A tunnel point cloud unfolding and imaging method, system, terminal device and computer readable storage medium of multi-channel feature fusion

Abstract

The application discloses a kind of multi-channel feature fusion's tunnel point cloud unfolding and imaging method.The method includes the following steps: S1, the original tunnel point cloud data obtained is processed with space alignment and polar coordinate transformation, and projection coordinate data and radial feature data are obtained;S2, the projection coordinate data is used to construct multi-channel physical canvas, and the pixel filling is carried out through mean projection and feature extraction logic, and the feature matrix of I, L, H, Z four dimensions is obtained;S3, the feature matrix is optimized and image fusion is executed, and three-channel feature fusion image is output.The application belongs to the field of three-dimensional point cloud processing and computer vision technology.The application solves the problems of significant diagonal stripe artifact, loss of geometric features and single optimization target in traditional projection method, realizes the deep coupling of material texture and geometric relief features, and improves the accuracy and reliability of tunnel component identification and safety evaluation.

Description

Technical Field

[0001] This invention relates to the fields of 3D point cloud processing and computer vision technology, and in particular to a method, system, terminal device, and computer-readable storage medium for multi-channel feature fusion tunnel point cloud unfolding and imaging. Background Technology

[0002] Tunnel point cloud data, as core data for infrastructure monitoring, directly impacts safety assessments through its processing efficiency and accuracy. Current preprocessing techniques for projecting 3D point clouds into 2D images face two major bottlenecks: First, traditional projection methods are affected by fluctuations in laser scanning linear velocity, often resulting in images with noticeable diagonal artifacts that interfere with feature extraction. Second, the optimization objective is singular; existing technologies primarily focus on reflection intensity information, neglecting the coupling characteristics of tunnel components in radial residuals and spatial elevation, leading to insufficient accuracy in identifying minute geometric defects. Therefore, a technical solution that balances multi-dimensional feature extraction and imaging optimization is urgently needed. Summary of the Invention

[0003] This application provides a method and system for tunnel point cloud unfolding and imaging using multi-channel feature fusion, in order to solve the problems of significant projection imaging artifacts and loss of geometric features in the prior art.

[0004] On the one hand, this application provides a method for tunnel point cloud unfolding and imaging based on multi-channel feature fusion, the method comprising: S1. Perform spatial alignment and polar coordinate transformation on the acquired raw tunnel point cloud data to obtain projected coordinate data and polar diameter data.

[0005] S2. Construct a multi-channel physical canvas based on the projected coordinate data, and fill pixels through mean projection and feature extraction logic to obtain feature matrices in three dimensions: reflection intensity channel I, elevation channel Z, and polar radius distance channel H.

[0006] S3. Perform channel optimization processing and image fusion on the feature matrix to output a visualization image of the three-channel feature fusion of the reflection intensity channel I, elevation channel Z, and polar radius distance channel H, as well as an independent channel image.

[0007] Step S1 includes: S11, extracting the three-dimensional spatial coordinates of the original tunnel point cloud data, performing spatial pose alignment using principal component analysis algorithm, and obtaining rotational feature values ​​of the point cloud in the tunnel extension direction, the horizontal direction of the cross section, and the vertical direction of the cross section.

[0008] S12. Calculate longitudinal mileage coordinates and azimuth angle based on the rotated eigenvalues. and the extreme radius distance from the point to the center axis of the tunnel .

[0009] Step S2 includes: S21, determining the length and width dimensions of the physical canvas based on the physical length, cross-sectional perimeter, and preset pixel size.

[0010] S22. Execute the mean projection fill logic. For multiple points falling into the same pixel unit, take the mean of their reflection intensity and fill it into the I channel canvas to suppress diagonal pattern artifacts.

[0011] S23. Extract the point with the maximum polar radius distance within the pixel unit, and fill its height feature and original elevation value into the H channel and Z channel canvas respectively.

[0012] Preferably, in step S11, the original tunnel point cloud data includes spatial coordinates, reflection intensity, and classification labels.

[0013] Preferably, step S2 further includes generating an index mapping matrix to record the correspondence between each pixel unit in the physical canvas and the original three-dimensional point cloud index, so as to realize pixel-level bidirectional tracing of two-dimensional image features and three-dimensional spatial coordinates.

[0014] Preferably, in step S3, the optimization processing of the I channel includes: performing linear normalization, Gaussian smoothing filtering, and contrast-limited adaptive histogram equalization on the intensity matrix obtained by mean projection to enhance the material texture details; Optimization of the Z channel: The elevation matrix obtained from depth projection is linearly normalized, mapping it to grayscale values ​​in the range of 0–255, preserving the original spatial distribution characteristics of the depth dimension, and ensuring spatial alignment consistency between depth information and polar radius and intensity information; Optimization of the H channel includes: Calculating the residual value of the polar radius distance of each pixel unit, using the formula: ,in The maximum polar radius of the current pixel. The global average radius is used; the residual value is linearly normalized and mapped to a gray value in the range of 0-255.

[0015] Preferably, in step S3, the image fusion logic is as follows: the optimized I-channel, Z-channel, and H-channel matrices are used as blue, green, and red components, respectively, to synthesize a three-channel IZH visual feature image.

[0016] Preferably, step S3 further includes a rotation processing step: determining the aspect ratio of the image; if the height is greater than the width, rotating all output channel images 90 degrees clockwise to unify the visual direction.

[0017] On the other hand, this application also provides a multi-channel feature fusion tunnel point cloud unfolding and imaging system, which includes: The spatial alignment module is used to read LAS format point cloud data and perform PCA spatial alignment and polar coordinate mapping. The feature projection module, with built-in mean projection unit and maximum polar radius extraction unit, is used to generate a standardized canvas in four dimensions: I, L, H, and Z. The channel optimization module includes a CLAHE enhancement unit, a height residual calculation unit, and a normalization unit, which realizes the mapping of physical attributes of each channel to image brightness; The image fusion and output module enables multi-channel matrix synthesis, automatic image rotation, and structured storage of multi-dimensional output files.

[0018] Preferably, the feature projection module supports dynamic adjustment of pixel size, automatically calculates the target image resolution based on the physical space span, and supports performing original value preservation mapping on classification labels.

[0019] On the other hand, this application also provides a terminal device, including: a memory and a processor, wherein the memory is used to store a multi-channel feature fusion tunnel point cloud unfolding and imaging program; when the imaging program is executed by the processor, it implements the steps of the multi-channel feature fusion tunnel point cloud unfolding and imaging method as described above.

[0020] On the other hand, this application also provides a computer-readable storage medium, wherein the storage medium stores a computer program, which, when executed, implements the tunnel point cloud unfolding and imaging method for multi-channel feature fusion as described above.

[0021] The multi-channel feature fusion tunnel point cloud unfolding and imaging method and system in this application have the following advantages: The solution provided in this application achieves automatic spatial pose alignment of tunnel point clouds through principal component analysis algorithm, and effectively eliminates the diagonal pattern artifacts caused by scanner linear velocity fluctuations by combining mean projection filling logic, significantly improving the visual clarity and feature continuity of the 2D unfolded image. Secondly, this application constructs a multi-channel feature system including strength, radial residual, and elevation. The radial residual is used to peel off the tunnel background curvature, making minor geometric defects such as segment misalignment and bolt loosening brightly visible, compensating for the shortcomings of single strength information in geometric feature expression. In addition, the introduced elevation information provides the model with the absolute prior position of components in vertical space. Combined with CLAHE texture enhancement technology, it greatly improves the recognition accuracy of deep learning models for key components in shadow areas or low-reflection areas, reducing the false negative rate. Finally, while significantly compressing the amount of 3D point cloud data, this method retains the maximum sub-millimeter level geometric fidelity, providing reliable data support for real-time, high-precision tunnel life cycle monitoring, and has broad engineering application prospects in fields such as subway inspection and railway maintenance. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A schematic diagram of the tunnel point cloud unfolding and imaging method for multi-channel feature fusion provided in an embodiment of this application; Figure 2 This is a flowchart illustrating the process of spatial alignment and polar coordinate transformation of the acquired raw tunnel point cloud data to obtain projected coordinate data and polar radius data, provided in an embodiment of this application. Figure 3 This is a flowchart illustrating the process of constructing a multi-channel physical canvas based on the coordinate data, filling pixels through mean projection and feature extraction logic, and obtaining feature matrices in four dimensions: I, L, H, and Z, as provided in the embodiments of this application. Figure 4 This is a schematic diagram illustrating the PCA spatial pose alignment principle provided in an embodiment of this application. Figure 5 This is a schematic diagram of multi-channel feature matrix synthesis and IZH image fusion provided in the embodiments of this application; Figure 6 This is a schematic diagram illustrating the pixel-level index mapping and bidirectional tracing principle provided in an embodiment of this application. Figure 7 This is a schematic diagram of the point cloud unfolding imaging result provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0025] Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0026] Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] refer to Figure 1 The diagram shown is a flowchart of a multi-channel feature fusion tunnel point cloud unfolding and imaging method provided in this application embodiment.

[0028] The method provided in this application includes: S1. Spatial alignment and polar coordinate transformation are performed on the acquired raw tunnel point cloud data to obtain projected coordinate data and radial feature data.

[0029] S2. Construct a multi-channel physical canvas based on the projected coordinate data, and fill pixels through mean projection and feature extraction logic, with feature matrices in three dimensions: reflection intensity channel I, elevation channel Z, and polar radius distance channel H. S3. Optimize the feature matrix and perform image fusion to output a three-channel feature fusion image of the reflection intensity channel I, elevation channel Z, and polar radius distance channel H.

[0030] In the embodiments of this application, after acquiring tunnel point cloud data using three-dimensional laser scanning technology (such as lidar scanning), the pose deviation of the scanner is corrected by spatial alignment, and a multi-channel feature map containing material, geometry and spatial position is constructed.

[0031] Furthermore, this fused image (i.e., multi-channel feature map) guides subsequent deep learning models to accurately identify tunnel components, achieving a balance between tunnel dimensionality reduction representation and feature fidelity.

[0032] In one embodiment, during the point cloud unfolding process, the tunnel principal axis direction is first determined by principal component analysis algorithm; then, the three-dimensional coordinates are mapped to a two-dimensional physical canvas by polar coordinate transformation; mean projection algorithm (weight ratio 0.7) is used to suppress diagonal artifacts, and maximum radial extraction logic (weight ratio 0.3) is used to capture geometric protrusions, finally outputting a high-fidelity IZH fused image.

[0033] refer to Figure 2 The diagram shows the process of spatial alignment and polar coordinate transformation of the acquired raw tunnel point cloud data in step S1 to obtain projected coordinate data and polar diameter data.

[0034] This step specifically includes: S11. Extract the three-dimensional spatial coordinates of the point cloud, perform spatial pose alignment using principal component analysis algorithm, and obtain the rotational feature values ​​of the point cloud in the tunnel extension direction, the horizontal direction of the cross section, and the vertical direction of the cross section.

[0035] In the embodiments of this application, the raw LAS format tunnel point cloud data is first read using a point cloud processing library, and spatial three-dimensional coordinates are extracted from it. point set matrix .

[0036] Since the coordinate system of the original point cloud is usually based on the scanner or geographic coordinate system, the main axis direction of the tunnel is often inconsistent with the coordinate axis.

[0037] This embodiment uses the principal component analysis algorithm to analyze the point set matrix. Processing: refer to Figure 4 As shown, calculate the point set. The centroid is determined, and the coordinates of the centroid are subtracted from all points to eliminate the influence of spatial translation on attitude calculation.

[0038] Construct a decentralized point cloud covariance matrix and obtain three mutually orthogonal eigenvectors through eigenvalue decomposition.

[0039] Main direction determination and projection: First principal component: The eigenvector with the largest corresponding eigenvalue, which is defined in this scenario as the extension direction of the tunnel (longitudinal principal axis).

[0040] The second and third principal components are the two eigenvectors with the second largest eigenvalues, which are defined in this scenario as the horizontal basis vector and the vertical basis vector within the tunnel cross-section.

[0041] Using the aforementioned eigenvectors, a rotation matrix is ​​constructed to rotate and map the original point cloud to a new local coordinate system based on the principal component eigenvectors, resulting in the transformed coordinate points. .

[0042] in, Let be the longitudinal mileage component of the point in the direction of tunnel extension. and This represents the local orthogonal component of the point within the tunnel cross-section.

[0043] S12. Calculate longitudinal mileage coordinates and azimuth angle based on the rotated eigenvalues. and the extreme radius distance from the point to the center axis of the tunnel .

[0044] In the embodiments of this application, after principal component analysis and spatial pose alignment in step S11, the point cloud data is converted into a new local three-dimensional coordinate system based on the tunnel center axis.

[0045] Let the coordinates of the rotated point cloud be... ,in Corresponding to the tunnel extension direction (i.e., the longitudinal main axis). and These correspond to the horizontal and vertical orthogonal directions within the tunnel cross section, respectively.

[0046] This step aims to convert the Cartesian coordinates to cylindrical coordinates to meet the requirements of two-dimensional unfolding.

[0047] Longitudinal mileage coordinate normalization calculation: Extract the coordinate values ​​of all point cloud points in the tunnel extension direction. .

[0048] To standardize the starting position of different scan segments, a minimum extreme value translation operation is performed on the coordinates in that direction.

[0049] Find the minimum vertical coordinate value of the current point cloud segment. Through formula The tunnel starting point is shifted to the zero-mileage position to obtain the normalized longitudinal mileage coordinates. .

[0050] This coordinate will determine the height mapping reference for the subsequent two-dimensional physical canvas.

[0051] Using the bivariate arctangent function within the tunnel cross-section, orthogonal coordinates Convert to circumferential azimuth.

[0052] The calculation formula is: in, The function can be based on and The symbol returns the point at that location. Precise rotation angle within the range.

[0053] This azimuth angle reflects the absolute position of the point cloud in the tunnel segment ring direction and is the benchmark for determining the width mapping of the subsequent two-dimensional physical canvas.

[0054] The polar radius distance characterizes the physical distance from the point cloud to the virtual central axis of the tunnel.

[0055] Based on orthogonal coordinates within the cross section The solution is obtained using the Euclidean distance formula: Calculated polar diameter It is not only used to help determine the physical perimeter of the unfolded plane, but also a core geometric feature value that characterizes the radial deformation, misalignment, and protrusions (such as bolts and pipes) of tunnel segments.

[0056] This value will serve as the basic input data for generating the H channel.

[0057] refer to Figure 3 The diagram shows the process in step S2 of constructing a multi-channel physical canvas based on the projected coordinate data, filling pixels with mean projection and feature extraction logic, and obtaining feature matrices in four dimensions: I, L, H, and Z.

[0058] This step specifically includes: S21. Determine the length and width dimensions of the physical canvas based on the physical length, cross-sectional perimeter, and preset pixel size.

[0059] refer to Figure 5As shown, this step aims to establish a precise mapping ratio between the three-dimensional physical space and the two-dimensional digital image pixel space, ensuring that the unfolded image does not suffer from proportional distortion in both the vertical (mileage) and horizontal (circumferential) directions.

[0060] Automated acquisition of physical range: The system first processes the normalized longitudinal mileage coordinates generated in step S12. and polar distance Perform a global scan.

[0061] Obtain the maximum vertical mileage, defined as the effective physical length of the current point cloud segment: Considering that the tunnel is not a perfect cylinder, in order for the pixel canvas to cover the entire toroidal surface without overlap, the system calculates the arithmetic mean of the epipolar diameters of all point clouds. The average physical circumference of the unfolded object is calculated using the circumference formula. .

[0062] Pixel aspect ratio and resolution adaptation: Introducing preset pixel size parameters (In this embodiment, it is set to 0.008 meters, that is, 1 pixel represents an 8mm * 8mm area in physical space).

[0063] This parameter can be dynamically adjusted according to the accuracy requirements for identifying tunnel defects.

[0064] Quantization and determination of canvas row and column dimensions: The physical canvas matrix is ​​determined based on the ratio of physical span to pixel size. Number of rows and columns: Image height (number of rows) )calculate: This dimension corresponds to the direction of tunnel mileage extension.

[0065] Image width (number of columns) )calculate: This dimension corresponds to the circumferential unfolding direction of the tunnel.

[0066] In this embodiment, the same pixel ratio is used in both the vertical and horizontal directions. .

[0067] This proportional sampling strategy ensures that circular components remain circular in the subsequently generated images, rather than being stretched into ellipses.

[0068] This provides a base map with metric attributes for subsequent shape recognition and geometric dimension measurement based on computer vision.

[0069] The system is based on the calculations obtained and Pre-initialize all-zero multichannel tensors in memory (e.g.) ), which serves as the physical carrier for feature filling in subsequent steps S22 and S23.

[0070] S22. Execute the mean projection fill logic. For multiple points falling into the same pixel unit, take the mean of their reflection intensity and fill it into the I channel canvas to suppress diagonal pattern artifacts.

[0071] During laser scanning, the original point cloud often exhibits a non-uniform striped spatial distribution due to the influence of the scanner's rotational linear speed, the shaking of the moving vehicle, or repeated scanning.

[0072] Using simple single-point projection will result in severe periodic diagonal artifacts in the generated two-dimensional image.

[0073] To address this issue, this embodiment employs a mean projection algorithm based on pixel unit statistics: Based on the canvas dimensions calculated in step S21, the spatial coordinates of each point cloud are... Discretize into corresponding pixel row and column indices .

[0074] The mapping formula is as follows: Using the formula above, all three-dimensional points are assigned to specific pixel units in the two-dimensional physical canvas.

[0075] This mean projection logic is physically equivalent to an adaptive spatial filter.

[0076] Multiple points falling into the same pixel unit represent multiple rounds of sampling or oversampling of the same physical location by the scanner. By taking the average value, the artificial brightness caused by the overlap of scan lines can be neutralized, and the random noise caused by the change of laser beam angle can be smoothed.

[0077] generated The channel matrix is ​​then fed into step S3 for contrast-limited adaptive histogram equalization.

[0078] Since mean projection has already eliminated large-amplitude impulse noise and stripe artifacts, adaptive histogram equalization can more smoothly stretch texture details, thereby outputting a tunnel lining image with coherent texture and clear contrast, providing a high-quality data foundation for subsequent crack detection.

[0079] S23. Extract the point with the maximum polar radius distance within the pixel unit, and fill its height feature and original elevation value into the H channel and Z channel canvas respectively.

[0080] This step aims to select the most representative "outer surface feature points" from the multiple redundant point clouds contained in each pixel unit in order to preserve the subtle geometric undulations of the tunnel wall.

[0081] To efficiently retrieve each pixel unit in massive point clouds To find the optimal feature points within the range, this embodiment employs a filtering algorithm based on multi-level index sorting: Indexed by pixel row Pixel column index and polar radius distance As the joint sorting key, all point clouds are sorted such that points with the same pixel index are grouped together, and within the same pixel group, points are sorted by their polar radius distance. Arranged from smallest to largest.

[0082] The index of the last sorted point in each pixel group is the point with the largest polar radius distance within that pixel unit.

[0083] The maximum polar distance selected Fill the H channel canvas matrix.

[0084] The H channel records the radial depth of the tunnel interior wall relative to the central axis.

[0085] In the subsequent S3 step, by calculating the local residual of this value, the circular outline of the tunnel can be stripped, so that the spalling defects of the steps or lining surfaces between the segments can show obvious grayscale differences in the image.

[0086] Simultaneously extract the maximum polar radius point in the original geographic coordinate system. The axis coordinate values ​​are filled into the Z channel canvas matrix.

[0087] The Z-channel reflects the absolute height variation of the tunnel in vertical space.

[0088] This passage provides spatial prior information for deep learning models, helping them distinguish between different components located at the top and bottom of the tunnel.

[0089] refer to Figure 6 As shown, in determining each pixel unit When the corresponding maximum polar radius point is reached, the system simultaneously extracts the global index number of that point in the original 3D point cloud array. And store it in a pre-built index mapping matrix. The corresponding position, that is .

[0090] This step establishes a "pixel-point cloud" index during feature projection, enabling subsequent identification of defects in the two-dimensional image to be queried. It instantly locates the precise coordinates in three-dimensional space.

[0091] While extracting the maximum extreme point, its original classification attributes are filled into the L channel canvas.

[0092] This ensures that the generated 2D image labels are strictly consistent with the "outermost surface" category in 3D space, eliminating label contamination caused by point cloud overlap.

[0093] S3. Optimize the feature matrix and perform image fusion to output a three-channel feature fusion image.

[0094] This step aims to convert the original feature matrix generated in step S2 into a standardized image that meets the input requirements of the computer vision model, and to highlight the details of the disease through feature enhancement techniques.

[0095] The original intensity matrix obtained by mean projection To eliminate noise and enhance the detail of the lining surface, the following treatments are performed: use or Gaussian kernel Convolution processing is performed to smooth out random noise between adjacent pixels.

[0096] Map the filtered floating-point intensity data to The basic grayscale image is obtained by using an 8-bit unsigned integer range.

[0097] A contrast-limited adaptive histogram equalization algorithm is introduced, and a contrast limit threshold is set.

[0098] This algorithm effectively solves the problem of local over-darkness caused by uneven tunnel illumination or differences in laser reflectivity by reconstructing histograms of local areas of the image, thus enabling the texture details of the lining material to be clearly presented.

[0099] In order to eliminate the overall cylindrical curvature of the tunnel and make minor defects such as segment misalignment geometrically visible, Channel residual processing is performed: Calculate the residual value of the epipolar radius distance of each pixel unit. The formula is: ,in The maximum polar radius of the current pixel extracted in step S23. It is the global average polar radius of all valid pixels within the currently processed point cloud segment.

[0100] The obtained residual values Perform a linear normalization mapping to transform its physical depth differences into... The grayscale value of the interval.

[0101] In this grayscale image, geometric anomalies that deviate from the standard circular plane will be manifested as significant high or low brightness.

[0102] To achieve deep coupling between material features and geometric features, this embodiment employs color fusion logic: the I channel matrix optimized by S3 is used as the blue component of the image; the Z channel matrix after normalization is used as the green component of the image; and the H channel matrix optimized by S32 is used as the red component of the image.

[0103] A three-channel IZH visualization feature image is generated using multi-channel synthesis technology.

[0104] This image can simultaneously represent the material texture, spatial elevation undulations, and surface geometric defects of a tunnel in a single image, greatly improving the information density for disease identification.

[0105] To ensure that the batch output images conform to engineering inspection habits, the system performs automatic pose determination: calculating the width of the generated images. With height .

[0106] If height is detected Greater than width If this occurs, the rotation logic will be automatically triggered, and all output results, including the I, L, H, Z, IZH fusion diagram, and index mapping matrix, will be rotated 90 degrees clockwise.

[0107] This processing step ensures that the long-distance tunnel unfolding diagram can be extended along the horizontal axis on the display terminal, facilitating manual continuous crack inspection.

[0108] refer to Figure 7 As shown, the system ultimately stores the optimized single-channel map, fused map, and index mapping matrix in the corresponding structured subdirectories according to the file base name, completing the complete unfolding process from 3D point cloud to multi-channel feature map.

[0109] Although preferred embodiments of this application have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0110] Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0111] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application.

[0112] Therefore, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A method for unfolding and imaging tunnel point clouds through multi-channel feature fusion, characterized in that, Includes the following steps: S1. Perform spatial alignment and polar coordinate transformation on the original tunnel point cloud data to obtain projected coordinate data and polar diameter data. S2. Construct a multi-channel physical canvas based on the projected coordinate data, and fill pixels through mean projection and feature extraction logic to obtain feature matrices in three dimensions: reflection intensity channel I, elevation channel Z, and polar radius distance channel H. S3. Perform channel optimization processing and image fusion on the feature matrix to output a visualization image of the three-channel feature fusion of the reflection intensity channel I, elevation channel Z, and polar radius distance channel H, as well as an independent channel image.

2. The tunnel point cloud unfolding and imaging method based on multi-channel feature fusion according to claim 1, characterized in that, Step S1 includes: S11. Extract the three-dimensional spatial coordinates of the original tunnel point cloud data, perform spatial pose alignment using principal component analysis algorithm, and obtain the rotational feature values ​​of the point cloud in the tunnel extension direction, the horizontal direction of the cross section, and the vertical direction of the cross section. S12. Calculate longitudinal mileage coordinates and azimuth angle based on the rotated eigenvalues. and the extreme radius distance from the point to the center axis of the tunnel ; Step S2 includes: S21. Determine the length and width dimensions of the physical canvas based on the physical length, cross-sectional perimeter, and preset pixel size; S22. Execute the mean projection fill logic. For multiple points falling into the same pixel unit, take the mean of their reflection intensity and fill it into the I channel canvas to suppress diagonal pattern artifacts. S23. Extract the point with the maximum polar radius distance within the pixel unit, and fill its height feature and original elevation value into the H channel and Z channel canvas respectively.

3. The tunnel point cloud unfolding and imaging method based on multi-channel feature fusion according to claim 2, characterized in that, In step S11, the original tunnel point cloud data includes spatial coordinates, reflection intensity, and classification labels.

4. The method for tunnel point cloud unfolding and imaging based on multi-channel feature fusion according to claim 2 or 3, characterized in that, Step S2 further includes generating an index mapping matrix to record the correspondence between each pixel unit in the physical canvas and the original three-dimensional point cloud index, so as to realize pixel-level bidirectional tracing of two-dimensional image features and three-dimensional spatial coordinates.

5. The method for tunnel point cloud unfolding and imaging based on multi-channel feature fusion according to claim 1 or 2, characterized in that, In step S3, the channel optimization process for the feature matrix includes: Optimization of the I channel: Linear normalization, Gaussian smoothing filtering, and contrast-limited adaptive histogram equalization are performed on the intensity matrix obtained by mean projection to enhance material texture details; Optimization of the Z channel: The elevation matrix obtained by depth projection is linearly normalized and mapped to grayscale values ​​in the range of 0 to 255, preserving the original spatial distribution characteristics of the depth dimension and ensuring the spatial alignment consistency between depth information and polar radius and intensity information; (It needs to be confirmed that this part is reflected in the specific implementation.) Optimization of the H channel: Calculate the residual value of the epipolar radius distance of each pixel unit, using the following formula: ,in The maximum polar radius of the current pixel. The global average radius is calculated, and the residual value is linearly normalized to convert it into a gray value in the range of 0 to 255.

6. The tunnel point cloud unfolding and imaging method based on multi-channel feature fusion according to claim 5, characterized in that, In step S3, the image fusion process of the feature matrix is ​​as follows: the optimized I-channel, Z-channel, and H-channel matrices are used as blue, green, and red components, respectively, and synthesized into a three-channel IZH visualization feature image.

7. The tunnel point cloud unfolding and imaging method based on multi-channel feature fusion according to claim 6, characterized in that, Step S3 further includes a rotation processing step: determining the aspect ratio of the image; if the height is greater than the width, rotating the visualization image fused with all three-channel features and the independent channel images clockwise by 90 degrees to unify the visual direction.

8. A tunnel point cloud unfolding and imaging system with multi-channel feature fusion, characterized in that, The system includes: The spatial alignment module is used to read LAS format point cloud data and perform PCA spatial alignment and polar coordinate mapping. The feature projection module, with built-in mean projection unit and maximum polar radius extraction unit, is used to generate a standardized canvas in four dimensions: I, L, H, and Z; (claims 1 to 7 only include I, H, and Z, and it is sufficient to confirm that there is no contradiction). The channel optimization module includes a CLAHE enhancement unit, a height residual calculation unit, and a normalization unit, which realizes the mapping of physical attributes of each channel to image brightness; The image fusion and output module enables multi-channel matrix synthesis, automatic image rotation, and structured storage of multi-dimensional output files.

9. The tunnel point cloud unfolding and imaging system with multi-channel feature fusion according to claim 8, characterized in that, The feature projection module supports dynamic adjustment of pixel size, automatically calculates the target image resolution based on the physical space span, and supports performing original value preservation mapping on classification labels.

10. A terminal device, characterized in that, It includes a memory and a processor; the memory is used to store a tunnel point cloud unfolding and imaging program for multi-channel feature fusion; when the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 7.

11. A computer-readable storage medium, characterized in that, The storage medium stores a computer program; when the computer program is executed, it implements the steps of the method as described in any one of claims 1 to 7.

Images