Laser scanning point cloud data processing method and device and electronic equipment

Abstract

The invention provides a laser scanning point cloud data processing method and device and electronic equipment, and relates to the field of data processing. The method comprises the following steps: acquiring original laser scanning point cloud data for a target entity; performing preliminary noise reduction on the original laser scanning point cloud data to obtain global point cloud data; performing multi-scale decomposition on the global point cloud data to obtain frequency domain enhanced point cloud data; performing non-rigid registration and modeling on the frequency domain enhanced point cloud data in combination with the reference model of the target entity to obtain a quantitative deformation field; and performing multi-modal verification and iteration on the quantitative deformation field to obtain a target optimization parameter so as to complete data processing of the original laser scanning point cloud data. By implementing the technical scheme provided by the invention, the accuracy of data processing is improved.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and specifically relates to a method, device, and electronic device for processing laser scanning point cloud data. Background Art

[0002] Laser scanning technology can quickly obtain three-dimensional point cloud data of the surface of an entity, so it is increasingly widely used in the fields of industrial inspection and structural health monitoring.

[0003] Currently, the laser scanning point cloud data obtained by using laser scanning technology needs to go through data processing processes such as noise reduction before it can be used. Traditional noise reduction methods, such as statistical outlier removal, rely on fixed thresholds and are difficult to distinguish different types of noise and micron-level real deformations, resulting in low accuracy of data processing.

[0004] Therefore, there is an urgent need for a method, device, and electronic device for processing laser scanning point cloud data. Summary of the Invention

[0005] This application provides a method, device, and electronic device for processing laser scanning point cloud data, which is convenient for improving the accuracy of data processing.

[0006] In the first aspect of this application, a method for processing laser scanning point cloud data is provided. The method includes: obtaining the original laser scanning point cloud data for a target entity; performing preliminary noise reduction on the original laser scanning point cloud data to obtain global point cloud data; performing multi-scale decomposition on the global point cloud data to obtain frequency-domain enhanced point cloud data; combining with the reference model of the target entity, performing non-rigid registration and modeling on the frequency-domain enhanced point cloud data to obtain a quantified deformation field; performing multi-modal verification and iteration on the quantified deformation field to obtain target optimization parameters, so as to complete the data processing of the original laser scanning point cloud data.

[0007] By adopting the above technical solutions, through preliminary noise reduction and frequency-domain enhancement processing, the signal-to-noise ratio and feature clarity of the point cloud data are effectively improved, providing a reliable data basis for subsequent analysis. The multi-scale decomposition technology is used to achieve cross-resolution feature extraction, which can not only retain the macroscopic structural features but also enhance the recognition accuracy of microscopic details. Combining with the non-rigid registration technology, it breaks through the limitations of traditional rigid registration, effectively processes non-linear changes such as entity deformation and displacement, and is applicable to dynamic scenarios such as organisms and flexible materials. Through deformation field modeling, the mathematical representation of structural changes is realized, and a quantifiable deformation evaluation index is established, providing a numerical basis for scenarios such as engineering monitoring and medical diagnosis. Introducing multi-source data cross-verification and parameter iterative optimization significantly improves the credibility and robustness of the results and avoids the limitations of single-modal analysis. Therefore, it is convenient for improving the accuracy of data processing.

[0008] Optionally, the obtaining of the original laser scanning point cloud data for the target entity specifically includes: receiving the multi-source point cloud data for the target entity sent by the laser scanner, where the multi-source point cloud data includes entity noise and environmental vibration noise; and fusing the multi-source point cloud data to obtain the original laser scanning point cloud data.

[0009] By adopting the above technical solution, the entity body noise and environmental vibration noise data are synchronously received, completely covering the physical interference sources of the target entity and the scene, avoiding the omission of key information caused by a single data source, and providing a comprehensive basis for subsequent noise reduction. Before data fusion, the entity noise and environmental noise are clearly distinguished, and a prior classification framework is established, significantly improving the efficiency of subsequent targeted noise reduction. Through the spatial alignment and temporal correction of multi-source data, the point position drift error caused by environmental vibration of the laser scanner is effectively compensated, enhancing the geometric consistency of the original data in complex scenes. The mapping relationship between the original multi-source data and the fusion parameters is retained, supporting the backtracking analysis of the deformation cause and meeting the verification requirements of high-precision monitoring scenarios. For scenarios such as industrial sites and geological exploration with mechanical vibration and airflow disturbance, the anti-interference ability and robustness of the data acquisition system are improved through the active fusion mechanism of environmental noise. This process can not only ensure the integrity of the original data but also provide a clear noise separation framework for subsequent noise reduction and deformation analysis, avoiding the risk of structural misjudgment caused by noise mixing in traditional methods.

[0010] Optionally, the preliminary noise reduction of the original laser scanning point cloud data to obtain the global point cloud data specifically includes: obtaining the IMU vibration data sent by the IMU located on the laser scanner; obtaining the preliminary noise reduction result through timestamp alignment and reverse compensation of the working noise of the laser scanner; adaptively adjusting the filtering radius according to the local curvature, and performing dynamic statistical filtering on the preliminary noise reduction result using the filtering radius to obtain the global point cloud data.

[0011] By adopting the above technical solution, through the hardware-level cooperation of the IMU vibration data and the laser scanning data, the source capture and compensation of physical vibration interference are realized, breaking through the accuracy bottleneck of traditional pure algorithm noise reduction, especially suitable for mobile laser scanning scenarios. The systematic influence of the scanner's own working noise on the point cloud coordinate system is directly eliminated by using the timestamp alignment and reverse compensation technology. Based on the local curvature-based dynamic adjustment technology of the filtering radius, a large radius filter is used to eliminate the granular noise in the flat area, and the radius is shrunk in the high-curvature feature area to protect the edge details, achieving an accurate balance between noise elimination and feature retention. First, the systematic hardware noise is eliminated, and then the environmental random noise is processed, avoiding the over-smoothing or under-noise reduction problems caused by noise mixing in traditional single-stage methods. The dynamic statistical filtering algorithm has strong adaptability to the change of point cloud density and can still maintain stable output in the scanning missing area, meeting the engineering requirements of large industrial component scanning.

[0012] Optionally, performing multi-scale decomposition on the global point cloud data to obtain frequency-domain enhanced point cloud data specifically includes: using three-dimensional wavelet transform to decompose the global point cloud data into high-frequency data, medium-frequency data, and low-frequency data; performing curvature mutation marking on the target entity through a curvature-strain model to obtain marking data; removing the high-frequency data matching according to the marking data to obtain detection data; generating the frequency-domain enhanced point cloud data based on the detection data, the medium-frequency data, and the low-frequency data.

[0013] By adopting the above technical solution, three-dimensional wavelet transform is used to realize spatial-frequency domain joint analysis, breaking through the limitation of traditional single spatial domain processing, accurately separating geometric features, structural details from noise / micro-deformation, and providing a data basis for targeted enhancement. The curvature mutation area is detected through a curvature-strain coupling model, and a strain energy threshold marking system is established to realize the automatic positioning of key vulnerable areas of the structure. Adaptive high-frequency filtering based on the marking data eliminates the vibration noise in non-critical areas while retaining the high-frequency details in the deformation-sensitive area. A fusion strategy is adopted to correct the overall deformation drift of the low-frequency layer, sharpen the engineering features such as assembly seams of the middle-frequency layer, and directionally restore the effective micro-deformation of the high-frequency layer in the frequency-domain reconstruction stage. A mathematical mapping relationship between the number of wavelet decomposition layers and the curvature calculation radius is established to realize the scale linkage between micro-defect detection and macro-deformation analysis.

[0014] Optionally, combining the reference model of the target entity to perform non-rigid registration and modeling on the frequency-domain enhanced point cloud data to obtain a quantified deformation field, specifically including: determining data key points according to the frequency-domain enhanced point cloud data; determining key model areas according to the reference model; adding hard constraints to the key model areas and the data key points to obtain registered and aligned points; calculating the displacement vector of the registered and aligned points relative to the reference model and mapping the displacement vector into a strain tensor; inputting the strain tensor and the local point cloud block corresponding to the strain tensor into a U-Net network to obtain the quantified deformation field.

[0015] By adopting the above technical solutions, the key regions of the benchmark model are forced to align with the key points of the measured data through a hard constraint mechanism, breaking through the local optimal trap of the traditional ICP algorithm under complex deformations. The displacement vector is transformed into a strain tensor for expression, realizing the intelligent conversion of geometric displacement data into mechanical parameters, and directly outputting engineering key indicators such as the principal strain direction and shear strain distribution, providing physical field data directly applicable to finite element analysis for structural safety assessment. The U-Net network is used to fuse the multi-modal features of the strain tensor and the local point cloud blocks, taking into account the strong engineering requirements and the rationality of natural deformations such as biological soft tissues. During the selection stage of the key regions, the tolerance zone data of the CAD benchmark model is integrated, and quality-sensitive regions such as high-precision machined surfaces and sealing mating surfaces are preferentially aligned, making the registration result more in line with the actual engineering acceptance standards.

[0016] Optionally, the multi-modal verification and iteration of the quantified deformation field to obtain the target optimization parameters specifically include: obtaining the DIC optical measurement results for the target entity; comparing the quantified deformation field with the DIC optical measurement results to obtain a first comparison result; based on the quantified deformation field, comparing the magnitude relationship between the deformation rate and a preset threshold in continuous scans to obtain a second comparison result; screening out the parameters that meet the optimization conditions from the first comparison result and the second comparison result to obtain the target optimization parameters.

[0017] By adopting the above technical solutions, through the cross-verification of the DIC optical measurement results and the laser point cloud deformation field, an optical-mechanical measurement data fusion evaluation system is established to solve the pain point of insufficient confidence in a single-sensor system, and double-guarantee the monitoring reliability of the system in dynamic scenarios such as bridge load tests and mechanical fatigue tests. A dual-objective optimization function is constructed, and through the Pareto front solution screening algorithm, the optimal parameter combination that takes into account both static accuracy and dynamic stability is automatically output. The verification results are injected back into the registration model parameter library to realize the adaptive iterative optimization of core parameters such as the weights of the U-Net network and the calculation coefficients of the strain tensor, enabling the system to maintain the error convergence characteristic during continuous monitoring.

[0018] Optionally, the method further includes: generating a data processing report according to the target optimization parameters; and displaying the data processing to the user through an AR device.

[0019] By adopting the above technical solution, this process combines the intelligent generation of interactive data processing reports with AR augmented reality display, constructing an all-chain advantage from data analysis to decision support: the structured report automatically generated based on the optimized parameters can accurately mark the over-limit deformation areas, intelligently match the industry safety thresholds and correlate with the historical data evolution curves, greatly improving the diagnosis efficiency and standardization; at the same time, with the help of AR devices, it realizes the three-dimensional deformation field perspective with millimeter-level spatial registration, gesture interactive cross-section analysis and remote expert annotation collaboration, increasing the complex structure damage assessment efficiency by more than 3 times, and supporting the real-time retrieval of material property databases for comparison and simulation verification of maintenance plans.

[0020] In the second aspect of the present application, a laser scanning point cloud data processing device is provided. The device includes an acquisition module and a processing module. Among them, the acquisition module is used to acquire the original laser scanning point cloud data for a target entity; the processing module is used to perform preliminary noise reduction on the original laser scanning point cloud data to obtain global point cloud data; the processing module is further used to perform multi-scale decomposition on the global point cloud data to obtain frequency-domain enhanced point cloud data; the processing module is further used to combine the reference model of the target entity to perform non-rigid registration and modeling on the frequency-domain enhanced point cloud data to obtain a quantified deformation field; the processing module is further used to perform multi-modal verification and iteration on the quantified deformation field to obtain target optimization parameters, so as to complete the data processing of the original laser scanning point cloud data.

[0021] In the third aspect of the present application, an electronic device is provided. The electronic device includes a processor, a memory, a user interface and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory, so that the electronic device executes the method as described above.

[0022] In the fourth aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores instructions, and when the instructions are executed, the method as described above is executed.

[0023] In summary, one or more technical solutions provided in the present application have at least the following technical effects or advantages: Through preliminary noise reduction and frequency-domain enhancement processing, the signal-to-noise ratio and feature clarity of point cloud data are effectively improved, providing a reliable data basis for subsequent analysis. The multi-scale decomposition technology is adopted to achieve cross-resolution feature extraction, which can not only retain the macroscopic structural features but also enhance the recognition accuracy of microscopic details. Combining with non-rigid registration technology, it breaks through the limitations of traditional rigid registration, effectively processes non-linear changes such as entity deformation and displacement, and is applicable to dynamic scenarios such as organisms and flexible materials. Through deformation field modeling, the mathematical representation of structural changes is realized, and a quantifiable deformation evaluation index is established, providing a numerical basis for scenarios such as engineering monitoring and medical diagnosis. Introducing multi-source data cross-validation and parameter iterative optimization significantly improves the credibility and robustness of the results, avoiding the limitations of single-modal analysis. Therefore, it is convenient to improve the accuracy of data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 FIG. is a schematic flow chart of a method for processing laser scanning point cloud data provided by an embodiment of the present application; Figure 2 FIG. is another schematic flow chart of a method for processing laser scanning point cloud data provided by an embodiment of the present application; Figure 3 FIG. is a schematic module diagram of a device for processing laser scanning point cloud data provided by an embodiment of the present application; Figure 4 FIG. is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

[0025] Description of the reference numerals: 31, acquisition module; 32, processing module; 41, processor; 42, communication bus; 43, user interface; 44, network interface; 45, memory. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0027] In the description of the embodiments of the present application, words such as "for example" or "for instance" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "for example" or "for instance" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of words such as "for example" or "for instance" is intended to present related concepts in a specific manner.

[0028] In the description of the embodiments of the present application, the term "a plurality of" means two or more. For example, a plurality of systems means two or more systems, and a plurality of screen terminals means two or more screen terminals. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly indicating the technical features indicated. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The terms "comprising", "including", "having" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0029] Laser scanning technology has become the core data acquisition means in the fields of industrial inspection and structural health monitoring due to its high-precision three-dimensional digital reconstruction ability.

[0030] However, the original point cloud data obtained based on this technology currently faces significant technical bottlenecks in the preprocessing stage: Traditional noise reduction algorithms (such as statistical outlier removal based on static thresholds) have three key defects. First, the rigid threshold setting cannot adapt to multi-source noise environments (such as the differential processing requirements of equipment vibration noise, environmental scattering noise, and micro-deformation of the entity surface). Second, there is a lack of a protection mechanism for sub-millimeter real structure responses, resulting in early damage signals such as micro-cracks being misjudged as noise and filtered out. Third, a single spatial domain processing method is difficult to decouple the frequency domain characteristics of high-frequency noise and low-frequency deformation. Therefore, the accuracy of data processing is relatively low.

[0031] To solve the above technical problems, the present application provides a method for processing laser scanning point cloud data. Refer to Figure 1 , Figure 1 which is a schematic flowchart of a method for processing laser scanning point cloud data provided by an embodiment of the present application. This method is applied to a server and includes steps S110 to S150. The above steps are as follows: S110. Obtain the original laser scanning point cloud data for the target entity.

[0032] Specifically, the server controls a lidar or structured light scanning device to emit a laser beam towards a target entity (such as an industrial part or a building structure), records the reflection time or phase difference, calculates the three-dimensional coordinates (X, Y, Z) of each point, and may additionally include reflection intensity or color information. The raw data is usually stored in formats such as.las,.ply, or.xyz, containing millions to billions of unordered points, and may include device noise, environmental interference points, and the real data of the target surface. The transmission methods include real-time transmission. For example, 5G / Wi-Fi 6 enables real-time synchronization between the scanning end and the server (such as in the scenario of construction quality monitoring). It also includes batch transmission. For example, after scanning is completed, large files are uploaded through the SFTP / SCP protocol (such as in the digitalization project of large cultural relics). After receiving the data, the server will verify the data integrity (such as by comparing MD5 hash values). The server then parses the point cloud header file to confirm the coordinate system (such as WGS-84 or a local engineering coordinate system) and the unit (millimeters / meters). Distributed storage (such as HDFS) processes terabyte-level data and supports multi-user concurrent access. Database indexing (such as MongoDB spatial indexing) accelerates subsequent queries.

[0033] For example, when the server is for the damage detection of an aeroengine blade, the target entity is a turbine blade (made of superalloy, with a complex curved surface, and microcracks and deformations need to be detected). A handheld laser scanner (such as Creaform HandySCAN 3D) is used, with an accuracy of ±0.025 mm and a scanning rate of 1,300,000 points per second. The blade is fixed to the detection tooling, and multiple-angle scanning is performed to obtain a complete point cloud, including information such as the cooling holes on the blade surface and edge wear. The scanner transmits the data to the cloud server in real time via Wi-Fi6. The data volume of a single scan is approximately 2 GB, and the transmission delay is <50 ms. The data is stored in an AWS S3 bucket, triggering a Lambda function for automatic preprocessing: converting the point cloud format to.ply and attaching material properties, and generating metadata (such as scanning time, operator ID, device model).

[0034] In a possible implementation manner, obtaining the original laser scanning point cloud data for a target entity specifically includes: receiving the multi-source point cloud data sent by a laser scanner for the target entity, where the multi-source point cloud data includes entity noise and environmental vibration noise; and fusing the multi-source point cloud data to obtain the original laser scanning point cloud data.

[0035] Specifically, the multi-source point cloud data is multiple sets of point cloud data sets about the same target entity obtained by multiple sensors (such as multiple laser scanners) or the same sensor under different spatio-temporal conditions. The entity noise is data distortion caused by the target surface characteristics (such as reflectivity, light absorption, roughness) or material defects (such as rust, oil stain). The environmental vibration noise is the point position drift caused by the jitter of the scanner itself, external mechanical vibration (such as the operation of factory equipment), or air flow disturbance. The entity noise includes local point cloud missing (such as the laser not returning due to specular reflection) and abnormal point clusters (such as pseudo points caused by oil stains). The vibration noise includes the overall distortion of the point cloud (such as the scanning trajectory being wavy) and random outlier points (such as the coordinate mutation during vibration). The server integrates the multi-source data into the original point cloud data with high integrity and high precision through spatio-temporal alignment, noise separation, and feature enhancement.

[0036] For example, taking the industrial part inspection as an example, the purpose is to detect the surface defects (scratches, burrs) of a stainless steel gear processed by a CNC machine tool. Three line laser scanners (A / B / C) are used to synchronously scan from different angles. A vibration sensor (IMU) is installed on the scanner bracket to record the environmental vibration data. Scanner A: Due to the air flow disturbance of the workshop fan, the edge of the point cloud is serrated (environmental vibration noise). Scanner B: The oil film on the gear surface causes local reflection, forming a hole in the point cloud (entity noise). Scanner C: Captures the real surface features (such as a 0.2 mm scratch at the tooth root). Align the timestamps of the three scanners through the NTP protocol (error < 1 ms). Based on the feature points of the calibration board (checkerboard corner points), convert the data of A / B / C to the same coordinate system. Use the ICP algorithm to optimize the registration, and control the residual within ±0.05 mm. Extract the vibration spectrum of the IMU (main frequency 25 Hz, corresponding to the workshop fan disturbance). Apply the Kalman filter to the data of Scanner A to suppress the 25 Hz vibration component. Adopt local curvature analysis for the data of Scanner B: The curvature mutation value in the oil film reflection area is less than 0.01 mm, which is marked as an invalid area. Use the normal data of Scanner C to fill the reflection hole area of Scanner B. Perform weighted fusion on the filtered data of Scanner A and the data of Scanner C (the weight of the tooth root scratch area is increased to 70%). The fused point cloud data completely retains the 0.2 mm scratch, and at the same time eliminates the air flow vibration noise and the reflection hole.

[0037] S120. Perform preliminary noise reduction on the original laser scan point cloud data to obtain the global point cloud data.

[0038] Specifically, the server processes the original scan data through algorithms, removes noise and retains valid information, and generates a complete and high-precision three-dimensional point cloud data set. Its core goal is to improve the data quality and provide a reliable basis for subsequent modeling and analysis.

[0039] In a possible implementation, the original laser scanning point cloud data is preliminarily denoised to obtain global point cloud data, which specifically includes: acquiring IMU vibration data sent by an IMU located on the laser scanner; obtaining a preliminary denoising result through timestamp alignment and reverse compensation of the working noise of the laser scanner; adaptively adjusting the filtering radius according to local curvature, and performing dynamic statistical filtering on the preliminary denoising result using the filtering radius to obtain global point cloud data.

[0040] Specifically, through the inertial measurement unit built into the laser scanner, the acceleration (X / Y / Z axes) and angular velocity (rotation around the three axes) data of the device are collected in real time to quantify the attitude changes caused by the vibration of the scanner itself (such as high-frequency jitter of the motor, bumps during mobile scanning). The time series data (sampling rate ≥ 100Hz) includes acceleration and angular velocity. For example, when handheld scanning the skin seam of an aircraft, the IMU detects the periodic vibration (main frequency 8Hz, amplitude ±0.3mm) of the scanner caused by the hand jitter of the operator. This data will be aligned with the timestamp of the laser point cloud for subsequent compensation. Next, the server establishes a time mapping relationship between the IMU data and the laser point cloud, and the accuracy needs to reach the millisecond level (such as through the PTP precision clock protocol). Ensure that the spatial coordinates of each laser point can be associated with the IMU vibration state at the corresponding moment. Among them, the server converts the IMU data into a 6-degree-of-freedom pose change (translation + rotation). And apply the inverse transformation matrix to each laser point to eliminate the coordinate offset caused by the vibration of the scanner.

[0041] Finally, the server calculates the root mean square curvature of the surface fitting within the k-neighborhood (such as 50 neighboring points) for each point. High-curvature regions: feature regions such as edges and holes. Low-curvature regions: planes and gentle slope regions. For each point, count the number of points within its neighborhood with a radius of r. If the number of points is lower than the threshold (such as the threshold = 3 when r = 1mm, and the threshold = 10 when r = 5mm), it is determined as a noise point and deleted.

[0042] Therefore, it has very positive application prospects in some industries. For example, precision manufacturing: protecting the edges of turbine blade cooling holes (φ0.2mm) to avoid aperture measurement deviation caused by traditional filtering. Geological exploration: eliminating bump vibration noise during vehicle-mounted mobile scanning and accurately reconstructing the fracture network of rock formations. Medical orthopedics: retaining the microporous structure (50 - 100μm) on the surface of implants to improve the accuracy of bone integration effect analysis. Through the synergistic effect of IMU vibration compensation and curvature adaptive filtering, this technology realizes the balance between "strong denoising" and "high fidelity" in dynamic scanning scenarios, providing a millimeter-level reliable data basis for industrial inspection.

[0043] S130. Perform multi-scale decomposition on the global point cloud data to obtain frequency-domain enhanced point cloud data.

[0044] Specifically, the server decomposes the point cloud into different spatial frequency components through an algorithm, enhances key features specifically, and finally synthesizes a high-fidelity three-dimensional model. Its core goal is to separate noise, strengthen microstructures, and optimize the overall shape.

[0045] For example, in the detection of micro-cracks on the blades of aero-engines, there are problems with the original data: fatigue cracks of 0.1 - 0.3 mm exist on the blade surface, but are masked by machining textures (Ra = 0.8 μm) and vibration noise. There is uneven point cloud density: 200 points / mm² in the tip region and 50 points / mm² in the root region. After applying this solution, first, multi-scale decomposition is carried out: Low-frequency layer (scale 5 mm): Characterizes the overall curvature and twist angle of the blade. Intermediate-frequency layer (scale 1 mm): Extracts the edges of cooling holes and surface machining textures. High-frequency layer (scale 0.2 mm): Contains micro-crack signals and equipment noise. Frequency-domain enhancement: High-frequency layer: Set the curvature threshold at 0.6 mm -1 , enhancing the curvature mutation region (the curvature at the crack reaches 1.2 mm -1 ). Intermediate-frequency layer: Bilateral filtering retains the sharp edges of cooling holes (tolerance ±0.05 mm). Low-frequency layer: Registered with the CAD model to compensate for thermal deformation errors (maximum correction amount 0.3 mm). Reconstruction result: The signal-to-noise ratio (SNR) of micro-cracks is increased from 1.2 dB to 8.5 dB, and the crack direction can be clearly identified.

[0046] Therefore, through multi-scale frequency-domain enhancement technology, industrial inspection is achieved. For micro-defects: The detection rate breaks through to the sub-micron level (0.1 μm), supporting zero-defect manufacturing. For macro-deformations: The reconstruction accuracy of the full-size deformation field reaches ±0.01 mm, ensuring the safety of large structures. Cross-scale correlation: Establish a mathematical model for the propagation of micro-cracks and the overall stress distribution to predict the remaining life of components.

[0047] In a possible implementation, multi-scale decomposition is performed on the global point cloud data to obtain frequency-domain enhanced point cloud data, specifically including: Using three-dimensional wavelet transform to decompose the global point cloud data into high-frequency data, intermediate-frequency data, and low-frequency data; Marking the curvature mutations of the target entity through a curvature-strain model to obtain marked data; Removing the high-frequency data matching according to the marked data to obtain detection data; Generating frequency-domain enhanced point cloud data based on the detection data, intermediate-frequency data, and low-frequency data.

[0048] Specifically, step 1: three-dimensional wavelet transform decomposition, function: to layer the point cloud according to the degree of detail, and to separate noise from key features. High-frequency data: to capture tiny details (such as scratches, burrs, noise), corresponding to areas with drastic changes in the point cloud (such as edges, uneven areas). Medium-frequency data: to retain medium-scale features (such as holes, seams, surface textures), corresponding to structural transition areas. Low-frequency data: to reflect the overall shape (such as surface curvature, large deformation), corresponding to a smooth geometric contour. Example (mechanical parts inspection): suppose a car engine cylinder head is scanned: High-frequency layer: contains casting burrs and machining tool marks of 0.1mm level, but also mixed with equipment vibration noise. Medium-frequency layer: displays the edge of the spark plug mounting hole (diameter 10mm) and the cooling water channel interface. Low-frequency layer: presents the overall flatness of the cylinder head (such as a flatness error of 0.5mm).

[0049] Step 2: Curvature-strain model marking, function: to identify vulnerable areas on the surface of the entity and distinguish between real defects and noise. Specific process: Curvature calculation: Analyze the degree of curvature of each point, and high curvature areas (such as edges and cracks) will be marked. Strain analysis: Combined with the deformation trend of the entity under force (such as the long-term deformation direction of the bridge), determine whether the sudden change in curvature is caused by real deformation. Marking rules: Areas with sudden changes in curvature and consistent strain directions are marked as "critical areas", and the rest are marked as "non-critical areas". Example (bridge crack detection): Assume that in the scanned point cloud of the bridge pier, the curvature of a certain area suddenly increases (possibly a crack), and the strain data shows that there is a tensile deformation trend in this area. The model marks this area as a "high-risk crack area", and another high curvature area (caused by surface decorative patterns) is marked as a "non-critical area" because there is no strain association.

[0050] Step 3: Matching and removal of high-frequency data, function: denoising in non-critical areas and retaining effective details in critical areas. Specific process: Non-critical area processing: For high-frequency data marked as "non-critical areas", use large-scale filtering to remove discrete noise (such as dust reflection points). Critical area protection: For high-frequency data marked as "critical areas", only remove abnormal points that do not match the curvature-strain model (such as outliers caused by equipment jitter) to retain the true defect characteristics. Example (aircraft turbine blade inspection): Non-critical area (blade plane): filter out 90% of high-frequency noise points (such as air scattering noise). Critical area (blade root crack): only delete a few incoherent abnormal points, and fully retain the 0.2mm level fatigue crack details.

[0051] Step 4: Generate frequency-domain enhanced point cloud data. Function: Integrate the advantages of each frequency band to generate high-fidelity and high-signal-to-noise ratio data. Specific process: Low-frequency correction: Calibrate the overall shape error (such as thermal deformation) using a reference model (such as a CAD drawing). Medium-frequency enhancement: Sharpen the structural edges (such as chamfers of bolt holes) to make features clearer. High-frequency fusion: Re-embed the high-frequency details of the key area after denoising to enhance the visibility of micro-defects. Example (weld inspection of high-speed rail tracks): Low-frequency layer: Correct the overall bending of the track caused by thermal expansion and contraction (correction amount: 2 mm). Medium-frequency layer: Enhance the fish-scale pattern features at the weld (width: 1 mm) for evaluating welding quality. High-frequency layer: Enhance the contrast of cracks at the 0.1 mm level to make them prominent in the point cloud. Result: Inspectors can directly locate micro-cracks in the enhanced data.

[0052] S140. Combine the reference model of the target entity to perform non-rigid registration and modeling on the frequency-domain enhanced point cloud data to obtain a quantified deformation field.

[0053] Specifically, the server realizes non-rigid registration by elastically aligning the reference model of the target entity (such as a high-precision CAD design model or a three-dimensional model in the initial non-destructive state) with the scanned point cloud that has been denoised and multi-scale enhanced: First, establish local correspondence relationships based on feature point matching (such as hole edges, extreme points of surface curvature), allowing the reference model to deform (such as stretching, bending, local expansion) to fit the actual scanned data, and finally generate a quantified deformation field - that is, record the displacement of each point relative to the reference model in vector form (such as moving 0.3 mm in the X / Y / Z direction), and calculate the corresponding strain distribution (such as the maximum principal strain in a certain area is 0.12%).

[0054] For example, in the inspection of aero-engine blades, the reference model is the original design model of the blade. The server discovers through non-rigid registration that the tip of the blade warps upward by 1.5 mm due to high-temperature creep, and there is a 0.08% tensile strain around the bolt holes at the root of the blade. Based on this, a three-dimensional strain cloud map is generated (the red high-strain area indicates the fatigue risk) to guide targeted maintenance or life prediction.

[0055] In a possible implementation manner, combining the reference model of the target entity to perform non-rigid registration and modeling on the frequency-domain enhanced point cloud data to obtain a quantified deformation field specifically includes: determining data key points according to the frequency-domain enhanced point cloud data; determining key regions of the model according to the reference model; adding hard constraints to the key regions of the model and the data key points to obtain registration alignment points; calculating the displacement vector of the registration alignment points relative to the reference model, and mapping the displacement vector to a strain tensor; inputting the strain tensor and the local point cloud block corresponding to the strain tensor into the U-Net network to obtain a quantified deformation field.

[0056] Specifically, the data key points are determined as extracting points with high geometric features (such as edges, holes, and corners) from the point cloud that has undergone multi-scale decomposition and noise removal. First, through curvature analysis, the Gaussian curvature of each point is calculated, and the regions with curvature mutations are selected (such as curvature > 0.5mm -1 ). And feature points are extracted in regions with obvious reflection or color changes (such as welding marks on metal parts). For example, in the point cloud of an aero-engine blade, the edges of the root bolt holes, the serrated cooling holes at the tip, and other high-curvature regions are extracted as key points.

[0057] Next, the server will determine the key regions of the model. In the reference model (such as the CAD design model), regions sensitive to precision are predefined (such as assembly interfaces, tolerance zones). High-precision regions are marked according to manufacturing requirements (such as the ±0.02mm tolerance zone of the sealing surface). Stress-sensitive regions are marked with the regions prone to deformation (such as thin-walled structures) through finite element analysis (FEA). For example, in the CAD model of a turbine blade, the root bolt holes (assembly key) and the thin-walled region in the middle of the blade body (stress-sensitive) are predefined as key regions.

[0058] Among them, adding hard constraints to generate registration alignment means that it is a mandatory requirement for the key points and key regions to be strictly aligned after registration, and the error approaches zero. The server binds the center points of the bolt holes to the corresponding points in the CAD model, restricting the displacement difference < 0.01mm. And the overall displacement direction of the thin-walled region is made consistent with the FEA prediction. For example, the actual scanned points of the root bolt holes are forced to align with the hole centers in the CAD model to ensure the reliability of the assembly analysis. The displacement vector is the displacement (Δx, Δy, Δz) of the aligned point relative to the reference model. The strain tensor is the Lagrangian strain (Green-Lagrange Strain) calculated based on the displacement gradient, characterizing the degree of local deformation. The strain tensor describes the tensile / compressive and shear deformations of material points in different directions. For example, for a point on the blade body with displacements (Δx = +0.3mm, Δy = -0.1mm), the maximum principal strain of 0.12% (indicating tensile deformation) is calculated.

[0059] Secondly, the U-Net network generates a quantized deformation field, which specifically includes the input data strain tensor: the strain value of each point (6 independent components), the local point cloud block: the 3D neighborhood point cloud containing geometric features (curvature, normal vector), and the network structure: Encoder: Extract multi-scale features of strain and geometry (such as 3×3×3 convolution kernels). Decoder: Upsample to restore the spatial resolution and output the exact strain value of each point. Example: Input the strain tensor and local point cloud of the root region, and the U-Net predicts the hidden microcracks (strain anomaly regions) around the bolt holes.

[0060] For example, key points are matched with key regions: extract the cooling hole edges (data key points) in the actual point cloud of the blade. Mark the designed positions of the cooling holes (model key regions) in the CAD model. Add hard constraints: the registration error of the cooling hole centers is forced to be ≤0.005 mm. Non-rigid registration and strain calculation: the overall displacement (+1.2 mm) occurs in the tip region due to centrifugal force, and local expansion (+0.15 mm) occurs at the blade root due to thermal expansion. The strain tensor shows that the maximum principal strain in the middle of the blade body is 0.1% (close to 30% of the material yield limit). Generation of the U-Net deformation field: Input: blade body strain data + local point cloud geometry. Output: a quantified deformation field shows the existence of hidden microcracks in the back region of the blade (the strain gradient mutation is 0.05% / mm). Result: internal damage at the 0.1 mm level missed by traditional methods is detected, triggering an early replacement warning.

[0061] Therefore, by using the hard constraint mechanism to force the alignment of the key regions of the reference model and the key points of the measured data, the local optimal trap of the traditional ICP algorithm under complex deformations is broken through. The displacement vector is transformed into a strain tensor for expression, realizing the intelligent conversion of geometric displacement data into mechanical parameters, and directly outputting engineering key indicators such as the principal strain direction and shear strain distribution, providing physical field data that can be directly used for finite element analysis for structural safety assessment. The U-Net network is used to fuse the multi-modal features of the strain tensor and the local point cloud block, taking into account the strong engineering requirements and the rationality of natural deformations such as biological soft tissues. The tolerance zone data of the CAD reference model is integrated in the selection stage of the key region, and high-precision machining surfaces, sealing mating surfaces and other quality-sensitive regions are preferentially aligned, making the registration result more in line with the actual engineering acceptance standards.

[0062] S150. Perform multi-modal verification and iteration on the quantified deformation field to obtain target optimization parameters to complete the data processing of the original laser scanning point cloud data.

[0063] Specifically, this process realizes the ultimate improvement of the accuracy of laser scanning data through multi-source data cross-validation and intelligent optimization. First, the server spatially aligns and numerically compares the quantified deformation field with the results of digital image correlation optical measurement (such as the displacement deviation of key bridge nodes ≤ 0.05 mm) to verify the physical consistency of the deformation field; synchronously accesses the time-series data of vibration sensors, analyzes the coincidence degree between the deformation rate curve and the theoretical model (such as the material creep constitutive equation), and screens out the periods of over-threshold fluctuations (such as the sudden increase in the strain rate by 300% in a certain period); then constructs a two-objective optimization function, which not only minimizes the DIC comparison error (static accuracy constraint) but also satisfies the dynamic deformation rate stability (dynamic process constraint), and searches for the optimal solution in the parameter space of hundreds of billions through the genetic algorithm (such as the number of U-Net network layers, the size of the convolutional kernel, the strain tensor weighting coefficient, etc.); after each iteration, the system automatically generates a new version of the deformation field and triggers secondary verification until the DIC deviation converges to ±0.01 mm and the strain rate fluctuation ≤ 5%; the finally output target optimization parameters will be injected into the preprocessing pipeline to guide the adaptive adjustment of the noise reduction intensity of subsequent scanning tasks (such as enhancing the IMU compensation weight when the vibration is severe), and the dynamic optimization of the multi-scale decomposition threshold (such as retaining more high-frequency details in high-curvature regions).

[0064] For example, in the deformation monitoring of the containment vessel of a nuclear power plant, the initial deformation field shows an abnormal displacement of 0.3 mm in a certain area, but no corresponding abnormality is found in the DIC optical measurement. After multi-modal verification, it is found that it is a server error caused by the condensation of the laser scanner lens; through 3 rounds of iterative optimization, the server autonomously adjusts the point cloud registration parameters and the frequency domain enhancement coefficient, and finally eliminates the pseudo-deformation signal, accurately captures the actually existing 0.12 mm material expansion area, and synchronously optimizes the parameter library to record the environmental characteristics (start the anti-condensation filtering mode when the humidity > 85%), providing a millimeter-level reliable data basis for structural health assessment.

[0065] In a possible implementation manner, multi-modal verification and iteration are performed on the quantified deformation field to obtain target optimization parameters, specifically including: obtaining the DIC optical measurement results for the target entity; comparing the quantified deformation field with the DIC optical measurement results to obtain a first comparison result; based on the quantified deformation field, comparing the magnitude relationship between the deformation rate in continuous scans and a preset threshold to obtain a second comparison result; screening out the parameters that meet the optimization conditions from the first comparison result and the second comparison result to obtain the target optimization parameters.

[0066] Specifically, this process realizes the parameter self-calibration and accuracy improvement of the laser scanning deformation detection system through multi-modal data cross-validation and intelligent iterative optimization: First, the server obtains the high-precision optical measurement results of the target entity from the Digital Image Correlation (DIC) system (such as the full-field displacement data of the key nodes of the bridge, with an accuracy of ±0.02 mm), and spatially aligns and point-by-point compares it with the quantified deformation field generated by laser scanning to generate the first comparison result - that is, to calculate the root mean square error of the static displacement of the two (such as the laser data displacement in a certain area is 1.5 mm, the DIC measures 1.48 mm, and the error is 0.02 mm), the local maximum deviation (such as the error at the bolt connection suddenly increases to 0.12 mm, exceeding the threshold of 0.05 mm), and the spatial consistency distribution map (marking the error concentration area).

[0067] Meanwhile, the server analyzes the deformation rate curve within consecutive scanning periods (such as collecting point clouds once an hour and calculating the displacement increment of key areas), and dynamically compares it with the preset safety threshold (such as the deformation rate threshold for concrete structures is set to 0.01 mm / day) to generate the second comparison result - including the over-limit period of the deformation rate (such as the rate soars to 0.03 mm / day due to a sudden temperature rise in a certain period), the fluctuation amplitude (whether the standard deviation exceeds 5%), and the trend prediction (judging whether the deformation will continue to deteriorate through time series analysis); Subsequently, the server constructs a multi-objective optimization model, with minimizing the static error in the first comparison result as the core constraint (such as requiring the DIC comparison error in 95% of the area ≤ 0.03 mm), and the dynamic stability in the second comparison result as the auxiliary constraint (such as the deformation rate fluctuation amplitude ≤ 8%), and uses the genetic algorithm to search for the optimal solution in the parameter space - these parameters include the filtering radius for point cloud noise reduction (such as adjusting the radius from 5 mm to 3 mm in areas with large vibration noise to retain details), the hard constraint weight for non-rigid registration (such as increasing the key point matching weight in the bolt hole area to 90% to enforce alignment), the number of neighborhood points for strain tensor calculation (increasing from 50 points to 80 points to improve the local strain smoothness), and the curvature threshold for frequency domain enhancement (from 0.5 mm -1 adjusted to 0.3 mm -1 to capture more subtle deformation features).

[0068] After each iteration, the server automatically generates a new version of the deformation field and triggers a new round of verification until the static error converges to the target interval and the dynamic fluctuation stabilizes within the safety threshold. Finally, a set of fixed parameters (such as a filtering radius of 3.2 mm, a registration weight of 88%, and 72 neighborhood points) are output. These parameters will be embedded in the data processing pipeline to guide the real-time adaptive adjustment of subsequent scanning tasks. For example, in the deformation monitoring of the containment vessel of a nuclear power plant, the deformation field under the initial parameters shows an abnormal displacement of 0.25 mm in a certain area, but the DIC optical measurement result only shows a displacement of 0.08 mm. The first comparison result marks this difference as a suspected false signal; further analysis of the continuous scanning data reveals that the deformation rate in this area is as high as 0.015 mm / h at night, but drops to 0.002 mm / h during the day. The second comparison result reveals that it is highly correlated with temperature changes (cooling and shrinking at night, heating and expanding during the day), but the rate fluctuation amplitude (650%) far exceeds the theoretical value of the material's thermal expansion coefficient (expected fluctuation ≤ 20%), indicating that the scanning server is sensitive to environmental interference; through 7 rounds of iterative optimization, the server gradually adjusts the temperature compensation coefficient in the noise reduction algorithm (from 0.5 to 1.2), the thermal expansion correction weight in the registration model (from being ignored to 30% weight), and dynamically expands the neighborhood range for strain calculation (from a local 50 mm area to 100 mm to smooth the influence of the temperature gradient). Finally, the DIC comparison error in this area is reduced to 0.03 mm, and the deformation rate fluctuation stabilizes at 18% (close to the theoretical value of 15%), while accurately capturing another area with a real 0.12 mm material creep (the DIC verification error in this area is 0.01 mm, and the deformation rate stabilizes at 0.008 mm / day).

[0069] In a possible implementation manner, referring to Figure 2 , Figure 2 is another flowchart of a method for processing laser scanning point cloud data provided by an embodiment of the present application, including steps S210 to S220. The above steps are as follows: S210, generate a data processing report according to the target optimization parameters; S220, display the data processing to the user through an AR device.

[0070] Specifically, this process realizes the efficient delivery and immersive analysis of data processing results through intelligent report generation and augmented reality visualization technology. The server automatically generates a structured detection report based on the optimized parameter set (such as a noise reduction filter radius of 3.2 mm, a registration weight of 88%, etc.). The report content includes core deformation indicators (such as a maximum displacement value of 1.2 mm, a strain in the dangerous area of 0.15%), a historical data comparison curve (showing the evolution of the deformation rate within three months), an over-limit warning list (such as marking that the displacement of bolt hole A3 exceeds the threshold by 0.05 mm), and the priority of maintenance suggestions (calculated based on the deformation field gradient and the material fatigue model, recommending to prioritize the treatment of high-risk areas). At the same time, through natural language generation technology, technical parameters are converted into engineering interpretations (such as "The strain in the blade root area is 0.12%, which has reached 85% of the design life. It is recommended to replace it within 6 months").

[0071] Subsequently, the server pushes the deformation field data and the 3D model to the AR device through the 5G network, and uses SLAM spatial positioning technology to accurately superimpose the virtual data on the entity (registration error < 0.5 mm). For example, when an engineer wears a HoloLens to scan an aircraft engine, a color strain heat map can be perspectively displayed on the real blade surface (red indicates the strain area above 0.1%). Gesture swiping can be used to call up the historical deformation animation (showing the tip warping trend in the past six months). The voice command "Show the data of the blade root bolt hole" immediately pops up the displacement vector arrow in this area (the length ratio is 1:1000, and the arrow direction indicates the deformation trend), and it supports remote experts to add maintenance guidelines in the AR field of view through the virtual annotation function (such as marking the high-risk crack area with a flashing red circle). For example, during the inspection of wind turbine blades, the detection report is automatically associated with the blade serial number to generate an electronic file, and the AR interface can display the three-dimensional positioning of micro-cracks with a level of 0.08 mm on the blade surface in real time (accurately positioned 2.3 cm from the edge of bolt hole No. 5). At the same time, the maintenance manual animation is superimposed to guide the disassembly steps. This data-report-AR closed-loop server is reshaping the industrial detection paradigm and realizing a minute-level response from data collection to decision execution.

[0072] This application also provides a laser scanning point cloud data processing device. Refer to Figure 3 , Figure 3It is a module schematic diagram of a laser scanning point cloud data processing device provided by an embodiment of the present application. The device is a server, and the server includes an acquisition module 31 and a processing module 32. The acquisition module 31 acquires the original laser scanning point cloud data for a target entity; the processing module 32 performs preliminary noise reduction on the original laser scanning point cloud data to obtain global point cloud data; the processing module 32 performs multi-scale decomposition on the global point cloud data to obtain frequency-domain enhanced point cloud data; the processing module 32 combines the reference model of the target entity to perform non-rigid registration and modeling on the frequency-domain enhanced point cloud data to obtain a quantified deformation field; the processing module 32 performs multi-modal verification and iteration on the quantified deformation field to obtain target optimization parameters, so as to complete the data processing of the original laser scanning point cloud data.

[0073] In a possible implementation manner, the acquisition module 31 acquires the original laser scanning point cloud data for a target entity, specifically including: the acquisition module 31 receives multi-source point cloud data for the target entity sent by a laser scanner, and the multi-source point cloud data includes entity noise and environmental vibration noise; the processing module 32 fuses the multi-source point cloud data to obtain the original laser scanning point cloud data.

[0074] In a possible implementation manner, the processing module 32 performs preliminary noise reduction on the original laser scanning point cloud data to obtain global point cloud data, specifically including: the acquisition module 31 acquires IMU vibration data sent by an IMU located on the laser scanner; the processing module 32 obtains a preliminary noise reduction result through timestamp alignment and reverse compensation of the working noise of the laser scanner; the processing module 32 adaptively adjusts the filtering radius according to local curvature, and performs dynamic statistical filtering on the preliminary noise reduction result using the filtering radius to obtain global point cloud data.

[0075] In a possible implementation manner, the processing module 32 performs multi-scale decomposition on the global point cloud data to obtain frequency-domain enhanced point cloud data, specifically including: the processing module 32 uses three-dimensional wavelet transform to decompose the global point cloud data into high-frequency data, medium-frequency data, and low-frequency data; the processing module 32 marks curvature mutations of the target entity through a curvature-strain model to obtain marked data; the processing module 32 matches and removes the high-frequency data according to the marked data to obtain detection data; the processing module 32 generates frequency-domain enhanced point cloud data based on the detection data, medium-frequency data, and low-frequency data.

[0076] In a possible implementation manner, the processing module 32 performs non-rigid registration and modeling on the frequency-domain enhanced point cloud data in combination with the reference model of the target entity to obtain a quantization deformation field, which specifically includes: the processing module 32 determines data key points according to the frequency-domain enhanced point cloud data; the processing module 32 determines key model regions according to the reference model; the processing module 32 adds hard constraints to the key model regions and the data key points to obtain registered and aligned points; the processing module 32 calculates the displacement vector of the registered and aligned points relative to the reference model and maps the displacement vector into a strain tensor; the processing module 32 inputs the strain tensor and the local point cloud block corresponding to the strain tensor into the U-Net network to obtain a quantization deformation field.

[0077] In a possible implementation manner, the processing module 32 performs multi-modal verification and iteration on the quantization deformation field to obtain target optimization parameters, which specifically includes: the processing module 32 obtains the DIC optical measurement results for the target entity; the processing module 32 compares the quantization deformation field with the DIC optical measurement results to obtain a first comparison result; the processing module 32 compares the magnitude relationship between the deformation rate and a preset threshold in continuous scans based on the quantization deformation field to obtain a second comparison result; the processing module 32 screens out the parameters that meet the optimization conditions from the first comparison result and the second comparison result to obtain target optimization parameters.

[0078] In a possible implementation manner, the processing module 32 generates a data processing report according to the target optimization parameters; the processing module 32 displays the data processing through an AR device to the user.

[0079] It should be noted that when the device provided in the above embodiments realizes its functions, only the above-mentioned division of each functional module is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0080] This application also provides an electronic device. Refer to Figure 4 , Figure 4 which is a schematic structural diagram of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 41, at least one network interface 44, a user interface 43, a memory 45, and at least one communication bus 42.

[0081] Among them, the communication bus 42 is used to realize the connection and communication between these components.

[0082] Among them, the user interface 43 may include a display screen and a camera. Optionally, the user interface 43 may also include standard wired interfaces and wireless interfaces.

[0083] Among them, the network interface 44 may optionally include standard wired interfaces and wireless interfaces (such as Wi-Fi interfaces).

[0084] Among them, the processor 41 may include one or more processing cores. The processor 41 connects various parts within the entire server through various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 45, and by calling data stored in the memory 45, it performs various functions of the server and processes data. Optionally, the processor 41 may be implemented in at least one of the hardware forms of digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor 41 may integrate one or a combination of several of the central processing unit (CPU), graphics processing unit (GPU), and modem. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 41 and may be implemented separately through a single chip.

[0085] Among them, the memory 45 may include random access memory (RAM) and may also include read-only memory. Optionally, the memory 45 includes a non-transitory computer-readable storage medium. The memory 45 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 45 may include a program storage area and a data storage area. Among them, the program storage area may store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned method embodiments, etc.; the data storage area may store the data involved in the above-mentioned method embodiments. Optionally, the memory 45 may also be at least one storage device located far from the aforementioned processor 41. For example Figure 4As shown, the memory 45, which is a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a method of processing laser scanning point cloud data.

[0086] In Figure 4 In the electronic device shown, the user interface 43 is mainly used to provide an interface for the user to input data and obtain the data input by the user; while the processor 41 can be used to call the application program for a method of processing laser scanning point cloud data stored in the memory 45. When executed by one or more processors, the electronic device performs the method as described in one or more of the above embodiments.

[0087] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0088] The present application also provides a computer-readable storage medium storing instructions. When executed by one or more processors, the electronic device performs the method as described in one or more of the above embodiments.

[0089] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0090] In several embodiments provided by the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some service interfaces. The indirect couplings or communication connections of the devices or units can be in electrical or other forms.

[0091] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0092] In addition, in each embodiment of the present application, each functional unit can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0093] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in each embodiment of the present application. The aforementioned memory includes various media that can store program codes, such as USB flash drives, mobile hard disks, magnetic disks, or optical discs.

[0094] The above are only exemplary embodiments of the present disclosure, and the scope of the present disclosure cannot be limited thereby. That is, all equivalent changes and modifications made in accordance with the teachings of the present disclosure still fall within the scope covered by the present disclosure. Those skilled in the art will readily think of other implementation manners of the present disclosure after considering the specification and the practice of the present disclosure. The present application aims to cover any variations, uses, or adaptive changes of the present disclosure, and these variations, uses, or adaptive changes follow the general principles of the present disclosure and include the common general knowledge or conventional technical means in the technical field not recorded in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A laser scanning point cloud data processing method, characterized in that: The method comprises: Obtaining original laser scanning point cloud data of the target entity; Performing preliminary noise reduction on the original laser scanning point cloud data to obtain global point cloud data; Performing multi-scale decomposition on the global point cloud data to obtain frequency domain enhanced point cloud data; In combination with the benchmark model of the target entity, non-rigid registration and modeling are performed on the frequency domain enhanced point cloud data to obtain a quantized deformation field; The quantized deformation field is subjected to multi-modal verification and iteration to obtain target optimization parameters to complete data processing of the original laser scanning point cloud data.

2. The laser scanning point cloud data processing method according to claim 1, characterized in that: The obtaining of original laser scanning point cloud data for the target entity specifically includes: Receiving multi-source point cloud data for the target entity sent by a laser scanner, wherein the multi-source point cloud data includes entity noise and environmental vibration noise; The multi-source point cloud data are fused to obtain the original laser scanning point cloud data.

3. The laser scanning point cloud data processing method according to claim 2, characterized in that: The performing preliminary noise reduction on the original laser scanning point cloud data to obtain global point cloud data specifically includes: Acquire IMU vibration data sent by an IMU located on the laser scanner; Obtaining a preliminary noise reduction result by aligning timestamps and inversely compensating for the working noise of the laser scanner; The filter radius is adaptively adjusted according to the local curvature, and the preliminary denoising result is dynamically statistically filtered using the filter radius to obtain the global point cloud data.

4. The laser scanning point cloud data processing method according to claim 1, characterized in that: The multi-scale decomposition of the global point cloud data to obtain frequency domain enhanced point cloud data specifically includes: Using three-dimensional wavelet transform, the global point cloud data is decomposed into high-frequency data, medium-frequency data and low-frequency data; Using a curvature-strain model, marking the curvature mutation of the target entity to obtain marking data; According to the marked data, the high-frequency data is matched and removed to obtain detection data; The frequency domain enhanced point cloud data is generated based on the detection data, the intermediate frequency data and the low frequency data.

5. The laser scanning point cloud data processing method according to claim 1, characterized in that: The non-rigid registration and modeling of the frequency domain enhanced point cloud data in combination with the benchmark model of the target entity to obtain a quantized deformation field specifically includes: Determining data key points according to the frequency domain enhanced point cloud data; According to the benchmark model, determining the key areas of the model; Adding hard constraints to the model key areas and the data key points to obtain registration alignment points; Calculating a displacement vector of the registration alignment point relative to the reference model, and mapping the displacement vector into a strain tensor; The strain tensor and the local point cloud block corresponding to the strain tensor are input into the U-Net network to obtain the quantized deformation field.

6. The laser scanning point cloud data processing method according to claim 1, characterized in that: The multi-modal verification and iteration of the quantized deformation field to obtain target optimization parameters specifically includes: Acquire a DIC optical measurement result for the target entity; Comparing the quantized deformation field with the DIC optical measurement result to obtain a first comparison result; Based on the quantized deformation field, comparing the magnitude relationship between the deformation rate in the continuous scanning and the preset threshold value to obtain a second comparison result; Parameters satisfying the optimization condition are screened out from the first comparison result and the second comparison result to obtain the target optimization parameters.

7. The laser scanning point cloud data processing method according to claim 1, characterized in that: The method further comprises: Generating a data processing report according to the target optimization parameters; Processing the data includes displaying it to a user through an AR device.

8. A laser scanning point cloud data processing device, characterized in that: The device comprises an acquisition module (31) and a processing module (32), wherein: The acquisition module (31) is used to acquire original laser scanning point cloud data for the target entity; The processing module (32) is used to perform preliminary noise reduction on the original laser scanning point cloud data to obtain global point cloud data; The processing module (32) is further used to perform multi-scale decomposition on the global point cloud data to obtain frequency domain enhanced point cloud data; The processing module (32) is further used to perform non-rigid registration and modeling on the frequency domain enhanced point cloud data in combination with the benchmark model of the target entity to obtain a quantized deformation field; The processing module (32) is also used to perform multi-modal verification and iteration on the quantified deformation field to obtain target optimization parameters to complete data processing of the original laser scanning point cloud data.

9. An electronic device, characterized in that: The electronic device comprises a processor (41), a memory (45), a user interface (43) and a network interface (44), wherein the memory (45) is used to store instructions, the user interface (43) and the network interface (44) are both used to communicate with other devices, and the processor (41) is used to execute the instructions stored in the memory (45) so that the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1 to 7 is performed.

Images