Depth camera-based damage identification method, system and equipment, and medium

Abstract

The invention relates to the technical field of machine vision, in particular to a damage identification method, system and device based on a depth camera and a medium. The method comprises the following steps: acquiring three-dimensional point cloud data of the surface of a recognition object; processing the three-dimensional point cloud data to extract morphology features of the surface of the recognition object; comparing the morphological characteristics, and calculating geometric distortion information to identify an abnormal region; acquiring the position of the abnormal area, and calculating the abnormal level to give an abnormal alarm; the dimension limitation of two-dimensional vision is broken through through three-dimensional point cloud data acquisition, space parameters of surface deformation can be directly measured, and accurate quantitative evaluation of complex surface defects is realized through multi-dimensional fusion analysis of geometric distortion information in a data processing link; the technical problems of three-dimensional parameter deficiency and non-uniform evaluation standards in flexible structure damage detection are effectively solved, non-contact high-precision detection is realized, and the objectivity and accuracy of defect identification are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, and in particular to a method, system, device and medium for damage identification based on a depth camera. Background Technology

[0002] Large storage tanks, pipelines, composite material wings, wind turbine blades and other flexible structures require damage detection. During use, these structures are prone to surface defects such as breakage, bulges, cracks and wrinkles due to factors such as stress, corrosion, fatigue or external impact. If these defects are not detected and assessed in time, they may lead to a decline in structural performance or even cause catastrophic accidents.

[0003] Currently, the inspection of flexible structures mainly relies on manual visual inspection or contact measurement. Manual inspection is not only inefficient and labor-intensive, but the results are also greatly affected by subjective factors, making it difficult to quantify and assess the severity of defects. Furthermore, it poses safety hazards for inspections at high altitudes or in dangerous environments.

[0004] Traditional machine vision inspection methods based on two-dimensional images, while capable of automation, are sensitive to changes in lighting conditions and cannot directly acquire three-dimensional geometric information of the structural surface. They are also difficult to accurately measure key defect parameters such as the height of bulges and the depth of cracks, resulting in insufficient reliability in defect identification and rating.

[0005] Therefore, there is an urgent need for a non-contact, high-precision, automated, and intelligent damage identification technology to overcome the limitations of existing technologies and improve detection efficiency and reliability. Summary of the Invention

[0006] The main objective of this invention is to provide a damage identification method based on a depth camera, which aims to solve the technical problem of missing three-dimensional parameters in the detection of damage to objects.

[0007] To achieve the above objectives, the present invention provides a damage identification method based on a depth camera, the method comprising the following steps: Acquire 3D point cloud data of the surface of the object to be identified; The 3D point cloud data is processed to extract the morphological features of the object's surface. By comparing morphological features and calculating geometric distortion information, abnormal regions can be identified. The location of the abnormal area is obtained, the abnormality level is calculated, and an abnormality alarm is triggered.

[0008] Furthermore, the processing of the 3D point cloud data specifically includes the following steps: The acquired 3D point cloud data is denoised and simplified, and isolated noise points are removed. Calculate the normal vector and curvature of the local neighborhood of each point in the processed point cloud data; Using the angle between curvature and normal vector as constraints, the entire point cloud is divided into multiple smooth sub-regions, and potential anomalous regions whose size or curvature exceeds the normal range are marked. Perform Euclidean clustering analysis on all potential anomalous regions, merge spatially adjacent and connected regions, and filter artifacts to obtain point cloud clusters of anomalous regions.

[0009] Furthermore, the comparison of morphological features and the calculation of geometric distortion information specifically include the following steps: Obtain design drawings or measured data of the surface of the object under healthy conditions to construct a digital twin reference model containing geometric topological information and material physical properties, and store the digital twin reference model as a benchmark; Obtain point cloud clusters of abnormal regions, and register and align these clusters with the digital twin reference model to identify abnormal regions. Calculate the normal distance deviation, local curvature change rate, and strain energy density of each point on the 3D point cloud data relative to the digital twin reference model, and fuse the normal distance deviation, local curvature change rate, and strain energy density into a multidimensional deformation feature vector; The damage level of the corresponding abnormal area is determined by multidimensional deformation feature vector, and a quantitative abnormality level is obtained based on the maximum value of the normal distance deviation and the area of ​​the abnormal area, which is defined as the damage level index.

[0010] Furthermore, in the process of determining the damage level of the corresponding abnormal region through multidimensional deformation feature vectors, a radial basis function is used as the kernel function, and the penalty coefficient and kernel parameters are optimized through grid search and cross-validation methods to achieve the distinction between different damage levels.

[0011] Furthermore, the method also includes: Establish a damage evolution database to store the location of abnormal areas, damage levels, and damage grade indices determined by historical detections in a time-series manner; Regression analysis was performed on the damage level index of each abnormal area to predict its development trend in one or more future inspection cycles; Before the predicted level of damage reaches a preset maintenance threshold, a proactive maintenance warning is generated.

[0012] Furthermore, the method also includes: In the registered and aligned digital twin reference model, abnormal areas are rendered to visualize the degree of damage; Simultaneously, the evolution curve of the damage level index of the abnormal area over time is obtained, and the evolution curve is bound to the abnormal area so that the corresponding evolution curve is displayed when the abnormal area is clicked.

[0013] To achieve the above objectives, the present invention also provides a damage identification system based on a depth camera, the system comprising: A depth perception module, which is used to acquire three-dimensional point cloud data of the surface of the object to be identified; An image processing and recognition module is used to process three-dimensional point cloud data to extract the morphological features of the surface of the object to be recognized. A 3D modeling and deformation analysis module is used to compare morphological features and calculate geometric distortion information to identify abnormal areas. An alarm and management terminal is used to obtain the location of abnormal areas, calculate the abnormality level, and issue an alarm for abnormality.

[0014] Furthermore, the depth perception module includes at least two structured light cameras arranged at a preset angle to form a binocular stereo vision system. The binocular stereo vision system is signal-connected to a compensation module, which is used to calculate and compensate for thermal drift error by photographing a standard checkerboard calibration board. The 3D modeling and deformation analysis module includes a machine learning classifier.

[0015] To achieve the above objectives, the present invention also provides a computer device including a memory and a processor, wherein the memory stores a computer program and the processor executes the computer program.

[0016] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, wherein a processor executes the computer program.

[0017] The beneficial effects that this invention can achieve are as follows: This invention is successful. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0019] Figure 1 This is a flowchart illustrating the identification method in Embodiment 1 of the present invention; Figure 2 This is a flowchart illustrating the identification system in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the overall architecture of the identification system in Embodiment 2 of the present invention; Figure 4 This is a flowchart of the image processing and recognition module in Embodiment 3 of the present invention; Figure 5 This is a flowchart of the three-dimensional modeling and deformation analysis module in Embodiment 3 of the present invention.

[0020] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] 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. Furthermore, the meaning of "and / or" throughout the text includes three parallel solutions; for example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied. Additionally, 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.

[0023] Example 1: As attached Figure 1 As shown, this embodiment provides a damage identification method based on a depth camera, the method including the following steps: Acquire 3D point cloud data of the surface of the object to be identified; The 3D point cloud data is processed to extract the morphological features of the object's surface. By comparing morphological features and calculating geometric distortion information, abnormal regions can be identified. The location of the abnormal area is obtained, the abnormality level is calculated, and an abnormality alarm is triggered.

[0024] Understandably, in existing technologies, the detection of damage to flexible structures has long relied on manual visual inspection or contact measurement, which suffers from low efficiency, strong subjectivity, and safety hazards. While machine vision methods based on two-dimensional images can achieve automation, they are significantly affected by lighting conditions and cannot acquire three-dimensional geometric information of the surface, making it difficult to accurately measure key parameters such as crack depth and bulge height. For example, in the inspection of wind turbine blades, manual inspection in high-altitude environments carries high risks, and traditional vision systems cannot quantify the degree of surface deformation, affecting the reliability of defect assessment.

[0025] To address the aforementioned issues, the lack of 3D geometric information was first observed as a core factor restricting detection accuracy. This led to the consideration of how to acquire complete spatial data of the object's surface non-contactly. Analysis of the technical characteristics of depth cameras revealed their ability to directly output 3D point cloud data, laying the foundation for geometric feature extraction. Further research into 3D point cloud processing algorithms was conducted, proposing the transformation of raw point clouds into a quantifiable and analyzable set of morphological features, establishing a complete technical path from data acquisition to feature analysis. Finally, addressing the lack of defect assessment standards, a quantitative calculation model based on geometric distortion information was designed to form an objective anomaly level determination mechanism.

[0026] Therefore, this embodiment proposes a technical solution including the following steps: acquiring three-dimensional point cloud data of the surface of the object to be identified; processing the three-dimensional point cloud data to extract morphological features; comparing the morphological features and calculating geometric distortion information to identify abnormal areas; obtaining the location of the abnormal area and calculating the abnormal level to trigger an alarm.

[0027] Specifically, after acquiring 3D point cloud data of an object's surface using a depth camera, the point cloud is processed for noise reduction and feature extraction to generate a shape feature set containing normal vectors and curvature information. The actually acquired shape features are spatially registered with a preset digital reference model, and the normal distance deviation and curvature change rate at each point are calculated. Anomaly region boundaries are determined by setting a threshold, and a damage level index is constructed by combining the area of ​​the anomaly region and the maximum deviation value. When the index exceeds a preset threshold, an alarm mechanism is triggered. For example, in tank inspection, by comparing the current point cloud with a reference model in good condition, millimeter-level dents can be accurately identified and their impact range calculated.

[0028] By acquiring 3D point cloud data, the dimensional limitations of 2D vision are overcome, enabling direct measurement of the spatial parameters of surface deformation. Compared to manual inspection, the use of automated feature extraction and quantitative calculation models eliminates subjective judgment errors. Compared to traditional contact measurement, non-contact detection methods can adapt to the operational needs of hazardous environments such as high altitudes and corrosion. In the data processing stage, through multi-dimensional fusion analysis of geometric distortion information, accurate quantitative assessment of complex surface defects is achieved.

[0029] The above technical solutions effectively solve the technical problems of missing three-dimensional parameters and inconsistent evaluation standards in the detection of damage to flexible structures, realize non-contact high-precision detection, significantly improve the objectivity and accuracy of defect identification, adapt to the needs of automated detection in complex environments, and provide a traceable evaluation basis for maintenance decisions through a quantitative calculation model of geometric distortion information, avoiding the subjective bias of human judgment.

[0030] In some embodiments, three-dimensional point cloud data refers to a set of discrete points containing three-dimensional coordinate information of an object's surface, collected by a depth sensing device. Specifically, it can be implemented using a structured light or time-of-flight depth camera, which helps to overcome the limitation of missing depth information in two-dimensional images.

[0031] In some embodiments, morphological feature extraction refers to converting the original three-dimensional data into feature parameters that characterize the surface geometry through point cloud processing algorithms. Specifically, this can be achieved by normal vector calculation and curvature analysis, which are used to establish a quantifiable and comparable geometric feature benchmark.

[0032] In some embodiments, anomaly level calculation refers to a quantitative evaluation model established based on the degree of geometric distortion and the area of ​​the abnormal region. Specifically, it can be implemented using a multi-dimensional feature vector fusion algorithm to generate an objective defect severity index.

[0033] In this embodiment, the processing of the 3D point cloud data specifically includes the following steps: The acquired 3D point cloud data is denoised and simplified, and isolated noise points are removed. Calculate the normal vector and curvature of the local neighborhood of each point in the processed point cloud data; Using the angle between curvature and normal vector as constraints, the entire point cloud is divided into multiple smooth sub-regions, and potential anomalous regions whose size or curvature exceeds the normal range are marked. Perform Euclidean clustering analysis on all potential anomalous regions, merge spatially adjacent and connected regions, and filter artifacts to obtain point cloud clusters of anomalous regions.

[0034] The 3D point cloud data is first processed using a statistical outlier removal algorithm. This algorithm calculates the average distance between each point and its k nearest neighbors and removes isolated points exceeding the standard deviation threshold. Subsequently, voxel mesh downsampling is used to reduce data density while preserving surface morphology. The processed point cloud then uses principal component analysis to calculate the local neighborhood covariance matrix of each point. The eigenvector corresponding to the smallest eigenvalue is taken as the normal vector direction, and the curvature value is obtained using the curvature calculation formula. The preferred region growing algorithm uses a normal vector angle not exceeding 5 degrees and a curvature change rate less than 0.1 mm. -1 As a growth condition, neighboring points that meet the condition are grouped into the same sub-region. After segmentation, areas exceeding 10 cm² are marked.2 Potentially abnormal regions with curvature values ​​exceeding a preset threshold are identified. Finally, Euclidean clustering is used to spatially aggregate these potential abnormal regions, with a clustering distance threshold of 5mm set to merge adjacent point cloud clusters while filtering out isolated artifact regions with fewer than 50 points.

[0035] In this embodiment, the comparison of morphological features and calculation of geometric distortion information specifically includes the following steps: Obtain design drawings or measured data of the surface of the object under healthy conditions to construct a digital twin reference model containing geometric topological information and material physical properties, and store the digital twin reference model as a benchmark; Obtain point cloud clusters of abnormal regions, and register and align these clusters with the digital twin reference model to identify abnormal regions. Calculate the normal distance deviation, local curvature change rate, and strain energy density of each point on the 3D point cloud data relative to the digital twin reference model, and fuse the normal distance deviation, local curvature change rate, and strain energy density into a multidimensional deformation feature vector; The damage level of the corresponding abnormal area is determined by multidimensional deformation feature vector, and a quantitative abnormality level is obtained based on the maximum value of the normal distance deviation and the area of ​​the abnormal area, which is defined as the damage level index.

[0036] In the implementation process, a high-precision digital twin model is first constructed based on the original data under healthy conditions. This model not only includes the geometric topology but also integrates physical properties such as the material's elastic modulus. When anomaly point clusters are detected, they are precisely aligned with the reference model using a registration algorithm to ensure spatial consistency in subsequent deformation parameter calculations. For each anomaly point, its offset in the normal direction, the deviation of local curvature relative to the reference value, and the strain energy density based on material properties are calculated simultaneously, forming a three-dimensional feature vector containing geometric distortion, surface curvature anomalies, and energy distribution. For example, a normal distance deviation exceeding a threshold indicates a bulge or depression, an abnormal rate of curvature change indicates a crack edge, and a sudden increase in strain energy density reflects a stress concentration area. Finally, the maximum normal deviation value and the area of ​​the anomaly region are weighted and calculated to generate a damage level index in the 0-1 range, achieving an objective quantitative assessment of the severity of defects.

[0037] In this embodiment, in determining the damage level of the corresponding abnormal region through multidimensional deformation feature vectors, a radial basis function is used as the kernel function, and the penalty coefficient and kernel parameters are optimized through grid search and cross-validation methods to distinguish different damage levels.

[0038] Understandably, when training a support vector machine (SVM) classification model, the multidimensional deformation feature vectors are first input into the model. Radial basis functions are then used to map the original feature space to a high-dimensional space, making data of different damage levels linearly separable in this high-dimensional space. Subsequently, within a preset penalty coefficient and kernel parameter range, all possible parameter combinations are generated using a grid search method. For each set of parameters, cross-validation is used to calculate the model's average classification accuracy. Finally, the parameter combination with the highest cross-validation accuracy is selected as the optimal parameter, and a classification model with the optimal decision boundary is trained. This model can accurately distinguish between different types of damage, such as cracks, bulges, and wrinkles, based on the multidimensional features fused from normal distance deviation, local curvature change rate, and strain energy density.

[0039] In this embodiment, the method further includes: Establish a damage evolution database to store the location of abnormal areas, damage levels, and damage grade indices determined by historical detections in a time-series manner; Regression analysis was performed on the damage level index of each abnormal area to predict its development trend in one or more future inspection cycles; Before the predicted level of damage reaches a preset maintenance threshold, a proactive maintenance warning is generated.

[0040] Understandably, by periodically collecting damage level indices from abnormal areas and storing them in a damage evolution database, a damage evolution trajectory containing a time dimension is formed. Regression analysis is performed on the historical data of each abnormal area, for example, using the least squares method to fit a time series curve, to predict the growth trend of the damage level index within future inspection cycles. When the predicted value exceeds a preset maintenance threshold, a maintenance early warning signal is triggered, for example, by pushing early warning information through a visual interface or generating a maintenance work order. In this process, the damage evolution database provides continuous time series data support for regression analysis, while the dynamic setting of the maintenance threshold avoids the problem of insufficient adaptability of fixed thresholds to complex damage scenarios.

[0041] In this embodiment, the method further includes: In the registered and aligned digital twin reference model, abnormal areas are rendered to visualize the degree of damage; Simultaneously, the evolution curve of the damage level index of the abnormal area over time is obtained, and the evolution curve is bound to the abnormal area so that the corresponding evolution curve is displayed when the abnormal area is clicked.

[0042] After the digital twin reference model is registered, the coordinates of abnormal areas in the 3D point cloud data are converted into corresponding positions in the virtual model. The abnormal areas are overlaid with a semi-transparent red overlay by a graphics rendering engine; the color depth of the overlay is positively correlated with the damage level index. Simultaneously, the damage level index generated for each detection is stored in a database with a timestamp, forming a time-series dataset. When a user clicks on an abnormal area in the 3D model, the event listening module extracts the area ID, queries the associated historical data, and calls the visualization library to generate a line chart, which is then displayed synchronously in a separate window or the model sidebar. The horizontal axis of this line chart represents the detection time point, the vertical axis represents the damage level index value, and the slope of the curve reflects the damage development trend.

[0043] Understandably, based on the above method, this embodiment effectively achieves the leap from two-dimensional qualitative to three-dimensional quantitative detection, with high detection accuracy and reliability. Traditional manual visual inspection or 2D vision methods cannot accurately obtain key geometric parameters such as the depth, height, and volume of defects. This invention directly acquires three-dimensional point clouds of the structural surface through a depth camera, which can quantitatively assess defects with millimeter-level or even sub-millimeter-level accuracy, providing a reliable data foundation for structural safety assessment.

[0044] This embodiment constructs a fully automated process from data acquisition to analysis and early warning, which significantly improves detection efficiency. This invention deeply integrates high-precision 3D perception, point cloud processing, 3D modeling, machine learning and other technologies to form a closed-loop automated system that can complete detection, analysis and report generation without manual intervention, greatly reducing labor intensity and improving detection frequency and response speed. It is especially suitable for structural monitoring in large-area or hazardous environments.

[0045] This embodiment introduces intelligent analysis based on digital twins and machine learning to achieve accurate classification and rating of defects. This invention is not satisfied with simply discovering defects, but rather uses precise comparison with a digital twin model of a "healthy" state and machine learning classifiers to analyze multidimensional deformation features. It can accurately distinguish different types of defects such as cracks, bulges, and dents, and give a quantitative severity level, making subsequent maintenance decisions more scientific and accurate.

[0046] This embodiment establishes a damage evolution prediction mechanism based on historical data, enabling predictive maintenance. By tracking the development history of defects and modeling the data, this system can predict the future trend of damage, thereby issuing early warnings before defects deteriorate to a dangerous state. This forward-looking maintenance strategy helps optimize maintenance resources, avoid unnecessary downtime, and ensure the safety of the structure throughout its entire life cycle.

[0047] Example 2: As attached Figure 2As shown, this embodiment provides a damage identification system based on a depth camera, the system comprising: A depth perception module, which is used to acquire three-dimensional point cloud data of the surface of the object to be identified; An image processing and recognition module is used to process three-dimensional point cloud data to extract the morphological features of the surface of the object to be recognized. A 3D modeling and deformation analysis module is used to compare morphological features and calculate geometric distortion information to identify abnormal areas. An alarm and management terminal is used to obtain the location of abnormal areas, calculate the abnormality level, and issue an alarm for abnormality.

[0048] The depth perception module includes at least two structured light cameras arranged at a preset angle to form a binocular stereo vision system. The binocular stereo vision system is connected to a compensation module, which is used to calculate and compensate for thermal drift error by photographing a standard checkerboard calibration board. The 3D modeling and deformation analysis module includes a machine learning classifier.

[0049] It should be noted that the core architecture of the system comprises four main signal connection modules: a depth sensing module, an image processing and recognition module, a 3D modeling and deformation analysis module, and an alarm and management terminal. Among these, the depth sensing module serves as the system's data input, and its core task is to accurately acquire the 3D geometric information of the surface of the flexible structure under test.

[0050] In this embodiment, the module preferably consists of at least two structured light cameras arranged at a preset angle to form a binocular stereo vision system. The structured light cameras project a grating of a specific pattern onto the object and calculate the depth based on the deformation of the grating. Using a dual-camera arrangement can expand the field of view of a single acquisition and reduce data blind spots caused by occlusion.

[0051] Example 3: In some embodiments, to ensure the highest accuracy, the system preferably automatically executes an online calibration program before each acquisition task. This program controls the camera to photograph a standard checkerboard calibration board, calculates the camera's internal and external parameters in real time, and can effectively compensate for thermal drift errors caused by changes in ambient temperature, ensuring that the final synthesized 3D point cloud data achieves sub-millimeter level global measurement accuracy.

[0052] Understandably, the image processing and recognition module is responsible for processing and refining the raw point cloud data collected by the depth perception module, extracting meaningful anomaly regions; such as Figure 3 As shown, its processing flow specifically includes: Point cloud preprocessing: The original point cloud data is large and contains noise. First, a voxel downsampling filter is applied to effectively reduce the data density while maintaining the point cloud features. Then, the statistical outlier removal (SOR) algorithm is combined to calculate the distance distribution from each point to its neighboring points and remove isolated noise points that do not conform to the distribution.

[0053] Geometric attribute calculation: In preparation for subsequent segmentation, the module calculates the normal vector and curvature of the local neighborhood for each point in the point cloud. The normal vector describes the orientation of the point, while the curvature describes the degree of curvature of the surface. These are the key geometric attributes that distinguish between flat and defective regions.

[0054] Region segmentation: A region growing segmentation algorithm is adopted. Starting from a seed point, the algorithm continuously merges points in the neighborhood with similar normal vector angles and curvature values ​​to form a smooth sub-region. By traversing all points, the entire point cloud can be segmented into multiple sub-regions, and potential abnormal regions with sizes or curvatures exceeding the normal range can be initially marked.

[0055] Anomaly Region Refinement: Euclidean clustering analysis is performed on potential anomaly regions to merge spatially close clusters of anomalous points into a single, complete anomaly region. Simultaneously, clusters with excessively small areas or point counts are filtered out as artifacts, ultimately resulting in an accurate output of the point cloud clusters representing the anomaly regions to be analyzed.

[0056] Example 4: The 3D modeling and deformation analysis module is the core analysis unit of the system, performing quantitative analysis and intelligent rating of refined abnormal areas. For example... Figure 4 With appendix Figure 5 As shown, the analysis process includes: Reference model construction: The system first needs a standard of "healthy" state. Based on the design drawings (such as CAD models) of the identified object (preferably a flexible structure) or the measured data in a completely new state, a high-precision digital twin reference model is constructed. This model not only contains geometric information, but also material physical properties, and is stored as a benchmark.

[0057] Data registration: The current point cloud data output by the image processing and recognition module is aligned with the digital twin reference model. In this embodiment, a non-rigid variant of the Iterative Closest Point (ICP) algorithm is used. This algorithm can handle the overall bending or stretching that may exist in flexible structures and achieve high-precision registration.

[0058] Deformation feature extraction: After registration, by comparing point by point, the normal distance deviation of each point on the current point cloud relative to the reference model (reflecting the height / depth of bulges or depressions), the rate of change of local curvature (reflecting sharp changes on the surface, such as cracks) and the strain energy density (a mechanical index that reflects the degree of local deformation) are calculated, and these three key indices are integrated into a multidimensional deformation feature vector.

[0059] Example 5: Intelligent recognition and rating: The multidimensional feature vector is input into a pre-trained machine learning classifier. In this embodiment, the classifier is preferably a Support Vector Machine (SVM) that uses a radial basis function (RBF) as its kernel function. Through learning from a large number of known samples, the SVM classifier can accurately determine the damage level of abnormal areas (e.g., classifying them as Type I cracks, Type II bulges, etc.) and output a damage level index, such as 1-5, based on the maximum value of the normal distance deviation and the area of ​​the abnormal area.

[0060] Example 6: The alarm and management terminal serves as the window for user interaction with the system. It displays analysis results in an intuitive graphical user interface (GUI). The terminal can render the identified abnormal areas on a three-dimensional visualization model of the flexible structure with highlighted colors (for example, using a gradient color spectrum from green to red to correspond to damage levels from low to high), thus achieving a "heat map" visualization of the severity of damage.

[0061] Simultaneously, users can click on any abnormal area, and the system will instantly and dynamically generate and display an evolution curve of the damage level index of that area over time since its initial discovery, helping users understand the development of the damage. When a new high-level defect is discovered, the system will trigger an audible and visual alarm or send a notification.

[0062] Example 7: This embodiment provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program.

[0063] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0064] Example 8: This embodiment provides a computer-readable storage medium on which a computer program is stored, and a processor executes the computer program.

[0065] This invention provides a processor for running a program, wherein the program executes the twin model construction method for tank decommissioning.

[0066] This invention provides an apparatus including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements at least the steps of the twin model construction method for tank decommissioning.

[0067] The device in this invention can be a server, PC, PAD, mobile phone, etc.

[0068] The present invention also provides a computer program product, which, when executed on a data processing device, is adapted to execute a program having at least the steps of the twin model construction method for tank decommissioning.

[0069] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0070] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0071] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0075] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0076] Computer-readable media include permanent and non-permanent, removable and non-removable media. Information storage can be achieved by any method or technology; the information can be computer-readable instructions, data structures, program modules or other data.

[0077] Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random-access memory (SRAM), dynamic random-access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media do not include temporary computer-readable media such as modulated data signals and carrier waves.

[0078] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A damage identification method based on a depth camera, characterized in that, The method includes the following steps: Acquire 3D point cloud data of the surface of the object to be identified; The 3D point cloud data is processed to extract the morphological features of the object's surface. By comparing morphological features and calculating geometric distortion information, abnormal regions can be identified. The location of the abnormal area is obtained, the abnormality level is calculated, and an abnormality alarm is triggered.

2. The damage identification method based on a depth camera as described in claim 1, characterized in that, The processing of the 3D point cloud data specifically includes the following steps: The acquired 3D point cloud data is denoised and simplified, and isolated noise points are removed. Calculate the normal vector and curvature of the local neighborhood of each point in the processed point cloud data; Using the angle between curvature and normal vector as constraints, the entire point cloud is divided into multiple smooth sub-regions, and potential anomalous regions whose size or curvature exceeds the normal range are marked. Perform Euclidean clustering analysis on all potential anomalous regions, merge spatially adjacent and connected regions, and filter artifacts to obtain point cloud clusters of anomalous regions.

3. The damage identification method based on a depth camera as described in claim 1, characterized in that, The comparison of morphological features and the calculation of geometric distortion information specifically include the following steps: Obtain design drawings or measured data of the surface of the object under healthy conditions to construct a digital twin reference model containing geometric topological information and material physical properties, and store the digital twin reference model as a benchmark; Obtain point cloud clusters of abnormal regions, and register and align these clusters with the digital twin reference model to identify abnormal regions. The steps for obtaining the location of the abnormal region and calculating the abnormality level include the following: Calculate the normal distance deviation, local curvature change rate, and strain energy density of each point on the 3D point cloud data relative to the digital twin reference model, and fuse the normal distance deviation, local curvature change rate, and strain energy density into a multidimensional deformation feature vector; The damage level of the corresponding abnormal area is determined by multidimensional deformation feature vector, and a quantitative abnormality level is obtained based on the maximum value of the normal distance deviation and the area of ​​the abnormal area, which is defined as the damage level index.

4. The damage identification method based on a depth camera as described in claim 3, characterized in that, In the process of determining the damage level of the corresponding abnormal region through multidimensional deformation feature vectors, radial basis functions are used as kernel functions, and grid search and cross-validation methods are used to optimize the penalty coefficient and kernel parameters in order to distinguish different damage levels.

5. The damage identification method based on a depth camera as described in claim 3, characterized in that, The method further includes: Establish a damage evolution database to store the location of abnormal areas, damage levels, and damage grade indices determined by historical detections in a time-series manner; Regression analysis was performed on the damage level index of each abnormal area to predict its development trend in one or more future inspection cycles; Before the predicted level of damage reaches a preset maintenance threshold, a proactive maintenance warning is generated.

6. The damage identification method based on a depth camera as described in claim 3, characterized in that, The method further includes: In the registered and aligned digital twin reference model, abnormal areas are rendered to visualize the degree of damage; Simultaneously, the evolution curve of the damage level index of the abnormal area over time is obtained, and the evolution curve is bound to the abnormal area so that the corresponding evolution curve is displayed when the abnormal area is clicked.

7. A damage identification system based on a depth camera, characterized in that, The system includes: A depth perception module, which is used to acquire three-dimensional point cloud data of the surface of the object to be identified; An image processing and recognition module is used to process three-dimensional point cloud data to extract the morphological features of the surface of the object to be recognized. A 3D modeling and deformation analysis module is used to compare morphological features and calculate geometric distortion information to identify abnormal areas. An alarm and management terminal is used to obtain the location of abnormal areas, calculate the abnormality level, and issue an alarm for abnormality.

8. The damage identification system based on a depth camera as described in claim 7, characterized in that, The depth perception module includes at least two structured light cameras arranged at a preset angle to form a binocular stereo vision system. The binocular stereo vision system is connected to a compensation module, which is used to calculate and compensate for thermal drift error by photographing a standard checkerboard calibration board. The 3D modeling and deformation analysis module includes a machine learning classifier.

9. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a damage identification method based on a depth camera as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and the processor executes the computer program to implement a damage identification method based on a depth camera as described in any one of claims 1-6.

Images