity of each pixel in the image; calculates the road mileage of the data collection point, supplements the mileage information of the uncollected place and performs mileage correction.

[0004] To achieve the above object, the present invention adopts the following technical solution: a method for locating the mileage pile number of a road inspection target highway, comprising the following steps:

[0005] Step S1: The hardware composition of the road detection system RIS includes a binocular stereo camera, a GNSS receiver, an inertial measurement unit IMU and a device host;

[0006] Step S2: Detection and optical character recognition of mileage posts. YOLOv5 is suitable for detecting and classifying mileage posts, and PaddleOCR is used to extract text information on mileage posts.

[0007] Step S3: generate a depth map in the actual scene, use semi-global block matching (SGBM) for binocular stereo matching, and use the UV disparity algorithm to calculate the disparity of each pixel in the image; calibrate the detection area of ​​the image based on the obtained depth map and data acquisition interval;

[0008] Step S4: Mileage positioning of road inspection, calculate the road mileage of the data collection point according to the mileage information on the mileage post and the actual depth information of the mileage post; supplement the road mileage of other collection points where the mileage post is not collected with reference to the data collection interval distance; perform mileage correction to make the actual distance between the two mileage posts consistent with the theoretical distance; after the mileage supplement and correction, output the road inspection mileage positioning.

[0009] In a preferred embodiment: step S2 specifically includes the following steps:

[0010] Step S21: Establish a database: A total of 10,000 pictures of urban roads and suburban highways are collected using a binocular road intelligent perception system; 8,000 pictures are used for model training, and the rest are used for verification. The image size is adjusted to 896*896;

[0011] Step S22: Model training: Import the training library data into the YOLOv5 model for model training. The training process starts with an initial learning rate of 0.00906; the exponential decay rate is 0.0005 for the first 50,000 cycles and drops to 0.00148 for the next 10,000 cycles to obtain stable convergence; after 300 rounds of training, the generalization performance of the model is verified;

[0012] Step S23: Text information extraction: The ultra-lightweight OCR system PaddleOCR is used to extract text information, and 700 road scene images collected by RIS are used for verification.

[0013] In a preferred embodiment: the step S3 specifically includes the following steps:

[0014] Step S31: Transect all pixels and mark the point with the maximum difference dmax_i) for each row of the matching area, as shown in Formulas 1 and 2; the maximum disparity in each row is used to fill the pixels without disparity value, as shown in Formula 3;

[0015]

[0016]

[0017]

[0018] Among them, (x1, y1) is the coordinate of the upper left corner of the matching area; (x2, y2) is the coordinate of the lower right corner of the matching area; δ ij is the disparity of the pixel in the i-th row and j-th column; Δ nm is the disparity matrix of the matching area; and Δ′ nm is the improved disparity matrix;

[0019] Step S32: Based on the similar triangle theory, the image coordinate system is converted into the camera coordinate system; the distance from the camera to the object on the imaging plane is calculated according to Formula 4; the world coordinates of the object are determined according to Formulas 5 and 6;

[0020]

[0021]

[0022]

[0023] Among them, Z W is the distance from the camera to the object on the image plane; b is the distance between the two cameras; f is the focal length; (x L ,y L ) are pixels on the left image.

[0024] In a preferred embodiment: the step S4 specifically includes the following steps:

[0025] Step S41: Uplink or downlink determination is as follows;

[0026] Step S411: when the kilometer mark appears on the collected picture for the first time, extract and record the road number and mileage information therein;

[0027] Step S412: Continue to search for mileage posts from the following pictures. When the first 100-meter mark appears, if the 100-meter mileage is "9", the detection is on the downhill road; if the 100-meter mileage is "1", the detection is on the uphill road;

[0028] Step S413: When the next kilometer post appears, if its road number is the same as the last one, it means that there is no road change here; if its road number is different from the last one, it means that there is a road change, the new road number is recorded, and the process returns to step S411;

[0029] Step S42: RIS collection point mileage is calculated as follows;

[0030] Step S421: Obtain mileage information at the road collection point

[0031] There is a distance between the collection point and the mileage post, so the mileage information on the mileage post cannot be used as the road mileage of the collection point; the road mileage of the mileage post obtained by the collection point is determined by referring to Formula 7;

[0032]

[0033] Where P m The road mileage of the mileage post obtained for the collection point; M is the kilometer mileage information on the kilometer sign; m is the hundred-meter mileage information on the hundred-meter sign; D is the distance between the collection point and the mileage post, obtained from the field depth map;

[0034] Step S422: Road collection points where mileage posts are not obtained

[0035] The collection points between two mileposts do not have mileposts; the road mileages of these collection points are determined based on the collection points with road mileages and the data collection interval, as shown in Formula 8;

[0036]

[0037] In the formula, The road mileage for which the mileage post is not captured at the collection point'; is the road mileage of the last collection point; I i is the data interval between two acquisition points;

[0038] Step S423: Road mileage of the collection point before the first mileage post detection

[0039] Reversely calculate the road mileage based on the previous mileage posts and data collection intervals, as shown in Formula 9;

[0040]

[0041] In the formula, The road mileage of the collection point before the first detection mileage post; is the road distance to the next collection point; I i is the data interval between two acquisition points;

[0042] Step S43: The mileage information is supplemented and corrected as follows;

[0043] Step S431: Supplement the missed inspection of mileage posts

[0044] Dim light or tree cover may cause mileage posts to be missed. If there is no road change and the mileages of the two detected mileage posts and the adjacent mileage posts are not continuous, it means that the mileage posts are missed. The location of the missed mileage posts is determined according to the movement distance of the RIS.

[0045] Step S432: Calculation of road mileage in the tunnel

[0046] If the number of received satellites is less than 5, the moving distance is calculated based on the initial speed at the beginning of the tunnel; when the number of received satellites is equal to or greater than 5, the distance trigger mode based on GNSS data is restored; the tunnel section needs to supplement the mileage piles for missed inspection;

[0047] Step S433: Calculation of road mileage in the tunnel

[0048] Correct the distance between the two mileposts to make the mileage positioning consistent with the theoretical distance of the mileposts; first, calculate the distance between the two mileposts according to the moving distance of the RIS, as shown in formula 10; then, calculate the correction coefficient through formula 11; finally, calibrate the road mileage of the collection point through formula 12;

[0049]

[0050]

[0051]

[0052] Where D RIS is the distance between two mileposts based on the RIS moving distance; n is the number of collection points between the two mileposts; i is the number of pictures; I i is the data interval between two acquisition points; c f is the correction factor; D m It refers to the theoretical distance between the text information on two adjacent mileposts; P' i m P is the corrected road mileage of the collection point; i m It is the uncorrected road mileage of the collection point.

[0053] Compared with the prior art, the present invention has the following beneficial effects: the present invention proposes a disparity map of left and right binocular images of a real road scene obtained by calculating based on the mileage post model recognition result and the SGBM stereo matching algorithm; the disparity within the mileage post recognition detection frame is retained, and the disparity outside the detection frame is assigned to zero; in view of the problems of inaccurate position of the mileage post recognition detection frame and the influence of noise on the matching algorithm, this example further improves the disparity map: all disparity values ​​are traversed within the rectangular detection frame of the target mileage post, the maximum disparity value is found, and then assigned to all pixels in the detection frame, thereby filling the disparity holes within the detection frame and obtaining a better disparity map. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A road foreground image which is an example image of a depth image in a preferred embodiment of the present invention;

[0055] Figure 2 A road depth map which is an example of a depth map in a preferred embodiment of the present invention;

[0056] Figure 3 This is a flow chart of road mileage calculation for collection points in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0057] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0058] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0059] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0060] A method for locating the mileage pile number of a road inspection target highway is as follows:

[0061] (1) Automatic identification of mileage posts based on the YOLOv5 model

[0062] This example uses the YOLOv5 model that achieves the best coordination between target detection accuracy and detection speed. Since YOLO can be used as a real-time target detector, it is suitable for road inspection. According to the analysis of the YOLO sequence, the performance of YOLOv5 is better than other products. With the increase of residual units and convolution kernels, the accuracy increases, while the speed decreases. However, according to the model size, only YOLOv5s and YOLOv 5m can be applied to ARM embedded development. Therefore, in order to obtain better recognition results, this study selects the high-precision YOLOv5m for mileage pile detection and classification.

[0063] (2) Improve the disparity calculation of road foreground images based on deep learning model combined with SGBM stereo matching algorithm

[0064] This example uses the SGBM algorithm to complete stereo matching of binocular images, thereby calculating the disparity of the left and right images; and based on the YOLOv5 model, mileage post recognition is performed to obtain the position information of the mileage post recognition detection frame; combining the mileage post recognition results and the disparity map, the disparity within the target mileage post recognition detection frame is retained, and the disparity outside the detection frame is assigned to zero; considering the inaccurate position of the detection frame and the influence of noise on the calculated disparity, the disparity map is further improved, all disparity values ​​are traversed within the detection frame, the maximum disparity value is found, and the maximum disparity value in the detection frame is assigned to all pixels in the detection frame to fill the disparity holes within the detection frame and improve the accuracy of mileage post positioning.

[0065] (3) Road detection mileage positioning

[0066] According to the road mileage of the collection point and the on-site depth map, the road mileage of the detection target in the image is calculated.

[0067] refer to Figures 1 to 3 , specifically including the following steps:

[0068] Step S1: The hardware composition of the road detection system (RIS) includes a binocular stereo camera, a GNSS receiver, an inertial measurement unit (IMU) and a device host;

[0069] Step S2: Detection and optical character recognition of mileage posts. YOLOv5 is suitable for detecting and classifying mileage posts, and PaddleOCR is used to extract text information on mileage posts.

[0070] Step S3: Depth map generation in the actual scene, semi-global block matching (SGBM) is used for binocular stereo matching, and UV disparity algorithm is used to calculate the disparity of each pixel in the image. According to the obtained depth map and data collection interval, the detection area of ​​the image is calibrated;

[0071] Step S4: Mileage positioning of road inspection. Calculate the road mileage of the data collection point based on the mileage information on the mileage post and the scene depth information of the mileage post. The road mileage of other collection points where the mileage post is not collected is supplemented with reference to the data collection interval distance. Perform mileage correction to make the actual distance between the two mileage posts consistent with the theoretical distance. After the mileage supplement and correction, output the road inspection mileage positioning.

[0072] In this example, the specific implementation is as follows:

[0073] The detection and optical character recognition of the mileage posts in step S2, YOLOv5 is suitable for detecting and classifying mileage posts, and Paddle OCR is used to extract text information on the mileage posts, including the following steps:

[0074] Step S21: Establish a database: A total of 10,000 pictures of urban roads and suburban highways are collected using a binocular road intelligent perception system. Among them, 8,000 pictures are used for model training and the rest are used for verification. The image size is adjusted to 896*896;

[0075] Step S22: Model training: The training library data is imported into the YOLOv5 model for model training. The training process starts with an initial learning rate of 0.00906. The exponential decay rate is 0.0005 for the first 50,000 cycles and drops to 0.00148 for the next 10,000 cycles to obtain stable convergence. After 300 rounds of training, the generalization performance of the model is verified.

[0076] Step S23: Text information extraction: The ultra-lightweight OCR system PaddleOCR is used to extract text information, and 700 road scene images collected by RIS are used for verification.

[0077] In step S3, the depth map of the actual scene is generated, the semi-global block matching (SGBM) is used for binocular stereo matching, and the UV disparity algorithm is used to calculate the disparity of each pixel in the image. According to the obtained depth map and data collection interval, the detection area of ​​the image is calibrated, including the following steps:

[0078] Step S31: Transect all pixels and mark each row of the matching area with the point with the maximum disparity (dmax_i), as shown in Formulas 1 and 2. The maximum disparity in each row is used to fill the pixels without disparity values, as shown in Formula 3.

[0079]

[0080]

[0081]

[0082] Among them, (x1, y1) is the coordinate of the upper left corner of the matching area; (x2, y2) is the coordinate of the lower right corner of the matching area; δ ij is the disparity of the pixel in the i-th row and j-th column; Δ nm is the disparity matrix of the matching area; and Δ′ nm is the improved disparity matrix.

[0083] Step S32: Based on the similar triangle theory, the image coordinate system is converted into the camera coordinate system. The distance from the camera to the object on the imaging plane is calculated according to Formula 4. The world coordinates of the object are determined according to Formulas 5 and 6.

[0084]

[0085]

[0086]

[0087] Among them, Z W is the distance from the camera to the object on the image plane; b is the distance between the two cameras; f is the focal length; (x L ,y L ) are pixels on the left image.

[0088] In step S4, the mileage positioning of the road inspection is performed by calculating the road mileage of the data collection point based on the mileage information on the mileage post and the current depth information of the mileage post. The road mileage of other collection points where the mileage post is not collected is supplemented with reference to the data collection interval distance. Mileage correction is performed to make the actual distance between the two mileage posts consistent with the theoretical distance. After the mileage supplement and correction, the output of the road inspection mileage positioning includes the following steps:

[0089] Step S41: Uplink or downlink determination is as follows.

[0090] (1) When the kilometer sign appears on the collected image for the first time, the road number and mileage information are extracted and recorded;

[0091] (2) Continue to search for mileage posts in the following picture. When the first 100-meter mark appears, if the 100-meter mileage is "9", the detection is on the downhill road; if the 100-meter mileage is "1", the detection is on the uphill road;

[0092] (3) When the next kilometer post appears, if its road number is the same as the last one, it means that there is no road change here. If its road number is different from the last one, it means that there is a road change, record the new road number, and return to (1).

[0093] Step S42: The RIS collection point mileage is calculated as follows.

[0094] (1) Obtaining mileage information at road collection points

[0095] There is a distance between the collection point and the mileage post, so the mileage information on the mileage post cannot be used as the road mileage of the collection point. The road mileage of the mileage post obtained by the collection point is determined by referring to Formula 7.

[0096]

[0097] Where P m The road mileage of the mileage post obtained for the collection point; M is the kilometer mileage information on the kilometer sign; m is the hundred-meter mileage information on the hundred-meter sign; D is the distance between the collection point and the mileage post, obtained from the field depth map.

[0098] (2) Road collection points without mileage posts

[0099] The collection points between two mileposts do not have mileposts. The road mileages of these collection points are determined based on the collection points with road mileages and the data collection interval, as shown in Formula 8.

[0100]

[0101] In the formula, The road mileage for which the mileage post is not captured at the collection point'; is the road mileage of the last collection point; I i is the data interval between two acquisition points.

[0102] (3) Road mileage at the collection point before the first mileage pile test

[0103] Normally, the road mileage of the collection points is calculated forward. The road mileage of most collection points is basically determined by referring to the mileage posts and data intervals. However, there are some collection points before the first detection mileage post, and there is no reference before them. Therefore, their road mileage is calculated backwards according to the previous mileage posts and data collection intervals, as shown in Formula 9.

[0104]

[0105] In the formula, The road mileage of the collection point before the first detection mileage post; is the road distance to the next collection point; I i is the data interval between two acquisition points.

[0106] Step S43: The mileage information is supplemented and corrected as follows.

[0107] (1) Supplementary mileage piles missed inspection

[0108] Dim light or tree cover may cause mileposts to be missed. If there is no road change and the mileages of the two detected mileposts and the adjacent mileposts are not continuous, it means that the mileposts are missed. The position of the missed milepost is determined according to the movement distance (data collection interval) of the RIS.

[0109] (2) Calculation of road mileage in tunnels

[0110] The satellite signal is weak in the tunnel. Therefore, the RIS distance trigger mode based on GNSS data is useless in the tunnel. In the present invention, if the number of received satellites is less than 5, the moving distance is calculated based on the initial speed at the beginning of the tunnel. When the number of received satellites is equal to or greater than 5, the distance trigger mode based on GNSS data is restored. The dim light in the tunnel may cause the mileage post to be missed. The tunnel section needs to supplement the mileage post missed detection.

[0111] (3) Calculation of road mileage in tunnels

[0112] The placement of mileage posts refers to the road design on the horizontal plane, but the undulating terrain increases the actual distance between two mileage posts. On the other hand, the construction error and positioning error of GNSS may increase or decrease the actual distance between the posts and the mileage posts. Therefore, it is necessary to correct the distance between the two mileage posts so that the mileage positioning in the present invention is consistent with the theoretical distance of the mileage posts. First, the distance between the two mileage posts is calculated according to the moving distance of RIS, as shown in Formula 10. Then, the correction coefficient is calculated by Formula 11. Finally, the road mileage of the collection point is calibrated by Formula 12.

[0113]

[0114]

[0115]

[0116] Where D RIS is the distance between two mileposts based on the RIS moving distance (based on the distance collected by the autonomous device); n is the number of collection points between the two mileposts; i is the number of images; I i is the data interval between two acquisition points; c f is the correction factor; D m It refers to the theoretical distance between the text information on two adjacent mileposts (the interval on the sign); P' i m P is the corrected road mileage of the collection point; i m It is the uncorrected road mileage of the collection point.

[0117] Preferably, this example combines the independently developed road detection system to propose an accurate, novel, multi-technology integrated road detection mileage positioning method, which improves the efficiency of handling abnormal objects on the road and reduces the risk of traffic accidents.

[0118] At the same time, this example fully considers the fact that the complex traffic road environment affects the robustness of the recognition model. This example specifically establishes a model training database of real road traffic scenes. The diversity of the database makes the mileage pile recognition model more generalizable.

[0119] In particular, this example aims at the fact that the disparity calculation of stereo matching in complex road traffic environments is seriously affected by noise, making it difficult to count UV disparity, which leads to difficulties in extracting disparity values ​​of road mileage posts. This example proposes a disparity map of the left and right binocular images of the real road scene calculated based on the recognition results of the mileage post model and the SGBM stereo matching algorithm; the disparity within the mileage post recognition detection frame is retained, and the disparity outside the detection frame is assigned to zero; in view of the inaccurate position of the mileage post recognition detection frame and the influence of noise on the matching algorithm, this example further improves the disparity map: traverse all disparity values ​​within the rectangular detection frame of the target mileage post, find the maximum disparity value, and then assign it to all pixels in the detection frame, thereby filling the disparity holes within the detection frame and obtaining a better disparity map.

[0120] This example is based on the target detection YOLOv5 model to obtain the detection area of ​​the mileage post; the coordinate information of the mileage post in the image is used to obtain the text information and the disparity values ​​in the horizontal and vertical directions of the corresponding UV disparity map; according to the road mileage of the collection point and the on-site depth map, the road mileage of the detection target in the image is calculated to obtain the road mileage information of the collection point.

[0121] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. A method for locating the mileage pile number of a road inspection target highway, characterized in that: The following steps are involved: Step S1: The hardware composition of the road detection system RIS includes a binocular stereo camera, a GNSS receiver, an inertial measurement unit IMU and a device host; Step S2: Detection and optical character recognition of mileage posts. YOLOv5 is suitable for detecting and classifying mileage posts, and Paddle OCR is used to extract text information on mileage posts. Step S3: generate a depth map in the actual scene, use semi-global block matching (SGBM) for binocular stereo matching, and use the UV disparity algorithm to calculate the disparity of each pixel in the image; calibrate the detection area of ​​the image based on the obtained depth map and data acquisition interval; Step S4: Mileage positioning of road inspection, according to the mileage information on the mileage post and the current depth information of the mileage post, the road mileage of the data collection point is calculated; the road mileage of other collection points where the mileage post is not collected is supplemented with reference to the data collection interval distance; the mileage correction is performed to make the actual distance between the two mileage posts consistent with the theoretical distance; after the mileage supplement and correction, the road inspection mileage positioning is output; the step S4 specifically includes the following steps: Step S41: Uplink or downlink determination is as follows; Step S411: when the kilometer mark appears on the collected picture for the first time, extract and record the road number and mileage information therein; Step S412: Continue to search for mileage posts from the following pictures. When the first 100-meter mark appears, if the 100-meter mileage is 9, the detection is on the downhill road; if the 100-meter mileage is 1, the detection is on the uphill road; Step S413: When the next kilometer post appears, if its road number is the same as the last one, it means that there is no road change here; if its road number is different from the last one, it means that there is a road change, the new road number is recorded, and the process returns to step S411; Step S42: RIS collection point mileage is calculated as follows; Step S421: Obtain mileage information at the road collection point There is a distance between the collection point and the mileage post, so the mileage information on the mileage post cannot be used as the road mileage of the collection point; the road mileage of the mileage post obtained by the collection point is determined by referring to Formula 7; Where P m The road mileage of the mileage post obtained for the collection point; M is the kilometer mileage information on the kilometer sign; m is the hundred-meter mileage information on the hundred-meter sign; D is the distance between the collection point and the mileage post, obtained from the field depth map; Step S422: Road collection points where mileage posts are not obtained The collection points between two mileposts do not have mileposts; the road mileages of these collection points are determined based on the collection points with road mileages and the data collection interval, as shown in Formula 8; In the formula, The road mileage for which the mileage post is not captured at the collection point'; is the road mileage of the last collection point; I i is the data interval between two acquisition points; Step S423: Road mileage of the collection point before the first mileage post detection Reversely calculate the road mileage based on the previous mileage posts and data collection intervals, as shown in Formula 9; In the formula, The road mileage of the collection point before the first detection mileage post; is the road distance to the next collection point; I i is the data interval between two acquisition points; Step S43: The mileage information is supplemented and corrected as follows; Step S431: Supplement the missed inspection of mileage posts Dim light or tree cover may cause mileage posts to be missed. If there is no road change and the mileages of the two detected mileage posts and the adjacent mileage posts are not continuous, it means that the mileage posts are missed. The location of the missed mileage posts is determined according to the movement distance of the RIS. Step S432: Calculation of road mileage in the tunnel If the number of received satellites is less than 5, the moving distance is calculated based on the initial speed at the beginning of the tunnel; when the number of received satellites is equal to or greater than 5, the distance trigger mode based on GNSS data is restored; the tunnel section needs to supplement the mileage piles for missed inspection; Step S433: Calculation of road mileage in the tunnel Correct the distance between the two mileposts to make the mileage positioning consistent with the theoretical distance of the mileposts; first, calculate the distance between the two mileposts according to the moving distance of the RIS, as shown in formula 10; then, calculate the correction coefficient through formula 11; finally, calibrate the road mileage of the collection point through formula 12; Where D RIS is the distance between two mileposts based on the RIS moving distance; n is the number of collection points between the two mileposts; i is the number of pictures; I i is the data interval between two acquisition points; c f is the correction factor; D m It refers to the theoretical distance between the text information on two adjacent mileposts; P' i m P is the corrected road mileage of the collection point; i m It is the uncorrected road mileage of the collection point.

2. A method for locating the mileage pile number of a road inspection target highway according to claim 1, characterized in that: Step S2 specifically includes the following steps: Step S21: Establish a database: A total of 10,000 pictures of urban roads and suburban highways are collected using a binocular road intelligent perception system; 8,000 pictures are used for model training, and the rest are used for verification. The image size is adjusted to 896*896; Step S22: Model training: Import the training library data into the YOLOv5 model for model training. The training process starts with an initial learning rate of 0.00906; the exponential decay rate is 0.0005 for the first 50,000 cycles and drops to 0.00148 for the next 10,000 cycles to obtain stable convergence; after 300 rounds of training, the generalization performance of the model is verified; Step S23: Text information extraction: The ultra-lightweight OCR system PaddleOCR is used to extract text information, and 700 road scene images collected by RIS are used for verification.

3. A method for locating the mileage pile number of a road inspection target highway according to claim 1, characterized in that: The step S3 specifically comprises the following steps: Step S31: Transect all pixels and mark each row of the matching area with the point dmax_i having the maximum disparity, as shown in Formulas 1 and 2; the maximum disparity in each row is used to fill the pixels without disparity value, as shown in Formula 3; Among them, (x1, y1) is the coordinate of the upper left corner of the matching area; (x2, y2) is the coordinate of the lower right corner of the matching area; δ ij is the disparity of the pixel in the i-th row and j-th column; Δ nm is the disparity matrix of the matching area; and Δ' nm is the improved disparity matrix; Step S32: Based on the similar triangle theory, the image coordinate system is converted into the camera coordinate system; the distance from the camera to the object on the imaging plane is calculated according to Formula 4; the world coordinates of the object are determined according to Formulas 5 and 6; Among them, Z W is the distance from the camera to the object on the image plane; b is the distance between the two cameras; f is the focal length; (x L ,y L ) are pixels on the left image.

Images