Tunnel panoramic image generation method and device based on SIFT and construction joint identification

Abstract

The application discloses a kind of based on SIFT and construction joint identification tunnel panoramic image generation method and device, comprising: using multiple image acquisition devices synchronously acquisition continuous inner wall image sequence of tunnel;Extract the SIFT feature point of adjacent inner wall image in continuous inner wall image sequence, determine the initial space geometric transformation parameter of adjacent inner wall image based on SIFT feature point, according to initial space geometric transformation parameter continuous inner wall image sequence is spliced into initial tunnel panoramic development image;Construction joint is identified in initial tunnel panoramic development image, and the observation mileage of construction joint feature point is determined in initial tunnel panoramic development image based on construction joint feature point pixel position, the actual mileage of construction joint feature point is acquired, and initial space geometric transformation parameter is optimized based on the difference between actual mileage and observation mileage;Continuous inner wall image sequence is spliced into tunnel panoramic development image based on the initial space geometric transformation parameter after optimization again.

Description

Technical Field

[0001] This invention relates to the field of tunnel inspection technology, and in particular to a method and apparatus for generating panoramic tunnel images based on SIFT and construction joint recognition. Background Technology

[0002] Surface inspection of tunnel lining is a crucial step in ensuring safe tunnel operation. Therefore, constructing a panoramic view of the tunnel is fundamental for identifying defects and analyzing deformation.

[0003] Currently, image sequences captured by cameras are typically stitched together to create a panoramic image of the tunnel, with the images aligned at their edges. However, due to the different installation positions and shooting angles of multiple cameras, this edge-aligned stitching method results in noticeable seams, causing image discontinuity and a mismatch between the actual tunnel image and the stitched tunnel image. Summary of the Invention

[0004] This invention provides a method and apparatus for generating panoramic tunnel images based on SIFT and construction joint recognition, which mainly improves the generation accuracy of tunnel images.

[0005] According to a first aspect of the present invention, a method for generating panoramic tunnel images based on SIFT and construction joint recognition is provided, comprising: Multiple image acquisition devices arranged circumferentially along the cross-section of the target tunnel are used to simultaneously acquire a continuous sequence of inner wall images along the axial direction of the target tunnel, wherein there is an overlapping area between adjacent inner wall images in the continuous inner wall image sequence; SIFT feature points are extracted from each adjacent inner wall image in the continuous inner wall image sequence. Initial spatial geometric transformation parameters for the corresponding adjacent inner wall images are determined based on the SIFT feature points. The continuous inner wall image sequence is then stitched together into an initial panoramic unfolded image of the tunnel according to the initial spatial geometric transformation parameters. The adjacent inner wall images include axial sequence adjacent inner wall images along the axial extension direction of the target tunnel, and circumferential adjacent inner wall images in the cross-sectional circumferential direction of the target tunnel. Multiple construction joints are identified in the initial panoramic image of the tunnel. Based on the pixel position information of the feature points of each construction joint in the initial panoramic image of the tunnel, the observation mileage of the feature points of the corresponding construction joint is determined. The actual mileage of the feature points of each construction joint is obtained. The initial spatial geometric transformation parameters are optimized based on the difference between the actual mileage and the corresponding observation mileage of the feature points of each construction joint. Based on the optimized initial spatial geometric transformation parameters, the continuous inner wall image sequence is re-stitched into a panoramic unfolded image of the tunnel.

[0006] Optionally, based on the SIFT feature points, initial spatial geometric transformation parameters are determined for corresponding adjacent inner wall images, and the continuous inner wall image sequence is stitched together into an initial panoramic unfolded tunnel image according to the initial spatial geometric transformation parameters, including: For the circumferential adjacent inner wall images in each circumferential single section of the target tunnel, circumferential feature matching is performed on the SIFT feature points in the overlapping area of ​​the circumferential adjacent inner wall images. Based on the circumferential feature matching results, the same circumferential SIFT feature points in the circumferential adjacent inner wall images are determined, and the initial circumferential spatial geometric transformation parameters between the circumferential adjacent inner wall images are determined based on the spatial geometric relationship between the same circumferential SIFT feature points. Based on the initial circumferential spatial geometric transformation parameters, the adjacent circumferential inner wall images are stitched together to form an initial single-section tunnel circumferential unfolded diagram. Axial feature matching is performed on the SIFT feature points of the overlapping area in the circumferential unfolded diagrams of adjacent initial single-section tunnels. Based on the axial feature matching results, the same axial SIFT feature points in the circumferential unfolded diagrams of adjacent initial single-section tunnels are determined. Based on the spatial geometric relationship between the same axial SIFT feature points, the initial axial spatial geometric transformation parameters between the circumferential unfolded diagrams of adjacent initial single-section tunnels are determined. Based on the initial axial spatial geometric transformation parameters, the adjacent initial single-section tunnel circumferential unfolded diagrams are stitched together to form the initial tunnel panoramic unfolded diagram.

[0007] Optionally, optimizing the initial spatial geometric transformation parameters based on the difference between the actual mileage and the corresponding observed mileage of each feature point of the construction joint includes: Based on the difference between the actual mileage and the corresponding observed mileage of each feature point of the construction joint, a global optimization function is constructed with the objective of minimizing the sum of squared errors of the feature point mileages of all construction joints. Using the geometric consistency of local feature matching as an optimization constraint, the global optimization function is used to jointly optimize each of the initial spatial geometric transformation parameters, wherein the initial spatial geometric transformation parameters include at least one of spatial translation parameters, spatial rotation parameters, spatial scaling parameters, and spatial perspective transformation parameters.

[0008] Optionally, before determining the observation mileage of the corresponding construction joint feature point based on the pixel position information of each construction joint feature point in the initial panoramic image of the tunnel, the method further includes: Each construction joint is taken as a target candidate construction joint, the geometric feature parameters of the target candidate construction joint are determined, and the geometric structure score of the target candidate construction joint is evaluated based on the geometric feature parameters. Obtain the groove region attribute parameters of the target candidate construction joint, and evaluate the texture purity score of the target candidate construction joint based on the groove region attribute parameters; The effective feature region ratio score of the target candidate construction joint is evaluated based on the construction joint image of the target candidate construction joint; The geometric structure score, the texture purity score, and the feature region proportion score are weighted and summed, and a target construction joint for optimizing the initial spatial geometric transformation parameters is selected in each construction joint based on the weighted summation result. The step of determining the observation mileage of the corresponding construction joint feature point based on the pixel position information of the feature point of each construction joint in the initial panoramic image of the tunnel includes: The observation mileage of the feature points of each target construction joint is determined based on the pixel position information of the feature points in the initial panoramic image of the tunnel.

[0009] Optionally, before extracting SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence, the method further includes: Each inner wall image in the continuous inner wall image sequence is taken as a target inner wall image. Construction joints are identified in the target inner wall images, and the pixel distance between each image pixel in the target inner wall image and the nearest construction joint pixel is determined. Personalized preprocessing parameters are set for the corresponding image pixels based on the pixel distance, wherein the preprocessing parameters include filter kernel size parameters and edge protection strength parameters; Based on the personalized preprocessing parameters, the corresponding image pixels are preprocessed, and the image composed of each preprocessed image pixel is used as the preprocessed target inner wall image. The step of extracting SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence includes: SIFT feature points are extracted from each adjacent inner wall image in the preprocessed continuous inner wall image sequence.

[0010] Optionally, after re-stitching the continuous inner wall image sequence into a panoramic unfolded tunnel image based on the optimized initial spatial geometric transformation parameters, the method further includes: The non-spatial attribute data and spatial geometric feature data of each construction joint in the panoramic unfolded image of the tunnel are determined, and the non-spatial attribute data and spatial geometric feature data of each construction joint are encapsulated into multi-dimensional attribute vector objects respectively. Using the panoramic image of the tunnel as the base layer, each of the multidimensional attribute vector objects is converted into a vector metadata layer, and the vector metadata layer is embedded into the base layer to generate a composite panoramic tunnel image file. A multidimensional spatial index structure is constructed in the runtime memory of the composite tunnel panoramic image file using the bounding box of the pixel coordinates of each construction joint in the panoramic image of the tunnel as the key and the memory address of the corresponding multidimensional attribute vector object as the value. A hash attribute inverse index structure is constructed in the runtime memory of the composite tunnel panoramic image file using the joint identifier of each construction joint as the key and the geometric center point of the bounding box of the corresponding construction joint and the storage address in the multidimensional spatial index structure as the value. The multidimensional spatial index structure and the hash attribute inverse index structure are associated with the composite tunnel panoramic image file as a spatial-attribute dual index structure.

[0011] Optionally, after re-stitching the continuous inner wall image sequence into a panoramic unfolded tunnel image based on the optimized initial spatial geometric transformation parameters, the method further includes: The radiometric attribute information of the panoramic unfolded image of the tunnel is determined, wherein the radiometric attribute information includes the local illuminance deficiency index and the global illuminance uniformity coefficient of the panoramic unfolded image of the tunnel. Based on the radiometric attribute information, the topological constraints of the mapping function of the pixel grayscale mapping function to be generated are determined, wherein the topological constraints of the mapping function include pixel value range constraints, monotonicity constraints of the pixel grayscale mapping function to be generated, and bidirectional mapping consistency constraints. Based on the topological constraints of the mapping function, a pixel grayscale mapping function is constructed to reconstruct the nonlinear radiometric response of the panoramic unfolded image of the tunnel. The radiometric response reconstruction of the tunnel panoramic unfolded image is performed using the pixel grayscale mapping function to obtain the quality-enhanced tunnel panoramic unfolded image.

[0012] According to a second aspect of the present invention, a tunnel panoramic image generation apparatus based on SIFT and construction joint recognition is provided, comprising: An image acquisition unit is used to simultaneously acquire a continuous sequence of inner wall images along the axial direction of the target tunnel using multiple image acquisition devices arranged circumferentially along the cross-section of the target tunnel, wherein there is an overlapping area between adjacent inner wall images in the continuous inner wall image sequence; An image stitching unit is used to extract SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence, determine the initial spatial geometric transformation parameters of the corresponding adjacent inner wall images based on the SIFT feature points, and stitch the continuous inner wall image sequence into an initial panoramic unfolded image of the tunnel according to the initial spatial geometric transformation parameters. The adjacent inner wall images include axial sequence adjacent inner wall images along the axial extension direction of the target tunnel, and circumferential adjacent inner wall images in the cross-sectional circumferential direction of the target tunnel. The parameter optimization unit is used to identify multiple construction joints in the initial panoramic image of the tunnel, determine the observation mileage of the feature points of the corresponding construction joint based on the pixel position information of the feature points of each construction joint in the initial panoramic image of the tunnel, obtain the actual mileage of the feature points of each construction joint, and optimize the initial spatial geometric transformation parameters based on the difference between the actual mileage and the corresponding observation mileage of the feature points of each construction joint. The image re-stitching unit is used to re-stitch the continuous inner wall image sequence into a panoramic unfolded image of the tunnel based on the optimized initial spatial geometric transformation parameters.

[0013] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described method for generating panoramic tunnel images based on SIFT and construction joint recognition.

[0014] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for generating panoramic tunnel images based on SIFT and construction joint recognition.

[0015] The present invention provides a method and apparatus for generating panoramic tunnel images based on SIFT and construction joint identification. Compared with the current method of directly stitching together image sequences acquired by cameras according to image edge alignment, the present invention uses multiple acquisition devices arranged in a circumferential manner to cover the entire circumference of the tunnel cross-section, avoiding blind spots from a single perspective and ensuring the integrity of the inner wall image. By calculating the geometric transformation relationship between images through overlapping areas, the image position can be calibrated more accurately, reducing stitching errors. By automatically calculating the initial transformation parameters between images through SIFT feature point matching, manual intervention is reduced and stitching efficiency is improved. By comparing the difference between the observed mileage and the actual mileage of the construction joint, the geometric transformation parameters of image stitching can be adjusted in reverse, eliminating long-distance cumulative errors during image stitching, thereby improving the generation accuracy of the panoramic tunnel image. At the same time, since the construction joint has obvious straight-line characteristics and its design location is known, the generation accuracy of the panoramic tunnel image can be further improved by using the construction joint as a reference object to eliminate cumulative errors. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This invention provides a flowchart of a method for generating panoramic tunnel images based on SIFT and construction joint recognition, according to an embodiment of the present invention. Figure 2 This invention provides a flowchart of another method for generating panoramic tunnel images based on SIFT and construction joint recognition, according to an embodiment of the present invention. Figure 3 This diagram illustrates the structure of a tunnel panoramic image generation device based on SIFT and construction joint recognition, according to an embodiment of the present invention. Figure 4 This invention provides a schematic diagram of another tunnel panoramic image generation device based on SIFT and construction joint recognition, according to an embodiment of the present invention. Figure 5 A schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0017] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0018] Currently, the method of directly stitching together image sequences captured by cameras according to the image edge alignment to form a panoramic image of the tunnel can lead to a mismatch between the actual tunnel image and the stitched tunnel image due to the different installation positions and shooting angles of multiple cameras.

[0019] To address the aforementioned problems, embodiments of the present invention provide a method for generating panoramic tunnel images based on SIFT and construction joint recognition, such as... Figure 1 As shown, the method includes: 101. Multiple image acquisition devices arranged circumferentially along the cross-section of the target tunnel are used to simultaneously acquire a continuous sequence of inner wall images along the axial direction of the target tunnel, wherein there is an overlapping area between adjacent inner wall images in the continuous inner wall image sequence.

[0020] In embodiments of the present invention, N image acquisition devices, which can be cameras, are uniformly or non-uniformly arranged along the circumference of the cross-section of the tunnel inspection equipment. The field of view of each acquisition device is pre-calibrated and adjusted to ensure that the optical axis angle between adjacent devices meets the requirement of covering the entire circumference of the tunnel inner wall, and that the edges of the field of view of any two adjacent acquisition devices form an overlapping imaging area of ​​a preset width on the tunnel inner wall. The inspection equipment is controlled to move along the tunnel axis at a constant speed or in a stepping manner. During the movement, all image acquisition devices are triggered to execute a global synchronous exposure strategy. That is, all cameras are exposed at the same time t. i Simultaneously capture the inner wall image of the current cross-section to generate the i-th frame image group. ,in, This represents the image captured by the k-th camera at time i. As the tunnel inspection equipment continues to move along the tunnel axis, the above synchronous acquisition process is repeated at fixed time intervals or fixed mileage intervals, thereby obtaining a continuous sequence of inner wall images extending along the tunnel axis. At the same time t i Within the acquired image group, adjacent camera images, such as and Pixel-level overlap exists at the seams to eliminate blind spots in the single-camera field of view and provide circumferential stitching features. At adjacent times, such as t... i and t i+1 In the image sequence acquired at any time, because the displacement of the tunnel detection device is less than the axial coverage length of a single frame image, there is a significant overlap between the previous frame image and the subsequent frame image in the axial dimension.

[0021] For example, a tunnel inspection vehicle is used, with six industrial area array cameras evenly spaced on a ring-shaped bracket at the front of the vehicle. The optical axes of the cameras point towards the inner wall of the tunnel, and the fields of view of adjacent cameras overlap by approximately 20%. The inspection vehicle travels at a speed of 60 km / h, and all cameras are triggered synchronously, acquiring 10 frames per second, each with a resolution of 4096×3000 pixels. Wheel speedometer pulses are recorded simultaneously during acquisition for preliminary mileage estimation. Radial distortion correction is performed on each frame (using pre-calibrated camera intrinsic parameters), and adaptive histogram equalization is applied to enhance contrast. This embodiment of the invention, through multiple acquisition devices arranged circumferentially, can cover the entire circumference of the tunnel cross-section, avoiding blind spots from a single viewpoint and ensuring the integrity of the inner wall images. By calculating the geometric transformation relationship between images through overlapping areas, image positions can be more accurately calibrated, reducing stitching errors.

[0022] 102. Extract SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence, determine the initial spatial geometric transformation parameters of the corresponding adjacent inner wall images based on the SIFT feature points, and stitch the continuous inner wall image sequence into an initial panoramic unfolded image of the tunnel according to the initial spatial geometric transformation parameters. The adjacent inner wall images include axial sequence adjacent inner wall images along the axial extension direction of the target tunnel, and circumferential adjacent inner wall images in the cross-sectional circumferential direction of the target tunnel.

[0023] SIFT feature points can include tunnel inner wall corners with scale-invariant feature transformation, distinctive spots, and points with significant features at the tunnel edge.

[0024] In this embodiment of the invention, adjacent inner wall images include circumferential adjacent inner wall images and axial sequential adjacent inner wall images. Specifically, circumferential adjacent inner wall images refer to the k-th and (k+1)-th inner wall images whose fields of view overlap, acquired by the circumferential array camera at the same acquisition time; axial sequential adjacent inner wall images refer to the i-th and (i+1)-th inner wall images continuously acquired along the tunnel axis from the same camera viewpoint. Subsequently, the SIFT algorithm is applied to all the aforementioned adjacent image pairs to detect scale-space extrema and generate a 128-dimensional feature descriptor subset with rotation, scale, and illumination invariance. A nearest neighbor distance ratio strategy is used to perform feature point matching search among the feature descriptor subsets of adjacent inner wall images, obtaining highly similar feature point pairs matched between adjacent images. To eliminate erroneous matches caused by repeated tunnel inner wall textures or abrupt illumination changes, a random sampling consensus algorithm is introduced to perform geometric consistency checks on the matched feature point pairs, eliminating outliers and retaining high-confidence matched feature point pairs as the benchmark for subsequent solutions. Based on the positional relationship between two points in the selected matching feature point pairs, an objective function is constructed to minimize the reprojection error. Considering the projection characteristics of the tunnel surface, the objective function can employ a homography matrix or affine transformation model. Through singular value decomposition or algorithms such as the Levonburg-Marquardt method, the initial spatial geometric transformation parameters (including translation vectors, rotation matrices, and scale factors) between each pair of adjacent images are calculated. These initial spatial geometric transformation parameters describe the relative pose relationship between adjacent inner wall images in the local coordinate system. Then, the first frame image in the inner wall image sequence is set as the global reference. Using the calculated initial spatial geometric transformation parameters, all circumferential and axial adjacent images are uniformly mapped to the same two-dimensional unfolding plane through chain transfer or global bundle adjustment strategies. Finally, weighted fusion processing is performed on the overlapping areas of the mapped images to eliminate stitching gaps and brightness differences, generating the initial panoramic unfolded image of the tunnel. This embodiment of the invention automatically calculates the initial transformation parameters between images through SIFT feature point matching, reducing manual intervention and improving stitching efficiency.

[0025] 103. Identify multiple construction joints in the initial panoramic image of the tunnel, and determine the observation mileage of the corresponding construction joint feature points based on the pixel position information of the feature points of each construction joint in the initial panoramic image of the tunnel. Obtain the actual mileage of the feature points of each construction joint, and optimize the initial spatial geometric transformation parameters based on the difference between the actual mileage and the corresponding observation mileage of the feature points of each construction joint.

[0026] In this embodiment of the invention, edge detection operators combined with morphological processing are used to automatically identify circumferential construction joints distributed along the axial direction in the initial panoramic unfolded image of the tunnel. The centerline or salient feature points (such as crack intersections and texture abrupt change points) of each construction joint are extracted as construction joint feature points, and their pixel coordinates in the initial panoramic unfolded image of the tunnel are recorded. Based on the pixel-physical size mapping relationship of the initial panoramic unfolded image of the tunnel (determined by camera calibration parameters and initial stitching scale), the pixel ordinates of the identified construction joint feature points are converted into observed mileage. This mileage reflects the relative positional distance derived solely from image feature matching. Simultaneously, based on pre-stored tunnel completion data, the actual mileage corresponding to the same construction joint feature point is obtained. This mileage represents the absolute geographical coordinates of the construction joint in the real world and serves as the "true value" benchmark for correction. The difference between the observed mileage and the actual mileage is taken as the mileage deviation at the corresponding construction joint. Then, a global error function is constructed with the goal of minimizing the mileage deviation. ,in, The initial spatial geometric transformation parameters are set as the parameters to be optimized. Then, the global longitudinal coordinates of the construction joint feature points are fixed, and the translation vector, rotation angle, and scale factor between adjacent image frames are adjusted. While maintaining the feature matching constraints between adjacent images (i.e., maintaining the geometric consistency of the original matching), the initial spatial geometric transformation parameters are iteratively solved with the goal of minimizing the global error function. The iteration stops when the required number of iterations or the mileage deviation meets the requirements, finally yielding the optimized initial spatial geometric transformation parameters. This embodiment of the invention compares the difference between the observed mileage and the actual mileage of the construction joint and adjusts the geometric transformation parameters of image stitching in reverse, thereby eliminating long-distance cumulative errors during image stitching and improving the generation accuracy of the tunnel panoramic image. Furthermore, since the construction joint has obvious straight-line characteristics and its design location is known, using the construction joint as a reference object to eliminate cumulative errors can further improve the generation accuracy of the tunnel panoramic image.

[0027] 104. Based on the optimized initial spatial geometric transformation parameters, the continuous inner wall image sequence is re-stitched into a panoramic unfolded image of the tunnel.

[0028] In this embodiment of the invention, for inner wall images acquired at the same time by different cameras in the same cross-section, using the first inner wall image as a reference, and based on the optimized initial spatial geometric transformation parameters between adjacent images, the images in the same cross-section acquired by other cameras are gradually transformed to a unified coordinate system, that is, adjacent images are stitched together to obtain a single-section circumferential unfolded image. This method is used to stitch together the internal images in multiple cross-sections to obtain a single-section circumferential unfolded image for each cross-section. For example, using a certain internal image as reference image 1, the adjacent inner wall image 2 is stitched into reference image 1 based on the optimized initial spatial geometric transformation parameters, and the adjacent image 3 is stitched into image 2 based on the optimized initial spatial geometric transformation parameters. By stitching the images sequentially in this way, a single-section circumferential unfolded image can be obtained. Furthermore, regarding the tunnel axial direction, using the first single-section circumferential unfolded image as a reference, the subsequent single-section circumferential unfolded images are gradually transformed to a unified coordinate system using the optimized initial spatial geometric transformation parameters between adjacent images. The adjacent single-section circumferential unfolded images are then stitched together to obtain the initial panoramic unfolded image of the tunnel.

[0029] The present invention provides a method for generating panoramic tunnel images based on SIFT and construction joint identification. Compared with the current method of directly stitching together image sequences acquired by cameras according to image edge alignment, the present invention uses multiple acquisition devices arranged in a circumferential manner to cover the entire circumference of the tunnel cross-section, avoiding blind spots from a single perspective and ensuring the integrity of the inner wall images. By calculating the geometric transformation relationship between images through overlapping areas, the image positions can be calibrated more accurately, reducing stitching errors. By automatically calculating the initial transformation parameters between images through SIFT feature point matching, manual intervention is reduced and stitching efficiency is improved. By comparing the difference between the observed mileage and the actual mileage of the construction joint, the geometric transformation parameters of image stitching can be adjusted in reverse, eliminating long-distance cumulative errors during image stitching, thereby improving the generation accuracy of the panoramic tunnel image. At the same time, since the construction joint has obvious straight-line characteristics and its design location is known, using the construction joint as a reference object to eliminate cumulative errors can further improve the generation accuracy of the panoramic tunnel image.

[0030] Furthermore, to better illustrate the process of the tunnel panoramic image generation method based on SIFT and construction joint recognition, as a refinement and extension of the above embodiments, this invention provides another tunnel panoramic image generation method based on SIFT and construction joint recognition, such as... Figure 2 As shown, the method includes: 201. Multiple image acquisition devices arranged circumferentially along the cross-section of the target tunnel are used to simultaneously acquire a continuous sequence of inner wall images along the axial direction of the target tunnel, wherein there is an overlapping area between adjacent inner wall images in the continuous inner wall image sequence.

[0031] Specifically, N high-definition cameras (e.g., N=6) are evenly arranged circumferentially along the tunnel cross-section on a mobile device, and images of the tunnel inner wall are acquired synchronously. Each camera continuously acquires an image sequence along the tunnel axis. The fields of view of adjacent cameras overlap to a certain extent to ensure circumferential stitching.

[0032] 202. Extract SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence, determine the initial spatial geometric transformation parameters of the corresponding adjacent inner wall images based on the SIFT feature points, and stitch the continuous inner wall image sequence into an initial panoramic unfolded image of the tunnel according to the initial spatial geometric transformation parameters. The adjacent inner wall images include axial sequence adjacent inner wall images along the axial extension direction of the target tunnel, and circumferential adjacent inner wall images in the cross-sectional circumferential direction of the target tunnel.

[0033] In this embodiment of the invention, to improve the quality of the inner wall image, each inner wall image needs to be preprocessed first. Based on this, the method includes: taking any inner wall image in the continuous inner wall image sequence as a target inner wall image; identifying construction joints in the target inner wall image and determining the pixel distance between each image pixel in the target inner wall image and the nearest construction joint pixel; setting personalized preprocessing parameters for the corresponding image pixels based on the pixel distance, wherein the preprocessing parameters include filter kernel size parameters and edge protection strength parameters; preprocessing the corresponding image pixels based on the personalized preprocessing parameters; and using the image composed of each preprocessed image pixel as the preprocessed target inner wall image.

[0034] Specifically, taking any one inner wall image from a continuous sequence of inner wall images as a target inner wall image is used as an example. The preprocessing method for all subsequent inner wall images is the same as that for the target inner wall image. Construction joints are identified in the target inner wall image using operators such as edge detection, and a binarized construction joint mask image is generated for the construction joint region. In this mask image, construction joint pixels are marked as 1, and background pixels are marked as 0. Based on the construction joint mask image, the Euclidean distance between each background pixel in the target inner wall image and the nearest construction joint pixel is calculated as the pixel distance. A pixel distance field image covering the entire image is constructed based on these pixel distances. This pixel distance field image quantifies the spatial proximity of any position in the inner wall image relative to the construction joint; a smaller distance value indicates closer proximity to the construction joint, and a larger distance value indicates farther distance from the construction joint. When the pixel distance is large, i.e., the area far from the construction joint in the inner wall image is set with a larger filter kernel size parameter and a smaller edge protection strength parameter. Then, the image area far from the construction joint is filtered using the larger filter kernel size parameter and the smaller edge protection strength parameter. The larger filter kernel size parameter enhances the smoothing ability for uniform noise on the concrete surface, while the smaller edge protection strength parameter reduces the protection intensity, allowing for greater smoothing operations. When the pixel distance is small, i.e., the area near the construction joint in the inner wall image is set with a smaller filter kernel size parameter and a higher edge protection strength parameter. Then, the image area near the construction joint is filtered using the smaller filter kernel size parameter and the higher edge protection strength parameter. The larger filter kernel size parameter avoids blurring the edge details of the construction joint, while the higher edge protection strength parameter ensures strict limitation of grayscale gradient diffusion during the filtering process, preventing crack widening or breakage. This embodiment of the invention, by dividing the region with the construction joint as a reference and using different preprocessing methods for different regions, ensures that the feature geometry of the construction joint and its adjacent pixels is not distorted, and effectively filters out sensor noise and artifacts caused by uneven illumination for pixels on large, flat wall surfaces.

[0035] Further, it is then necessary to construct an initial panoramic unfolded image of the tunnel based on the preprocessed inner wall images. Therefore, step 202 specifically includes: for each circumferentially adjacent inner wall image in each circumferential single section of the target tunnel, performing circumferential feature matching on the SIFT feature points in the overlapping areas of the circumferentially adjacent inner wall images; determining the same circumferential SIFT feature points in the circumferentially adjacent inner wall images based on the circumferential feature matching results; and determining the initial circumferential spatial geometric transformation parameters between the circumferentially adjacent inner wall images based on the spatial geometric relationship between the same circumferential SIFT feature points; and based on the initial circumferential spatial geometric transformation parameters... The adjacent circumferential inner wall images are stitched together to form an initial single-section tunnel circumferential unfolded image. Axial feature matching is performed on the SIFT feature points in the overlapping areas of the adjacent initial single-section tunnel circumferential unfolded images. Based on the axial feature matching results, the same axial SIFT feature points in the adjacent initial single-section tunnel circumferential unfolded images are determined. Based on the spatial geometric relationship between the same axial SIFT feature points, the initial axial spatial geometric transformation parameters between the adjacent initial single-section tunnel circumferential unfolded images are determined. Based on the initial axial spatial geometric transformation parameters, the adjacent initial single-section tunnel circumferential unfolded images are stitched together to form the initial tunnel panoramic unfolded image.

[0036] Specifically, for any circumferential single-section of the target tunnel, SIFT feature points are extracted from the overlapping areas of all adjacent circumferential inner wall images within that section. Using feature descriptor similarity calculations (such as Euclidean distance ratio testing), high-confidence pairs of identical circumferential SIFT feature points are selected. Then, based on the pixel coordinates of the successfully matched feature point pairs in their respective images, initial circumferential spatial geometric transformation parameters between adjacent images are determined, including circumferential translation, rotation angle, and local scale factor. Next, using these initial circumferential spatial geometric transformation parameters, all inner wall images within the single section are projected onto a unified cylindrical or planar coordinate system and weighted fusion is performed to generate a geometrically continuous initial single-section tunnel circumferential unfolded image, thus completing the 360-degree image loop stitching of the tunnel cross-section. Further, along the tunnel's travel direction (axial direction), two adjacent initial single-section tunnel circumferential unfolded images are selected, and SIFT feature points are extracted again in the overlapping area of ​​the two images. Axial feature matching is then performed to determine identical axial SIFT feature point pairs spanning the sections. Based on the spatial distribution relationship of axial feature point pairs, the initial axial spatial geometric transformation parameters between adjacent single-section tunnel circumferential unfolded images are calculated. These parameters characterize the longitudinal displacement deviation, pitch / yaw angle error, and longitudinal scale expansion caused by vehicle speed fluctuations. Then, according to these initial axial spatial geometric transformation parameters, all initial single-section tunnel circumferential unfolded images are sequentially stitched together in mileage order. Through global coordinate transformation and image fusion, misalignment and breaks between sections are eliminated, ultimately generating an initial panoramic unfolded image covering the entire detection section. The layered stitching strategy of this invention effectively reduces the complexity of large-scale image matching.

[0037] 203. Identify multiple construction joints in the initial panoramic image of the tunnel, and determine the observation mileage of the corresponding construction joint feature points based on the pixel position information of the feature points of each construction joint in the initial panoramic image of the tunnel, and obtain the actual mileage of the feature points of each construction joint.

[0038] In this embodiment of the invention, multiple construction joints are first identified in the initial panoramic image of the tunnel. Then, to ensure the quality of the construction joints, a high-quality construction joint needs to be selected from each construction joint. Based on this, the method includes: taking any construction joint from each of the construction joints as a target candidate construction joint; determining the geometric feature parameters of the target candidate construction joint; evaluating the geometric structure score of the target candidate construction joint based on the geometric feature parameters; obtaining the groove region attribute parameters of the target candidate construction joint; evaluating the texture purity score of the target candidate construction joint based on the groove region attribute parameters; evaluating the effective feature area proportion score of the target candidate construction joint based on the construction joint image of the target candidate construction joint; performing a weighted summation of the geometric structure score, the texture purity score, and the feature area proportion score; and selecting a target construction joint from each of the construction joints based on the weighted summation result for optimizing the initial spatial geometric transformation parameters.

[0039] Among them, geometric feature parameters include, but are not limited to, circumferential continuity, straightness, width uniformity, and closure; the groove region refers to the dark gap inside the construction joint, and the attribute parameters include, but are not limited to, grayscale mean, variance, contrast, and gradient difference with the surrounding concrete background; the effective feature area refers to the area where there are clear and usable effective feature points.

[0040] Specifically, each construction joint is treated as a target candidate construction joint for scoring purposes. The scoring method for all subsequent construction joints is the same as that for the target candidate construction joint. First, geometric feature parameters such as circumferential continuity, straightness, width uniformity, and closure of the target candidate construction joint are extracted. Based on these parameters, a geometric structure score is generated. For example, a high score is given to a circumferentially continuous construction joint with a gradual width change, while points are deducted for construction joints with breaks, severe bends, or abrupt width changes. Simultaneously, attribute parameters such as grayscale mean, variance, contrast, and gradient difference with the surrounding concrete background are extracted from the groove area of ​​the target candidate construction joint. The texture purity of the groove area is evaluated based on these attribute parameters. For example, if the groove has uniform grayscale, no debris filling, and clear and sharp boundaries with the side walls, the texture purity score is high; if there are water seepage marks, soil accumulation, or light and shadow interference causing texture chaos within the groove, the texture purity score is low. This process generates a texture purity score. Simultaneously, within the local image window of the target candidate construction joint, the number and distribution density of effective feature points that can be used for stable matching, such as corner points and high gradient points, are statistically analyzed. Based on the number, the proportion of the area covered by effective feature points to the total area of ​​the construction joint is calculated. The higher the proportion, the higher the score of the effective feature area proportion; the lower the proportion, the lower the score. Then, according to actual needs, the weighting coefficients for geometric structure score, texture purity score, and feature area proportion score are determined. Based on these weighting coefficients, the geometric structure score, texture purity score, and feature area proportion score are weighted and summed to obtain the comprehensive health score of the target candidate construction joint. Thus, the comprehensive health score of each construction joint can be determined in the above manner. Then, construction joints with a comprehensive health score greater than a preset score set according to actual needs are selected as target construction joints for optimizing the initial spatial geometric transformation parameters. This embodiment of the invention, by selecting construction joints that meet the health requirements for optimizing spatial geometric transformation parameters, avoids a decrease in overall splicing accuracy due to identification errors or feature blurring of individual construction joints.

[0041] Furthermore, after selecting the target construction joint, for the selected target construction joint, the pixel coordinates of its feature points in the initial panoramic image of the tunnel are extracted, such as the intersection of the construction joint centerlines, corner points, or texture salient points. Based on the generation logic of the initial panoramic image, a preset pixel-to-mileage conversion coefficient (i.e., the physical distance corresponding to the image's vertical resolution) is used to convert the vertical pixel coordinates of the feature points into the observation mileage. Specifically, the observation mileage can be determined according to the following formula:

[0042] in, For the observation mileage, This refers to the known absolute mileage value in the real physical world of the starting point (i.e., the first frame or the first cross-section) corresponding to the initial panoramic view of the tunnel. The vertical pixel coordinates of the feature point. The pixel-to-mileage conversion factor represents the actual physical distance represented by each pixel in the image along the vertical direction. This factor is determined by the camera resolution, shooting distance, lens distortion correction, and average scale estimation during initial stitching. The specific value can be set according to actual needs. Simultaneously, the actual mileage of corresponding construction joint feature points is obtained based on design drawings, external sensors, and other tools.

[0043] 204. Based on the difference between the actual mileage and the corresponding observed mileage of each construction joint feature point, a global optimization function is constructed with the objective of minimizing the sum of squared errors of feature point mileages of all construction joints.

[0044] 205. Using the geometric consistency of local feature matching as the optimization constraint, the global optimization function is used to jointly optimize each initial spatial geometric transformation parameter, wherein the initial spatial geometric transformation parameter includes at least one of spatial translation parameter, spatial rotation parameter, spatial scaling parameter, and spatial perspective transformation parameter.

[0045] 206. Based on the optimized initial spatial geometric transformation parameters, the continuous inner wall image sequence is re-stitched into a panoramic unfolded image of the tunnel.

[0046] Specifically, based on the actual and observed mileage of each construction joint feature point obtained in the previous steps, a global optimization objective function is constructed. This function is defined as the weighted sum of squares of the mileage errors of all valid construction joint feature points. To prevent simple mileage fitting from causing texture tearing or structural distortion within the image, local feature matching geometric consistency is introduced as a strong constraint. For example, stable feature point pairs between adjacent image frames are extracted, and the optimized transformation parameters are constrained to ensure that the reprojection error of these feature point pairs remains within a threshold range, ensuring that the topological structure of the local image is not deformed and the continuity is not destroyed. The initial spatial geometric transformation parameters to be optimized are decoupled into two levels of sub-parameter sets, corresponding to different stitching stages: cross-section image stitching parameters (circumferential layer): for the unfolding process of a single-loop tunnel cross-section, the optimized parameters include spatial rotation parameters, spatial perspective transformation parameters, and local spatial scaling parameters. This level of optimization can ensure the closure and geometric regularity of the single-loop cross-section. Axial image stitching parameters (longitudinal layer): For the sequential stitching process of multiple cross-sections along the tunnel axis, the optimized parameters include spatial translation parameters, global spatial scaling parameters, and minute spatial rotation parameters. This hierarchical optimization aims to eliminate accumulated deviations during long-distance extension. A nonlinear least squares or bundle adjustment framework is used to solve the objective function and geometric consistency constraints iteratively. In each iteration, the mileage residual gradient and local feature reprojection gradient are calculated simultaneously. The transformation parameters of the cross-section layer and the axial layer are updated synchronously until the objective function converges to a preset accuracy threshold set according to actual needs. Using the finally converged optimal parameter set, a unified geometric transformation and resampling are performed on the original image sequence to generate an optimized panoramic tunnel unfolded image. This result ensures both clear and distortion-free local details and accurate alignment with the actual mileage at the macro scale, providing a reliable data foundation for subsequent high-precision defect quantitative analysis.

[0047] Furthermore, to further improve the image quality of the panoramic unfolded image of the tunnel, it is necessary to perform quality enhancement processing on the panoramic unfolded image of the tunnel. Based on this, the method includes: determining the radiometric attribute information of the panoramic unfolded image of the tunnel, wherein the radiometric attribute information includes the local illuminance deprivation index and the global illumination uniformity coefficient of the panoramic unfolded image of the tunnel; based on the radiometric attribute information, determining the topological constraints of the mapping function of the pixel gray-level mapping function to be generated, wherein the topological constraints of the mapping function include pixel value range constraints, monotonicity constraints of the pixel gray-level mapping function to be generated, and bidirectional mapping consistency constraints; based on the topological constraints of the mapping function, constructing a pixel gray-level mapping function for nonlinear radiometric response reconstruction of the panoramic unfolded image of the tunnel; and using the pixel gray-level mapping function to perform radiometric response reconstruction of the panoramic unfolded image of the tunnel to obtain the quality-enhanced panoramic unfolded image of the tunnel.

[0048] Specifically, a sliding window mechanism is used to traverse the panoramic unfolded image. The deviation ratio between the average gray value of each local area of ​​the image and the preset standard illuminance threshold set according to actual needs is used as a local illuminance deficiency index. This index is used to accurately locate "dark areas" and "shadow areas" caused by tunnel lighting malfunctions or obstructions. The statistical moments (such as skewness and kurtosis) of the gray-level histogram of the entire panoramic unfolded image or the global coefficient of variation based on block variance is calculated as the global illumination uniformity coefficient. This coefficient quantifies the uniformity of the overall brightness distribution of the image. Based on the aforementioned radiometric attribute information, the pixel value range constraint, monotonicity constraint, and bidirectional mapping consistency constraint of the pixel grayscale mapping function to be generated are dynamically determined. Among them, the pixel value range constraint sets the output value range of the mapping function to be strictly limited to the effective dynamic range to prevent overexposure or blackness caused by truncation, and ensures that all enhanced pixel values ​​are legal and valid. The monotonicity constraint sets the mapping function to remain strictly monotonically increasing throughout the entire domain. This constraint ensures that the original brightness and darkness levels of the image are not reversed, avoiding artifacts or "negative" effects that violate human visual habits. The bidirectional mapping consistency constraint requires the mapping function to have reversibility or near-reversibility, that is, the existence of an inverse function that makes the original information traceable. This constraint ensures that the radiometric correction process does not destroy the radiometric linearity of the image. Then, under the premise of satisfying the above topological constraints, a nonlinear pixel grayscale mapping function adapted to the current image characteristics is constructed. For example, if the local illumination deficiency index is high, the function exhibits a steep nonlinear stretching characteristic in the low grayscale range to significantly improve the visibility of details in dark areas; if the global illumination uniformity coefficient is low (extremely uneven illumination), the function introduces an adaptive piecewise linear or sigmoid curve shape to smoothly transition the gain in different brightness ranges, suppressing excessive enhancement in bright areas and compensating for low-brightness areas. Spline interpolation or parameterized sigmoid family functions are used to fit the curve that satisfies the constraints, and the optimal function parameters are solved by minimizing the local contrast loss function, thereby constructing the pixel grayscale mapping function. Furthermore, the constructed pixel grayscale mapping function is used... For each pixel in the panoramic image of the tunnel The form of pointwise radiative response reconstruction is shown below:

[0049] in, For the pixel grayscale values ​​of the panoramic image of the tunnel, The grayscale value is the corresponding pixel value after reconstructing the radiometric response. After reconstructing the grayscale value of each pixel in the unfolded image, the enhanced panoramic unfolded image of the tunnel can be obtained.

[0050] Furthermore, in order to output the panoramic unfolded image and facilitate subsequent human interaction, it is also necessary to generate a tunnel panoramic image file that is easy to interact with. Based on this, the method further includes: determining the non-spatial attribute data and spatial geometric feature data of each construction joint in the tunnel panoramic unfolded image, and encapsulating the non-spatial attribute data and spatial geometric feature data of each construction joint into multi-dimensional attribute vector objects; using the tunnel panoramic unfolded image as a base layer, converting each multi-dimensional attribute vector object into a vector metadata layer, and embedding the vector metadata layer into the base layer to generate a composite tunnel panoramic image file; Using the bounding box of the pixel coordinates of each construction joint in the panoramic image of the tunnel as the key and the memory address of the corresponding multidimensional attribute vector object as the value, a multidimensional spatial index structure is constructed in the runtime memory of the composite tunnel panoramic image file. Using the joint identifier of each construction joint as the key and the geometric center point of the bounding box of the corresponding construction joint and the storage address in the multidimensional spatial index structure as the value, a hash attribute reverse index structure is constructed in the runtime memory of the composite tunnel panoramic image file. The multidimensional spatial index structure and the hash attribute reverse index structure are associated with the composite tunnel panoramic image file as a spatial-attribute dual index structure.

[0051] Non-spatial attribute data includes, but is not limited to, the unique identifier (ID) of the construction joint, the actual mileage station number, the quantitative value of the crack width / length, the disease level, the detection time, and the remarks text; spatial geometric feature data includes, but is not limited to, the pixel coordinate bounding box of the construction joint in the image, the coordinate sequence of the skeleton line, the geometric center point, and the fitted inclination angle.

[0052] The panoramic image of the tunnel is used as the underlying raster data (base layer) to provide an intuitive visual background. The "multi-dimensional attribute vector objects" of all construction joints are serialized and then overlaid and encapsulated into an independent vector metadata layer. The base layer and the vector metadata layer are logically linked through a unified coordinate reference system to obtain a composite tunnel panoramic image file, forming a fused data structure where the base image is visible and the upper layers are searchable, supporting layer visibility control and independent parsing. To achieve rapid data retrieval and bidirectional positioning, two complementary indexing mechanisms (spatial-attribute dual indexing structure) are dynamically constructed when the file is loaded into runtime memory. These include a multi-dimensional spatial index structure, i.e., a forward index, which functions as a lookup of data from the image. The bounding box of the pixel coordinates of the construction joints in the image is used as the key, and the memory address pointer of the corresponding multi-dimensional attribute vector object is used as the value. This allows users to select any area on the panoramic image, and the system can quickly locate the attribute data of all construction joints within that area through the spatial index, achieving a "what you see is what you get" interactive query. The hash attribute reverse index structure, also known as the reverse index, is primarily used for map lookup. It uses the unique joint identifier (ID) or mileage marker of the construction joint as the key, and the coordinates of the bounding box geometric center point of the construction joint and its storage address in the multi-dimensional spatial index structure as the value. Users can input specific IDs or mileage markers, and the system directly locates the position using hash table complexity, immediately obtaining its spatial location to drive automatic view navigation and highlighting. Furthermore, this spatial-attribute dual index structure is embedded or associated as metadata header information in the composite tunnel panoramic image file. In subsequent inspection report generation, defect statistical analysis, and operation and maintenance management system calls, this dual index allows for one-click reverse navigation from report data to the specific pixel location in the panoramic image. Moreover, when adding or modifying construction joint attributes, only the corresponding vector object and index mapping need to be updated, without re-rendering the entire panoramic image, significantly improving system response efficiency.

[0053] According to another method for generating panoramic tunnel images based on SIFT and construction joint identification provided by the present invention, compared with the current method of directly stitching together image sequences acquired by cameras according to the image edge alignment, the present invention uses multiple acquisition devices arranged in a circumferential manner to cover the entire circumference of the tunnel cross section, avoiding blind spots from a single perspective and ensuring the integrity of the inner wall image; by calculating the geometric transformation relationship between images through overlapping areas, the image position can be calibrated more accurately, reducing stitching errors; by automatically calculating the initial transformation parameters between images through SIFT feature point matching, manual intervention is reduced and stitching efficiency is improved; by comparing the difference between the observed mileage and the actual mileage of the construction joint, the geometric transformation parameters of image stitching can be adjusted in reverse, which can eliminate long-distance cumulative errors during image stitching, thereby improving the generation accuracy of the panoramic tunnel image; at the same time, since the construction joint has obvious straight-line characteristics and its design location is known, the generation accuracy of the panoramic tunnel image can be further improved by using the construction joint as a reference object to eliminate cumulative errors.

[0054] Furthermore, as Figure 1 In specific implementation, embodiments of the present invention provide a tunnel panoramic image generation device based on SIFT and construction joint recognition, such as... Figure 3 As shown, the device includes: an image acquisition unit 31, an image stitching unit 32, a parameter optimization unit 33, and an image re-stitching unit 34.

[0055] The image acquisition unit 31 can be used to simultaneously acquire a continuous inner wall image sequence along the axial direction of the target tunnel using multiple image acquisition devices arranged circumferentially along the cross-section of the target tunnel, wherein there is an overlapping area between adjacent inner wall images in the continuous inner wall image sequence.

[0056] The image stitching unit 32 can be used to extract SIFT feature points of each adjacent inner wall image in the continuous inner wall image sequence, determine the initial spatial geometric transformation parameters of the corresponding adjacent inner wall images based on the SIFT feature points, and stitch the continuous inner wall image sequence into an initial panoramic unfolded image of the tunnel according to the initial spatial geometric transformation parameters. The adjacent inner wall images include axial sequence adjacent inner wall images along the axial extension direction of the target tunnel, and circumferential adjacent inner wall images in the cross-sectional circumferential direction of the target tunnel.

[0057] The parameter optimization unit 33 can be used to identify multiple construction joints in the initial panoramic image of the tunnel, determine the observation mileage of the feature points of the corresponding construction joint based on the pixel position information of the feature points of each construction joint in the initial panoramic image of the tunnel, obtain the actual mileage of the feature points of each construction joint, and optimize the initial spatial geometric transformation parameters based on the difference between the actual mileage and the corresponding observation mileage of the feature points of each construction joint.

[0058] The image re-stitching unit 34 can be used to re-stitch the continuous inner wall image sequence into a panoramic unfolded image of the tunnel based on the optimized initial spatial geometric transformation parameters.

[0059] In specific application scenarios, in order to stitch together a sequence of continuous inner wall images into an initial panoramic unfolded image of the tunnel, such as... Figure 4 As shown, the image stitching unit 32 includes a determining module 321 and an image stitching module 322.

[0060] The determining module 321 can be used to perform circumferential feature matching on SIFT feature points in overlapping areas of circumferential adjacent inner wall images in each circumferential single cross-section of the target tunnel, determine the same circumferential SIFT feature points in the circumferential adjacent inner wall images based on the circumferential feature matching results, and determine the initial circumferential spatial geometric transformation parameters between the circumferential adjacent inner wall images based on the spatial geometric relationship between the same circumferential SIFT feature points.

[0061] The image stitching module 322 can be used to stitch together the images of adjacent inner walls in the circumferential direction into an initial single-section tunnel circumferential unfolded diagram based on the initial circumferential spatial geometric transformation parameters.

[0062] The determining module 321 can also be used to perform axial feature matching on SIFT feature points in overlapping areas of adjacent initial single-section tunnel circumferential unfolded diagrams, determine the same axial SIFT feature points in the adjacent initial single-section tunnel circumferential unfolded diagrams based on the axial feature matching results, and determine the initial axial spatial geometric transformation parameters between the adjacent initial single-section tunnel circumferential unfolded diagrams based on the spatial geometric relationship between the same axial SIFT feature points.

[0063] The image stitching module 322 can also be used to stitch the adjacent initial single-section tunnel circumferential unfolded images into the initial tunnel panoramic unfolded image based on the initial axial spatial geometric transformation parameters.

[0064] In specific application scenarios, in order to optimize the initial spatial geometric transformation parameters, the parameter optimization unit 33 includes a function construction module 331 and a parameter optimization module 332.

[0065] The function construction module 331 can be used to construct a global optimization function with the objective of minimizing the sum of squared errors of the feature point mileages of all construction joints, based on the difference between the actual mileage and the corresponding observed mileage of each feature point of the construction joint.

[0066] The parameter optimization module 332 can be used to optimize each initial spatial geometric transformation parameter by using the global optimization function with local feature matching geometric consistency as the optimization constraint. The initial spatial geometric transformation parameters include at least one of spatial translation parameters, spatial rotation parameters, spatial scaling parameters, and spatial perspective transformation parameters.

[0067] In specific application scenarios, in order to screen high-quality construction joints, the device also includes a construction joint screening unit 35.

[0068] The construction joint screening unit 35 can be used to treat any construction joint in each of the construction joints as a target candidate construction joint, determine the geometric feature parameters of the target candidate construction joint, evaluate the geometric structure score of the target candidate construction joint based on the geometric feature parameters, obtain the groove region attribute parameters of the target candidate construction joint, evaluate the texture purity score of the target candidate construction joint based on the groove region attribute parameters, evaluate the effective feature area ratio score of the target candidate construction joint based on the construction joint image of the target candidate construction joint, perform a weighted summation of the geometric structure score, the texture purity score, and the feature area ratio score, and select a target construction joint for optimizing the initial spatial geometric transformation parameters in each of the construction joints based on the weighted summation result.

[0069] In specific application scenarios, in order to determine the observation mileage, the parameter optimization unit 33 can be used to determine the observation mileage of the feature points of the corresponding target construction joint based on the pixel position information of the feature points of each target construction joint in the initial panoramic image of the tunnel.

[0070] In specific application scenarios, the device further includes an image processing unit 36 ​​for preprocessing the inner wall image.

[0071] The image processing unit 36 ​​can be used to take any one of the inner wall images in the continuous inner wall image sequence as a target inner wall image, identify construction joints in the target inner wall image, and determine the pixel distance between each image pixel in the target inner wall image and the nearest construction joint pixel; set personalized preprocessing parameters for the corresponding image pixels based on the pixel distance, wherein the preprocessing parameters include filter kernel size parameters and edge protection strength parameters; preprocess the corresponding image pixels based on the personalized preprocessing parameters, and take the image composed of each preprocessed image pixel as the preprocessed target inner wall image.

[0072] In specific application scenarios, in order to extract image feature points, the image stitching unit 32 can be used to extract SIFT feature points of each adjacent inner wall image in the preprocessed continuous inner wall image sequence.

[0073] In specific application scenarios, in order to generate tunnel panoramic image files with a spatial-attribute dual index structure for image output and viewing, the device also includes a file generation unit 37.

[0074] The file generation unit 37 can be used to determine the non-spatial attribute data and spatial geometric feature data of each construction joint in the panoramic unfolded image of the tunnel, and encapsulate the non-spatial attribute data and spatial geometric feature data of each construction joint into multi-dimensional attribute vector objects; using the panoramic unfolded image of the tunnel as a base layer, converting each multi-dimensional attribute vector object into a vector metadata layer, and embedding the vector metadata layer into the base layer to generate a composite panoramic image file; constructing a multi-dimensional spatial index structure in the running memory of the composite panoramic image file using the bounding box of the pixel coordinates of each construction joint in the panoramic unfolded image of the tunnel as the key and the memory address of the corresponding multi-dimensional attribute vector object as the value; and constructing a hash attribute reverse index structure in the running memory of the composite panoramic image file using the joint identifier of each construction joint as the key and the geometric center point of the bounding box of the corresponding construction joint and the storage address in the multi-dimensional spatial index structure as the value; and associating the multi-dimensional spatial index structure and the hash attribute reverse index structure as a spatial-attribute dual index structure to the composite panoramic image file.

[0075] In specific application scenarios, in order to enhance the quality of the panoramic image of the tunnel, the device also includes an image enhancement unit 38.

[0076] The image enhancement unit 38 can be used to determine the radiometric attribute information of the panoramic unfolded image of the tunnel, wherein the radiometric attribute information includes the local illuminance deprivation index and the global illumination uniformity coefficient of the panoramic unfolded image of the tunnel; based on the radiometric attribute information, determine the topological constraints of the mapping function of the pixel grayscale mapping function to be generated, wherein the topological constraints of the mapping function include pixel value range constraints, monotonicity constraints of the pixel grayscale mapping function to be generated, and bidirectional mapping consistency constraints; based on the topological constraints of the mapping function, construct a pixel grayscale mapping function for nonlinear radiometric response reconstruction of the panoramic unfolded image of the tunnel; and use the pixel grayscale mapping function to perform radiometric response reconstruction of the panoramic unfolded image of the tunnel to obtain the enhanced panoramic unfolded image of the tunnel.

[0077] It should be noted that other corresponding descriptions of the functional modules involved in the tunnel panoramic image generation device based on SIFT and construction joint recognition provided in this embodiment of the invention can be found in the following references. Figure 1 The corresponding description of the method shown will not be repeated here.

[0078] Based on the above, Figure 1The method shown, correspondingly, also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the following steps: synchronously acquiring a continuous sequence of inner wall images along the axial direction of the target tunnel using multiple image acquisition devices arranged circumferentially along the cross-section of the target tunnel, wherein adjacent inner wall images in the continuous inner wall image sequence have overlapping regions; extracting SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence, determining initial spatial geometric transformation parameters for the corresponding adjacent inner wall images based on the SIFT feature points, and stitching the continuous inner wall image sequence into an initial panoramic unfolded image of the tunnel according to the initial spatial geometric transformation parameters, wherein... The adjacent inner wall images include axial sequence adjacent inner wall images along the axial extension direction of the target tunnel, and circumferential adjacent inner wall images along the cross-section of the target tunnel. Multiple construction joints are identified in the initial panoramic unfolded tunnel image, and the observation mileage of the corresponding construction joint feature points is determined based on the pixel position information of each construction joint feature point in the initial panoramic unfolded tunnel image. The actual mileage of each construction joint feature point is obtained, and the initial spatial geometric transformation parameters are optimized based on the difference between the actual mileage and the corresponding observation mileage of each construction joint feature point. Based on the optimized initial spatial geometric transformation parameters, the continuous inner wall image sequence is re-stitched into a panoramic unfolded tunnel image.

[0079] Based on the above, Figure 1 The method shown and as Figure 3 The embodiment of the device shown in the invention also provides a physical structure diagram of a computer device, such as... Figure 5As shown, the computer device includes a processor 41, a memory 42, and a computer program stored in the memory 42 and executable on the processor. Both the memory 42 and the processor 41 are mounted on a bus 43. When the processor 41 executes the program, it performs the following steps: simultaneously acquiring a continuous sequence of inner wall images along the axial direction of the target tunnel using multiple image acquisition devices arranged circumferentially along the cross-section of the target tunnel, wherein adjacent inner wall images in the continuous inner wall image sequence have overlapping regions; extracting SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence; determining initial spatial geometric transformation parameters for the corresponding adjacent inner wall images based on the SIFT feature points; and stitching the continuous inner wall image sequence together according to the initial spatial geometric transformation parameters. The initial panoramic image of the tunnel is generated, wherein the adjacent inner wall images include axial sequence adjacent inner wall images extending along the axial direction of the target tunnel, and circumferential adjacent inner wall images along the cross-section of the target tunnel. Multiple construction joints are identified in the initial panoramic image, and the observation mileage of the corresponding feature points of each construction joint is determined based on the pixel position information of the feature points in the initial panoramic image. The actual mileage of the feature points of each construction joint is obtained, and the initial spatial geometric transformation parameters are optimized based on the difference between the actual mileage and the corresponding observation mileage of the feature points of each construction joint. The continuous inner wall image sequence is then re-stitched into a panoramic image of the tunnel based on the optimized initial spatial geometric transformation parameters.

[0080] Through the technical solution of this invention, multiple acquisition devices arranged in a ring can cover the entire circumference of the tunnel cross-section, avoiding blind spots from a single perspective and ensuring the integrity of the inner wall image; by calculating the geometric transformation relationship between images through overlapping areas, the image position can be calibrated more accurately, reducing stitching errors; by automatically calculating the initial transformation parameters between images through SIFT feature point matching, manual intervention is reduced and stitching efficiency is improved; by comparing the difference between the observed mileage and the actual mileage of the construction joint, the geometric transformation parameters of image stitching can be adjusted in reverse, eliminating long-distance cumulative errors during image stitching, thereby improving the generation accuracy of the tunnel panoramic image; at the same time, since the construction joint has obvious straight-line characteristics and its design location is known, by using the construction joint as a reference object to eliminate cumulative errors, the generation accuracy of the tunnel panoramic image can be further improved.

[0081] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating panoramic tunnel images based on SIFT and construction joint recognition, characterized in that, include: Multiple image acquisition devices arranged circumferentially along the cross-section of the target tunnel are used to simultaneously acquire a continuous sequence of inner wall images along the axial direction of the target tunnel, wherein there is an overlapping area between adjacent inner wall images in the continuous inner wall image sequence; SIFT feature points are extracted from each adjacent inner wall image in the continuous inner wall image sequence. Initial spatial geometric transformation parameters for the corresponding adjacent inner wall images are determined based on the SIFT feature points. The continuous inner wall image sequence is then stitched together into an initial panoramic unfolded image of the tunnel according to the initial spatial geometric transformation parameters. The adjacent inner wall images include axial sequence adjacent inner wall images along the axial extension direction of the target tunnel, and circumferential adjacent inner wall images in the cross-sectional circumferential direction of the target tunnel. Multiple construction joints are identified in the initial panoramic image of the tunnel. Based on the pixel position information of the feature points of each construction joint in the initial panoramic image of the tunnel, the observation mileage of the feature points of the corresponding construction joint is determined. The actual mileage of the feature points of each construction joint is obtained. The initial spatial geometric transformation parameters are optimized based on the difference between the actual mileage and the corresponding observation mileage of the feature points of each construction joint. Based on the optimized initial spatial geometric transformation parameters, the continuous inner wall image sequence is re-stitched into a panoramic unfolded image of the tunnel.

2. The method according to claim 1, characterized in that, Based on the SIFT feature points, initial spatial geometric transformation parameters are determined for corresponding adjacent inner wall images. Then, the continuous inner wall image sequence is stitched together into an initial panoramic unfolded tunnel image according to the initial spatial geometric transformation parameters, including: For the circumferential adjacent inner wall images in each circumferential single section of the target tunnel, circumferential feature matching is performed on the SIFT feature points in the overlapping area of ​​the circumferential adjacent inner wall images. Based on the circumferential feature matching results, the same circumferential SIFT feature points in the circumferential adjacent inner wall images are determined, and the initial circumferential spatial geometric transformation parameters between the circumferential adjacent inner wall images are determined based on the spatial geometric relationship between the same circumferential SIFT feature points. Based on the initial circumferential spatial geometric transformation parameters, the adjacent circumferential inner wall images are stitched together to form an initial single-section tunnel circumferential unfolded diagram. Axial feature matching is performed on the SIFT feature points of the overlapping area in the circumferential unfolded diagrams of adjacent initial single-section tunnels. Based on the axial feature matching results, the same axial SIFT feature points in the circumferential unfolded diagrams of adjacent initial single-section tunnels are determined. Based on the spatial geometric relationship between the same axial SIFT feature points, the initial axial spatial geometric transformation parameters between the circumferential unfolded diagrams of adjacent initial single-section tunnels are determined. Based on the initial axial spatial geometric transformation parameters, the adjacent initial single-section tunnel circumferential unfolded diagrams are stitched together to form the initial tunnel panoramic unfolded diagram.

3. The method according to claim 1, characterized in that, The optimization of the initial spatial geometric transformation parameters based on the difference between the actual mileage and the corresponding observed mileage of each feature point of the construction joint includes: Based on the difference between the actual mileage and the corresponding observed mileage of each feature point of the construction joint, a global optimization function is constructed with the objective of minimizing the sum of squared errors of the feature point mileages of all construction joints. Using the geometric consistency of local feature matching as an optimization constraint, the global optimization function is used to jointly optimize each of the initial spatial geometric transformation parameters, wherein the initial spatial geometric transformation parameters include at least one of spatial translation parameters, spatial rotation parameters, spatial scaling parameters, and spatial perspective transformation parameters.

4. The method according to claim 1, characterized in that, Before determining the observation mileage of the corresponding construction joint feature point based on the pixel position information of the feature point in the initial panoramic image of the tunnel, the method further includes: Each construction joint is taken as a target candidate construction joint, the geometric feature parameters of the target candidate construction joint are determined, and the geometric structure score of the target candidate construction joint is evaluated based on the geometric feature parameters. Obtain the groove region attribute parameters of the target candidate construction joint, and evaluate the texture purity score of the target candidate construction joint based on the groove region attribute parameters; The effective feature region ratio score of the target candidate construction joint is evaluated based on the construction joint image of the target candidate construction joint; The geometric structure score, the texture purity score, and the feature region proportion score are weighted and summed, and a target construction joint for optimizing the initial spatial geometric transformation parameters is selected in each construction joint based on the weighted summation result. The step of determining the observation mileage of the corresponding construction joint feature point based on the pixel position information of the feature point of each construction joint in the initial panoramic image of the tunnel includes: The observation mileage of the feature points of each target construction joint is determined based on the pixel position information of the feature points in the initial panoramic image of the tunnel.

5. The method according to claim 1, characterized in that, Before extracting SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence, the method further includes: Each inner wall image in the continuous inner wall image sequence is taken as a target inner wall image. Construction joints are identified in the target inner wall images, and the pixel distance between each image pixel in the target inner wall image and the nearest construction joint pixel is determined. Personalized preprocessing parameters are set for the corresponding image pixels based on the pixel distance, wherein the preprocessing parameters include filter kernel size parameters and edge protection strength parameters; Based on the personalized preprocessing parameters, the corresponding image pixels are preprocessed, and the image composed of each preprocessed image pixel is used as the preprocessed target inner wall image. The step of extracting SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence includes: SIFT feature points are extracted from each adjacent inner wall image in the preprocessed continuous inner wall image sequence.

6. The method according to claim 1, characterized in that, After re-stitching the continuous inner wall image sequence into a panoramic unfolded image of the tunnel based on the optimized initial spatial geometric transformation parameters, the method further includes: The non-spatial attribute data and spatial geometric feature data of each construction joint in the panoramic unfolded image of the tunnel are determined, and the non-spatial attribute data and spatial geometric feature data of each construction joint are encapsulated into multi-dimensional attribute vector objects respectively. Using the panoramic image of the tunnel as the base layer, each of the multidimensional attribute vector objects is converted into a vector metadata layer, and the vector metadata layer is embedded into the base layer to generate a composite panoramic tunnel image file. A multidimensional spatial index structure is constructed in the runtime memory of the composite tunnel panoramic image file using the bounding box of the pixel coordinates of each construction joint in the panoramic image of the tunnel as the key and the memory address of the corresponding multidimensional attribute vector object as the value. A hash attribute inverse index structure is constructed in the runtime memory of the composite tunnel panoramic image file using the joint identifier of each construction joint as the key and the geometric center point of the bounding box of the corresponding construction joint and the storage address in the multidimensional spatial index structure as the value. The multidimensional spatial index structure and the hash attribute inverse index structure are associated with the composite tunnel panoramic image file as a spatial-attribute dual index structure.

7. The method according to claim 1, characterized in that, After re-stitching the continuous inner wall image sequence into a panoramic unfolded image of the tunnel based on the optimized initial spatial geometric transformation parameters, the method further includes: The radiometric attribute information of the panoramic unfolded image of the tunnel is determined, wherein the radiometric attribute information includes the local illuminance deficiency index and the global illuminance uniformity coefficient of the panoramic unfolded image of the tunnel. Based on the radiometric attribute information, the topological constraints of the mapping function of the pixel grayscale mapping function to be generated are determined, wherein the topological constraints of the mapping function include pixel value range constraints, monotonicity constraints of the pixel grayscale mapping function to be generated, and bidirectional mapping consistency constraints. Based on the topological constraints of the mapping function, a pixel grayscale mapping function is constructed to reconstruct the nonlinear radiometric response of the panoramic unfolded image of the tunnel. The radiometric response reconstruction of the tunnel panoramic unfolded image is performed using the pixel grayscale mapping function to obtain the quality-enhanced tunnel panoramic unfolded image.

8. A device for generating panoramic tunnel images based on SIFT and construction joint recognition, characterized in that, include: An image acquisition unit is used to simultaneously acquire a continuous sequence of inner wall images along the axial direction of the target tunnel using multiple image acquisition devices arranged circumferentially along the cross-section of the target tunnel, wherein there is an overlapping area between adjacent inner wall images in the continuous inner wall image sequence; An image stitching unit is used to extract SIFT feature points from each adjacent inner wall image in the continuous inner wall image sequence, determine the initial spatial geometric transformation parameters of the corresponding adjacent inner wall images based on the SIFT feature points, and stitch the continuous inner wall image sequence into an initial panoramic unfolded image of the tunnel according to the initial spatial geometric transformation parameters. The adjacent inner wall images include axial sequence adjacent inner wall images along the axial extension direction of the target tunnel, and circumferential adjacent inner wall images in the cross-sectional circumferential direction of the target tunnel. The parameter optimization unit is used to identify multiple construction joints in the initial panoramic image of the tunnel, determine the observation mileage of the feature points of the corresponding construction joint based on the pixel position information of the feature points of each construction joint in the initial panoramic image of the tunnel, obtain the actual mileage of the feature points of each construction joint, and optimize the initial spatial geometric transformation parameters based on the difference between the actual mileage and the corresponding observation mileage of the feature points of each construction joint. The image re-stitching unit is used to re-stitch the continuous inner wall image sequence into a panoramic unfolded image of the tunnel based on the optimized initial spatial geometric transformation parameters.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images