Concrete coarse aggregate angle measurement system based on digital image processing

Abstract

The invention relates to the technical field of building material performance detection, in particular to a concrete coarse aggregate corner angle measurement system based on digital image processing, which comprises a multi-view image acquisition module, a three-dimensional point cloud reconstruction module, a corner angle feature vector extraction module and a corner angle quantitative calculation module. A multi-view imaging system is constructed through a rotary objective table and a binocular camera, surface images of aggregate at different angles are obtained, and high-precision three-dimensional point cloud data are generated in combination with stereo matching and a point cloud dense reconstruction technology. And further constructing an edge feature vector based on the volume compression ratio and the surface roughness, inputting a pre-trained support vector regression model, outputting a normalized edge index, and realizing objective quantitative evaluation of the coarse aggregate edge. The concrete aggregate quality grading and screening device has the advantages of being high in precision, high in automation, wide in adaptability and the like and is suitable for quality grading and screening of concrete aggregates.

Description

Technical Field

[0001] This invention relates to the field of building material performance testing technology, and in particular to a system for measuring the angularity of coarse aggregates in concrete based on digital image processing. Background Technology

[0002] The angular characteristics of coarse aggregates in concrete are one of the important indicators determining the workability, interfacial bonding performance, and mechanical strength of concrete mixtures. The higher the angularity of the aggregate particles, the stronger the mechanical interlocking force between them and the cement paste, which helps to form a high-strength, high-durability aggregate structure. This is particularly suitable for applications with high requirements for aggregate performance, such as highways, hydraulic structures, and ultra-high performance concrete. Traditional methods for evaluating the angularity of coarse aggregates mainly rely on manual visual experience or two-dimensional image methods for approximate contour evaluation. However, these methods are limited by the complexity of particle shape, irregular surface structure, and imaging angle limitations, making it difficult to comprehensively and objectively reflect the three-dimensional angular morphology of the aggregates.

[0003] In existing technologies, although some studies have attempted to introduce 3D laser scanners and industrial CT for morphology reconstruction, these methods suffer from problems such as high equipment costs, low acquisition efficiency, and limited applicable particle size ranges. Furthermore, 3D reconstructed data often lacks effective coupling with angularity measurement models, resulting in subsequent angularity evaluation processes remaining at a subjective grading stage based on expert experience, lacking standardized indices that can be generalized. Particularly concerning are insufficient point cloud reconstruction accuracy, poor robustness of curvature calculation, and weak generalization ability of angularity prediction models. A complete, efficient, and automated angularity quantification scheme for coarse aggregates has not yet been established, hindering its widespread application in engineering material evaluation and concrete design optimization. Summary of the Invention

[0004] This invention provides a system for measuring the angularity of coarse aggregates in concrete based on digital image processing.

[0005] A system for measuring the edge angle of coarse aggregate in concrete based on digital image processing includes a multi-view image acquisition module, a three-dimensional point cloud reconstruction module, an edge angle feature vector extraction module, and an edge angle quantization calculation module, wherein;

[0006] The multi-view image acquisition module is configured to place the coarse aggregate sample on a rotating stage, illuminate it uniformly with a ring light source array, and simultaneously acquire images of the aggregate surface from multiple fixed angles using a binocular camera, outputting a multi-view image set.

[0007] The 3D point cloud reconstruction module is configured to perform stereo matching and dense reconstruction on the multi-view image set to generate 3D point cloud data of the aggregate.

[0008] The corner feature vector extraction module is configured to calculate the ratio of the aggregate entity to its convex hull volume as the volume compression ratio based on the three-dimensional point cloud data, and at the same time extract the feature quantity of the point cloud curvature distribution as the surface roughness, forming a corner feature vector composed of the volume compression ratio and the surface roughness.

[0009] The edge angle quantization calculation module is configured to input the edge angle feature vector into a pre-trained edge angle prediction model and output a normalized edge angle index.

[0010] Optionally, in the multi-view image acquisition module, the fixed viewing angles include four azimuth angles: 0°, 45°, 90°, and 135°, and the ring light source array consists of eight groups of LED diffuse light sources.

[0011] Optionally, the execution process of the multi-view image acquisition module includes the following steps:

[0012] Coarse aggregate sample placement: Fix the coarse aggregate sample to be tested at the geometric center of the rotating stage;

[0013] Rotating stage positioning: Control the rotating stage to rotate in preset angular increments so that the surface of the coarse aggregate sample to be collected is aligned with the center of the field of view of the binocular camera;

[0014] Ring light source array illumination: Activate the ring light source array to diffuse and uniformly illuminate the surface of the coarse aggregate sample, eliminate shadows and maintain illuminance uniformity ≥95%;

[0015] Synchronous acquisition by binocular camera: When the rotating stage is rotated to four fixed viewing angles of 0°, 45°, 90° and 135°, the binocular camera is triggered to synchronously acquire surface images of the coarse aggregate sample and generate a multi-view image set.

[0016] Optionally, the binocular camera acquires two images from each viewpoint, namely a left-eye image and a right-eye image, generating a total of eight surface images from the four views.

[0017] Optionally, the execution process of the 3D point cloud reconstruction module includes the following steps:

[0018] Multi-view image set input: Receives a multi-view image set from the multi-view image acquisition module. The multi-view image set includes the left and right eye images acquired by the binocular camera at four azimuth angles.

[0019] Stereo matching: Stereo matching is performed on the left and right images from the same viewpoint, and the initial 3D point cloud of that viewpoint is generated by calculating pixel disparity;

[0020] Dense Reconstruction: Dense reconstruction is performed on the generated initial 3D point cloud. Through multi-view point cloud fusion and hole filling, a dense 3D point cloud is generated.

[0021] 3D point cloud data output: After curvature filtering optimization of dense 3D point clouds and removal of outliers, 3D point cloud data that meets the resolution requirements is output.

[0022] Optionally, meeting the resolution requirement means that the point cloud density of the 3D point cloud data is not less than 50 points / mm. 2 And the point cloud position error is ≤0.05mm.

[0023] Optionally, the execution process of the angle feature vector extraction module includes the following steps:

[0024] 3D point cloud data input: Receives 3D point cloud data from the 3D point cloud reconstruction module;

[0025] Volume compression ratio calculation: Based on the three-dimensional point cloud data, a point cloud convex hull is generated through a fast convex hull algorithm. The ratio of the actual point cloud volume of the aggregate to the convex hull volume is calculated to obtain the volume compression ratio.

[0026] Surface roughness calculation: Based on the three-dimensional point cloud data, the local surface is fitted by moving least squares method and the Gaussian curvature value is calculated. The standard deviation of all point cloud Gaussian curvature values ​​is calculated to obtain the surface roughness.

[0027] Edge feature vector generation: The volume compression ratio and the surface roughness are combined into a two-dimensional vector to form an edge feature vector composed of the volume compression ratio and the surface roughness.

[0028] Optionally, the execution process of the edge angle quantization calculation module includes the following steps:

[0029] Corner feature vector input: Receives corner feature vectors from the corner feature vector extraction module;

[0030] Loading the pre-trained edge angle prediction model: Load the pre-trained edge angle prediction model from memory. The model is based on a support vector regression mechanism, and the training data includes granite, basalt and limestone aggregate samples.

[0031] Edge angle index prediction calculation: Input the edge angle feature vector into the pre-trained edge angle prediction model, and perform prediction calculation through support vector regression machine to generate initial edge angle values;

[0032] Normalized edge angle index output: The initial edge angle value is linearly normalized, and the output is a normalized edge angle index with a value range of 0 to 1.

[0033] Optionally, the value of the edge angle index is positively correlated with the sharpness of the aggregate edges.

[0034] The beneficial effects of this invention are:

[0035] This invention automates the entire process of coarse aggregate edge feature analysis, from image acquisition to quantitative evaluation, by constructing a complete measurement system comprising a multi-view image acquisition module, a 3D point cloud reconstruction module, an edge feature vector extraction module, and an edge angle quantification calculation module. By acquiring omnidirectional surface information of the aggregate through multi-view acquisition and binocular imaging technology, combined with 3D point cloud reconstruction and curvature analysis, it effectively avoids the dependence on edge characteristics in traditional subjective evaluation methods, significantly improving the stability and repeatability of edge measurement.

[0036] This invention introduces a support vector regression-based edge angle prediction model and combines it with a linear normalization strategy to output a normalized edge angle index with uniform dimensions (range [0,1]), which is applicable to the edge angle characterization of common aggregate types such as granite, basalt, and limestone. The proposed index is significantly positively correlated with the sharpness of the aggregate surface morphology, has clear physical meaning and engineering guidance value, and provides a basis for concrete strength grading.

[0037] This invention achieves a density of no less than 50 points / mm by using binocular parallax calculation, multi-view fusion, and aperture compensation in the point cloud reconstruction module. 2 High-fidelity 3D point cloud data with a point position error of no more than 0.05 mm is obtained, taking into account both coarse-scale volumetric configuration and micro-scale surface undulation features. Combined with moving least squares fitting and Gaussian curvature standard deviation calculation, it not only captures the macroscopic edges and corners of the aggregates, but also quantitatively extracts the microscopic ruggedness, ensuring the integrity and resolution of the edge and corner feature description. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the system flow according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of the edge angle quantization calculation module according to an embodiment of the present invention. Detailed Implementation

[0041] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0042] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0043] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.

[0044] like Figure 1-2 As shown, a system for measuring the edge angle of coarse aggregate in concrete based on digital image processing includes a multi-view image acquisition module, a three-dimensional point cloud reconstruction module, an edge angle feature vector extraction module, and an edge angle quantization calculation module, wherein;

[0045] The multi-view image acquisition module is configured to place the coarse aggregate sample on a rotating stage, illuminate it uniformly with a ring light source array, and simultaneously acquire images of the aggregate surface from multiple fixed angles using a binocular camera, outputting a multi-view image set.

[0046] The 3D point cloud reconstruction module is configured to perform stereo matching and dense reconstruction on multi-view image sets to generate 3D point cloud data of the aggregate.

[0047] The corner feature vector extraction module is configured to calculate the ratio of the aggregate entity to its convex hull volume as the volume compression ratio based on 3D point cloud data, and at the same time extract the feature quantity of the point cloud curvature distribution as the surface roughness, forming a corner feature vector composed of the volume compression ratio and the surface roughness.

[0048] The edge angle quantization calculation module is configured to input the edge angle feature vector into the pre-trained edge angle prediction model and output a normalized edge angle index.

[0049] The execution process of the multi-view image acquisition module includes:

[0050] The multi-view image acquisition module is used to acquire high-quality images of coarse aggregate samples from multiple azimuth angles to support subsequent 3D point cloud reconstruction and edge feature extraction processes. The module consists of a rotating stage, a ring light source array, and a binocular camera, and the acquisition process is automated through an integrated control system.

[0051] The rotating stage is driven by a high-precision stepper motor, providing an angular positioning accuracy of ±0.5°. The coarse aggregate sample is placed at the geometric center of the rotating stage and fixed by a mechanical clamping mechanism to prevent displacement or shaking during rotation.

[0052] The fixed viewing angles are set to four azimuth angles: 0°, 45°, 90°, and 135°. Each azimuth angle corresponds to the rotation angle of the rotating stage. The rotation control program is set to rotate the rotating stage in 45° increments and pause filming at each fixed viewing angle.

[0053] The ring light source array consists of eight groups of uniformly distributed LED diffuse light sources arranged around the binocular camera to form a shadowless, uniform ring illumination field. Each group of LED light sources is equipped with a diffuser and a constant current drive unit maintains stable illuminance, ensuring a consistent brightness distribution on the surface of the coarse aggregate sample from all viewing angles.

[0054] Actual measurements show that the illumination uniformity of this system reaches over 95%, effectively suppressing image information loss caused by local shadows.

[0055] The binocular camera uses an industrial-grade imaging device with a resolution of 1920×1080. The baseline length between the left and right cameras is 60mm. It is equipped with a synchronous trigger function and achieves high-precision synchronous imaging through hardware-level time locking. Each shot acquires the left and right eye images separately, for a total of 8 images (4 sets of binocular images), forming a multi-view image set.

[0056] The data collection process is as follows:

[0057] Placement of coarse aggregate samples: Place the pretreated concrete coarse aggregate samples (e.g., with a particle size of 10-20mm) in the center of the rotating platform, ensuring that the samples are unobstructed and the surface is dry.

[0058] Rotating stage positioning: The rotating stage is controlled to rotate sequentially to four fixed viewing angles: 0°, 45°, 90° and 135°. The system automatically completes braking positioning at each target angle, with a positioning error not exceeding ±0.5°.

[0059] Ring light source array illumination: At each viewing angle, the ring light source array is automatically activated, with an output luminous flux of 1200lm and an illumination distance of 300mm, ensuring that the entire surface of the coarse aggregate sample is uniformly illuminated.

[0060] Synchronous acquisition by binocular cameras: After the illumination stabilizes, the system triggers the binocular cameras to complete synchronous imaging of the left and right perspectives via TTL signals, acquires the left and right eye images, and automatically saves them in a standard image format (such as TIFF), labeling the perspective and timestamp.

[0061] Example explanation: Taking sample A as an example, when the rotating stage is rotated to a 45° viewing angle, the binocular cameras acquire the following image data:

[0062] Left eye image:

[0063] Right eye image:

[0064] The final result is the following multi-view image set:

[0065]

[0066] in: Represents a multi-view image set. This represents the left eye image acquired at an azimuth angle θ. This represents the right eye image acquired at an azimuth angle θ, where θ∈{0°,45°,90°,135°}.

[0067] The above image set will be fed into the 3D point cloud reconstruction module as input, supporting subsequent stereo reconstruction processing.

[0068] The multi-view image acquisition module can significantly improve the accuracy of 3D reconstruction, ensuring comprehensive sampling of the surface contour and microstructure of coarse aggregates, and providing a high-fidelity image basis for the extraction of corner feature vectors.

[0069] The execution process of the 3D point cloud reconstruction module includes:

[0070] The 3D point cloud reconstruction module first receives a multi-view image set from the multi-view image acquisition module, and each pair of images... All images are stereo images taken at the same azimuth angle θ. The input image resolution is set to 1920×1080, the bit depth is 8-bit grayscale, and the pixel size is 4.5μm.

[0071] Stereo matching was performed on each group of adjacent left and right eye images (4 groups in total) using a block matching and disparity search window method, and the disparity map D was calculated. θ (x,y), and based on this, backproject to generate the initial 3D point cloud P. θ .

[0072] The specific parallax-to-depth formula is as follows:

[0073] Where Z(x,y) represents the depth value corresponding to the pixel, f is the focal length of the stereo camera, B is the baseline length of the stereo camera, d(x,y) corresponds to the disparity value in the disparity map, and (x,y) represents the pixel coordinates in the image. The initial point cloud format is as follows: Four initial 3D point clouds were generated.

[0074] To improve the integrity and density of point clouds, a multi-view fusion strategy is adopted. First, four initial point cloud sets P0, P... 45 ,P 90 ,P 135 Coordinate system alignment was performed using a SIFT-3D-based point cloud registration method to achieve rigid registration with >2000 registration point pairs. Subsequently, dense reconstruction based on a voxel grid was executed.

[0075] Point cloud fusion uses a spatial grid with a voxel resolution of 0.05 mm for downsampling and reconstruction;

[0076] The hole filling method employs an interpolation approach based on nearest neighbor interpolation and curvature continuity constraints, and fills holes with an area ≤1mm². 2 area;

[0077] The formula for evaluating point cloud density is:

[0078] Where ρ is the point cloud density, N is the number of points per unit area, and A is the area of ​​the region;

[0079] After the above processing, the overall density of the generated dense 3D point cloud reaches 62-75 points / mm. 2 .

[0080] Even after processing, dense point clouds may still contain a small number of mismatched or jittery points. Therefore, the following optimization operations are performed:

[0081] Curvature filtering: The principal radius of curvature at each point is calculated using a 3×3 neighborhood. If an outlier in the curvature exceeds a set threshold (e.g., k > 1.5 mm), the filter is applied. -1 If ) is identified as an anomaly, it should be removed.

[0082] Outlier removal: The Statistical Outlier Removal (SOR) algorithm is used, with a standard deviation factor of 2, to remove low-density outliers;

[0083] Point position error control: The optimized point cloud is verified by comparison with real-size blocks. The point position error is controlled by registering and comparing the measured data of the standard sample block with the point cloud reconstruction model. The maximum point position error is controlled within ε≤0.05mm.

[0084] The final output 3D point cloud data is in standard PLY format or XYZ point coordinate format, and has the following performance indicators:

[0085] Point cloud density ρ ≥ 50 points / mm 2 ;

[0086] Position error ε≤0.05mm;

[0087] Void rate ≤ 2%;

[0088] Example: Taking coarse aggregate sample B as an example, after completing the 3D reconstruction using this system, the sampling area is 10×10mm. 2 The total number of points in the point cloud is 5420. The calculation yields:

[0089]

[0090] The location error was determined by comparing the key point positions of the solid model and the reconstructed model. The maximum error was 0.038 mm, which meets the requirements.

[0091] The execution process of the angle feature vector extraction module includes:

[0092] Receives standard 3D point cloud data output from the 3D point cloud reconstruction module. The point cloud format is... Where N represents the total number of point clouds. The data density satisfies ρ ≥ 50 points / mm. 2 The spatial error does not exceed 0.05mm, providing a basis for high-precision geometric analysis.

[0093] Based on 3D point cloud data, the QuickHull algorithm is used to generate the convex hull surface mesh of the point cloud. The steps are as follows:

[0094] (1) Point cloud volume calculation: A closed surface is constructed from the original point cloud using a triangulation algorithm (such as Delaunay triangulation), and the actual point cloud volume V is calculated using polyhedral volume integration. actual .

[0095]

[0096] in Let be the vertex vector of the k-th triangle face.

[0097] (2) Convex Hull Volume Calculation: The QuickHull algorithm is used to construct the convex hull of the point cloud, generating the minimum outer mesh volume, and the convex hull volume V is calculated. hull (The method is the same as that used for point cloud volume calculation.)

[0098] (3) The formula for calculating volume compressibility is: (where 0 < γ ≤ 1);

[0099] This index is used to measure the degree of concavity or compactness of coarse aggregate shape. The smaller the value, the more irregular the sample edge and the stronger the angularity.

[0100] Surface roughness is used to characterize the fluctuation of the micro-curvature of the point cloud surface. The calculation process is as follows:

[0101] (1) For each point p in the 3D point cloud i Using a set of points within a neighborhood of a fixed radius r = 1 mm as input, the Moving Least Squares (MLS) method is applied to fit its local quadratic surface.

[0102] (2) Gaussian curvature calculation: On each fitted local surface, the Gaussian curvature is calculated based on the principal curvatures (k1, k2):

[0103] (3) Statistical feature extraction: For all K i The set of curvature values Perform standard deviation statistics:

[0104]

[0105] in Let be the average Gaussian curvature of the entire point cloud.

[0106] Ultimately, σ K Defined as the surface roughness of the coarse aggregate sample, the larger the value, the more complex the microstructure and angular protrusions on the surface.

[0107] The volume compressibility γ and the surface roughness σ K Combined to form a two-dimensional feature vector:

[0108]

[0109] The angular feature vector It contains both coarse-scale global geometric information (described by γ) and micro-scale local curvature information (described by σ). K (Description), providing high-quality input for subsequent edge angle quantization calculations.

[0110] Example explanation:

[0111] After processing the 3D point cloud data extracted from a coarse aggregate sample C, the following results were obtained:

[0112] Actual point cloud volume: V actual =312.4mm 3 ;

[0113] Convex hull volume: V hull=386.1mm 3 ;

[0114] Gaussian curvature standard deviation: σ K =0.0176mm -2 ;

[0115] but:

[0116]

[0117] This feature vector will be passed as input to the subsequent edge angle quantization calculation module.

[0118] The edge angle quantification module inputs the two-dimensional feature vector output by the edge angle feature vector extraction module into the pre-trained model to achieve quantitative prediction and standardized characterization of the edge angle properties of coarse aggregates. This module builds a prediction model based on support vector regression and, combined with a normalization strategy, outputs a normalized edge angle index that can be used for aggregate screening and quality grading, specifically including:

[0119] It receives the edge feature vector output from the previous module, and calls a pre-trained edge angle prediction model from local non-volatile memory or an embedded model library. This model is based on a support vector regression machine (SVR), and the core algorithm uses a radial basis function (RBF).

[0120]

[0121] The model training samples include standard prepared granite, basalt and limestone aggregate samples. The training set contains 300 sets of data pairs with manually labeled edge angle levels, and the output labels are standardized edge angle levels (0-100).

[0122] The training uses the following configuration:

[0123] Kernel function type: RBF (Radial Basis Function);

[0124] Error tolerance parameter: ε = 0.01;

[0125] Regularization coefficient: C = 10;

[0126] Cross-validation method: 5-fold cross-validation;

[0127] Model evaluation metric: Root mean square error (RMSE) < 3.2;

[0128] After training, the model is stored in .pkI format and can be loaded and invoked through the model calling interface.

[0129] The input angular feature vector The input is fed into the loaded support vector regression model to perform prediction calculations and obtain the initial edge angle value A of the aggregate sample. raw :

[0130]

[0131] Among them, the predicted output A raw The numerical range is based on the distribution of training data, and is generally in the range of [0, 100]. The larger the value, the stronger the angularity and the higher the sharpness of the aggregate.

[0132] To improve the interpretability and adaptability of the indicators, the system will use the initial edge angle value A. raw Perform linear normalization to obtain the final output normalized edge angle index A. norm The value range is [0,1], and its calculation formula is:

[0133]

[0134] Among them, A min A max These are the historical minimum and maximum values ​​in the training sample set (set as A in this invention). min =20,A max =95);

[0135] When A raw min When, take A norm =0;

[0136] When A raw >A max When, take A norm =1.

[0137] The model training phase used a total of 300 sets of sample data, covering three common coarse aggregate types: granite, basalt, and limestone. The sharpness of the sample edges was physically verified through manual image annotation and mechanical stripping. Combined with expert grading labels, initial edge angle evaluation data was generated. The value range was concentrated between 20 and 95, with no obvious extreme values ​​or outliers. Therefore, this invention sets A... min =20,A max =95.

[0138] The closer the normalized edge angle index is to 1, the sharper the edges and the more prominent the edges of the aggregate particles; conversely, the closer the index is to 0, the more rounded the aggregate surface tends to be.

[0139] Example explanation: Taking a coarse aggregate sample D as an example, the extracted corner feature vector is:

[0140] ​

[0141] The prediction model calculates: A raw =87.6

[0142] After normalization:

[0143] The final output normalized edge angle index is 0.901, which can be used to determine that the sample has significant edge angle characteristics and is suitable for high-strength concrete applications where edge angle requirements are high.

[0144] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0145] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A system for measuring the angularity of coarse aggregate in concrete based on digital image processing, characterized in that, It includes a multi-view image acquisition module, a 3D point cloud reconstruction module, an edge feature vector extraction module, and an edge angle quantization calculation module, among which; The multi-view image acquisition module is configured to place the coarse aggregate sample on a rotating stage, illuminate it uniformly with a ring light source array, and simultaneously acquire images of the aggregate surface from multiple fixed angles using a binocular camera, outputting a multi-view image set. The 3D point cloud reconstruction module is configured to perform stereo matching and dense reconstruction on the multi-view image set to generate 3D point cloud data of the aggregate. The corner feature vector extraction module is configured to calculate the ratio of the aggregate entity to its convex hull volume as the volume compression ratio based on the three-dimensional point cloud data, and at the same time extract the feature quantity of the point cloud curvature distribution as the surface roughness, forming a corner feature vector composed of the volume compression ratio and the surface roughness. The edge angle quantization calculation module is configured to input the edge angle feature vector into a pre-trained edge angle prediction model and output a normalized edge angle index.

2. The system for measuring the angularity of coarse aggregate in concrete based on digital image processing according to claim 1, characterized in that, In the multi-view image acquisition module, the fixed viewing angles include four azimuth angles: 0°, 45°, 90°, and 135°, and the ring light source array consists of eight groups of LED diffuse light sources.

3. The system for measuring the edge angle of coarse aggregate in concrete based on digital image processing according to claim 2, characterized in that, The execution process of the multi-view image acquisition module includes the following steps: Coarse aggregate sample placement: Fix the coarse aggregate sample to be tested at the geometric center of the rotating stage; Rotating stage positioning: Control the rotating stage to rotate in preset angular increments so that the surface of the coarse aggregate sample to be collected is aligned with the center of the field of view of the binocular camera; Ring light source array illumination: Activate the ring light source array to diffuse and uniformly illuminate the surface of the coarse aggregate sample, eliminate shadows and maintain illuminance uniformity ≥95%; Synchronous acquisition by binocular camera: When the rotating stage is rotated to four fixed viewing angles of 0°, 45°, 90° and 135°, the binocular camera is triggered to synchronously acquire surface images of the coarse aggregate sample and generate a multi-view image set.

4. The system for measuring the angularity of coarse aggregate in concrete based on digital image processing according to claim 1, characterized in that, The binocular camera acquires two images from each viewpoint, namely the left eye image and the right eye image, generating a total of eight surface images from the four views.

5. The system for measuring the angularity of coarse aggregate in concrete based on digital image processing according to claim 4, characterized in that, The execution process of the 3D point cloud reconstruction module includes the following steps: Multi-view image set input: Receives a multi-view image set from the multi-view image acquisition module. The multi-view image set includes the left and right eye images acquired by the binocular camera at four azimuth angles. Stereo matching: Stereo matching is performed on the left and right images from the same viewpoint, and the initial 3D point cloud of that viewpoint is generated by calculating pixel disparity; Dense Reconstruction: Dense reconstruction is performed on the generated initial 3D point cloud. Through multi-view point cloud fusion and hole filling, a dense 3D point cloud is generated. 3D point cloud data output: After curvature filtering optimization of dense 3D point clouds and removal of outliers, 3D point cloud data that meets the resolution requirements is output.

6. The system for measuring the angularity of coarse aggregate in concrete based on digital image processing according to claim 5, characterized in that, The resolution requirement refers to a point cloud density of 3D point cloud data of not less than 50 points / mm². 2 And the point cloud position error is ≤0.05mm.

7. The system for measuring the angularity of coarse aggregate in concrete based on digital image processing according to claim 6, characterized in that, The execution process of the angle feature vector extraction module includes the following steps: 3D point cloud data input: Receives 3D point cloud data from the 3D point cloud reconstruction module; Volume compression ratio calculation: Based on the three-dimensional point cloud data, a point cloud convex hull is generated through a fast convex hull algorithm. The ratio of the actual point cloud volume of the aggregate to the convex hull volume is calculated to obtain the volume compression ratio. Surface roughness calculation: Based on the three-dimensional point cloud data, the local surface is fitted by moving least squares method and the Gaussian curvature value is calculated. The standard deviation of all point cloud Gaussian curvature values ​​is calculated to obtain the surface roughness. Edge feature vector generation: The volume compression ratio and the surface roughness are combined into a two-dimensional vector to form an edge feature vector composed of the volume compression ratio and the surface roughness.

8. The system for measuring the angularity of coarse aggregate in concrete based on digital image processing according to claim 7, characterized in that, The execution process of the edge angle quantization calculation module includes the following steps: Corner feature vector input: Receives corner feature vectors from the corner feature vector extraction module; Loading the pre-trained edge angle prediction model: Load the pre-trained edge angle prediction model from memory. The model is based on a support vector regression mechanism, and the training data includes granite, basalt and limestone aggregate samples. Edge angle index prediction calculation: Input the edge angle feature vector into the pre-trained edge angle prediction model, and perform prediction calculation through support vector regression machine to generate initial edge angle values; Normalized edge angle index output: The initial edge angle value is linearly normalized, and the output is a normalized edge angle index with a value range of 0 to 1.

9. A system for measuring the angularity of coarse aggregate in concrete based on digital image processing according to claim 8, characterized in that, The value of the edge angle index is positively correlated with the sharpness of the aggregate edges.

Images