Tunnel submillimeter imaging method based on linear array camera

Abstract

The invention discloses a tunnel submillimeter imaging method based on line-scan digital cameras. The tunnel submillimeter imaging method comprises the following steps: S1, configuring a plurality of line-scan digital cameras and light source modules; s2, starting a mobile platform control system, moving at a constant speed and synchronously acquiring images; s3, performing lens shadow correction, calculating a gain coefficient and performing direction correction; s4, analyzing the inclination angle of the scanning line, and performing inclination correction; s5, performing orthographic correction based on the geometric model, and re-projecting the curved surface image to the UV coordinate system; s6, extracting an overlapping region, performing feature extraction and matching, and calculating relative position offset; s7, performing dynamic planning to determine an optimal suture line, and performing weighted fusion processing on an overlapping region; and S8, brightness, contrast and details are enhanced, and a panoramic image of the inner wall of the tunnel is generated. According to the invention, through cooperative work of a plurality of linear array cameras, dynamic planning splicing and ortho-rectification, efficient and accurate tunnel disease detection and panoramic image generation are realized.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a tunnel submillimeter imaging method based on a linear array camera. Background Art

[0002] As an important transportation infrastructure, tunnels are widely used in railways, highways, subways and other fields. With the continuous increase in tunnel construction, the importance of its maintenance and management has become increasingly prominent. During the long-term use of tunnels, due to the influence of various factors such as the external environment, geological structure, and traffic load, various diseases such as cracks, leakage, block loss, and corrosion are prone to occur. If these diseases are not discovered and treated in time, they will not only affect the service life of the tunnel, but may also pose a serious threat to traffic safety. Therefore, timely and accurate detection of tunnel wall defects is of great significance to ensure the safe operation of the tunnel and extend its service life.

[0003] Traditional tunnel defect detection methods mainly include manual inspection and laser scanning technologies. Manual inspection, as the most traditional and commonly used method, relies on manual inspection of the inner wall of the tunnel, combined with observation, handheld devices, etc. to record and detect defects. Although this method is simple and intuitive, it has great limitations. First, it is inefficient. Manual inspection usually requires a lot of time and manpower to complete the inspection of large-scale tunnels; second, due to the influence of human subjective factors, the identification of defects may not be accurate enough, and it is easy to miss or misdetect, and it is difficult to carry out comprehensive monitoring in a long-term and continuous manner. Therefore, manual inspection has gradually failed to meet the needs of efficient and accurate detection in large-scale tunnel systems.

[0004] In order to improve the efficiency and accuracy of detection, laser scanning technology has been introduced into tunnel disease detection. Laser scanning technology can quickly and accurately obtain the three-dimensional point cloud data of the inner wall of the tunnel through the principle of laser ranging, so as to carry out detailed geometric analysis and disease detection. Compared with traditional manual inspection, laser scanning has higher measurement accuracy and faster scanning speed, and can provide more detailed tunnel inner wall data. However, laser scanning technology also has some shortcomings. First, its equipment cost is high, special laser scanning instruments are required, and operators need to have certain professional skills. Secondly, the measurement results of laser scanning are usually three-dimensional point cloud data. Although it can reflect the geometric morphology of the tunnel structure, the specific type and location of the disease still need further analysis and processing. In addition, the use of laser scanning instruments is limited by the structure and spatial conditions of the tunnel. In some complex or narrow tunnel environments, the application of laser scanning may be limited.

[0005] Existing tunnel image detection methods often lack effective control over image quality. During the image acquisition process, factors such as lighting conditions, camera resolution, and lens distortion will affect image quality. Most existing processing technologies can only rely on post-image enhancement and restoration methods to optimize image quality, and cannot improve image clarity and accuracy from the source. Especially in tunnel environments, due to complex lighting conditions and wall geometry, noise and distortion often occur in the image, which greatly limits the accuracy of disease detection.

[0006] Therefore, how to provide a tunnel submillimeter imaging method based on a linear array camera is a problem that technicians in this field need to solve urgently. Summary of the invention

[0007] One purpose of the present invention is to propose a tunnel submillimeter imaging method based on a linear array camera. The present invention makes full use of linear array cameras, image stitching technology, dynamic programming algorithm and weighted fusion algorithm. Through the collaborative work of multiple cameras and combined with image processing technology, it describes in detail how to efficiently realize tunnel disease detection and accurate tunnel inner wall panoramic image stitching. This method can solve the problems of insufficient field of view of a single camera, unnatural stitching seams and unstable image quality, and has the advantages of high efficiency, high precision and low cost.

[0008] The tunnel sub-millimeter imaging method based on a linear array camera according to an embodiment of the present invention comprises the following steps:

[0009] S1. Configure multiple linear array cameras and light source modules, install the cameras on the bracket of the mobile platform, set the resolution, scanning frequency, exposure time and gain parameters, and adjust the angle and intensity of the light source module;

[0010] S2, start the mobile platform control system to make the platform move at a constant speed along the tunnel track or the specified path, start the linear array camera and the light source module to synchronously collect the original image data of the inner wall of the tunnel, and store the image data;

[0011] S3, performing lens shading correction on the collected original image data, eliminating the problem of uneven brightness distribution by calculating the gain coefficient of each pixel, and performing direction correction;

[0012] S4, based on the corrected image data, analyzing the tilt angle of the camera scan line, and performing tilt correction on the image data through oblique shear transformation;

[0013] S5. Based on the geometric model of the tunnel cross section, the tilt-corrected image data is orthorectified, and the surface image is reprojected into a unified UV coordinate system using a projection algorithm;

[0014] S6, extracting overlapping areas in the orthorectified images, identifying matching points between adjacent images through feature extraction and matching algorithms, and calculating relative position offset parameters between the images;

[0015] S7, using a dynamic programming algorithm to determine the optimal stitching line according to the pixel energy distribution of the overlapping area, and combining the weighted fusion algorithm to process the overlapping area to generate a panoramic image of the tunnel inner wall;

[0016] S8. Adjust the brightness, optimize the contrast, and enhance the details of the generated panoramic image of the tunnel inner wall.

[0017] Optionally, the S3 specifically includes:

[0018] S31, selecting a reference target with uniform brightness, and using a line array camera to collect reference image data under fixed exposure conditions as a reference for correcting image brightness distribution;

[0019] S32, analyzing the brightness value of each row of pixels in the reference image data, calculating the ratio between the actual brightness value of each pixel and the maximum brightness value of the row, and generating a corresponding gain coefficient lookup table:

[0020]

[0021] Among them, G(i,j) is the spatial adaptive gain coefficient matrix, μ ideal is the ideal average brightness, I ref (i, j) is the actual brightness value of the pixel in the i-th row and j-th column in the reference image, and ε is a stabilization term to avoid the denominator being zero;

[0022] S33, for each row of pixels in the original image data, matching the corresponding gain coefficient according to the position, and eliminating the uneven brightness distribution in the image caused by the lens shading effect by adjusting the brightness value;

[0023] S34. By analyzing the edge features of the original image and the geometric features of the tunnel, it is determined whether the image direction deviates from the geometric coordinate system of the inner wall of the tunnel:

[0024]

[0025] Among them, θ is the angle that the image needs to be rotated, R(θ) is the rotation matrix, and is the coordinate of the matching point pair, M is the total number of feature point pairs, and k is the number of the feature point pair currently being processed;

[0026] S35. According to the analysis results, the image is subjected to necessary rotation and mirror transformation to keep it consistent with the geometric coordinate system of the inner wall of the tunnel.

[0027] Optionally, the S4 specifically includes:

[0028] S41, extracting tunnel annular gap pixel points from the corrected image, and using a feature extraction algorithm to obtain the precise coordinates of the annular gap pixel points;

[0029] S42, based on the extracted annular seam pixel points, the inclination angle of the annular seam is fitted using the least square method:

[0030]

[0031] Among them, tan(α) is the tangent value of the tilt angle, n is the total number of feature points extracted, and x i is the horizontal coordinate of the i-th feature point in the image coordinate system, is the average value of the horizontal coordinates of all extracted feature points, y i is the ordinate of the i-th feature point in the image coordinate system, is the average value of the ordinates of all extracted feature points;

[0032] S43, based on the inclination angle of the ring seam obtained by fitting, construct a shear transformation matrix for geometric transformation of the image data:

[0033]

[0034] Among them, T shear is the skew transformation matrix;

[0035] S44, applying a shear transformation to each pixel in the image data, and calculating the coordinates of the pixel after the transformation:

[0036]

[0037] in, represents the position of the pixel in the two-dimensional image, Indicates the new horizontal and vertical coordinates of the pixel point after the oblique shearing transformation;

[0038] S45, adjusting the pixel positions of all scan lines in the image, so that the image scan lines after the oblique shearing transformation are consistent with the geometric features of the inner wall of the tunnel, and the tilt correction is completed.

[0039] Optionally, the S5 specifically includes:

[0040] S51, according to the geometric characteristics of the tunnel cross section, establish a mathematical model of the tunnel inner wall, fit the tunnel cross section into a regular geometric shape, and calculate the position parameters of each point on the tunnel surface, including the center coordinates, radius and arch height;

[0041] S52, extracting the coordinates of each pixel in the tilt-corrected image, and mapping each pixel to a physical point on the tunnel surface model according to the internal and external parameters of the camera and the spatial projection relationship;

[0042] S53, for the physical points projected onto the tunnel surface model, reproject them into a unified UV coordinate system according to the axial and circumferential coordinates of the tunnel, and calculate the coordinates of the image points in the UV plane:

[0043]

[0044] V = Z·cos(φ)+Y·sin(φ);

[0045] Among them, U is the transverse coordinate in the UV plane, V is the longitudinal coordinate in the UV plane, R is the radius of the tunnel, arctan2(Y,X) represents the angle value calculated according to the transverse and longitudinal coordinates of the object point, ΔX and ΔZ are the transverse and longitudinal sampling intervals respectively, Z is the longitudinal height coordinate of the object point in the three-dimensional coordinates of the tunnel, Y represents the longitudinal component of the point in the cross section, and φ represents the local tilt state of the tunnel at a specific acquisition position;

[0046] S54, performing interpolation processing on the image data, resampling the original image data into the UV coordinate system, and filling the pixel gap caused by the curved surface projection;

[0047] S55. Complete the orthorectification of the image, generate a high-resolution orthorectified image that is consistent with the tunnel cross-section geometry model, and eliminate the problem of uneven resolution caused by the curved surface structure.

[0048] Optionally, the S6 specifically includes:

[0049] S61, determining the overlapping area of ​​adjacent images in the orthorectified image, separating the pixel data of the overlapping area by using an image segmentation algorithm, and recording the initial range coordinates of the overlapping area;

[0050] S62, extracting features from the overlapping area, extracting feature points using an accelerated robust feature algorithm, and calculating descriptors of the feature points;

[0051] S63, matching the extracted feature point descriptors, using a nearest neighbor matching method to find similar feature point pairs in the overlapping area of ​​adjacent images to form a matching point set;

[0052] S64, performing robustness screening on the matching point set, eliminating mismatching points using a random sampling consistency algorithm, and optimizing the matching point set by calculating the ratio of internal points;

[0053] S65. Calculate the relative position offset parameters between adjacent images based on the optimized matching point set, and provide the calculated relative position offset parameters to the image stitching algorithm:

[0054]

[0055] Among them, ΔX is the horizontal position offset parameter of two adjacent images, Δy is the vertical position offset parameter of two adjacent images, T is the total number of matching points in the matching point set, x i,2 -x i,1 Indicates the lateral offset, y i,2 -y i,1 represents the longitudinal offset, and i represents the number of the matching point currently calculated in the matching point set.

[0056] Optionally, the S7 specifically includes:

[0057] S71, extracting the overlapping area of ​​adjacent images, and establishing a pixel energy distribution matrix of the overlapping area, wherein the energy value of each pixel is calculated based on the grayscale gradient and brightness difference:

[0058]

[0059] Among them, E(i,j) is the pixel energy distribution matrix, I1(i,j) and I2(i,j) are the gray values ​​of the corresponding pixels in the overlapping area of ​​the two images, τ is the weight parameter, is the gradient magnitude of the pixel;

[0060] S72. Consider the energy distribution matrix as the weight matrix of the graph model, and construct the optimal stitching path from the top to the bottom of the overlapping area based on the dynamic programming algorithm:

[0061]

[0062] Among them, P is the optimal suture path, (x k ,y k ) represents the coordinates of the kth pixel on the path, argmin P It means to find the path that minimizes the objective function among all possible paths. n is the total number of pixels on the optimal stitching path. E(x k ,y k ) is the energy value of the kth pixel on the stitching path;

[0063] S73, segmenting the overlapping area of ​​adjacent images along the optimal stitching path, and determining the pixel areas retained in each image after segmentation;

[0064] S74, combining the weighted fusion algorithm, performing smooth transition processing on the pixel values ​​in the overlapping area:

[0065] I fusion (i,j)=w1(i,j)·I1(i,j)+w2(i,j)·I2(i,j);

[0066] Among them, I fusion(i, j) represents the fusion result of the overlapping area, w1(i, j) represents the credibility of the pixel point from the first image, w2(i, j) represents the credibility of the pixel point from the second image, I1(i, j) represents the pixel value of the first image in the overlapping area, and I2(i, j) represents the pixel value of the second image in the overlapping area;

[0067] S75. Stitch the weighted fusion processed images into a panoramic image to generate a complete panoramic image of the tunnel inner wall.

[0068] The beneficial effects of the present invention are:

[0069] The beneficial effects of the present invention are reflected in many aspects. First, the method of using multiple linear array cameras to work together solves the problem of limited field of view of a single camera. Through multi-angle and multi-view image acquisition, it can ensure that every part of the tunnel wall can be covered, thereby avoiding the blind spot problem that may be caused by single camera scanning. This multi-camera collaborative scanning method greatly improves the coverage and comprehensiveness of tunnel disease detection, and provides more accurate data support for tunnel health management.

[0070] Secondly, the present invention determines the optimal stitching line based on the pixel energy distribution in the overlapping area by introducing a dynamic programming algorithm. Traditional image stitching is often plagued by problems such as unnatural seams in the stitching area and image distortion. The present invention can effectively calculate the optimal stitching path through the optimization method of dynamic programming, ensure the visual continuity of the stitching area, and thus improve the quality of tunnel image stitching. This not only reduces the need for manual intervention, but also makes the stitching results smoother and more seamless, avoiding obvious defects in traditional stitching methods and ensuring the accuracy of disease detection results.

[0071] In addition, the weighted fusion algorithm is used to process the overlapping areas, making the transition between the pixels after the images are stitched more natural. By dynamically adjusting the weights according to the distance from the pixel to the stitching path, a smooth transition between the parts of the image can be achieved, avoiding the visual abruptness of the stitching and improving the overall quality of the image. This weighted fusion method not only optimizes the image quality, but also enhances the applicability of the stitched images in practical applications, especially in complex tunnel environments, and can effectively improve the accuracy of disease detection.

[0072] Finally, the present invention further improves the clarity and accuracy of the collected images through orthorectification technology and image quality enhancement methods. Orthorectification technology can convert curved images into flat images, eliminate image distortion caused by the complexity of the tunnel structure, and ensure the geometric consistency of the image. At the same time, through image quality enhancement, noise caused by factors such as uneven lighting and camera distortion can be removed, further improving the accuracy of disease detection.

[0073] In summary, the present invention achieves efficient and accurate tunnel disease detection through the combination of a series of innovative methods such as collaborative scanning of multiple linear array cameras, dynamic planning stitching algorithm, weighted fusion technology and image quality enhancement, significantly improves the quality and stitching accuracy of panoramic images of the tunnel inner wall, and has significant advantages such as high efficiency, accuracy and cost control, providing strong technical support for intelligent monitoring and health management of tunnels. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0075] Figure 1 This is a flow chart of the tunnel submillimeter imaging method based on a linear array camera proposed in the present invention;

[0076] Figure 2 This is a schematic diagram of the dynamic programming algorithm in the image stitching process of the tunnel submillimeter imaging method based on linear array cameras proposed in the present invention. DETAILED DESCRIPTION

[0077] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0078] refer to Figure 1-2 , a tunnel submillimeter imaging method based on a linear array camera includes the following steps:

[0079] S1. Configure multiple linear array cameras and light source modules, install the cameras on the bracket of the mobile platform, set the resolution, scanning frequency, exposure time and gain parameters, and adjust the angle and intensity of the light source module;

[0080] S2, start the mobile platform control system to make the platform move at a constant speed along the tunnel track or the specified path, start the linear array camera and the light source module to synchronously collect the original image data of the inner wall of the tunnel, and store the image data;

[0081] S3, performing lens shading correction on the collected original image data, eliminating the problem of uneven brightness distribution by calculating the gain coefficient of each pixel, and performing direction correction;

[0082] S4, based on the corrected image data, analyzing the tilt angle of the camera scan line, and performing tilt correction on the image data through oblique shear transformation;

[0083] S5. Based on the geometric model of the tunnel cross section, the tilt-corrected image data is orthorectified, and the surface image is reprojected into a unified UV coordinate system using a projection algorithm;

[0084] S6, extracting overlapping areas in the orthorectified images, identifying matching points between adjacent images through feature extraction and matching algorithms, and calculating relative position offset parameters between the images;

[0085] S7, using a dynamic programming algorithm to determine the optimal stitching line according to the pixel energy distribution of the overlapping area, and combining the weighted fusion algorithm to process the overlapping area to generate a panoramic image of the tunnel inner wall;

[0086] S8. Adjust the brightness, optimize the contrast, and enhance the details of the generated panoramic image of the tunnel inner wall.

[0087] In this implementation, S3 specifically includes:

[0088] S31, selecting a reference target with uniform brightness, and using a line array camera to collect reference image data under fixed exposure conditions as a reference for correcting image brightness distribution;

[0089] S32, analyzing the brightness value of each row of pixels in the reference image data, calculating the ratio between the actual brightness value of each pixel and the maximum brightness value of the row, and generating a corresponding gain coefficient lookup table:

[0090]

[0091] Among them, G(i,j) is the spatial adaptive gain coefficient matrix, μ ideal is the ideal average brightness, I ref (i, j) is the actual brightness value of the pixel in the i-th row and j-th column in the reference image, and ∈ is a stabilization term to avoid the denominator being zero;

[0092] S33, for each row of pixels in the original image data, matching the corresponding gain coefficient according to the position, and eliminating the uneven brightness distribution in the image caused by the lens shading effect by adjusting the brightness value;

[0093] S34. By analyzing the edge features of the original image and the geometric features of the tunnel, it is determined whether the image direction deviates from the geometric coordinate system of the inner wall of the tunnel:

[0094]

[0095] Among them, θ is the angle that the image needs to be rotated, R(θ) is the rotation matrix, and is the coordinate of the matching point pair, M is the total number of feature point pairs, and k is the number of the feature point pair currently being processed;

[0096] S35. According to the analysis results, the image is subjected to necessary rotation and mirror transformation to keep it consistent with the geometric coordinate system of the inner wall of the tunnel.

[0097] In this implementation manner, the S4 specifically includes:

[0098] S41, extracting tunnel annular gap pixel points from the corrected image, and using a feature extraction algorithm to obtain the precise coordinates of the annular gap pixel points;

[0099] S42, based on the extracted annular seam pixel points, the inclination angle of the annular seam is fitted using the least square method:

[0100]

[0101] Among them, tan(α) is the tangent value of the tilt angle, n is the total number of feature points extracted, and x i is the horizontal coordinate of the i-th feature point in the image coordinate system, is the average value of the horizontal coordinates of all extracted feature points, y i is the ordinate of the i-th feature point in the image coordinate system, is the average value of the ordinates of all extracted feature points;

[0102] S43, based on the inclination angle of the ring seam obtained by fitting, construct a shear transformation matrix for geometric transformation of the image data:

[0103]

[0104] Among them, T shear is the skew transformation matrix;

[0105] S44, applying a shear transformation to each pixel in the image data, and calculating the coordinates of the pixel after the transformation:

[0106]

[0107] in, represents the position of the pixel in the two-dimensional image, Indicates the new horizontal and vertical coordinates of the pixel point after the oblique shearing transformation;

[0108] S45, adjusting the pixel positions of all scan lines in the image, so that the image scan lines after the oblique shearing transformation are consistent with the geometric features of the inner wall of the tunnel, and the tilt correction is completed.

[0109] In this implementation manner, S5 specifically includes:

[0110] S51, according to the geometric characteristics of the tunnel cross section, establish a mathematical model of the tunnel inner wall, fit the tunnel cross section into a regular geometric shape, and calculate the position parameters of each point on the tunnel surface, including the center coordinates, radius and arch height;

[0111] S52, extracting the coordinates of each pixel in the tilt-corrected image, and mapping each pixel to a physical point on the tunnel surface model according to the internal and external parameters of the camera and the spatial projection relationship;

[0112] S53, for the physical points projected onto the tunnel surface model, reproject them into a unified UV coordinate system according to the axial and circumferential coordinates of the tunnel, and calculate the coordinates of the image points in the UV plane:

[0113]

[0114] V = Z·cos(φ)+Y·sin(φ);

[0115] Among them, U is the transverse coordinate in the UV plane, V is the longitudinal coordinate in the UV plane, R is the radius of the tunnel, arctan2(Y,X) represents the angle value calculated according to the transverse and longitudinal coordinates of the object point, ΔX and ΔZ are the transverse and longitudinal sampling intervals respectively, Z is the longitudinal height coordinate of the object point in the three-dimensional coordinates of the tunnel, Y represents the longitudinal component of the point in the cross section, and φ represents the local tilt state of the tunnel at a specific acquisition position;

[0116] S54, performing interpolation processing on the image data, resampling the original image data into the UV coordinate system, and filling the pixel gap caused by the curved surface projection;

[0117] S55. Complete the orthorectification of the image, generate a high-resolution orthorectified image that is consistent with the tunnel cross-section geometry model, and eliminate the problem of uneven resolution caused by the curved surface structure.

[0118] In this implementation manner, S6 specifically includes:

[0119] S61, determining the overlapping area of ​​adjacent images in the orthorectified image, separating the pixel data of the overlapping area by using an image segmentation algorithm, and recording the initial range coordinates of the overlapping area;

[0120] S62, extracting features from the overlapping area, extracting feature points using an accelerated robust feature algorithm, and calculating descriptors of the feature points;

[0121] S63, matching the extracted feature point descriptors, using a nearest neighbor matching method to find similar feature point pairs in the overlapping area of ​​adjacent images to form a matching point set;

[0122] S64, performing robustness screening on the matching point set, eliminating mismatching points using a random sampling consistency algorithm, and optimizing the matching point set by calculating the ratio of internal points;

[0123] S65. Calculate the relative position offset parameters between adjacent images based on the optimized matching point set, and provide the calculated relative position offset parameters to the image stitching algorithm:

[0124]

[0125] Among them, ΔX is the horizontal position offset parameter of two adjacent images, Δy is the vertical position offset parameter of two adjacent images, T is the total number of matching points in the matching point set, x i,2 -x i,1 Indicates the lateral offset, y i,2 -y i,1 represents the longitudinal offset, and i represents the number of the matching point currently calculated in the matching point set.

[0126] In this implementation manner, the S7 specifically includes:

[0127] S71, extracting the overlapping area of ​​adjacent images, and establishing a pixel energy distribution matrix of the overlapping area, wherein the energy value of each pixel is calculated based on the grayscale gradient and brightness difference:

[0128]

[0129] Among them, E(i,j) is the pixel energy distribution matrix, I1(i,j) and I2(i,j) are the gray values ​​of the corresponding pixels in the overlapping area of ​​the two images, τ is the weight parameter, is the gradient magnitude of the pixel;

[0130] S72. Consider the energy distribution matrix as the weight matrix of the graph model, and construct the optimal stitching path from the top to the bottom of the overlapping area based on the dynamic programming algorithm:

[0131]

[0132] Among them, P is the optimal stitching path, (xk, y k ) represents the coordinates of the kth pixel on the path, argmin P It means to find the path that minimizes the objective function among all possible paths. n is the total number of pixels on the optimal stitching path. E(xk, y k ) is the energy value of the kth pixel on the stitching path;

[0133] S73, segmenting the overlapping area of ​​adjacent images along the optimal stitching path, and determining the pixel areas retained in each image after segmentation;

[0134] S74, combining the weighted fusion algorithm, performing smooth transition processing on the pixel values ​​in the overlapping area:

[0135] I fusion (i,j)=w1(i,j)·I1(i,j)+w2(i,j)·I2(i,j);

[0136] Among them, I fusion (i, j) represents the fusion result of the overlapping area, w1(i, j) represents the credibility of the pixel point coming from the first image, w2(i, j) represents the credibility of the pixel point coming from the second image, I1(i, j) represents the pixel value of the first image in the overlapping area, and I2(i, j) represents the pixel value of the second image in the overlapping area;

[0137] S75. Stitch the weighted fusion processed images into a panoramic image to generate a complete panoramic image of the tunnel inner wall.

[0138] Embodiment 1:

[0139] In order to verify the feasibility of the present invention in implementation, the present invention was applied to a high-speed railway tunnel in a mountainous area. The tunnel is about 5 kilometers long and is designed as a curved structure. The lighting conditions in some areas are poor, and the application effect of traditional manual inspection and laser scanning methods in this environment is limited. To ensure the safe operation of the tunnel, the tunnel operating company decided to apply tunnel disease detection technology based on linear array cameras to achieve efficient and accurate disease identification and panoramic image stitching.

[0140] This tunnel is located in a mountainous area. The inner wall of the tunnel has a complex shape and contains multiple bends and curved surfaces. Traditional detection methods are prone to missed detection and false detection problems in this special tunnel environment. Especially in the curves and poorly illuminated areas of the tunnel, manual inspections and laser scanning cannot achieve full coverage and cannot provide high-resolution disease data. Therefore, the tunnel operator decided to introduce the method of the present invention and use multiple linear array cameras to work together to perform high-resolution image acquisition and disease detection on the inner wall of the tunnel. In order to achieve accurate stitching of tunnel panoramic images and disease detection, five high-resolution linear array cameras were installed at different locations in the tunnel. The resolution of each camera is 0.2mm / pixel, and there is an overlap in the acquisition areas of adjacent cameras. .

[0141] After collecting image data, all images will be processed through feature extraction and matching algorithms to determine the matching points between adjacent images, and the optimal stitching line will be calculated using a dynamic programming algorithm. The overlapping areas will be processed using a weighted fusion algorithm to ensure that the transition of the stitching area is natural and there are no obvious seams or ghosting. After the stitching is completed, the orthorectification technology is used to perform geometric correction on the image to eliminate the image distortion caused by the curved surface structure of the tunnel, and finally generate a complete panoramic image of the tunnel inner wall.

[0142] The tunnel is located in a mountainous area, with a total length of about 5 kilometers. The widest part of the tunnel wall is 8 meters, the narrowest part is 4 meters, and the maximum curve radius is 120 meters. It is a typical complex tunnel structure. The resolution of each linear array camera is 0.2mm / pixel. Five cameras have collected a total of about 100,000 high-resolution image data, covering the entire length of the tunnel. The overlapping area between the cameras ensures image acquisition without blind spots. Through the spliced ​​images, the present invention detects various diseases on the inner wall of the tunnel, including cracks (maximum width of about 0.8cm), falling blocks (maximum area of ​​about 0.4㎡) and leakage (leakage area of ​​about 6㎡). Compared with traditional manual inspection, the method of the present invention improves the disease recognition rate by about 45%, especially in the tunnel turns and areas with poor lighting, the diseases that the traditional method failed to find are accurately identified. The method of the present invention was used to conduct a comprehensive inspection of a 5-kilometer-long tunnel, and the time used was 60 hours, which saved about 55% of the time compared with the traditional manual inspection method, and significantly improved the detection efficiency.

[0143] Table 1 Tunnel disease detection comparison table

[0144]

[0145] Table 2 Comparison of tunnel disease detection time

[0146]

[0147] Table 1 shows the comparison between the present invention and the traditional manual detection method in terms of the types of defects, the number of detections and the detection efficiency. The method of the present invention significantly increases the number of detections of defects such as cracks, falling blocks and leakage, and the efficiency is increased by about 50%. In particular, in the turning parts of the tunnel and the areas with poor lighting, the traditional method has the problem of missed detection, while the present invention greatly reduces the missed detection rate, from 43 times of the traditional method to 7 times, and the difference in missed detection rate reaches 83.72%. This shows that the method of the present invention has significant advantages in the comprehensiveness and accuracy of tunnel disease detection.

[0148] Table 2 shows the advantages of the method of the present invention in terms of detection time. In the detection of different areas, traditional manual detection takes a long time, while the method of the present invention significantly saves detection time. For example, the traditional method takes 4 hours to detect 1 kilometer, while the method of the present invention only takes 1.2 hours, saving 70% of the time. For a 5-kilometer tunnel, the time saving effect is even more significant, proving the great advantage of the method of the present invention in improving detection efficiency.

[0149] First, in terms of the accuracy and comprehensiveness of disease detection, the present invention is significantly superior to traditional manual detection methods, with a significantly reduced missed detection rate and a significantly improved disease recognition rate. Second, in terms of detection efficiency, the tunnel disease detection time using the present invention is greatly shortened, saving a lot of manpower and time costs. This provides strong technical support for the intelligent maintenance and long-term operation management of tunnels, and has broad application prospects.

[0150] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A tunnel submillimeter imaging method based on a linear array camera, characterized in that: The steps include: S1. Configure multiple linear array cameras and light source modules, install the cameras on the bracket of the mobile platform, set the resolution, scanning frequency, exposure time and gain parameters, and adjust the angle and intensity of the light source module; S2, start the mobile platform control system to make the platform move at a constant speed along the tunnel track or the specified path, start the linear array camera and the light source module to synchronously collect the original image data of the inner wall of the tunnel, and store the image data; S3, performing lens shading correction on the collected original image data, eliminating the problem of uneven brightness distribution by calculating the gain coefficient of each pixel, and performing direction correction; S4, based on the corrected image data, analyzing the tilt angle of the camera scan line, and performing tilt correction on the image data through oblique shear transformation; S5. Based on the geometric model of the tunnel cross section, the tilt-corrected image data is orthorectified, and the surface image is reprojected into a unified UV coordinate system using a projection algorithm; S6, extracting overlapping areas in the orthorectified images, identifying matching points between adjacent images through feature extraction and matching algorithms, and calculating relative position offset parameters between the images; S7, using a dynamic programming algorithm to determine the optimal stitching line according to the pixel energy distribution of the overlapping area, and combining the weighted fusion algorithm to process the overlapping area to generate a panoramic image of the tunnel inner wall; S8. Adjust the brightness, optimize the contrast, and enhance the details of the generated panoramic image of the tunnel inner wall.

2. The tunnel submillimeter imaging method based on a linear array camera according to claim 1, characterized in that: The S3 specifically includes: S31, selecting a reference target with uniform brightness, and using a line array camera to collect reference image data under fixed exposure conditions as a reference for correcting image brightness distribution; S32, analyzing the brightness value of each row of pixels in the reference image data, calculating the ratio between the actual brightness value of each pixel and the maximum brightness value of the row, and generating a corresponding gain coefficient lookup table: Among them, G(i,j) is the spatial adaptive gain coefficient matrix, μ ideal is the ideal average brightness, I ref (i, j) is the actual brightness value of the pixel in the i-th row and j-th column in the reference image, and ε is a stabilization term to avoid the denominator being zero; S33, for each row of pixels in the original image data, matching the corresponding gain coefficient according to the position, and eliminating the uneven brightness distribution in the image caused by the lens shading effect by adjusting the brightness value; S34. By analyzing the edge features of the original image and the geometric features of the tunnel, it is determined whether the image direction deviates from the geometric coordinate system of the inner wall of the tunnel: Among them, θ is the angle that the image needs to be rotated, R(θ) is the rotation matrix, and is the coordinate of the matching point pair, M is the total number of feature point pairs, and k is the number of the feature point pair currently being processed; S35. According to the analysis results, the image is subjected to necessary rotation and mirror transformation to keep it consistent with the geometric coordinate system of the inner wall of the tunnel.

3. The tunnel submillimeter imaging method based on a linear array camera according to claim 1, characterized in that: The S4 specifically includes: S41, extracting tunnel annular gap pixel points from the corrected image, and using a feature extraction algorithm to obtain the precise coordinates of the annular gap pixel points; S42, based on the extracted annular seam pixel points, the inclination angle of the annular seam is fitted using the least square method: Among them, tan(α) is the tangent value of the tilt angle, n is the total number of feature points extracted, and x i is the horizontal coordinate of the i-th feature point in the image coordinate system, is the average value of the horizontal coordinates of all extracted feature points, y i is the ordinate of the i-th feature point in the image coordinate system, is the average value of the ordinates of all extracted feature points; S43, based on the inclination angle of the ring seam obtained by fitting, construct a shear transformation matrix for geometric transformation of the image data: Among them, T shear is the skew transformation matrix; S44, applying a shear transformation to each pixel in the image data, and calculating the coordinates of the pixel after the transformation: in, represents the position of the pixel in the two-dimensional image, Indicates the new horizontal and vertical coordinates of the pixel point after the oblique shearing transformation; S45, adjusting the pixel positions of all scan lines in the image, so that the image scan lines after the oblique shearing transformation are consistent with the geometric features of the inner wall of the tunnel, and the tilt correction is completed.

4. The tunnel submillimeter imaging method based on a linear array camera according to claim 1, characterized in that: The S5 specifically includes: S51, according to the geometric characteristics of the tunnel cross section, establish a mathematical model of the tunnel inner wall, fit the tunnel cross section into a regular geometric shape, and calculate the position parameters of each point on the tunnel surface, including the center coordinates, radius and arch height; S52, extracting the coordinates of each pixel in the tilt-corrected image, and mapping each pixel to a physical point on the tunnel surface model according to the internal and external parameters of the camera and the spatial projection relationship; S53, for the physical points projected onto the tunnel surface model, reproject them into a unified UV coordinate system according to the axial and circumferential coordinates of the tunnel, and calculate the coordinates of the image points in the UV plane: V = Z·cos(φ)+Y·sin(φ); Among them, U is the transverse coordinate in the UV plane, V is the longitudinal coordinate in the UV plane, R is the radius of the tunnel, arctan2(Y,X) represents the angle value calculated according to the transverse and longitudinal coordinates of the object point, ΔX and ΔZ are the transverse and longitudinal sampling intervals respectively, Z is the longitudinal height coordinate of the object point in the three-dimensional coordinates of the tunnel, Y represents the longitudinal component of the point in the cross section, and φ represents the local tilt state of the tunnel at a specific acquisition position; S54, performing interpolation processing on the image data, resampling the original image data into the UV coordinate system, and filling the pixel gap caused by the curved surface projection; S55. Complete the orthorectification of the image, generate a high-resolution orthorectified image consistent with the tunnel cross-section geometry model, and eliminate the problem of uneven resolution caused by the curved surface structure.

5. The tunnel submillimeter imaging method based on a linear array camera according to claim 1, characterized in that: The S6 specifically includes: S61, determining the overlapping area of ​​adjacent images in the orthorectified image, separating the pixel data of the overlapping area by using an image segmentation algorithm, and recording the initial range coordinates of the overlapping area; S62, extracting features from the overlapping area, extracting feature points using an accelerated robust feature algorithm, and calculating descriptors of the feature points; S63, matching the extracted feature point descriptors, using a nearest neighbor matching method to find similar feature point pairs in the overlapping area of ​​adjacent images to form a matching point set; S64, performing robustness screening on the matching point set, eliminating mismatching points using a random sampling consistency algorithm, and optimizing the matching point set by calculating the ratio of internal points; S65. Calculate the relative position offset parameters between adjacent images based on the optimized matching point set, and provide the calculated relative position offset parameters to the image stitching algorithm: Among them, ΔX is the horizontal position offset parameter of two adjacent images, Δy is the vertical position offset parameter of two adjacent images, T is the total number of matching points in the matching point set, x i,2 -x i,1 Indicates the lateral offset, y i,2 -y i,1 represents the longitudinal offset, and i represents the number of the matching point currently calculated in the matching point set.

6. The tunnel submillimeter imaging method based on a linear array camera according to claim 1, characterized in that: The S7 specifically includes: S71, extracting the overlapping area of ​​adjacent images, and establishing a pixel energy distribution matrix of the overlapping area, wherein the energy value of each pixel is calculated based on the grayscale gradient and brightness difference: Among them, E(i,j) is the pixel energy distribution matrix, I1(i,j) and I2(i,j) are the gray values ​​of the corresponding pixels in the overlapping area of ​​the two images, τ is the weight parameter, is the gradient magnitude of the pixel; S72. Consider the energy distribution matrix as the weight matrix of the graph model, and construct the optimal stitching path from the top to the bottom of the overlapping area based on the dynamic programming algorithm: Among them, P is the optimal suture path, (x k ,y k ) represents the coordinates of the kth pixel on the path, argmin P It means to find the path that minimizes the objective function among all possible paths. n is the total number of pixels on the optimal stitching path. E(x k ,y k ) is the energy value of the kth pixel on the stitching path; S73, segmenting the overlapping area of ​​adjacent images along the optimal stitching path, and determining the pixel areas retained in each image after segmentation; S74, combining the weighted fusion algorithm, performing smooth transition processing on the pixel values ​​in the overlapping area: I fusion (i,j)=w1(i,j)·I1(i,j)+w2(i,j)·I2(i,j); Among them, I fusion (i, j) represents the fusion result of the overlapping area, w1(i, j) represents the credibility of the pixel point from the first image, w2(i, j) represents the credibility of the pixel point from the second image, I1(i, j) represents the pixel value of the first image in the overlapping area, and I2(i, j) represents the pixel value of the second image in the overlapping area; S75. Stitch the weighted fusion processed images into a panoramic image to generate a complete panoramic image of the tunnel inner wall.

Images