Concrete structure surface defect detection and positioning three-dimensional reconstruction method based on monocular vision

Through monocular vision technology combined with YOLO deep learning and three-dimensional Gaussian rendering, the problems of traditional low detection efficiency and difficulty in three-dimensional positioning are solved, and high-precision three-dimensional reconstruction and quantitative measurement of surface defects of concrete structures are realized, providing effective support for structural safety assessment. .

CN118918285APending Publication Date: 2024-11-08HEFEI UNIV
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202410924634.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-11
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Traditional concrete structure surface defect detection relies on manual visual inspection, is inefficient and difficult to accurately measure three-dimensional characteristics. The existing automation technology has limitations in three-dimensional positioning and quantification.

Method used

Using a monocular vision-based method, multi-angle images are collected through machine vision cameras, combined with YOLO deep learning network and three-dimensional Gaussian sputtering rendering technology, image preprocessing, defect recognition, marking and three-dimensional structure reconstruction are carried out to generate high-precision three-dimensional points Cloud model.

Benefits of technology

It realizes accurate quantification and three-dimensional reconstruction of concrete structure surface defects, provides effective technical support for structural safety assessment and maintenance, and can accurately measure defects, save costs, and is suitable for the inspection of concrete structures such as bridges and buildings.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918285A_ABST
    Figure CN118918285A_ABST
Patent Text Reader

Abstract

The invention is suitable for the technical field of automatic detection, and provides a concrete structure surface defect detection and positioning three-dimensional reconstruction method based on monocular vision, and the method comprises the following steps: collecting multi-angle images of a target concrete structure through a machine vision camera; performing image space information preprocessing according to the obtained image to obtain an image set of the target concrete structure and a reconstructed three-dimensional point cloud; on the basis of the obtained image set, recognizing concrete surface defects by using YOLO deep learning network detection to obtain a structure picture marked with the defects; and performing concrete structure three-dimensional reconstruction based on a three-dimensional Gaussian sputtering rendering technology in combination with the three-dimensional point cloud and the structure picture to obtain a three-dimensional structure model with high-precision surface textures and defect marks. The method can be used for surface defect detection and positioning of concrete structures such as bridges and buildings, the result is accurate, and the cost is saved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of automated detection technology, and in particular to a method for detecting and positioning three-dimensional reconstruction of concrete structure surface defects based on monocular vision. Background Art

[0002] Reinforced concrete is the most commonly used material for the construction of modern buildings, bridges and other structures. Reinforced concrete is composed of cement, sand, gravel and steel bars. Concrete is prone to cracks due to its material properties. The shrinkage caused by the cement reaction and the load during use will cause concrete cracks. These cracks will allow water, air and chemicals to invade the concrete, causing rust and corrosion, and ultimately leading to structural failure. Before the concrete structure fails, cracks, peeling and corrosion can be observed on the surface of the material. Regular inspection of its surface defects can promptly detect damage to the concrete structure and deal with it in time to prevent serious structural damage.

[0003] Traditional surface defect detection methods rely on manual visual inspection, which is inefficient, easily affected by subjective factors, and difficult to accurately measure the three-dimensional characteristics of surface defects. With the development of computer vision technology, automated surface defect detection methods have gradually become a research hotspot. However, existing technologies still have limitations in the three-dimensional positioning and quantification of surface defects. Therefore, a three-dimensional reconstruction method for concrete structure surface defect detection and positioning based on monocular vision is proposed to solve the above problems. Summary of the invention

[0004] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a method for detecting and positioning three-dimensional reconstruction of surface defects of concrete structures based on monocular vision, so as to solve the problems existing in the above-mentioned background technology.

[0005] The present invention is implemented by a method for detecting and positioning three-dimensional reconstruction of concrete structure surface defects based on monocular vision, the method comprising the following steps:

[0006] The multi-angle images of the target concrete structure are collected by a machine vision camera, wherein the machine vision camera is a monocular camera;

[0007] Performing image spatial information preprocessing on the acquired images to obtain an image set of the target concrete structure and a reconstructed three-dimensional point cloud;

[0008] Based on the obtained image set and using the YOLO deep learning network detection, the surface defects of the concrete are identified and marked to obtain a structural image with defects marked;

[0009] Based on the 3D Gaussian sputtering rendering technology combined with 3D point cloud and structural pictures, the concrete structure is reconstructed to obtain a 3D structural model with high-precision surface texture.

[0010] As a further solution of the present invention: the step of preprocessing the image spatial information according to the obtained image specifically includes:

[0011] Image feature extraction, detecting feature points in each image;

[0012] Feature matching, which uses a set of feature points to find similar feature points in different images to establish correspondence between images;

[0013] Geometric verification verifies whether the matched feature points are indeed mapped to the same scene point. This step maps feature points between image pairs by estimating the transformation matrix;

[0014] Incremental reconstruction, which uses verified image correspondences to build a scene graph with images as nodes and verified image pairs as edges, and matches points on the 2D image to the 3D space through triangulation, thereby expanding the scene point set;

[0015] Optimization, in the incremental reconstruction process, the bundle adjustment method is used to jointly optimize the camera parameters and the scene point set. This process improves the accuracy of the model by minimizing the reprojection error;

[0016] Point cloud generation, output the reconstructed 3D point cloud model.

[0017] As a further solution of the present invention: the step of identifying and marking the concrete surface defects based on the acquired image set and using the YOLO deep learning network detection specifically includes:

[0018] Prepare datasets and train YOLO deep learning networks based on image collections;

[0019] Image cropping processing: cropping the concrete structure images in the image collection to obtain better image processing speed and accuracy;

[0020] Use the YOLO deep learning network to identify concrete structural defects after image cropping;

[0021] Image stitching processing, after image cropping and defect identification marking, the cropped image is stitched back to its original size.

[0022] As a further solution of the present invention: the steps of preparing a data set and training a YOLO deep learning network based on an image set specifically include:

[0023] Acquire an image dataset of the concrete structure based on the image set, wherein the image dataset should include various types of defects;

[0024] Preprocess the collected images, including resizing the images, normalizing pixel values, and data augmentation;

[0025] Accurately annotate the preprocessed images to mark the location and type of defects. This step is done by drawing bounding boxes and assigning class labels.

[0026] The YOLO deep learning network is trained using a set of labeled images.

[0027] As a further solution of the present invention: the step of reconstructing the concrete structure based on the three-dimensional Gaussian sputtering rendering technology combined with the three-dimensional point cloud and the structure picture specifically includes:

[0028] Gaussian weight calculation and Gaussian point projection: assign a Gaussian weight to each pixel in the structure image based on the reconstructed 3D point cloud, and use the Gaussian weight to project each pixel in the image into 3D space, which is used to convert the 2D image coordinates into 3D coordinates and perform weighted averaging according to the Gaussian weight.

[0029] Point cloud fusion, which merges points whose projections overlap into a single point while preserving their color and intensity information;

[0030] Denoising and smoothing: denoising and smoothing of 3D point clouds;

[0031] The final model output is the dense 3D point cloud obtained after processing in the above steps.

[0032] As a further solution of the present invention: the steps of Gaussian weight calculation and Gaussian point projection specifically include:

[0033] Generate a point cloud of structure-from-motion based on a set of images, then construct a 3D Gaussian for each point with the position of each point cloud as the center, and then project the 3D Gaussian onto the image plane with the help of camera extrinsics;

[0034] The transformation matrix of the entire Gaussian is approximated using the transformation matrix of the center point coordinates of each three-dimensional Gaussian. The transformation of each Gaussian is regarded as an affine transformation.

[0035] Adjust the opacity of the projected 2D Gaussian according to the opacity and covariance matrix of the 3D Gaussian;

[0036] Optimization is performed using the stochastic gradient descent technique.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] ① The present invention combines 3D Gaussian sputtering rendering technology with the YOLO deep learning network to achieve quantification and 3D reconstruction of concrete structure surface defects, providing effective technical support for structural safety assessment and maintenance;

[0039] ② After obtaining the three-dimensional structural model including the defect marks on the concrete structure, the present invention can quantitatively measure the defects on the three-dimensional structural model;

[0040] ③The present invention can be used for surface defect detection and positioning of concrete structures such as bridges and buildings, with accurate results and cost savings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of the three-dimensional reconstruction method for concrete structure surface defect detection and positioning based on monocular vision.

[0042] Figure 2 A schematic diagram of generating a three-dimensional point cloud of a concrete structure based on a picture provided in an embodiment of the present invention. Figure 3 A group of pictures of concrete structures after detection using the surface defect detection and positioning three-dimensional reconstruction method of the present invention. Figure 4 A comparison chart between the surface defect detection and positioning three-dimensional reconstruction method of the present invention and the traditional recognition method. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0044] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0045] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting and positioning three-dimensional reconstruction of concrete structure surface defects based on monocular vision, and the method comprises the following steps:

[0046] The multi-angle images of the target concrete structure are collected by a machine vision camera, wherein the machine vision camera is a monocular camera;

[0047] Performing image spatial information preprocessing on the acquired images to obtain an image set of the target concrete structure and a reconstructed three-dimensional point cloud;

[0048] Based on the obtained image set and using the YOLO deep learning network detection, the surface defects of the concrete are identified and marked to obtain a structural image with defects marked;

[0049] Based on the 3D Gaussian sputtering rendering technology combined with the 3D point cloud and structural pictures, the concrete structure is reconstructed to obtain a 3D structural model.

[0050] It should be noted that the machine vision camera is fixed on a drone or a tripod to shoot videos of the concrete structure to be tested from different angles. The captured video includes all angles of the concrete structure to be tested without blind spots to ensure good results of the final three-dimensional reconstruction.

[0051] In the embodiment of the present invention, the present invention realizes the quantification and three-dimensional reconstruction of the surface defects of the concrete structure by combining the three-dimensional Gaussian sputtering rendering technology and the deep learning network, and provides effective technical support for the structural safety assessment and maintenance. The present invention can also be used for the surface defect detection and positioning of concrete structures such as bridges and buildings, with accurate results and cost savings. In addition, after obtaining the three-dimensional structural model including the defect identification on the concrete structure, the present invention can perform quantitative measurement of the defects on the three-dimensional structural model.

[0052] As a preferred embodiment of the present invention, the image spatial information preprocessing comprises the following steps:

[0053] Step 1: Image feature extraction, detect feature points in each image, which are unique points in the image that can describe the location. These feature points are invariant to lighting and geometric transformations, making them uniquely identifiable in different images.

[0054] Step 2: Feature matching, using a set of feature points to find the same scene parts seen in different images. This step finds similar feature points based on the appearance description of image features, thereby establishing a correspondence between images.

[0055] Step 3: Geometric verification verifies whether the matched feature points are indeed mapped to the same scene point. Feature points are mapped between image pairs by estimating the transformation matrix (such as homography matrix or basis matrix). The purpose of this step is to ensure the accuracy of matching and improve the robustness of reconstruction.

[0056] Step 4: Incremental reconstruction, using the verified image pairs to build a scene graph, where the images are nodes and the verified image pairs are edges. Then, the points on the 2D image are matched to 3D space through triangulation to expand the scene point set. This process is iterative, and new images are continuously added to the existing model and optimized through bundle adjustment to improve the accuracy and completeness of the overall reconstruction.

[0057] Step 5: Optimization, during the entire reconstruction process, the bundle adjustment method is used to jointly optimize the camera parameters and the scene point set. This process improves the accuracy of the model by minimizing the reprojection error. The bundle adjustment method is a nonlinear optimization problem, which is usually solved using the Levenberg-Marquardt algorithm or other numerical optimization methods.

[0058] Step 6: Point cloud generation, the output is the reconstructed 3D point cloud. Users can export this data for further analysis or visualization, such as Figure 2 shown.

[0059] As a preferred embodiment of the present invention, the use of YOLO deep learning network detection to identify and mark concrete surface defects includes the following steps:

[0060] Step 1: Dataset preparation and YOLO neural network training

[0061] Obtain an image dataset of concrete structures, which should contain various types of defects, such as cracks, honeycombs, pockmarks, exposed reinforcement, etc.; preprocess the collected images, including resizing, normalizing pixel values, and data enhancement (such as rotation, flipping, scaling, color transformation, etc.) to improve the generalization ability of the model; accurately annotate the preprocessed images to mark the location and type of defects, which is usually done by drawing bounding boxes and assigning category labels; select a suitable YOLO version (such as YOLOv3, YOLOv4, YOLOv5, YOLOv8, etc.), and adjust and optimize it according to project requirements and computing resources; train the YOLO model using the annotated dataset; during the training process, the model will learn how to identify and locate defects from images.

[0062] Step 2: Image cropping

[0063] The captured target concrete structure images (i.e., the images in the above-mentioned image dataset) are cropped to obtain better image processing speed and accuracy. For example, if the original material is a video shot at 1080P or higher resolution, the video frame images are extracted according to the requirements, and then the images are cropped into small blocks of 512*512 pixels for concrete structure surface defect detection.

[0064] Step 3: Identify concrete structural defects using neural networks

[0065] YOLOv8 is used to identify and mark concrete structure defects on a small image of 512*512 pixels, and to segment and color the defects in the image.

[0066] Step 4: Image stitching

[0067] After image cropping and defect identification marking, small blocks of 512*512 pixels are stitched together to restore the original size of the image.

[0068] As a preferred embodiment of the present invention, the step of reconstructing the concrete structure based on the three-dimensional Gaussian sputtering rendering technology combined with the three-dimensional point cloud and the structure picture specifically includes:

[0069] Step 1: Triangulation

[0070] The sparse point set obtained from the above steps of geometry verification and feature matching needs further processing. Triangulation is the process of converting these points into a continuous surface in three-dimensional space. This step is achieved by creating triangles between the matched feature points, thus forming a mesh structure.

[0071] Step 2: Point cloud generation

[0072] After triangulation is complete, each triangle vertex can be considered a point in a 3D point cloud. These points together form a rough representation of the entire scene. However, in order to obtain a more detailed surface model, operations need to be performed on each pixel.

[0073] Step 3: Gaussian weight calculation

[0074] Each pixel is assigned a Gaussian weight based on its distance from the triangle vertex. This weight reflects the contribution of the pixel to the final surface. Pixels that are closer have higher weights, while pixels that are farther away have lower weights.

[0075] Step 4: Gaussian Point Projection

[0076] Each pixel in the image is projected into 3D space using Gaussian weights. This step involves converting the 2D image coordinates into 3D coordinates and taking a weighted average of these points according to the Gaussian weights. Each pixel contributes to one or more points in 3D space.

[0077] The three-dimensional Gaussian function in three-dimensional Gaussian rendering is defined as:

[0078]

[0079] Where x is a point in space x = [x, y, z] T , μ is the center (mean) of the Gaussian, Σ is the covariance matrix Σ = RSS T R T , R is the rotation matrix and S is the scaling matrix.

[0080] S3: Use the transformation matrix of the center point coordinates of each three-dimensional Gaussian to approximate the transformation matrix of the entire Gaussian. In this way, the transformation of each Gaussian can be regarded as an affine transformation. This particular transformation matrix is ​​the Jacobian matrix

[0081]

[0082] J is a matrix that transforms a 3D vector into a 2D vector. The center position and color of the projected 2D Gaussian can be directly obtained from the parameters of the 3D Gaussian.

[0083] Projecting the 3D Gaussian to the 2D screen space requires calculating the inverse of the covariance matrix in the camera space: Σ′=JWΣW T J T , where W is the view transformation matrix, which includes affine rotation and translation.

[0084] The opacity of the projected 2D Gaussian needs to be adjusted according to the opacity and covariance matrix of the 3D Gaussian. Specifically, the opacity α′ of the projected 2D Gaussian can be calculated using the following formula:

[0085]

[0086] Given a pixel position x, its distance to all overlapping Gaussians, i.e. the depth of these Gaussians, can be calculated through the view transformation W to form a sorted Gaussian list. Alpha synthesis is then used to calculate the final color of this pixel.

[0087]

[0088] c i is the color, and opacity is

[0089] Optimization is performed using stochastic gradient descent techniques, and the loss function L may include L1 loss and the deep structural similarity (D-SSIM) loss term, L = (1-tadpole)L1+L D-SSIM . Where ∈ is the weight used to balance the two terms. For sparse regions, the Gaussian is cloned and moved along the position gradient direction. For over-reconstructed regions, the Gaussian is split into two new Gaussians.

[0090] Step 5: Point cloud fusion

[0091] As all pixels are projected, a large number of points are obtained, which may overlap in three-dimensional space. The purpose of point cloud fusion is to merge these overlapping points into a single point while retaining its color and intensity information. This is usually achieved by calculating the average position and color of all overlapping points.

[0092] Step 6: Denoising and Smoothing

[0093] In order to improve the quality of the final model, it may be necessary to denoise and smooth the 3D point cloud. This can be achieved through various filtering techniques, such as using a median filter to remove outliers, or applying a Gaussian filter to smooth the point cloud.

[0094] Step 7: Final model output and concrete surface defect identification

[0095] After the above steps, the dense 3D point cloud or surface model can be used for further analysis, editing or rendering. Figure 3 As shown in the figure, the reconstructed 3D model has high-precision surface texture and can be used to identify surface defects of structural concrete. After scaling the 3D model, you can see the comparison between the local concrete surface defect marking results and the 2D image results, as shown in the figure. Figure 4 These models can be exported to common 3D file formats for use in other software.

[0096] The above only describes in detail the preferred embodiments of the present invention, which is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0097] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0098] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0099] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the disclosure in the specification and examples. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. 3D reconstruction method for surface defect detection and positioning of concrete structures based on monocular vision, It is characterized in that The method comprises the following steps: The multi-angle images of the target concrete structure are collected by a machine vision camera, wherein the machine vision camera is a monocular camera; Performing image spatial information preprocessing on the acquired images to obtain an image set of the target concrete structure and a reconstructed three-dimensional point cloud; Based on the obtained image set and using the YOLO deep learning network detection, the surface defects of the concrete are identified and marked to obtain a structural image with defects marked; Based on the 3D Gaussian sputtering rendering technology combined with the 3D point cloud and structural pictures, the concrete structure is reconstructed to obtain a 3D structural model.

2. According to claim 1, the method for detecting and positioning three-dimensional reconstruction of concrete structure surface defects based on monocular vision, It is characterized in that The step of preprocessing the image spatial information according to the obtained image specifically includes: Image feature extraction, detecting feature points in each image; Feature matching, which uses a set of feature points to find similar feature points in different images to establish correspondence between images; Geometric verification verifies whether the matched feature points are indeed mapped to the same scene point. This step maps feature points between image pairs by estimating the transformation matrix; Incremental reconstruction, which uses verified image correspondences to build a scene graph with images as nodes and verified image pairs as edges, and matches points on the 2D image to the 3D space through triangulation, thereby expanding the scene point set; Optimization, in the incremental reconstruction process, the bundle adjustment method is used to jointly optimize the camera parameters and the scene point set. This process improves the accuracy of the model by minimizing the reprojection error; Point cloud generation, output the reconstructed 3D point cloud model.

3. The method for detecting and positioning three-dimensional reconstruction of concrete structure surface defects based on monocular vision according to claim 1, It is characterized in that The step of identifying and marking concrete surface defects based on the acquired image set and using the YOLO deep learning network detection specifically includes: Prepare datasets and train YOLO deep learning networks based on image collections; Image cropping processing: cropping the concrete structure images in the image collection to obtain better image processing speed and accuracy; Use the YOLO deep learning network to identify concrete structural defects after image cropping; Image stitching processing, after image cropping and defect identification marking, the cropped image is stitched back to its original size.

4. The method for detecting and positioning three-dimensional reconstruction of concrete structure surface defects based on monocular vision according to claim 3, It is characterized in that The steps of preparing a data set and training a YOLO deep learning network based on an image set specifically include: Acquire an image dataset of the concrete structure based on the image set, wherein the image dataset should include various types of defects; Preprocess the collected images, including resizing the images, normalizing pixel values, and data augmentation; Accurately annotate the preprocessed images to mark the location and type of defects. This step is done by drawing bounding boxes and assigning class labels. The YOLO deep learning network is trained using a set of labeled images.

5. The method for detecting and positioning three-dimensional reconstruction of concrete structure surface defects based on monocular vision according to claim 1, It is characterized in that The step of reconstructing the concrete structure based on the three-dimensional Gaussian sputtering rendering technology combined with the three-dimensional point cloud and the structure picture specifically includes: Gaussian weight calculation and Gaussian point projection: assign a Gaussian weight to each pixel in the structure image based on the reconstructed 3D point cloud, and use the Gaussian weight to project each pixel in the image into 3D space, which is used to convert the 2D image coordinates into 3D coordinates and perform weighted averaging according to the Gaussian weight. Point cloud fusion, which merges points whose projections overlap into a single point while preserving their color and intensity information; Denoising and smoothing: denoising and smoothing of 3D point clouds; The final model output is the dense 3D point cloud obtained after processing in the above steps.

6. The method for detecting and positioning three-dimensional reconstruction of concrete structure surface defects based on monocular vision according to claim 5, It is characterized in that The steps of Gaussian weight calculation and Gaussian point projection specifically include: Generate a point cloud of structure-from-motion based on a set of images, then construct a 3D Gaussian for each point with the position of each point cloud as the center, and then project the 3D Gaussian onto the image plane with the help of camera extrinsics; The transformation matrix of the entire Gaussian is approximated using the transformation matrix of the center point coordinates of each three-dimensional Gaussian. The transformation of each Gaussian is regarded as an affine transformation. Adjust the opacity of the projected 2D Gaussian according to the opacity and covariance matrix of the 3D Gaussian; Optimization is performed using the stochastic gradient descent technique.

Citation Information

Cited By

  • Construction site measuring system based on image processing

    CN119206099A

  • Intelligent camera system for three-dimensional modeling, displacement detection and defect identification of bridge structure

    CN121691640A

  • Monocular microscopic imaging surface micro-nano defect 3D reconstruction detection device and method

    CN121703102A

  • A monocular microscopic imaging surface micro-nano defect 3D reconstruction detection device and method

    CN121703102B