A tunnel point cloud registration processing method, system and platform fusing static noise features

Abstract

The application discloses a tunnel point cloud registration processing method, system and platform fusing static noise features, generates and acquires first data corresponding to a tunnel, processes the first data by combining acceleration and curvature filtering, generates second data corresponding to the first data, adjusts the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling, extracts and processes feature data corresponding to the second data, constructs a corresponding auxiliary point pair set based on noise auxiliary points, establishes a corresponding initial matching relationship by combining global depth consistency, iteratively optimizes and processes the initial matching by using a weighted geometric optimization algorithm, simultaneously searches for global parameters by using a particle swarm optimization algorithm, implements residual static noise filtering processing, generates corresponding third data, and generates a corresponding system and platform, thereby providing an efficient and robust solution for tunnel point cloud registration, and meeting high-precision construction quality detection and long-term maintenance requirements.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and 3D point cloud processing technology, specifically relating to a tunnel point cloud registration processing method, system and platform that integrates static noise features. Background Technology

[0002] Currently, construction quality inspection is crucial during the final acceptance phase of highway tunnels. SLAM technology plays a key role in this process, constructing a high-precision tunnel model by aligning multi-view 3D point cloud data. This model is then compared with the design model to verify whether the geometric dimensions of the tunnel walls, arches, and other main structures, such as length and curvature, meet design requirements, while also enabling effective detection of tunnel deformation. However, using handheld mobile LiDAR scanners to collect point clouds of highway tunnels presents several significant challenges:

[0003] Equipment sampling error issues: Because GPS signals cannot be received inside the tunnel, RTK technology cannot be used to locate and deploy control points. Furthermore, the long sampling time of handheld devices inevitably leads to accumulated errors. Additionally, invalid data such as abnormal acceleration and sudden increases in tunnel curvature need to be removed, resulting in a reduction in the amount of effective data.

[0004] The environmental noise is complex: noise in the tunnel environment includes both dynamic and static noise. Dynamic noise mainly originates from workers and moving equipment within the tunnel; static noise comes from waste materials, stationary equipment, lighting fixtures, and electrical installations. This complex noise environment severely interferes with the accuracy of point cloud registration.

[0005] Low overlap rate of effective point cloud fragments: After removing invalid or erroneous data, the overlap rate of effective point cloud data fragments is usually below 25%, resulting in a severe shortage of corresponding point pairs. This often leads to registration failure when traditional registration methods process such point cloud data, making it impossible to accurately construct a tunnel model.

[0006] Poor geometric distinguishability of tunnels: Most highway tunnels have tubular or cylindrical structures, resulting in low distinguishability of their local geometric features. When the overlapping area of ​​point clouds is small, establishing reliable matching relationships based on geometric features becomes extremely difficult, greatly increasing the difficulty of point cloud registration.

[0007] Low lighting conditions affect color features: The lighting conditions inside highway tunnels are often poor, and the colors are relatively uniform, which significantly weakens the reliability of color features in point cloud processing. As one of the important bases for point cloud matching, the reduced reliability of color features directly affects the detection accuracy of overlapping areas.

[0008] Existing point cloud registration techniques all exhibit significant shortcomings when dealing with SLAM scenarios in highway tunnels:

[0009] Iterative Closest Point (ICP): The ICP method is highly dependent on the initial pose. In scenarios with partial overlap and sampling errors, coarse registration is difficult to provide accurate initial alignment, while fine registration is prone to getting trapped in local optima, resulting in inaccurate registration results.

[0010] Scale-Invariant Feature Transform (SIFT): SIFT mainly relies on the detection of key points in an image and the matching of descriptors, and performs well in the processing of textured 2D images. However, in highway tunnel scenes, the low lighting and monochromatic environment make key point detection difficult, and the simple geometric structure of the point cloud further reduces the robustness of matching.

[0011] Accelerated Robust Features (SURF): SURF detects keypoints using the Hessian matrix, offering a slight improvement in computational efficiency compared to SIFT. However, in highway tunnel point cloud processing, it remains severely hampered by low lighting and noise, and feature matching is easily influenced by single geometric structures, resulting in a high false-match rate. Furthermore, SURF performs poorly in handling dynamic noise.

[0012] Difference of Gaussian (DoG): DoG detects key points based on scale-space difference, but in highway tunnel point clouds, dynamic noise and low illumination weaken feature stability. Furthermore, DoG offers limited support for the geometric discriminative power of 3D point clouds.

[0013] Therefore, in order to address the technical problems and deficiencies mentioned above, there is an urgent need to design and develop a tunnel point cloud registration processing method, system, and platform that integrates static noise characteristics. Summary of the Invention

[0014] To overcome the shortcomings and difficulties of the existing technologies, this invention provides a tunnel point cloud registration processing method, system, and platform that integrates static noise features. By utilizing static noise to assist registration and combining adaptive voxel downsampling and particle swarm optimization techniques, the registration accuracy of point cloud post-processing is improved, effectively overcoming the aforementioned technical challenges.

[0015] The first objective of this invention is to provide a tunnel point cloud registration processing method that integrates static noise features; the second objective of this invention is to provide a tunnel point cloud registration processing system that integrates static noise features; and the third objective of this invention is to provide a tunnel point cloud registration processing platform that integrates static noise features.

[0016] The first objective of this invention is achieved as follows: the method comprises the following steps:

[0017] First data corresponding to the tunnel is generated and acquired, and the first data is processed by combining acceleration and curvature filtering to generate second data corresponding to the first data; wherein, the first data is the original point cloud data of the tunnel; and the second data is the point cloud data after cropping.

[0018] The point cloud resolution corresponding to the second data is dynamically adjusted by combining adaptive voxel downsampling, and the feature data corresponding to the second data is extracted and processed; wherein, the feature data includes HSV color feature data, normal vector feature data and curvature feature data;

[0019] Construct a set of corresponding auxiliary point pairs based on the noise auxiliary points, and establish the corresponding initial matching relationship by combining global depth consistency;

[0020] The initial matching is iteratively optimized using a weighted geometric optimization algorithm, while a particle swarm optimization algorithm is used simultaneously for global parameter search. Static noise filtering of residuals is implemented, and corresponding third data is generated. The third data is the final registration result data.

[0021] Furthermore, the process of generating and acquiring first data corresponding to the tunnel, processing the first data by combining acceleration and curvature filtering, and generating second data corresponding to the first data further includes:

[0022] The fourth and fifth data corresponding to the data acquisition device are calculated and generated respectively; wherein, the fourth data is the inter-frame acceleration of the device; and the fifth data is the local curvature of the tunnel.

[0023] The system marks and processes frames with abnormal device acceleration and prunes the entire frame; it also marks and processes tunnel curvature anomalies corresponding to the tunnel and removes them.

[0024] Furthermore, the step of dynamically adjusting the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling and extracting and processing the feature data corresponding to the second data also includes:

[0025] The corresponding voxel resolution is dynamically adjusted based on the point cloud density and curvature; the calculation formula is as follows:

[0026] (11)

[0027] In the formula, The voxel resolution of point i is represented by . It is the basic voxel resolution; and It is the weight of density and curvature; Let i be the local density. This represents the global average point cloud density. Let i be the local curvature. The global average curvature;

[0028] Dynamic noise generated by workers and moving equipment is removed based on the inter-frame density change rate.

[0029] Furthermore, the step of dynamically adjusting the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling and extracting and processing the feature data corresponding to the second data also includes:

[0030] The point cloud RGB colors are converted to HSV space, and the brightness is normalized. At the same time, the corresponding hue and saturation third-order color moments are calculated and generated.

[0031] Normal vector data and curvature feature data are extracted and processed separately, and high curvature characteristic data corresponding to tunnel tubular geometry and static noise are captured and processed.

[0032] Based on geometric distance, curvature, and density, noise-perceptual weighted least squares is used to calculate weights, enhancing noise resistance and distinguishing normal points from noise points;

[0033] The HSV color, normal vector, and curvature feature vector are fused with different weights; and the corresponding feature vectors after fusion are reduced in dimensionality.

[0034] Furthermore, the step of constructing a corresponding set of auxiliary point pairs based on noise-assisted points and establishing a corresponding initial matching relationship in conjunction with global depth consistency also includes:

[0035] Initial pose data is generated based on the depth histogram and principal axis alignment; and matching scores for color distance and geometric distance are calculated; the calculation formula is as follows:

[0036] (26)

[0037] In the formula, This represents the matching score between point p and point q; the larger the value, the more reliable the point is. This represents the color distance between points p and q. and These are the standard deviations of color distance and geometric distance, respectively. It is the geometric distance weighting coefficient;

[0038] The system employs dynamic weighted similarity measurement, bidirectional nearest neighbor matching, and Huber loss filtering.

[0039] Furthermore, the iterative optimization of the initial matching using a weighted geometric optimization algorithm, the simultaneous global parameter search using a particle swarm optimization algorithm, the implementation of residual static noise filtering, and the generation of corresponding third data also include:

[0040] Key point data is adaptively selected based on the point cloud overlap, and the geometric relationship is optimized using weights.

[0041] The particle swarm optimization algorithm is used to update parameters in real time, and the parameter space is searched by iteratively adjusting the particle velocity and position; the calculation formula is as follows:

[0042] (34)

[0043] In the formula, It is the velocity of particle i in the (t+1)th iteration, which controls the direction of parameter update; It is inertial weight. and It is the cognitive and social coefficient. and It is a random factor. and These are the individual optimal and global optimal parameters for particle i, respectively.

[0044] The second objective of this invention is achieved as follows: the system is used to implement the tunnel point cloud registration processing method that fuses static noise features, the system comprising:

[0045] A first data generation unit is used to generate and acquire first data corresponding to the tunnel, and to process the first data by combining acceleration and curvature filtering, and to generate second data corresponding to the first data; wherein, the first data is the original point cloud data of the tunnel; the second data is the point cloud data after cropping; a first data processing unit is used to dynamically adjust the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling, and to extract and process the feature data corresponding to the second data; wherein, the feature data includes HSV color feature data, normal vector feature data, and curvature feature data; a matching relationship creation unit is used to construct a corresponding auxiliary point pair set based on noise auxiliary points, and to establish a corresponding initial matching relationship by combining global depth consistency; a second data generation unit is used to iteratively optimize the initial matching by using a weighted geometric optimization algorithm, simultaneously using a particle swarm optimization algorithm to perform global parameter search, implement residual static noise filtering, and generate corresponding third data; wherein, the third data is the final registration result data.

[0046] Furthermore, the first data generation unit further includes: a first generation module, used to calculate and generate fourth data and fifth data corresponding to the data acquisition device respectively; wherein the fourth data is the device inter-frame acceleration; the fifth data is the tunnel local curvature; a first processing module, used to mark and process abnormal frames of device acceleration and crop the entire frame; and mark and process abnormal points of tunnel curvature corresponding to the tunnel and remove them;

[0047] The first data processing unit further includes: a second processing module, used to dynamically adjust the corresponding voxel resolution based on the point cloud density and curvature; wherein the calculation formula is:

[0048] (11)

[0049] In the formula, The voxel resolution of point i is represented by . It is the basic voxel resolution; and It is the weight of density and curvature; Let i be the local density. This represents the global average point cloud density. Let i be the local curvature. The global average curvature;

[0050] The third processing module is used to remove dynamic noise generated by workers and moving equipment based on the inter-frame density change rate.

[0051] The matching relationship creation unit further includes: a second generation module, used to generate corresponding initial pose data based on the depth histogram and principal axis alignment; and to calculate matching score data for color distance and geometric distance; wherein the calculation formula is:

[0052] (26)

[0053] In the formula, This represents the matching score between point p and point q; the larger the value, the more reliable the point is. This represents the color distance between points p and q. and These are the standard deviations of color distance and geometric distance, respectively. It is the geometric distance weighting coefficient;

[0054] The fourth processing module is used for bidirectional nearest neighbor matching through dynamic weighted similarity measurement and Huber loss filtering.

[0055] The second data generation unit further includes: a fifth processing module, used to adaptively select key point data based on point cloud overlap and optimize geometric relationships using weights; and a sixth processing module, used to update parameters in real time using a particle swarm optimization algorithm and perform parameter space search by iteratively adjusting particle velocity and position; wherein the calculation formula is:

[0056] (34)

[0057] In the formula, It is the velocity of particle i in the (t+1)th iteration, which controls the direction of parameter update; It is inertial weight. and It is the cognitive and social coefficient. and It is a random factor. and These are the individual optimal and global optimal parameters for particle i, respectively.

[0058] Furthermore, the first data processing unit further includes: a third generation module, used to convert the RGB colors of the point cloud into HSV space, normalize the brightness, and simultaneously calculate and generate the corresponding third-order color moments of hue and saturation; a seventh processing module, used to extract and process the normal vector data and curvature feature data respectively, and capture and process the high curvature characteristic data corresponding to the tunnel tubular geometry and static noise; an eighth processing module, used to calculate weights based on geometric distance, curvature, and density using noise-perceptual weighted least squares, to enhance noise resistance and distinguish normal points from noise points; and a ninth processing module, used to fuse HSV colors, normal vectors, and curvature feature vectors with different weights, and to perform dimensionality reduction processing on the corresponding fused feature vectors.

[0059] The third objective of this invention is achieved as follows: the platform includes a processor, a memory, and a tunnel point cloud registration processing platform control program that integrates static noise features; wherein the tunnel point cloud registration processing platform control program that integrates static noise features is executed on the processor, the tunnel point cloud registration processing platform control program that integrates static noise features is stored in the memory, and the tunnel point cloud registration processing platform control program that integrates static noise features implements the tunnel point cloud registration processing method that integrates static noise features.

[0060] This invention generates and acquires first data corresponding to a tunnel through a method, and processes the first data by combining acceleration and curvature filtering to generate second data corresponding to the first data. The first data is the original point cloud data of the tunnel; the second data is the point cloud data after clipping. The method dynamically adjusts the point cloud resolution corresponding to the second data using adaptive voxel downsampling, and extracts and processes feature data corresponding to the second data. The feature data includes HSV color feature data, normal vector feature data, and curvature feature data. A corresponding auxiliary point pair set is constructed based on noise-assisted points, and a corresponding initial matching relationship is established by combining global depth consistency. The initial matching is iteratively optimized using a weighted geometric optimization algorithm, and a particle swarm optimization algorithm is simultaneously used for global parameter search, implementing residual static noise filtering, and generating corresponding third data. The third data is the final registration result data. The invention also includes a corresponding system and platform, providing an efficient and robust solution for tunnel point cloud registration, meeting the needs of high-precision construction quality inspection and long-term maintenance.

[0061] In other words, the present invention effectively removes dynamic noise by comprehensively utilizing technologies such as acceleration and curvature filtering, adaptive voxel downsampling, noise-assisted point pair matching, noise robust weighted optimization, geometric curvature estimation, weighted geometric optimization, and particle swarm optimization. It also uses static noise to assist point cloud registration, successfully solving the problem of insufficient registration accuracy caused by factors such as partial overlap of point clouds, single geometric structure of tunnels, complex noise environment, low illumination conditions, and equipment sampling errors. This ensures high-precision registration and provides strong technical support for the completion and acceptance of highway tunnels. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a schematic diagram of the process steps of a tunnel point cloud registration method that integrates static noise features according to the present invention.

[0064] Figure 2 This is a schematic flowchart of an embodiment of the tunnel point cloud registration processing method that integrates static noise features according to the present invention.

[0065] Figure 3 This is a schematic diagram of the tunnel point cloud registration processing system architecture that integrates static noise features according to the present invention.

[0066] Figure 4 This is a schematic diagram of the tunnel point cloud registration processing platform architecture that integrates static noise features according to the present invention. Detailed Implementation

[0067] To facilitate a clearer understanding of the objectives, technical solutions, and advantages of this invention, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments. Those skilled in the art can easily understand other advantages and effects of this invention from the content disclosed in this specification.

[0068] This invention can also be implemented or applied through other different specific examples, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of this invention.

[0069] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0070] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Secondly, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0071] Preferably, the tunnel point cloud registration processing method integrating static noise features of the present invention is applied in one or more terminals or servers. The terminal is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0072] The terminal can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal can interact with the customer via a keyboard, mouse, remote control, touchpad, or voice control device.

[0073] This invention provides a method, system, and platform for tunnel point cloud registration processing that integrates static noise features.

[0074] like Figure 1 The diagram shown is a flowchart of a tunnel point cloud registration processing method that integrates static noise features, provided in an embodiment of the present invention.

[0075] In this embodiment, the tunnel point cloud registration processing method that integrates static noise features can be applied to terminals or fixed terminals with display functions. The terminals are not limited to personal computers, smartphones, tablets, desktop computers or all-in-one computers with cameras, etc.

[0076] The tunnel point cloud registration processing method that integrates static noise features can also be applied to a hardware environment consisting of a terminal and a server connected to the terminal via a network. The network includes, but is not limited to, a wide area network (WAN), a metropolitan area network (MAN), or a local area network (LAN). The tunnel point cloud registration processing method that integrates static noise features in this embodiment of the invention can be executed by the server, by the terminal, or by both the server and the terminal.

[0077] For example, for a tunnel point cloud registration processing terminal that requires fusing static noise features, the tunnel point cloud registration processing function for fusing static noise features provided by the method of this invention can be directly integrated into the terminal, or a client for implementing the method of this invention can be installed. Alternatively, the method provided by this invention can also run on servers or other devices in the form of a Software Development Kit (SDK), providing an interface for the tunnel point cloud registration processing function for fusing static noise features. Terminals or other devices can then implement the tunnel point cloud registration processing function for fusing static noise features through the provided interface. The invention will be further described below with reference to the accompanying drawings.

[0078] like Figures 1-2 As shown, this invention provides a tunnel point cloud registration processing method that integrates static noise features. The method includes the following steps:

[0079] S01. Generate and acquire first data corresponding to the tunnel, and combine acceleration and curvature filtering to crop the first data, and generate second data corresponding to the first data; wherein, the first data is the original point cloud data of the tunnel; the second data is the cropped point cloud data;

[0080] S02. Dynamically adjust the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling, and extract and process the feature data corresponding to the second data; wherein, the feature data includes HSV color feature data, normal vector feature data and curvature feature data;

[0081] S03. Construct a corresponding set of auxiliary points based on the noise auxiliary points, and establish the corresponding initial matching relationship in combination with global depth consistency;

[0082] S04. The initial matching is iteratively optimized using a weighted geometric optimization algorithm, and a global parameter search is performed simultaneously using a particle swarm optimization algorithm. Static noise filtering of residuals is implemented, and corresponding third data is generated; wherein, the third data is the final registration result data.

[0083] The process of generating and acquiring first data corresponding to the tunnel, processing the first data by combining acceleration and curvature filtering, and generating second data corresponding to the first data further includes:

[0084] S011. Calculate and generate the fourth data and the fifth data corresponding to the data acquisition device respectively; wherein, the fourth data is the inter-frame acceleration of the device; and the fifth data is the local curvature of the tunnel;

[0085] S012, mark and process abnormal frames of device acceleration and crop the entire frame; and mark and process abnormal points of tunnel curvature corresponding to the tunnel and remove them.

[0086] The step of dynamically adjusting the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling and extracting and processing the feature data corresponding to the second data further includes:

[0087] S021. Dynamically adjust the corresponding voxel resolution based on the point cloud density and curvature; the calculation formula is as follows:

[0088] (11)

[0089] In the formula, The voxel resolution of point i is represented by . It is the basic voxel resolution; and It is the weight of density and curvature; Let i be the local density. This represents the global average point cloud density. Let i be the local curvature. The global average curvature;

[0090] S022. Dynamic noise generated by workers and moving equipment is removed based on the inter-frame density change rate.

[0091] The method of dynamically adjusting the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling and extracting and processing the feature data corresponding to the second data further includes: S023, converting the RGB color of the point cloud to HSV space and normalizing the brightness, while calculating and generating the corresponding hue and saturation third-order color moments; S024, extracting and processing the normal vector data and curvature feature data respectively, and capturing and processing the high curvature characteristic data corresponding to the tunnel tubular geometry and static noise; S025, based on geometric distance, curvature, and density, using noise-perceptual weighted least squares to calculate weights, enhancing noise resistance and distinguishing normal points from noise points; S026, fusing the HSV color, normal vector, and curvature feature vector with different weights; and reducing the dimensionality of the corresponding feature vector after fusion processing.

[0092] The step of constructing a corresponding set of auxiliary points based on noise-assisted points and establishing a corresponding initial matching relationship in conjunction with global depth consistency also includes:

[0093] S031. Generate corresponding initial pose data based on the depth histogram and principal axis alignment; and calculate matching scores for color distance and geometric distance; wherein the calculation formula is:

[0094] (26)

[0095] In the formula, This represents the matching score between point p and point q; the larger the value, the more reliable the point is. This represents the color distance between points p and q. and These are the standard deviations of color distance and geometric distance, respectively. It is the geometric distance weighting coefficient;

[0096] S032. Through dynamic weighted similarity measurement, bidirectional nearest neighbor matching is performed, and Huber loss is used for filtering.

[0097] The process of iteratively optimizing the initial matching using a weighted geometric optimization algorithm, simultaneously employing a particle swarm optimization algorithm for global parameter search, implementing residual static noise filtering, and generating corresponding third data also includes:

[0098] S041. Select key point data adaptively based on the point cloud overlap, and optimize the geometric relationship using weights;

[0099] S042. Parameters are updated in real time using a particle swarm optimization algorithm, and parameter space search is achieved by iteratively adjusting particle velocity and position; the calculation formula is as follows:

[0100] (34)

[0101] In the formula, It is the velocity of particle i in the (t+1)th iteration, which controls the direction of parameter update; It is inertial weight. and It is the cognitive and social coefficient. and It is a random factor. and These are the individual optimal and global optimal parameters for particle i, respectively.

[0102] Specifically, this invention relates to a tunnel point cloud registration method specifically designed for Simultaneous Localization and Mapping (SLAM) technology. This method is primarily applicable to the processing of point cloud data collected by handheld mobile LiDAR scanners during the construction quality inspection of highway tunnels during the final acceptance phase. It focuses on accurately verifying whether the dimensions of the tunnel's main structure meet design requirements, such as achieving centimeter-level accuracy in verifying the length, curvature, and other geometric dimensions of tunnel walls and arches. This invention comprehensively utilizes techniques such as acceleration and curvature filtering, adaptive voxel downsampling, noise-assisted point pair matching, noise robust weighted optimization, geometric curvature estimation, weighted geometric optimization, and particle swarm optimization to effectively remove dynamic noise and assist point cloud registration with static noise. This successfully solves the problem of insufficient registration accuracy caused by factors such as partial overlap of point clouds, the single geometric structure of the tunnel, complex noise environments, low lighting conditions, and equipment sampling errors, ensuring high-precision registration and providing strong technical support for the final acceptance of highway tunnels.

[0103] In other words, this invention provides a precise point cloud registration technology for highway tunnels based on static noise assistance, aiming to solve the problems of insufficient registration accuracy caused by handheld LiDAR in highway tunnel completion acceptance due to low overlap rate of effective point cloud segments, complex dynamic and static noise, poor distinguishability inside the tunnel, limited color information in low light, and equipment sampling errors. This method removes dynamic noise through acceleration and curvature filtering, utilizes the characteristics of static noise such as waste materials, stationary equipment, and lighting devices to assist point pair matching, and combines adaptive voxel downsampling, noise robust weighted optimization, and global particle swarm optimization to generate a high-precision 3D model that meets the stringent standards for tunnel construction quality inspection.

[0104] The core innovation of this invention lies in utilizing the high curvature and density characteristics of static noise as registration auxiliary features, breaking through the dependence of traditional methods on initial pose or rich texture. The method first prunes drifting data through acceleration and curvature filtering, and dynamically adjusts the point cloud resolution by combining adaptive voxel downsampling, balancing computational efficiency and detail preservation. Next, HSV color features, normal vectors, and curvature features are extracted, and noise-robust weighted optimization enhances the feature robustness in low-light and single-geometric scenes. Noise-assisted point pair matching utilizes static noise to generate high-quality point pair sets, and combines global depth consistency priors to improve the registration reliability of partially overlapping scenes. Finally, through weighted geometry optimization and particle swarm optimization, the registration results are refined and residual static noise is removed, achieving synergy between local adaptation and global optimization, ensuring registration accuracy and resilience. This invention is applicable to the completion and acceptance of various tunnel types such as highways, railways, and subways, as well as deformation monitoring, maintenance, and inspection scenarios, and is adapted to the portability of handheld LiDAR and the operational needs in complex environments. The originality of this invention lies in static noise-assisted registration, multi-feature fusion, and local-global collaborative mechanisms, which provide an efficient and robust solution for tunnel point cloud registration, meeting the needs of high-precision construction quality inspection and long-term maintenance.

[0105] The specific technical solution involves acceleration and curvature filtering: Existing technologies, such as ICP, typically lack a preprocessing mechanism for drifting data, while methods like SIFT, SURF, and DoG are not suitable for handling dynamic anomalies in point clouds. The acceleration and curvature filtering algorithm of this invention combines a dual dynamic filtering mechanism of device inter-frame acceleration and tunnel local curvature, enabling the processing of invalid data such as IMU acceleration anomalies and sudden increases in measured tunnel curvature that occur during data acquisition by handheld devices. The specific steps are as follows:

[0106] (1) Calculate the device inter-frame acceleration: First, extract the pose information of the point cloud of the device in each frame, including the rotation matrix. and position vector This information can be estimated by IMU or SLAM. Next, inter-frame acceleration is calculated using the following formula:

[0107] (1)

[0108] In the formula, Let be the acceleration vector of the device at time t during the measurement process;

[0109] The velocity vector of the device at time t during the measurement process is calculated using the following formula:

[0110] (2)

[0111] This represents the inter-frame time interval between time t and time t-1.

[0112] If the norm of the acceleration vector is greater than a set threshold, or the dot product of the acceleration vector and the velocity vector is less than 0, then the corresponding point is marked as an outlier. The conditions for an outlier are:

[0113] or (3)

[0114] In the formula, express The L2 norm is calculated using the following formula:

[0115] (4)

[0116] The threshold for outliers in acceleration data is calculated using the following formula:

[0117] (5)

[0118] In the formula, It is all the median; The standard deviation is the correlation coefficient; It is all The absolute deviation of the median of the L2 norm is calculated using the following formula:

[0119] (6)

[0120] (2) Evaluate the local curvature of the tunnel: Calculate the local curvature of the tunnel for each frame of the point cloud using the geometrically constrained PCA method with k nearest neighbors (the value of k is determined by the particle swarm optimization algorithm). The curvature calculation formula is as follows:

[0121] (7)

[0122] In the formula, Represents the local curvature of the i-th point at time t; These are the eigenvalues ​​of the three principal components obtained through PCA analysis, which represent the variance of the point cloud in the three principal directions at that point; It is the minimum of these three eigenvalues, reflecting the direction of curvature.

[0123] The conditions for detecting a sudden increase in curvature are:

[0124] (8)

[0125] In the formula, The threshold for outliers in curvature data is calculated using the following formula:

[0126] (9)

[0127] In the formula, For all the median; The standard deviation is the correlation coefficient; It is all The median absolute deviation.

[0128] (3) Smoothing of outliers: Mark frames with abnormal device acceleration and crop the entire frame; mark and remove tunnel curvature outliers. For cropped boundary frames, a weighted average pose method is used for smooth transition, with the following formula:

[0129] (10)

[0130] In the formula, , This is the smoothing coefficient.

[0131] (4) Verification and output: Check the continuity of the point cloud after cropping, requiring that the inter-frame displacement is less than the threshold. ( The value is determined by the particle swarm optimization algorithm. If the condition is met, the filtered point cloud data is output.

[0132] Adaptive Voxel Downsampling: SLAM for highway tunnels requires efficient downsampling and noise filtering to support dimensional checks during final acceptance. However, the actual measured point cloud density and curvature distribution are uneven, and a fixed voxel resolution may lose key features or retain too much noise. Adaptive voxel downsampling dynamically adjusts the voxel resolution based on the point cloud density and curvature, efficiently preserving static noise to support noise-guided matching, while filtering dynamic noise from workers and moving equipment. The specific steps are as follows:

[0133] (1) Calculate adaptive voxel resolution: The voxel resolution is dynamically adjusted based on the point cloud density and curvature. The formula is as follows:

[0134] (11)

[0135] In the formula, The voxel resolution of point i is represented by . It is the basic voxel resolution; and It is the weight of density and curvature; Let i be the local density. This represents the global average point cloud density. Let i be the local curvature. The global average curvature.

[0136] (2) Filtering dynamic noise: Calculating voxel density Based on the inter-frame density change rate, dynamic noise generated by workers and moving equipment is removed, i.e.:

[0137] (12)

[0138] like Then the point is determined to be a noise point, where It is the density change threshold.

[0139] (3) Comprehensive Distance Metric: This metric considers both spatial distance and color distance for subsequent point-to-point matching. The calculation formula is as follows:

[0140] (13)

[0141] In the formula, Let Euclidean distance be the representative point. The hue-saturation-lightness (HSV) color distance; and These are the weighting coefficients.

[0142] Hybrid Feature Extraction and Noise Detection: This invention extracts hybrid features for each representative point, including normalized HSV color features, normal vector features, and curvature features. Noise robustness weighted optimization enhances the algorithm's noise resistance, and dimensionality reduction techniques are employed to ultimately detect static noise, strengthening its adaptability under low illumination and overlapping rate fluctuations.

[0143] (1) Extract HSV color features; convert the RGB colors of the point cloud to HSV space, normalize the brightness, calculate the third-order color moments of hue and saturation, and adapt to low-light environments by using the following conversion method:

[0144] Input RGB values ​​(R,G,B∈[0,255]) and normalize them to [0,1]. The specific formula is as follows:

[0145] (14)

[0146] Next, calculate the maximum value: M=max(R′,G′,B′), the minimum value: m=min(R′,G′,B′), and the difference: C=M−m.

[0147] Next, calculate the brightness: V=M.

[0148] Next, we calculate the saturation:

[0149] (15)

[0150] Then calculate the hue:

[0151] (16)

[0152] Final output: H∈[0,360], S∈[0,1], V∈[0,1].

[0153] (2) Normalized HSV color characteristics: Convert the RGB color values ​​of representative points to the HSV color space. Perform global normalization on the lightness V, calculated using the following formula:

[0154] (17)

[0155] In the formula, The normalized brightness, Minimum brightness Maximum brightness.

[0156] Calculate the third-order color moments of hue H and saturation S, including mean, standard deviation, and skewness, to generate a 6-dimensional feature vector:

[0157] (18)

[0158] In the formula, mean represents the data mean, std represents the data standard deviation, and skew represents the data skewness.

[0159] (3) Calculation of normal vector and curvature: Extract the normal vector and curvature features to capture the high curvature characteristics of the tunnel's tubular geometry and static noise. Calculate the normal vector using principal component analysis with k nearest neighbors (the value of k is determined by the particle swarm optimization algorithm). The formula is:

[0160] (19)

[0161] In the formula, Let i represent the normal vector of point i; This represents the smallest eigenvalue of the neighborhood covariance matrix; It corresponds to the smallest eigenvalue eigenvectors.

[0162] Combined with the previously calculated curvature Generate a 4-dimensional eigenvector containing the three components of the normal vector and the curvature. .

[0163] (4) Noise-Robust Weighted Optimization: Based on geometric distance, curvature, and density, noise-perceived weighted least squares (NP-WLS) is used to calculate weights, enhancing noise resistance and distinguishing normal points from noise points. The formula is as follows:

[0164] (20)

[0165] In the formula, It is the weight of point i; the larger the value, the more reliable the point is. Let be the Euclidean distance from point to its nearest neighbor. Let be the local curvature of the i-th voxel. The threshold for outliers in the curvature data; Let be the point cloud density of the i-th voxel. The threshold for density; It is a distance threshold, defined as:

[0166] (twenty one)

[0167] in, The average point cloud density for all voxels. and The correlation coefficient.

[0168] (5) Weighted Feature Fusion: The HSV color, normal vector, and curvature feature vector are fused with different weights. The calculation formula is as follows:

[0169] (twenty two)

[0170] In the formula, It is the fused feature vector; It is an HSV color feature vector. These are the normal vector and the curvature eigenvector. and These are weights, which will be determined later using the particle swarm optimization algorithm.

[0171] (6) Feature dimensionality reduction: Reduce the dimensionality of the fused feature vectors to a fixed dimension. This dimension is determined by the particle swarm optimization algorithm.

[0172] (7) Static noise detection: Based on the geometrically constrained PCA curvature, static noise generated by waste materials, stationary machinery, lighting devices, and electrical equipment is identified. The formula is:

[0173] (twenty three)

[0174] In the formula, The curvature of the geometric constraint at point i is represented; These are the eigenvalues ​​of the three principal components obtained through PCA analysis, representing the variance of the point cloud in the three principal directions at that point; It is the minimum value among these three eigenvalues; It is the normal vector of point i. It is the principal axis direction vector of the tunnel. It is a coefficient.

[0175] when and When identified as static noise, the static noise point set is marked. , This provides crucial information for subsequent noise-guided point pair matching.

[0176] Noise-assisted point pair matching: Static noise in highway tunnel SLAM has unique geometric features, which can enhance matching for small overlaps. A comprehensive point pair set is generated by utilizing global depth consistency priors, color and geometric consistency priors, and static noise feature matching.

[0177] (1) Global depth consistency prior: The initial pose is generated by aligning the depth histogram with the principal axis, and the formula is:

[0178] (twenty four)

[0179] (25)

[0180] In the formula, This represents the initial translation vector; N is the number of intervals used to divide the data range in the histogram. It is the depth histogram value of point cloud P in the i-th interval; It was translated The depth histogram value of the point cloud Q in the i-th interval; The initial rotation matrix is ​​defined by minimizing the principal axis direction vector of the source point cloud P. The principal axis direction vector of the target point set Q Calculate the difference between them; rotation matrix Used to align the main axis directions of two point clouds.

[0181] (2) Color and geometric consistency matching: Calculate the matching score for color distance and geometric distance, using the following formula:

[0182] (26)

[0183] In the formula, This represents the matching score between point p and point q; the larger the value, the more reliable the point is. This represents the color distance between points p and q. and These are the standard deviations of color distance and geometric distance, respectively. It is the geometric distance weighting coefficient; The distance between points p and q in the geometric feature space is represented by the following formula:

[0184] (27)

[0185] In the formula, , , and These are the normal vector components and curvature of point p, respectively; , , and These are the normal vector components and curvature of point q, respectively. It is the curvature correlation coefficient.

[0186] (3) Noise-guided feature matching: Calculate the similarity of noise point pairs using the following formula:

[0187] (28)

[0188] In the formula, Represents a noise-assisted distance metric. and It is the weighting coefficient.

[0189] (4) Point-to-point filtering: Through dynamic weighted similarity measurement, bidirectional nearest neighbor matching, Huber loss filtering, and setting a threshold of Further, mismatched point pairs are removed to ensure the accuracy and reliability of the final point pair set.

[0190] Registration optimization and static noise removal: The registration ratio is optimized through adaptive keypoint sampling, and static noise is removed by combining weighted geometric optimization. The specific steps are as follows:

[0191] (1) Adaptive key point sampling: Key points are adaptively selected based on the overlap of the point cloud to improve registration efficiency and accuracy. The calculation formula is as follows:

[0192] (29)

[0193] In the formula, This represents the set of keypoints sampled from the initial set of matching point pairs C′; It is the minimum sampling ratio. It is the overlap rate adjustment coefficient. It is the threshold of the overlap rate. It is the estimated overlap rate.

[0194] Next, dynamic weighted point-to-point optimization is performed, with the following formula:

[0195] (30)

[0196] In the formula, For rotation matrix, As weight; point The position vector in the point cloud P For point The position vector in the point cloud Q.

[0197] (2) Weighted geometric optimization: The geometric relationship is optimized using this weight to further improve the registration accuracy. The formula is as follows:

[0198] (31)

[0199] In the formula, These are the weights used in weighted geometric optimization; and They are points and The geometric constraint curvature of the point pair, It is a weighting scaling factor. It is the curvature threshold; It is the local density at point i. It is the average local density.

[0200] (3) Static noise removal: Remove high curvature or low density points, if the points Greater than the set curvature threshold or the density of points less than the set density threshold If the point is identified as a static noise point and removed, a clean registration point cloud and accurate transformation parameters are output, providing high-quality data for subsequent verification of the tunnel's main structure dimensions and deformation analysis.

[0201] Registration optimization and static noise removal: Particle swarm optimization is used, with design drawings or high-precision LiDAR point clouds as references, to optimize relevant weights and thresholds to maximize registration accuracy. The specific steps are as follows:

[0202] (1) Define the parameter space: The parameter space includes weight parameters, threshold parameters, and other parameters. Weight parameters include , , , , , , , , , , , , , , Threshold parameters include , , , , , , , , , , Other parameters include , , , , , , , These parameters play a crucial role in the entire point cloud registration process, and the appropriateness of their values ​​directly affects the registration accuracy.

[0203] (2) Define the objective function: The objective function is used to measure the registration accuracy and is calculated using the root mean square error (RMSE) or Hausdorff distance. When the reference data is a high-precision lidar point cloud, the root mean square error is used:

[0204] (32)

[0205] In the formula, N is the number of point cloud points involved in the calculation. and These are the coordinates of the registration point cloud points and the nearest point in the reference point cloud, respectively.

[0206] If the reference data is the grid of the design drawing, then the Hausdorff distance is used, and the calculation formula is as follows:

[0207] (33)

[0208] In the formula, P and Q are the registration point cloud and the reference mesh level, respectively. Let p represent the Euclidean distance between points p and q, sup denote the supremum, and inf denote the infimum.

[0209] (3) Particle Swarm Optimization Algorithm: The parameters are updated using the particle swarm optimization algorithm. By continuously updating the velocity and position of the particles, the algorithm gradually converges to the optimal solution. The calculation formula is as follows:

[0210] (34)

[0211] In the formula, It is the velocity of particle i in the (t+1)th iteration, which controls the direction of parameter update; It is inertial weight. and It is the cognitive and social coefficient. and It is a random factor. and These are the individual optimal and global optimal parameters for particle i, respectively.

[0212] Then update the parameter values ​​of particle i:

[0213] (35)

[0214] In the formula, and These are the parameter values ​​of particle i in the t-th and t+1-th iterations, respectively.

[0215] (4) Parameter Constraints: To ensure the rationality of the parameters and the stability of the algorithm, constraints are imposed on various parameters. The weight parameters are limited to the [0,1] interval; the threshold parameters are set with corresponding ranges based on their physical meaning, for example... exist Values ​​can be taken within a range.

[0216] (5) Output results: After iterative calculation by the particle swarm optimization algorithm, the optimized parameter set is output, and the final registration point cloud is generated using the parameter set to achieve high-precision point cloud registration.

[0217] To achieve the above objectives, the present invention also provides a tunnel point cloud registration processing system that integrates static noise features, such as... Figure 3 As shown, the system is used to implement the tunnel point cloud registration processing method that fuses static noise features. The system specifically includes:

[0218] The first data generation unit is used to generate and acquire first data corresponding to the tunnel, and to process the first data by combining acceleration and curvature filtering, and to generate second data corresponding to the first data; wherein, the first data is the original point cloud data of the tunnel; and the second data is the point cloud data after processing.

[0219] The first data processing unit is used to dynamically adjust the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling, and to extract and process the feature data corresponding to the second data; wherein, the feature data includes HSV color feature data, normal vector feature data and curvature feature data;

[0220] The matching relationship creation unit is used to construct a corresponding set of auxiliary point pairs based on the noise auxiliary points, and to establish the corresponding initial matching relationship in combination with global depth consistency;

[0221] The second data generation unit is used to iteratively optimize the initial matching using a weighted geometric optimization algorithm, simultaneously perform global parameter search using a particle swarm optimization algorithm, implement residual static noise filtering, and generate corresponding third data; wherein, the third data is the final registration result data.

[0222] The first data generation unit further includes: a first generation module, used to calculate and generate fourth data and fifth data corresponding to the data acquisition device respectively; wherein, the fourth data is the device inter-frame acceleration; and the fifth data is the local curvature of the tunnel;

[0223] The first processing module is used to mark and process frames with abnormal acceleration of the processing device and crop the entire frame; and to mark and process tunnel curvature anomalies corresponding to the tunnel and remove them.

[0224] The first data processing unit further includes: a second processing module, used to dynamically adjust the corresponding voxel resolution based on the point cloud density and curvature; wherein the calculation formula is:

[0225] (11)

[0226] In the formula, The voxel resolution of point i is represented by . It is the basic voxel resolution; and It is the weight of density and curvature; Let i be the local density. This represents the global average point cloud density. Let i be the local curvature. The global average curvature;

[0227] The third processing module is used to remove dynamic noise generated by workers and moving equipment based on the inter-frame density change rate.

[0228] The matching relationship creation unit further includes: a second generation module, used to generate corresponding initial pose data based on the depth histogram and principal axis alignment; and to calculate matching score data for color distance and geometric distance; wherein the calculation formula is:

[0229] (26)

[0230] In the formula, This represents the matching score between point p and point q; the larger the value, the more reliable the point is. This represents the color distance between points p and q. and These are the standard deviations of color distance and geometric distance, respectively. It is the geometric distance weighting coefficient;

[0231] The fourth processing module is used for bidirectional nearest neighbor matching through dynamic weighted similarity measurement and Huber loss filtering.

[0232] The second data generation unit further includes: a fifth processing module, used to adaptively select key point data based on point cloud overlap and optimize geometric relationships using weights; and a sixth processing module, used to update parameters in real time using a particle swarm optimization algorithm and perform parameter space search by iteratively adjusting particle velocity and position; wherein the calculation formula is:

[0233] (34)

[0234] In the formula, It is the velocity of particle i in the (t+1)th iteration, which controls the direction of parameter update; It is inertial weight. and It is the cognitive and social coefficient. and It is a random factor. and These are the individual optimal and global optimal parameters for particle i, respectively.

[0235] The first data processing unit further includes: a third generation module, used to convert the RGB colors of the point cloud into HSV space, normalize the brightness, and simultaneously calculate and generate the corresponding third-order color moments of hue and saturation; a seventh processing module, used to extract and process the normal vector data and curvature feature data respectively, and capture and process the high curvature characteristic data corresponding to the tunnel tubular geometry and static noise; an eighth processing module, used to calculate weights based on geometric distance, curvature, and density using noise-perceptual weighted least squares, to enhance noise resistance and distinguish normal points from noise points; and a ninth processing module, used to fuse HSV colors, normal vectors, and curvature feature vectors with different weights, and to perform dimensionality reduction processing on the corresponding fused feature vectors.

[0236] In the system solution embodiment of the present invention, the specific details of the method steps involved in the tunnel point cloud registration processing that integrates static noise features have been described above. That is to say, the functional modules in the system are used to implement the steps or sub-steps in the above method embodiment, and will not be repeated here.

[0237] To achieve the above objectives, the present invention also provides a tunnel point cloud registration processing platform that integrates static noise features, such as... Figure 4 As shown, the system includes a processor, a memory, and a control program for a tunnel point cloud registration processing platform that integrates static noise features. The processor executes the control program, which is stored in the memory. This control program implements the steps of the tunnel point cloud registration processing method that integrates static noise features. For example:

[0238] S01. Generate and acquire first data corresponding to the tunnel, and combine acceleration and curvature filtering to crop the first data, and generate second data corresponding to the first data; wherein, the first data is the original point cloud data of the tunnel; the second data is the cropped point cloud data;

[0239] S02. Dynamically adjust the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling, and extract and process the feature data corresponding to the second data; wherein, the feature data includes HSV color feature data, normal vector feature data and curvature feature data;

[0240] S03. Construct a corresponding set of auxiliary points based on the noise auxiliary points, and establish the corresponding initial matching relationship in combination with global depth consistency;

[0241] S04. The initial matching is iteratively optimized using a weighted geometric optimization algorithm, and a global parameter search is performed simultaneously using a particle swarm optimization algorithm. Static noise filtering of residuals is implemented, and corresponding third data is generated; wherein, the third data is the final registration result data.

[0242] The specific details of the steps have been explained above and will not be repeated here.

[0243] In this embodiment of the invention, the tunnel point cloud registration processing platform integrating static noise features has a built-in processor, which can be composed of integrated circuits. For example, it can be composed of a single packaged integrated circuit, or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor connects to various components using various interfaces and lines, and executes programs or units stored in the memory, as well as calling data stored in the memory, to perform various functions and process data for tunnel point cloud registration integrating static noise features. The memory is used to store program code and various data, is installed in the tunnel point cloud registration processing platform integrating static noise features, and achieves high-speed and automatic access to programs or data during operation. The memory includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0244] This invention generates and acquires first data corresponding to a tunnel through a method, and processes the first data by combining acceleration and curvature filtering to generate second data corresponding to the first data. The first data is the original point cloud data of the tunnel; the second data is the point cloud data after clipping. The method dynamically adjusts the point cloud resolution corresponding to the second data using adaptive voxel downsampling, and extracts and processes feature data corresponding to the second data. The feature data includes HSV color feature data, normal vector feature data, and curvature feature data. A corresponding auxiliary point pair set is constructed based on noise-assisted points, and a corresponding initial matching relationship is established by combining global depth consistency. The initial matching is iteratively optimized using a weighted geometric optimization algorithm, and a particle swarm optimization algorithm is simultaneously used for global parameter search, implementing residual static noise filtering, and generating corresponding third data. The third data is the final registration result data. The invention also includes a corresponding system and platform, providing an efficient and robust solution for tunnel point cloud registration, meeting the needs of high-precision construction quality inspection and long-term maintenance.

[0245] In other words, the present invention effectively removes dynamic noise by comprehensively utilizing technologies such as acceleration and curvature filtering, adaptive voxel downsampling, noise-assisted point pair matching, noise robust weighted optimization, geometric curvature estimation, weighted geometric optimization, and particle swarm optimization. It also uses static noise to assist point cloud registration, successfully solving the problem of insufficient registration accuracy caused by factors such as partial overlap of point clouds, single geometric structure of tunnels, complex noise environment, low illumination conditions, and equipment sampling errors. This ensures high-precision registration and provides strong technical support for the completion and acceptance of highway tunnels.

[0246] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for registering tunnel point clouds by incorporating static noise features, characterized in that, The method includes the following steps: First data corresponding to the tunnel is generated and acquired, and the first data is processed by combining acceleration and curvature filtering to generate second data corresponding to the first data; wherein, the first data is the original point cloud data of the tunnel; and the second data is the point cloud data after cropping. The point cloud resolution corresponding to the second data is dynamically adjusted by combining adaptive voxel downsampling, and the feature data corresponding to the second data is extracted and processed; wherein, the feature data includes HSV color feature data, normal vector feature data and curvature feature data; Construct a corresponding set of auxiliary point pairs based on the noise auxiliary points, and establish the corresponding initial matching relationship by combining global depth consistency; The initial matching is iteratively optimized using a weighted geometric optimization algorithm, while a particle swarm optimization algorithm is used simultaneously for global parameter search. Static noise filtering of residuals is implemented, and corresponding third data is generated. The third data is the final registration result data.

2. The tunnel point cloud registration processing method based on the fusion of static noise features according to claim 1, characterized in that, The process of generating and acquiring first data corresponding to the tunnel, processing the first data by combining acceleration and curvature filtering, and generating second data corresponding to the first data further includes: The fourth and fifth data corresponding to the data acquisition device are calculated and generated respectively; wherein, the fourth data is the inter-frame acceleration of the device; and the fifth data is the local curvature of the tunnel. The system marks and processes frames with abnormal device acceleration and prunes the entire frame; it also marks and processes tunnel curvature anomalies corresponding to the tunnel and removes them.

3. The tunnel point cloud registration processing method based on the fusion of static noise features according to claim 1, characterized in that, The step of dynamically adjusting the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling and extracting and processing the feature data corresponding to the second data further includes: The corresponding voxel resolution is dynamically adjusted based on the point cloud density and curvature; the calculation formula is as follows: （11） In the formula, The voxel resolution of point i is represented by . It is the basic voxel resolution; and It is the weight of density and curvature; Let i be the local density. This represents the global average point cloud density. Let i be the local curvature. The global average curvature; Dynamic noise generated by workers and moving equipment is removed based on the inter-frame density change rate.

4. A tunnel point cloud registration processing method based on the fusion of static noise features according to claim 1 or 3, characterized in that, The step of dynamically adjusting the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling and extracting and processing the feature data corresponding to the second data further includes: The point cloud RGB colors are converted to HSV space, and the brightness is normalized. At the same time, the corresponding hue and saturation third-order color moments are calculated and generated. Normal vector data and curvature feature data are extracted and processed separately, and high curvature characteristic data corresponding to tunnel tubular geometry and static noise are captured and processed. Based on geometric distance, curvature, and density, noise-perceptual weighted least squares is used to calculate weights, enhancing noise resistance and distinguishing normal points from noise points; The HSV color, normal vector, and curvature feature vector are fused with different weights; and the corresponding feature vectors after fusion are reduced in dimensionality.

5. The tunnel point cloud registration processing method based on the fusion of static noise features according to claim 1, characterized in that, The step of constructing a corresponding set of auxiliary points based on noise-assisted points and establishing a corresponding initial matching relationship in conjunction with global depth consistency also includes: Initial pose data is generated based on the depth histogram and principal axis alignment; and matching scores for color distance and geometric distance are calculated; the calculation formula is as follows: （26） In the formula, This represents the matching score between point p and point q; the larger the value, the more reliable the point is. This represents the color distance between points p and q. and These are the standard deviations of color distance and geometric distance, respectively. It is the geometric distance weighting coefficient; The system employs dynamic weighted similarity measurement, bidirectional nearest neighbor matching, and Huber loss filtering.

6. The tunnel point cloud registration processing method according to claim 1, characterized in that, The process of iteratively optimizing the initial matching using a weighted geometric optimization algorithm, simultaneously employing a particle swarm optimization algorithm for global parameter search, implementing residual static noise filtering, and generating corresponding third data also includes: Key point data is adaptively selected based on the point cloud overlap, and the geometric relationship is optimized using weights. The particle swarm optimization algorithm is used to update parameters in real time, and the parameter space is searched by iteratively adjusting the particle velocity and position; the calculation formula is as follows: （34） In the formula, It is the velocity of particle i in the (t+1)th iteration, which controls the direction of parameter update; It is inertial weight. and It is the cognitive and social coefficient. and It is a random factor. and These are the individual optimal and global optimal parameters for particle i, respectively.

7. A tunnel point cloud registration processing system that integrates static noise features, characterized in that, The system is used to implement the tunnel point cloud registration processing method for fusing static noise features as described in any one of claims 1-6, the system comprising: The first data generation unit is used to generate and acquire first data corresponding to the tunnel, and to process the first data by combining acceleration and curvature filtering, and to generate second data corresponding to the first data; wherein, the first data is the original point cloud data of the tunnel; and the second data is the point cloud data after processing. The first data processing unit is used to dynamically adjust the point cloud resolution corresponding to the second data by combining adaptive voxel downsampling, and to extract and process the feature data corresponding to the second data; wherein, the feature data includes HSV color feature data, normal vector feature data and curvature feature data; The matching relationship creation unit is used to construct a corresponding set of auxiliary point pairs based on the noise auxiliary points, and to establish the corresponding initial matching relationship in combination with global depth consistency; The second data generation unit is used to iteratively optimize the initial matching using a weighted geometric optimization algorithm, simultaneously perform global parameter search using a particle swarm optimization algorithm, implement static noise filtering of residuals, and generate corresponding third data; wherein, the third data is the final registration result data.

8. The tunnel point cloud registration processing system based on the fusion of static noise features according to claim 7, characterized in that, The first data generation unit further includes: The first generation module is used to calculate and generate fourth data and fifth data corresponding to the data acquisition device, respectively; wherein, the fourth data is the device inter-frame acceleration; and the fifth data is the local curvature of the tunnel. The first processing module is used to mark and process frames with abnormal acceleration of the processing device and crop the entire frame; and to mark and process tunnel curvature anomalies corresponding to the tunnel and remove them. The first data processing unit further includes: The second processing module is used to dynamically adjust the corresponding voxel resolution based on the point cloud density and curvature; the calculation formula is as follows: （11） In the formula, The voxel resolution of point i is represented by . It is the basic voxel resolution; and It is the weight of density and curvature; Let i be the local density. This represents the global average point cloud density. Let i be the local curvature. The global average curvature; The third processing module is used to remove dynamic noise generated by workers and moving equipment based on the inter-frame density change rate. The matching relationship creation unit further includes: The second generation module is used to generate corresponding initial pose data based on the depth histogram and principal axis alignment; and to calculate matching scores for color distance and geometric distance; wherein the calculation formula is: （26） In the formula, This represents the matching score between point p and point q; the larger the value, the more reliable the point is. This represents the color distance between points p and q. and These are the standard deviations of color distance and geometric distance, respectively. It is the geometric distance weighting coefficient; The fourth processing module is used for bidirectional nearest neighbor matching through dynamic weighted similarity measurement and Huber loss filtering. The second data generation unit further includes: The fifth processing module is used to adaptively select key point data based on the overlap of point clouds and to optimize the geometric relationship using weights. The sixth processing module is used to update parameters in real time using a particle swarm optimization algorithm, achieving parameter space search by iteratively adjusting particle velocity and position; the calculation formula is as follows: （34） In the formula, It is the velocity of particle i in the (t+1)th iteration, which controls the direction of parameter update; It is inertial weight. and It is the cognitive and social coefficient. and It is a random factor. and These are the individual optimal and global optimal parameters for particle i, respectively.

9. A tunnel point cloud registration processing system for fusing static noise features according to claim 7 or 8, characterized in that, The first data processing unit further includes: The third generation module is used to convert the RGB colors of the point cloud into the HSV space, normalize the brightness, and calculate and generate the corresponding third-order color moments of hue and saturation. The seventh processing module is used to extract and process normal vector data and curvature feature data respectively, and to capture and process high curvature characteristic data corresponding to tunnel tubular geometry and static noise. The eighth processing module is used to calculate weights based on geometric distance, curvature, and density using noise-perceptual weighted least squares, thereby enhancing noise resistance and distinguishing normal points from noise points. The ninth processing module is used to fuse HSV color, normal vector and curvature feature vector with different weights; and to perform dimensionality reduction on the corresponding feature vector after fusion.

10. A tunnel point cloud registration processing platform that integrates static noise features, characterized in that, The system includes a processor, a memory, and a control program for a tunnel point cloud registration processing platform that integrates static noise features. The control program is executed on the processor and stored in the memory. The control program implements the tunnel point cloud registration processing method that integrates static noise features as described in any one of claims 1 to 6.

Images