Turnout three-dimensional visualization method, device, equipment and medium

Abstract

The invention discloses a turnout three-dimensional visualization method, device and equipment and a medium, and the method comprises the steps: collecting and obtaining video data of a plurality of visual angles of turnout parts, and carrying out the preprocessing, and obtaining a plurality of pieces of image data; calculating the multiple pieces of image data to obtain initialized three-dimensional point cloud data; training the three-dimensional point cloud data by using a three-dimensional Gaussian algorithm to generate a three-dimensional Gaussian model; and carrying out real-time visual rendering on the three-dimensional Gaussian model, and checking rendering images of the turnout parts at multiple different visual angles. According to the invention, the shot turnout part is represented by using the three-dimensional Gaussian model, and the images of the turnout part at multiple angles are rendered in real time, so that a user can detect the turnout part without on-site exploration; the detection efficiency is improved, manpower and material resources are saved, and the detection cost is reduced; and the detection precision and the detection timeliness of the turnout parts are further ensured, and the method is more suitable for application requirements in dynamic scenes.

Description

Technical Field

[0001] The present invention relates to the field of railway and three-dimensional reconstruction technology, and in particular to a three-dimensional visualization method, device, equipment and medium for a turnout. Background Art

[0002] Turnouts are an indispensable part of rail transit lines, but they are also the weak link of the lines. Defects in turnout components will directly affect the operational safety of urban rail transit vehicles. Therefore, it is of great significance to timely detect whether there are defects in the components of turnouts.

[0003] At present, the commonly used inspection method for turnout parts mainly relies on the combination of manual field inspection and small monitoring instruments. This inspection method not only consumes a lot of manpower and material resources, has low inspection efficiency and high inspection cost, but also has poor real-time performance due to the fixed inspection time, and cannot detect sudden faults of turnouts in time. Summary of the invention

[0004] The present invention provides a three-dimensional visualization method, device, equipment and medium for a turnout, which solves the problem that the turnout component detection method in the prior art mainly relies on the combination of manual field inspection and small monitoring instruments, which not only consumes a lot of manpower and material resources, has low detection efficiency and high detection cost, but also has poor real-time performance due to the fixed detection time, and cannot timely detect sudden diseases of the turnout.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is to provide a three-dimensional visualization method for turnouts, comprising the steps of: acquiring video data of multiple perspectives of turnout components, preprocessing the video data to obtain image data of multiple different perspectives; calculating the multiple image data to obtain initialized three-dimensional point cloud data; using a three-dimensional Gaussian algorithm to train the three-dimensional point cloud data to generate a corresponding three-dimensional Gaussian model; performing real-time visualization rendering of the three-dimensional Gaussian model to view the rendered images of the turnout components from multiple different perspectives.

[0006] In some embodiments, the step of calculating the multiple image data to obtain initialized three-dimensional point cloud data specifically includes: extracting feature points of the multiple image data respectively; performing feature matching on the feature points of the image data to obtain feature matching results; performing motion structure recovery on the feature matching results to generate a motion trajectory of the camera; calculating the feature points based on the feature matching results to obtain corresponding three-dimensional point coordinates; and reconstructing and generating the three-dimensional point cloud data using the three-dimensional point coordinates and the motion trajectory of the camera.

[0007] In some embodiments, the step of performing motion structure recovery on the feature matching results to generate the motion trajectory of the camera specifically includes: calculating the basic matrix and the intrinsic matrix of the camera based on the feature matching results; performing singular value decomposition on the intrinsic matrix to obtain the rotation matrix and translation vector of the camera; and using the rotation matrix and translation vector to generate the motion trajectory of the camera.

[0008] In some embodiments, the step of reconstructing and generating the three-dimensional point cloud data using the three-dimensional point coordinates and the motion trajectory of the camera specifically includes: defining a reprojection error function, adjusting the three-dimensional point coordinates and the external parameters of the camera until the value of the reprojection error function converges to a first threshold; and reconstructing and generating the three-dimensional point cloud data based on the adjusted three-dimensional point coordinates and the motion trajectory of the camera.

[0009] In some embodiments, the step of using a three-dimensional Gaussian algorithm to train the three-dimensional point cloud data and generate a corresponding three-dimensional Gaussian model specifically includes: initializing the three-dimensional point cloud data, generating a set of corresponding Gaussian ellipsoid sets according to each point in the three-dimensional point cloud data, and obtaining an initial three-dimensional Gaussian model; projecting the initial three-dimensional Gaussian model onto a two-dimensional plane at a target angle, and performing rasterization rendering to obtain a corresponding two-dimensional image; using a target loss function, comparing the two-dimensional image with its corresponding real image, and based on the comparison result, back-propagating to optimize the parameters of the initial three-dimensional Gaussian model until the value of the target loss function converges to obtain the three-dimensional Gaussian model.

[0010] In some embodiments, the step of projecting the initial three-dimensional Gaussian model onto a two-dimensional plane at a target angle and performing rasterization rendering to obtain a corresponding two-dimensional image specifically includes: projecting the initial three-dimensional Gaussian model onto a two-dimensional plane at a target angle to obtain a plurality of different projection ellipses; dividing the two-dimensional plane into a plurality of projection blocks, and sorting all of the Gaussian ellipsoids according to the external parameters of the camera and the distances between the plurality of Gaussian ellipsoids in the initial three-dimensional Gaussian model and the two-dimensional plane; and calculating the color of each pixel in the projection block to obtain the corresponding two-dimensional image.

[0011] In some embodiments, the video data is preprocessed, and the preprocessing includes at least one of the following operations: denoising, lens distortion correction, and enhancement.

[0012] The present invention also provides a three-dimensional visualization device for a turnout, comprising: a data acquisition module, used to collect and acquire video data of multiple perspectives of turnout components, and pre-process the video data to obtain image data of multiple different perspectives; a point cloud initialization module, used to calculate the multiple image data to obtain initialized three-dimensional point cloud data; a three-dimensional Gaussian model generation module, used to train the three-dimensional point cloud data using a three-dimensional Gaussian algorithm to generate a corresponding three-dimensional Gaussian model; a visualization module, used to perform real-time visualization rendering of the three-dimensional Gaussian model to view the rendered images of the turnout components from multiple different perspectives.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method when executing the computer program.

[0014] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and is characterized in that when the computer program is executed by a processor, the steps of the method described above are implemented.

[0015] The beneficial effects of the present invention are as follows: the present invention discloses a three-dimensional visualization method, device, equipment and medium for turnouts, the method comprising the steps of: acquiring video data of multiple perspectives of turnout components, preprocessing to obtain multiple image data; calculating multiple image data to obtain initialized three-dimensional point cloud data; using a three-dimensional Gaussian algorithm to train the three-dimensional point cloud data to generate a three-dimensional Gaussian model; performing real-time visualization rendering of the three-dimensional Gaussian model to view the rendered images of the turnout components from multiple different perspectives. The present invention uses a three-dimensional Gaussian model to represent the photographed turnout components and renders images of multiple angles in real time, so that users can detect turnout components without field investigation; it not only improves detection efficiency, saves human and material resources, and reduces the cost required for detection; it also further ensures the detection accuracy and timeliness of turnout components, and is more adaptable to application requirements in dynamic scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the process of a three-dimensional visualization method of a turnout in the present application; Figure 2 It is a specific flow chart of S2 in a three-dimensional visualization method of a turnout in the present application; Figure 3 It is a specific flow chart of S3 in a three-dimensional visualization method of a turnout in the present application; Figure 4 It is a schematic diagram of a rendered image of a left-view direction of one of the turnout components in a three-dimensional turnout visualization method of the present application; Figure 5 It is a schematic diagram of a rendered image of a right-view direction of one of the turnout components in a three-dimensional turnout visualization method of the present application; Figure 6 It is a schematic diagram of a rendered image in a front-view direction of one of the turnout components in a three-dimensional turnout visualization method of the present application; Figure 7 It is a schematic diagram of a rendered image in a rear-view direction of one of the turnout components in a three-dimensional turnout visualization method of the present application; Figure 8 It is a schematic diagram of a rendered image of a turnout component magnified in an oblique direction in a turnout three-dimensional visualization method of the present application; Fig. 9 This is a schematic diagram of a rendered image of a turnout component in the oblique direction in a turnout 3D visualization method of the present application. Figure 1 ; Fig.10 This is a schematic diagram of a rendered image of a turnout component in the oblique direction in a turnout 3D visualization method of the present application. Figure 2 ; Fig.11 This is a schematic diagram of a rendered image of a turnout component in the oblique direction in a turnout 3D visualization method of the present application. Figure 3 ; Fig.12 It is a structural block diagram of an embodiment of a three-dimensional visualization device for a turnout of the present application; Fig.13 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present application; Fig.14 It is a schematic block diagram of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0017] In order to facilitate the understanding of the present invention, the present invention is described in more detail below in conjunction with the accompanying drawings and specific embodiments. Preferred embodiments of the present invention are provided in the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in this specification. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

[0018] It should be noted that, unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used in this specification includes any and all combinations of one or more related listed items.

[0019] Figure 1 The implementation flow chart of the three-dimensional visualization method of the turnout provided by the present application is shown, including the steps of: S1: Collect and obtain video data of turnout components from multiple perspectives, and pre-process the video data to obtain image data from multiple different perspectives.

[0020] S2: Calculate multiple image data to obtain initialized three-dimensional point cloud data.

[0021] S3: Use the 3D Gaussian algorithm to train the 3D point cloud data and generate the corresponding 3D Gaussian model.

[0022] S4: Real-time visualization rendering of the 3D Gaussian model to view rendered images of the turnout components from multiple different perspectives.

[0023] This application uses a three-dimensional Gaussian model to represent the photographed turnout components and renders images from multiple angles in real time, allowing users to detect turnout components without on-site inspections. This not only improves detection efficiency, saves manpower and material resources, and reduces detection costs; it also further ensures the detection accuracy and timeliness of turnout components, and is more adaptable to application needs in dynamic scenarios.

[0024] Combine the following Figures 1 to 14 , the present application is further described in detail with specific embodiments.

[0025] like Figure 1 As shown, the three-dimensional visualization method of the turnout of the present application includes the following steps: S1: Collect and obtain video data of turnout components from multiple perspectives, and pre-process the video data to obtain image data from multiple different perspectives.

[0026] First, multiple PTZ (Pan Tilt Zoom) cameras are set up on both sides of the turnout in the rail transit line, and the shooting posture and camera internal parameters of each PTZ camera are configured and adjusted to ensure that the images captured by all PTZ cameras can cover all parts and multiple angles of the turnout.

[0027] It should be noted that PTZ cameras can move in all directions (left / right / up / down) and control the zoom and zoom of the lens. Camera intrinsic parameters refer to the parameters that describe the internal properties of the PTZ camera. These parameters are related to the geometric properties and posture of the PTZ camera, including focal length, principal point (optical center) coordinates, distortion coefficient, etc. Camera intrinsic parameters are the bridge between the camera coordinate system and the image coordinate system. Through the camera intrinsic parameters, the three-dimensional point in the camera coordinate system can be projected to the two-dimensional point in the image coordinate system.

[0028] In this embodiment, by using a plurality of pre-set PTZ cameras, video data of a plurality of different perspectives of the same turnout component is collected, and a certain degree of overlap is maintained when shooting the same component at different angles, so that the video data obtained at different angles contain a certain number of the same feature points. Among them, the plurality of different perspectives refers to perspectives that can capture the complete structure of the turnout component, such as left view, right view, front view, rear view, top view and top view.

[0029] Acquiring relevant data of turnout components through PTZ cameras is not only convenient and fast, but also saves costs. The resolution of the acquired video data is also high, which helps to process textures and details, further improving the visual presentation effect and information richness.

[0030] Furthermore, the acquired video data is preprocessed.

[0031] Specifically, the video data is first split into two frames per second, so that a set of multiple continuous images of turnout components shot along a certain motion trajectory can be obtained, and the blurred images and those with insufficient overlap are deleted to obtain image data from multiple different perspectives.

[0032] After deletion, lens distortion correction processing is performed on multiple continuous image data to eliminate the distortion of the image data; at the same time, a filtering algorithm (such as Gaussian filtering and median filtering) is used to remove noise in the image data; and the image data is enhanced, that is, the contrast and brightness of the image data are enhanced to improve the quality of the image data for subsequent processing.

[0033] S2: Calculate multiple image data to obtain initialized three-dimensional point cloud data.

[0034] Specifically, if Figure 2 As shown, S2 includes the following sub-steps: S21: Extract feature points of multiple image data respectively.

[0035] In this embodiment, the feature point consists of two parts: a key point and a descriptor.

[0036] Among them, key points are points found in an image, and their function is to find landmark points (representative points) in an image. For each key point, there are three pieces of information: position, scale, and direction. The function of the descriptor is to match the corresponding key points in the two images when the key points of the two images are found, usually vectors.

[0037] A descriptor, or feature descriptor, is a vector or a set of eigenvalues ​​used to describe the features of the area around a feature point. Feature descriptors are used to characterize local feature information in images, point clouds or other data, and are usually used for tasks such as matching, recognition and tracking. Feature descriptors need to have properties such as robustness, uniqueness, compactness and computational efficiency to ensure that they are invariant to a variety of geometric and illumination transformations. At the same time, the feature descriptor of an area can be well distinguished from other areas, and the feature can be robustly and uniquely described using only a low-dimensional vector, and the feature descriptor can be quickly calculated and efficiently compared.

[0038] Furthermore, in the process of feature point extraction, for each image, key points are first extracted, and then the descriptors corresponding to the key points are calculated. A descriptor is established for each key point so that it does not change with various changes, such as changes in lighting, changes in viewing angle, etc., and the descriptor has a high uniqueness, so as to increase the probability of correct matching of feature points.

[0039] Here, commonly used feature descriptors include but are not limited to SIFT and SURF, which play an important role in computer vision tasks such as target detection, image matching and 3D reconstruction. Among them, SIFT (Scale Invariant Feature Transform) feature extraction algorithm is a widely used feature extraction method in computer vision, which can detect and describe local features in images. In addition, SIFT feature extraction algorithm can extract stable feature points at different scales and rotation angles, and has strong robustness to interference such as illumination and noise. SIFT feature extraction algorithm includes constructing Difference of Gaussian (DOG) scale space, key point search and positioning, direction assignment and key point descriptor generation. The extracted SIFT features have a large amount of information and are suitable for fast and accurate matching in massive databases. SIFT features are based on some local appearance interest points on objects, which are independent of the size and rotation of the image and have a high tolerance to changes in light, noise and perspective.

[0040] Furthermore, in this embodiment, the SIFT feature extraction algorithm is used to extract feature points of the image data preprocessed in S1. Among them, the local features of the image data are extracted, and these feature points remain invariant to the rotation, scale scaling, brightness change, etc. of the image data. The scale invariance is to ensure that the key points can be detected from far to near, and the rotation invariance is to ensure that the key points can still be detected after the image is rotated. At the same time, a certain degree of stability is maintained for changes in viewing angle, affine transformation, and noise, so that effective feature extraction and feature matching can be performed in complex environments. Among them, feature extraction is the basis of feature matching, and the feature points extracted by the SIFT feature extraction algorithm can be further used in subsequent feature matching.

[0041] S22: Perform feature matching on the feature points of the image data to obtain feature matching results.

[0042] In this embodiment, for a feature point in one image, two feature points that are most similar to its descriptor are searched in another image, and feature matching is achieved by calculating the distance between the corresponding descriptors in the two images. At the same time, the reliability of the match is judged by a ratio test, specifically: comparing the distance ratio between the nearest neighbor and the next nearest neighbor of the descriptor. If the distance ratio between the nearest neighbor and the next nearest neighbor is less than the ratio threshold, the two corresponding feature points are matched as a feature point pair, and the match is considered reliable, which also indicates that there is a correlation between the corresponding two images, and all reliable matches are retained; if the distance ratio between the nearest neighbor and the next nearest neighbor is greater than or equal to the ratio threshold, the match is considered unreliable, and the unreliable match is eliminated, thereby obtaining a feature matching result. In this process, by comparing the distance ratio between the nearest neighbor and the next nearest neighbor, the correctness of the match can be judged, thereby improving the accuracy and efficiency of the match.

[0043] Furthermore, the FLANN feature matching algorithm can be used to perform feature matching on the feature points of the image data to obtain feature matching results. Among them, the FLANN (Fast Library for Approximate Nearest Neighbors) feature matching algorithm can achieve fast and efficient matching. The FLANN feature matching algorithm significantly reduces the complexity of searching for nearest neighbors in high-dimensional space by constructing data structures such as KD trees and KMeans trees. Compared with traditional linear search methods, the FLANN feature matching algorithm can reduce the search complexity from O (N) to O (log N), greatly improving the search efficiency.

[0044] S23: Perform motion structure recovery on the feature matching results to generate a motion trajectory of the camera.

[0045] Specifically, the basic matrix and the essential matrix of the camera are calculated based on the feature matching results (i.e., the matched feature point pairs) obtained in S22. That is, the essential matrix or the basic matrix is ​​solved by matching the feature points in different images, the relative position and posture between the cameras are estimated, and the three-dimensional structure in the scene is restored.

[0046] The computational method includes: First, define the problem: given 8 matching feature point pairs, each feature point pair corresponds to a point in both images, and these feature point pairs satisfy the epipolar constraint, that is, the corresponding point of a point in one image in the other image must be on the epipolar line. Second, construct equations: for each pair of feature points, an equation about the essential matrix or fundamental matrix can be established. Both the essential matrix and the fundamental matrix are 3×3 matrices used to describe the epipolar geometric relationship between the two images. Third, combine equations: combine all 8 equations into a linear system of equations. Fourth, solve equations: if the rank of the matrix of the linear system of equations is 8, then the system of equations has a unique solution; if the rank is less than 8, there may be multiple solutions. By solving this linear system of equations, an approximate solution to the essential matrix or fundamental matrix can be obtained. Fifth, verify the uniqueness of the solution: the correctness of the solution can be further confirmed by checking the uniqueness of the solution (for example, by calculating the rank of the solution or checking whether it satisfies certain geometric conditions). It should be noted that the fundamental matrix and the essential matrix are two important concepts in the field of computer vision. They respectively describe the relationship between images of the same scene (such as a switch component) taken from different perspectives. The fundamental matrix is ​​a 3×3 matrix with a rank of 2. It expresses the correspondence between feature points in a feature point pair. Specifically, if there are two images, taken from two different cameras or camera positions, the fundamental matrix can be used to find the corresponding points of the feature points in one image in the other image. The fundamental matrix describes the geometric constraints of the corresponding points between two different images and is the basis for solving stereo vision problems in computer vision. The essential matrix is ​​a special case of the fundamental matrix. The essential matrix describes the relationship between the projections of the same feature point in space in two different camera coordinate systems. The difference between it and the fundamental matrix is ​​that the essential matrix takes into account the internal parameters and relative position of the camera (that is, the external parameters of the camera: rotation matrix and translation vector), while the fundamental matrix focuses more on describing the geometric relationship between two different images without involving specific camera parameters.

[0047] Furthermore, the obtained essential matrix is ​​subjected to singular value decomposition to obtain the rotation matrix and translation vector (i.e., the extrinsic parameters of the camera) of the camera.

[0048] Among them, singular value decomposition (SVD) is a matrix factorization method and a concept in linear algebra. Any matrix can be expressed as the product (factorization) of three matrices. The singular value decomposition of a matrix must exist, but it is not unique. It can be regarded as a method of matrix data compression, that is, to approximate the original matrix by factorization. This approximation is the optimal approximation in the sense of square loss.

[0049] ‌Rotation matrices and translation vectors are two basic transformations used to describe the position and orientation of an object (such as a camera) in three-dimensional space. ‌Rotation matrices‌ are special matrices used to describe the rotation of an object in three-dimensional space. ‌Translation vectors‌ are used to describe the translational motion of an object in three-dimensional space. A translation vector represents the displacement from one coordinate system to another. It is a vector that represents the direction and magnitude starting from the origin and ending at another point. A translation vector is used to move an object from one position to another. A translation vector is independent of the coordinate system and will not change due to changes in the coordinate system. The rotation matrix and translation vector together constitute the transformation matrix, which is used to fully describe the changes in the position and orientation of an object in three-dimensional space.

[0050] Furthermore, the rotation matrix and translation vector are used to generate the motion trajectory of the camera.

[0051] Specifically, the position and posture of the camera in each image (i.e., the orientation of the camera) can be determined through the camera's rotation matrix and translation vector, and the camera's motion trajectory can be further obtained. The camera's motion trajectory reflects the camera's movement path and direction during the shooting process, i.e., the changes in the camera's position and posture over time.

[0052] Specifically, using the eight-point algorithm to solve the essential matrix or the basic matrix requires at least 8 matching feature point pairs. The core idea of ​​the eight-point algorithm is to solve the essential matrix or the basic matrix through 8 matching feature point pairs, which is used to reconstruct the three-dimensional scene from multiple images.

[0053] S24: Calculate the feature points according to the feature matching results to obtain the corresponding three-dimensional point coordinates.

[0054] In this embodiment, the method for calculating the corresponding three-dimensional point coordinates includes: When the feature points and the camera's projection matrix (at least 2 pairs) are given, the corresponding 3D point coordinates can be calculated by direct linear transformation or least squares method. Ideally, the coordinates of the 3D point can be obtained by the intersection of two rays, but due to calculation or detection errors, the intersection of the two rays may not be guaranteed in practice, so a new mathematical model (such as least squares method) needs to be established for solution. The implementation process involves mathematical derivations such as vector dot multiplication and vector cross multiplication. Vector dot multiplication geometrically reflects the similarity of two vectors, while vector cross multiplication is a binary operation in three-dimensional space, which is used to calculate the outer product of two vectors.

[0055] In this embodiment, the pixel coordinates of the feature points are converted into rays in the camera coordinate system according to the pixel coordinates of the feature matching results (i.e., the matched feature point pairs) obtained in S22 and the corresponding intrinsic and extrinsic parameters of the camera. Each feature point corresponds to a ray in three-dimensional space, and the goal of triangulation is to find the intersection of these rays. Using the least squares method, the optimal estimate of the intersection of all rays is calculated, and the coordinates of each feature point in three-dimensional space are calculated to minimize the reprojection error between rays at different angles. Finally, the obtained three-dimensional point coordinates are evaluated to check whether the reprojection error is within an acceptable range, that is, to make the position of the matched feature point in three-dimensional space as consistent as possible with the position in the two-dimensional image.

[0056] Specifically, a triangulation algorithm can also be used. The triangulation algorithm is a method of recovering the coordinates of a three-dimensional point from the projection coordinates and transformation matrix (i.e., rotation matrix and translation vector) of two cameras. By recovering the coordinates of a three-dimensional point from the projection coordinates and transformation matrix of two cameras, information recovery from a two-dimensional image to a three-dimensional space is achieved, which is of great significance for tasks such as building three-dimensional models and scene understanding. The basic principle of the triangulation algorithm is based on epipolar geometry and vector operations. By calculating the projection matrix at two different angles and using the constraints at the two angles, the coordinates of the three-dimensional point are solved.

[0057] S25: Reconstruct and generate three-dimensional point cloud data using the three-dimensional point coordinates and the motion trajectory of the camera.

[0058] Specifically, a reprojection error function is defined, and bundle adjustment is used to adjust the three-dimensional point coordinates and the camera's extrinsic parameters (i.e., the camera's position and posture) to reduce the reprojection error and cumulative error until the value of the reprojection error function converges to a first threshold, and the adjusted three-dimensional point coordinates and the camera's extrinsic parameters are saved.

[0059] In this embodiment, for each matched feature point, its position (i.e., projection point) projected onto the two-dimensional image by the camera under the three-dimensional point coordinates is calculated; the value of the reprojection error function is defined as the difference between the feature point and the projection point; the bundle adjustment is used to adjust the camera's external parameters and the three-dimensional point coordinates so that the difference between the feature point and the projection point gradually decreases, even if the value of the reprojection error function gradually decreases, and it is iteratively updated until the error converges to a first threshold (i.e., a preset threshold), thereby improving the accuracy of the camera's motion trajectory and the three-dimensional point coordinates by minimizing the reprojection error.

[0060] It should be noted that the reprojection error function is an important indicator used to measure the accuracy of camera calibration results. Specifically, the reprojection error usually involves the following steps: defining the reprojection error; calculating the reprojection error; optimizing and adjusting. It measures the accuracy of calibration by comparing the difference between the theoretical projection position and the actual measured position. By adjusting the camera's external parameters through bundle adjustment to minimize this error, the accuracy of calibration and image processing can be improved.

[0061] Bundle Adjustment (‌BA) is a technique widely used in computer vision and photogrammetry to optimize the reconstruction of a 3D scene. This process involves adjusting the camera's extrinsic parameters and 3D point coordinates to minimize the difference between the actual image and the image predicted by the model.

[0062] Furthermore, based on the adjusted 3D point coordinates and the motion trajectory of the camera, the 3D structure of the entire scene (i.e., the turnout) is restored, the position and shape information of the objects in the scene (i.e., the components of the turnout) are obtained, and the corresponding 3D point cloud data is acquired.

[0063] S3: Use the 3D Gaussian algorithm to train the 3D point cloud data and generate the corresponding 3D Gaussian model.

[0064] Among them, 3D Gaussian Splatting is a technology for real-time differentiable rendering. It describes the scene through 3D Gaussian distribution and uses these 3D Gaussian distributions for efficient rasterization to achieve high-quality 3D rendering. This method models the 3D point cloud data in the scene by converting it into the form of a Gaussian function, combining the advantages of implicit and explicit radiation fields, aiming to achieve high-quality rendering while having faster training and real-time performance. The implementation process of the 3D Gaussian Splatting includes several key steps: starting from the obtained sparse 3D point cloud data, using 3D Gaussian to represent the scene; cross-optimizing and density controlling the 3D Gaussian; and using a fast visual perception rendering algorithm for fast training and rendering. The advantages of the 3D Gaussian Splatting include fast training and rendering speeds, while meeting or exceeding the quality of other methods.

[0065] Specifically, if Figure 3 As shown, S3 includes the following sub-steps: S31: Initialize the three-dimensional point cloud data, generate a set of corresponding Gaussian ellipsoid sets according to each point in the three-dimensional point cloud data, and obtain an initial three-dimensional Gaussian model.

[0066] Specifically, the three-dimensional point cloud data is initialized using a three-dimensional Gaussian algorithm, and a three-dimensional Gaussian is used to represent the scene according to each point in the three-dimensional point cloud data to generate a set of corresponding Gaussian ellipsoid sets.

[0067] The Gaussian ellipsoid set includes multiple Gaussian ellipsoids, each of which has parameters such as opacity, position, color, covariance, etc. An initial three-dimensional Gaussian model is created based on the generated Gaussian ellipsoid set.

[0068] S32: Projecting the initial three-dimensional Gaussian model onto a two-dimensional plane at a target angle, and performing rasterization rendering to obtain a corresponding two-dimensional image.

[0069] In this embodiment, the target angle refers to the angle of the camera currently used for training; specifically, the Splatting algorithm is used to project the initial three-dimensional Gaussian model onto a two-dimensional plane at the target angle, and raster rendering is performed to obtain a corresponding two-dimensional image.

[0070] Among them, the ‌Splatting algorithm‌, also known as the footprint method, is different from the ray casting method in that it generates images by repeatedly calculating the projection superposition effect of volume elements. The Splatting algorithm uses a function called footprint to calculate the influence range of each volume element projection, and uses a Gaussian function to define the intensity distribution of points or small area pixels, so as to calculate its overall contribution to the image and synthesize it to form the final image. This method imitates the phenomenon of a diffuse trace left by a snowball thrown on a wall, so it is named "snowball throwing method". The key advantage of the Splatting algorithm is that it can effectively retain the excellent properties of the continuous volume radiation field while avoiding unnecessary calculations in blank space. By optimizing various properties of three-dimensional Gaussians (such as position, opacity, covariance, and color), the Splatting algorithm supports anisotropic snowball throwing, which not only speeds up the training process, but also maintains high quality for real-time rendering.

[0071] Specifically, the initial three-dimensional Gaussian model is projected onto a two-dimensional plane at a target angle to obtain a plurality of different projection ellipses.

[0072] In this embodiment, the angle of one of the cameras is used as the target angle, and multiple Gaussian ellipsoids in the initial three-dimensional Gaussian model are projected onto a two-dimensional plane in a perspective projection manner to obtain multiple different projected ellipses. The two-dimensional covariance matrix of the projected ellipse can be expressed by the following formula: ;

[0073] in, represents the two-dimensional covariance matrix, represents the Jacobian matrix, represents the weight, represents the transposed matrix, represents the rotation matrix represented by a quaternion, Represents the scaling matrix.

[0074] It should be noted that the relationship between the two-dimensional covariance matrix and the rotation matrix and the scaling matrix is: for the scaling matrix, the larger the variance value, the more dispersed the distribution on the axis, that is, the points are distributed more widely in this direction. For the rotation matrix, if the non-diagonal elements (covariance) of the two-dimensional covariance matrix are not zero, it means that there is a correlation between the two axes. This correlation makes the distribution show a certain tilt or rotation, and this tilt is caused by the linear relationship between the two axes.

[0075] Furthermore, the two-dimensional plane is divided into a plurality of projection blocks, and the projection blocks are sorted according to the external parameters of the camera and the distances between a plurality of Gaussian ellipsoids in the initial three-dimensional Gaussian model and the two-dimensional plane.

[0076] Specifically, the two-dimensional plane (i.e., the projection screen) is divided into multiple projection blocks of 16×16 pixels, and only the projection ellipses with 99% confidence that intersect the viewing cone are retained, and the projection ellipses close to the viewing plane or outside the viewing cone are removed. It should be noted that the viewing cone is a visible cone range defined by the camera's field of view and the distance between the near clipping plane and the far clipping plane. The viewing cone has six faces in three-dimensional space: top, bottom, left, right, near, and far. Only objects within the viewing cone will be rendered, thereby improving rendering efficiency.

[0077] The projection blocks intersecting with the projection ellipse are identified, wherein one projection ellipse may intersect with multiple projection blocks at the same time. The projection blocks intersecting with the projection ellipse are copied and assigned unique identifiers so that each projection block can be associated with multiple projection ellipses, thereby effectively reducing the complexity of the calculation while ensuring efficiency.

[0078] According to the external parameters of the camera, the distance between multiple Gaussian ellipsoids and the two-dimensional plane is calculated under the current camera viewing angle, and all Gaussian ellipsoids projected on the current projection block are sorted according to the distance from the two-dimensional plane to form a depth list.

[0079] Furthermore, the Alpha Blending algorithm is used to calculate the color of each pixel in the projection block to obtain the corresponding two-dimensional image.

[0080] Among them, the ‌Alpha Blending algorithm‌ is a technology that makes 3D objects transparent. It is achieved by calculating the actual visible pixel value of a semi-transparent pixel superimposed on another pixel. This algorithm involves calculating the red, green, and blue channels of the pixel while taking into account the transparency (i.e., ‌Alpha) value to determine the color and transparency of the final pixel.

[0081] In this embodiment, the color of each pixel in each projection block is calculated by the Alpha Blending algorithm, and the algorithm stops when the color of each pixel reaches saturation. The final color of each pixel can be expressed by the following formula:

[0082] in, represents the color obtained through training, represents opacity, Represents peace, Represents continuous multiplication.

[0083] After obtaining the color of each pixel in each projection block, the corresponding two-dimensional image can be obtained.

[0084] S33: Using the target loss function, the two-dimensional image is compared with its corresponding real image, and according to the comparison result, back propagation is performed to optimize the parameters of the initial three-dimensional Gaussian model until the value of the target loss function converges to obtain the three-dimensional Gaussian model.

[0085] Among them, the convergence of the value of the objective loss function means that the objective loss function tends to a certain value at every point in the domain of definition.

[0086] In this embodiment, the target loss function is expressed by the following formula:

[0087] in, represents the absolute error loss, represents the loss function based on structural similarity, represents the regularization parameter, Represents the target loss function.

[0088] Furthermore, the acquired two-dimensional image and its corresponding real image are brought into the target loss function for solution, and based on the obtained value, reverse propagation is performed to optimize the parameters of the initial three-dimensional Gaussian model (such as the position, opacity, color and covariance of the Gaussian ellipsoid) so that the generated two-dimensional image is closer to the real image, until the value of the target loss function converges to obtain the required three-dimensional Gaussian model.

[0089] In this embodiment, an adaptive density control method is also used to adaptively control the number and density of Gaussian ellipsoids per unit volume to train the initial three-dimensional Gaussian model.

[0090] Specifically, at the beginning of the optimization process, a densification process is performed every 100 iterations and Gaussian ellipsoids with transparency below a threshold and too large in the view space are deleted, where the densification process includes cloning and splitting. In the point densification stage, the density of the Gaussian ellipsoids is adaptively increased to better capture the details of the scene, including cloning small Gaussian ellipsoids in insufficiently reconstructed areas or splitting large Gaussian ellipsoids in over-reconstructed areas. For cloning, clones of the Gaussian are created and moved toward the position gradient. For splitting, a large Gaussian ellipsoid is replaced with two smaller Gaussian ellipsoids, and their scales are reduced by a specific factor. This step aims to seek the optimal distribution and representation of Gaussian ellipsoids in 3D space and enhance the overall quality of the reconstructed model.

[0091] In this embodiment, the final three-dimensional Gaussian model can be expressed by the following formula:

[0092] Among them, the covariance matrix , Represents the variance on the x, y, and z axes respectively, , represents the covariance on the x and y axes, , represents the covariance on the x and z axes, , represents the covariance on the y and z axes, Represents the transposed matrix.

[0093] S4: Real-time visualization rendering of the 3D Gaussian model to view rendered images of the turnout components from multiple different perspectives.

[0094] Specifically, the obtained three-dimensional Gaussian model is rendered in real time through the Gaussian point cloud viewer to obtain rendering images of the turnout components from multiple different perspectives. The surface details of the turnout components at different angles can be viewed based on the rendered images to determine whether there are defects, and to complete online inspection and measurement of key turnout components.

[0095] The images of turnout components rendered by the three-dimensional Gaussian model are sufficiently realistic, and the details of the rendered images can reflect the appearance of actual turnout components. The rendering frame rate is very high, and can be used by users to check whether there are defects or other problems in the turnout components.

[0096] like Figures 4 to 11 As shown, a schematic diagram of the rendered images of the turnout components at multiple different angles obtained by the method is shown (taking the left view, right view, front view, rear view, enlarged view, and views in three oblique directions as examples). The figure intuitively shows the specific condition of the turnout, and the surfaces of different components of the turnout and the same component at multiple different angles can be directly observed and analyzed to further determine whether there are defects and other problems, so as to facilitate timely discovery and maintenance of problems.

[0097] Based on the same inventive concept, Fig.12 As shown, the present application also provides a three-dimensional visualization device for a turnout, the device comprising: The data acquisition module 101 is used to collect and acquire video data of the turnout components from multiple viewing angles, and pre-process the video data to obtain image data from multiple different viewing angles.

[0098] The point cloud initialization module 102 is used to calculate the multiple image data to obtain initialized three-dimensional point cloud data.

[0099] The three-dimensional Gaussian model generation module 103 is used to train the three-dimensional point cloud data using the three-dimensional Gaussian algorithm to generate a corresponding three-dimensional Gaussian model.

[0100] The visualization module 104 is used to perform real-time visualization rendering of the three-dimensional Gaussian model and view the rendered images of the turnout components from multiple different perspectives.

[0101] Other technical features of the three-dimensional visualization device for turnouts are the same as those disclosed in the above-mentioned embodiment methods. Please refer to the description in the above-mentioned corresponding method embodiments and will not be repeated here.

[0102] Based on the same inventive concept, the present application also provides an electronic device, which includes a processor, a memory and a communication circuit, and the processor is connected to the memory and the communication circuit respectively; wherein the communication circuit is used for communication connection, the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-mentioned surface mesh generation method.

[0103] See also Fig.13 The electronic device described in the embodiment of the present application may specifically include a processor 210 and a memory 220. The memory 220 is coupled to the processor 210.

[0104] The processor 210 is used to control the operation of the electronic device. The processor 210 may also be referred to as a CPU (Central Processing Unit). The processor 210 may be an integrated circuit chip with signal processing capabilities. The processor 210 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor 210 may also be any conventional processor, etc.

[0105] The memory 220 is used to store computer programs, and may be a RAM, a ROM, or other types of storage terminals. Specifically, the memory 220 may include one or more computer-readable storage media, which may be non-transitory or transient. The memory 220 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage terminals, flash memory storage terminals. In some embodiments, the non-transitory computer-readable storage medium in the memory 220 is used to store at least one program code.

[0106] The processor 210 is used to execute the computer program stored in the memory 220 to implement the methods described in the various method embodiments of the present application.

[0107] In some embodiments, the electronic device may further include: a peripheral terminal interface 230 and at least one peripheral terminal. The processor 210, the memory 220 and the peripheral terminal interface 230 may be connected via a bus or a signal line. Each peripheral terminal may be connected to the peripheral terminal interface 230 via a bus, a signal line or a circuit board. Specifically, the peripheral terminal includes: at least one of a radio frequency circuit 240, a display screen 250, an audio circuit 260 and a power supply 270.

[0108] The peripheral terminal interface 230 may be used to connect at least one peripheral terminal related to I / O (Input / output) to the processor 210 and the memory 220. In some embodiments, the processor 210, the memory 220, and the peripheral terminal interface 230 are integrated on the same chip or circuit board; in some other implementations, any one or two of the processor 210, the memory 220, and the peripheral terminal interface 230 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0109] The radio frequency circuit 240 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 240 communicates with the communication network and other IoT devices through electromagnetic signals, and the radio frequency circuit 240 is the communication circuit of the electronic device. The radio frequency circuit 240 converts the electrical signal into an electromagnetic signal for transmission, or converts the received electromagnetic signal into an electrical signal. Optionally, the radio frequency circuit 240 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, an operator identity module card, and the like. The radio frequency circuit 240 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes, but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 240 may also include circuits related to NFC (Near Field Communication), which is not limited in this application.

[0110] The display screen 250 is used to display the UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 250 is a touch display screen, the display screen 250 also has the ability to collect touch signals on the surface or above the surface of the display screen 250. The touch signal can be input to the processor 210 as a control signal for processing. At this time, the display screen 250 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 250 can be one, set on the front panel of the electronic device; in other embodiments, the display screen 250 can be at least two, respectively set on different surfaces of the electronic device or in a folding design; in other embodiments, the display screen 250 can be a flexible display screen, set on a curved surface or a folding surface of the electronic device. Even, the display screen 250 can also be set as a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 250 can be made of materials such as LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode, organic light-emitting diode).

[0111] The audio circuit 260 may include a microphone and a speaker. The microphone is used to collect sound waves from the operator and the environment, and convert the sound waves into electrical signals and input them into the processor 210 for processing, or input them into the radio frequency circuit 240 to achieve voice communication. For the purpose of stereo acquisition or noise reduction, there may be multiple microphones, which are respectively arranged in different parts of the electronic device. The microphone may also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signal from the processor 210 or the radio frequency circuit 240 into sound waves. The speaker may be a traditional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for purposes such as ranging. In some embodiments, the audio circuit 260 may also include a headphone jack.

[0112] The power supply 270 is used to power various components in the electronic device. The power supply 270 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 270 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery that is charged through a wired line, and a wireless rechargeable battery is a battery that is charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0113] For a detailed description of the functions and execution processes of each functional module or component in the electronic device embodiment of the present application, reference can be made to the description in the above-mentioned method embodiments of the present application, which will not be repeated here.

[0114] In the several embodiments provided in the present application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the various embodiments of the electronic devices described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0115] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0116] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0117] Based on the same inventive concept, the present application also provides a computer-readable storage medium storing a computer program, which can be executed by a processor to implement the above-mentioned three-dimensional visualization method of turnout.

[0118] See also Fig.14 , if the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium 300. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions / computer programs to enable an IoT device (which can be a personal computer, server, or network terminal, etc.) or a processor (processor) to execute all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks or optical disks, and electronic terminals such as computers, mobile phones, laptops, tablet computers, cameras, etc. having the above-mentioned storage media.

[0119] The description of the execution process of the program data in the computer-readable storage medium can refer to the description in the above-mentioned method embodiments of the present application, and will not be repeated here.

[0120] It can be seen that the present invention discloses a three-dimensional visualization method, device, equipment and medium for turnouts, which includes the steps of: collecting and acquiring video data of multiple perspectives of turnout components, performing preprocessing, and obtaining multiple image data; calculating multiple image data to obtain initialized three-dimensional point cloud data; using a three-dimensional Gaussian algorithm to train the three-dimensional point cloud data and generate a three-dimensional Gaussian model; performing real-time visualization rendering of the three-dimensional Gaussian model to view the rendered images of the turnout components from multiple different perspectives. The present invention uses a three-dimensional Gaussian model to represent the photographed turnout components and renders images of multiple angles in real time, so that users can detect turnout components without field exploration; it not only improves the detection efficiency, saves human and material resources, and reduces the cost required for detection; it also further ensures the detection accuracy and timeliness of turnout components, and is more adaptable to application requirements in dynamic scenes.

[0121] The above are only embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in other related technical fields, are included in the patent protection scope of the present invention.

Claims

1. A three-dimensional visualization method for turnouts, It is characterized in that Includes steps: Collecting and acquiring video data of multiple viewing angles of turnout components, and preprocessing the video data to obtain image data of multiple different viewing angles; Calculating the plurality of image data to obtain initialized three-dimensional point cloud data; Using a three-dimensional Gaussian algorithm to train the three-dimensional point cloud data to generate a corresponding three-dimensional Gaussian model; The three-dimensional Gaussian model is rendered in real time for visual viewing, and the rendered images of the turnout components from multiple different perspectives are viewed.

2. The three-dimensional visualization method of turnout according to claim 1, It is characterized in that The step of calculating the plurality of image data to obtain the initialized three-dimensional point cloud data specifically includes: Extracting a plurality of feature points of the image data respectively; Performing feature matching on the feature points of the image data to obtain a feature matching result; Performing motion structure recovery on the feature matching results to generate a motion trajectory of the camera; According to the feature matching result, the feature point is calculated to obtain the corresponding three-dimensional point coordinates; The three-dimensional point cloud data is reconstructed and generated using the three-dimensional point coordinates and the motion trajectory of the camera.

3. The three-dimensional visualization method of turnout according to claim 2, It is characterized in that The step of performing motion structure recovery on the feature matching result to generate the motion trajectory of the camera specifically includes: According to the feature matching result, a basic matrix and an essential matrix of the camera are calculated; Performing singular value decomposition on the essential matrix to obtain a rotation matrix and a translation vector of the camera; The motion trajectory of the camera is generated using the rotation matrix and the translation vector.

4. The three-dimensional visualization method of turnout according to claim 2, It is characterized in that The step of reconstructing and generating the three-dimensional point cloud data by using the three-dimensional point coordinates and the motion trajectory of the camera specifically includes: Define a reprojection error function, and adjust the three-dimensional point coordinates and the external parameters of the camera until the value of the reprojection error function converges to a first threshold; The three-dimensional point cloud data is reconstructed and generated according to the adjusted three-dimensional point coordinates and the motion trajectory of the camera.

5. The three-dimensional visualization method of turnout according to claim 1, It is characterized in that The step of using a three-dimensional Gaussian algorithm to train the three-dimensional point cloud data to generate a corresponding three-dimensional Gaussian model specifically includes: Initializing the three-dimensional point cloud data, generating a set of corresponding Gaussian ellipsoid sets according to each point in the three-dimensional point cloud data, and obtaining an initial three-dimensional Gaussian model; Projecting the initial three-dimensional Gaussian model onto a two-dimensional plane at a target angle, and performing rasterization rendering to obtain a corresponding two-dimensional image; The two-dimensional image is compared with its corresponding real image by using the target loss function, and according to the comparison result, back propagation is performed to optimize the parameters of the initial three-dimensional Gaussian model until the value of the target loss function converges to obtain the three-dimensional Gaussian model.

6. The three-dimensional visualization method of turnout according to claim 5, It is characterized in that The step of projecting the initial three-dimensional Gaussian model onto a two-dimensional plane at a target angle and performing rasterization rendering to obtain a corresponding two-dimensional image specifically includes: Projecting the initial three-dimensional Gaussian model onto a two-dimensional plane at a target angle to obtain a plurality of different projection ellipses; Divide the two-dimensional plane into a plurality of projection blocks, and sort all the Gaussian ellipsoids according to the external parameters of the camera and the distances between the plurality of Gaussian ellipsoids in the initial three-dimensional Gaussian model and the two-dimensional plane; The color of each pixel in the projection block is calculated to obtain the corresponding two-dimensional image.

7. The three-dimensional visualization method of turnout according to claim 1, It is characterized in that The preprocessing of the video data includes at least one of the following operations: De-noising, lens distortion correction and enhancement.

8. A three-dimensional visualization device for turnouts, It is characterized in that include: A data acquisition module is used to collect and acquire video data of multiple viewing angles of turnout components, and pre-process the video data to obtain image data of multiple different viewing angles; A point cloud initialization module, used to calculate the plurality of image data to obtain initialized three-dimensional point cloud data; A three-dimensional Gaussian model generation module is used to train the three-dimensional point cloud data using a three-dimensional Gaussian algorithm to generate a corresponding three-dimensional Gaussian model; The visualization module is used to perform real-time visualization rendering of the three-dimensional Gaussian model and view the rendering images of the turnout components from multiple different perspectives.

9. An electronic device, It is characterized in that The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program. It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images