A rock lump laser indicating device

Abstract

The present application relates to the technical field of rock blockiness identification, and particularly relates to a rock blockiness laser indication device, which comprises a laser projector configured with L groups and respectively installed at different positions of a mechanical arm of a breaking hammer; a collection module used for acquiring grid deformation size data in cooperation with the laser projector; and a central processing unit used for acquiring the grid deformation size data and analyzing and processing the grid deformation size data to obtain rock blockiness indexes. The present application synchronously projects laser grid light in multiple directions and collects deformation size data from different visual angles, can realize the complementation of different direction visual angles, thereby significantly reducing the "dead angle" and information missing problem caused by the mutual stacking and shielding of rocks, improving the coverage rate and reconstruction integrity of the overall point cloud, and obtaining a point cloud model with higher completeness and richer edge information, which is beneficial to subsequent blockiness identification based on a machine learning model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rock block size identification, in particular to a rock block size laser indicating device. BACKGROUND

[0002] In an open-pit mine, the particle size, i.e., the block size, is generally strictly controlled before the rock is loaded onto a vehicle to improve the crushing efficiency of the crusher. Large oversized rocks are generally broken down using a hydraulic breaking hammer. At present, the block size is determined by the naked eye of the breaking hammer operator, and there is a phenomenon of block size being too large or too small. When the block size is too large, the crushing efficiency of the crusher is affected, and when the block size is too small, the crushing cost is increased.

[0003] Chinese patent application No. CN112285107B discloses a method for measuring three-dimensional block size distribution by plane photographing, which comprises the following steps: obtaining the total volume of all block sizes and the volume occupied by the blast pile after blasting, and calculating the loose coefficient Kc after blasting; obtaining the shape and total surface area S of all rock blocks in the two-dimensional plane direction of the blast pile by processing the photos of the rock blocks on the surface of the blast pile; obtaining the total sum H of the cumulative rock height in the third dimension of all rock blocks by dividing the total volume of the block size by the total surface area S 总 ; according to the fixed ratio of the length, width, and height of the rock, after knowing the total sum in the height direction and the number of rock blocks, the block size distribution in the third dimension is calculated.

[0004] The prior art, as described above, mainly obtains the shape and total surface area S of the rock blocks in the two-dimensional plane direction, and calculates the block size distribution in the third dimension. However, there are still some deficiencies. Firstly, it only relies on single-angle scanning, and when facing a scene where rock blocks are densely stacked and severely blocked, "dead angle" areas are easily generated, affecting the identification accuracy, and it cannot adapt to dynamic working environments and cannot be used for real-time detection with construction equipment. SUMMARY

[0005] To solve the above problems, the present application provides a rock block size laser indicating device.

[0006] The present application adopts the following technical solution: a rock block size laser indicating device, comprising:

[0007] A laser projector is configured with L groups, L≥3, and is installed at different positions of the mechanical arm of the breaking hammer.

[0008] A collection module is used to obtain grid deformation size data in cooperation with the laser projector.

[0009] A central processing unit is used to obtain grid deformation size data and analyze and process the grid deformation size data to obtain rock block size indicators. The central processing unit comprises:

[0010] a geometric measurement module, which calibrates and registers the deformation size data of each angle grid using an internal and external parameter calibration method to obtain a geometric contour model of the rock pile;

[0011] a segmentation module, which performs spatial clustering processing on the geometric contour model of the rock pile to segment the geometric contour model of the rock pile into i rock blocks;

[0012] a rock marking module, which performs occlusion rate analysis on the i rock blocks, predefines an occlusion rate threshold, removes the rock blocks with an occlusion rate greater than the threshold, and marks the remaining rock blocks as target rock blocks;

[0013] a block size prediction module, which performs block size identification on the target rock blocks, predefines a block size threshold, and obtains the number and size of the rock blocks exceeding the block size threshold.

[0014] As a further description of the above technical solutions, the method of performing occlusion rate analysis on the i rock blocks, predefining an occlusion rate threshold, and removing the rock blocks with an occlusion rate greater than the threshold includes:

[0015] performing occlusion analysis on each rock block spatial coordinate point set, and calculating the occlusion rate by using the ratio of visible surface points to theoretical envelope area;

[0016] when the occlusion rate is greater than the preset occlusion rate threshold, removing the rock block;

[0017] when the occlusion rate is less than or equal to the preset occlusion rate threshold, retaining the rock block and marking it as a target rock block.

[0018] As a further description of the above technical solutions, the method of performing occlusion analysis on each rock block spatial coordinate point set and calculating the occlusion rate by using the ratio of visible surface points to theoretical envelope area includes:

[0019] filtering the rock block spatial coordinate point set to remove noise and abnormal isolated points;

[0020] performing wrapping on the target rock block using a geometric envelope algorithm, and constructing a model of multiple triangular facets for representing the complete theoretical shell of the rock block;

[0021] summing up the areas of all triangular facets in the complete theoretical shell of the rock block to obtain the theoretical complete surface area of the rock block;

[0022] performing normal vector estimation on all points in the rock block spatial coordinate point set, and screening out visible surface points based on a preset screening condition to obtain a visible point set of the rock block, wherein the screening condition is to obtain a normal vector of a point outward by line-of-sight projection detection, and there is no other occlusion point nearby or to obtain a region point located at the edge of the point cloud by curvature detection;

[0023] The visible point set is input into the surface reconstruction algorithm, and the visible surface constructed is a triangular mesh model. The areas of all triangular facets are added together to obtain the visible surface area.

[0024] The occlusion ratio is calculated using the theoretical complete surface area and visible surface area of ​​the rock block.

[0025] As a further description of the above technical solution: the method for obtaining the rock pile geometric contour model includes:

[0026] Step S01: Using a standard checkerboard pattern calibration plate, collect grid deformation size data at multiple known angles, extract the checkerboard corner points in the grid deformation size data, and use the Zhang Zhengyou calibration algorithm to calculate the focal length, principal point position, and distortion coefficient of each camera to obtain the intrinsic parameter matrix of each camera;

[0027] Step S02: Using the same feature points appearing in the fields of view of different cameras using a standard checkerboard pattern calibration plate, the rotation matrix R and translation vector T of each camera relative to the world coordinate system or the main camera are solved to complete the extrinsic calibration of all cameras;

[0028] Step S03: A laser projector emits a regular orthogonal grid of light onto the surface of the rock pile. Due to the unevenness of the rock surface, the grid is distorted in the image. Multiple industrial cameras synchronously collect grid deformation size data at the current angle.

[0029] Step S04: filtering, edge detection, and sub-pixel fitting are performed on the grid deformation size data to extract the position coordinates of each deformed grid line and grid intersections in the grid deformation size data as feature points;

[0030] Step S05: Using the triangulation geometric relationship between each industrial camera and laser projector, the pixel points corresponding to each grid intersection are back-projected into three-dimensional coordinate points, the intrinsic parameter matrix and extrinsic parameter calibration are obtained, and the intersection point with the laser plane is calculated using the three-dimensional reconstruction formula to obtain the three-dimensional coordinates of the corresponding point in the actual space. These are recorded as the target three-dimensional coordinate points. Steps S03-S05 are repeated to obtain the dense point cloud data under the current viewing angle.

[0031] Step S06: Align the point cloud data from multiple perspectives (perspectives captured by multiple industrial cameras) into the same coordinate system to obtain a complete and continuous rock pile geometric contour model.

[0032] As a further description of the above technical solution: the method of aligning point cloud data from multiple perspectives into the same coordinate system to obtain a complete and continuous rock pile geometric contour model includes:

[0033] Step S061: using the rotation matrix R and translation vector T obtained in the camera calibration process, performing a rigid transformation on each local point cloud under each view to convert it to a unified global coordinate system;

[0034] Step S062: selecting the point cloud data seen by the front industrial camera as the main point cloud, calculating the nearest neighbor point pair set for each other sub-point cloud data, calculating the rigid transformation matrix by minimizing the Euclidean distance error, updating the source point cloud position by applying the rigid transformation matrix, and repeating the iteration until the maximum iteration number is reached;

[0035] Step S063: after all sub-point cloud data is registered with the main point cloud, merging all point cloud sets, removing redundant points in overlapping areas using weighted fusion, and fusing into a continuous point cloud, and performing triangular mesh reconstruction on the fused point cloud to obtain a rock pile geometry contour model.

[0036] As a further description of the above technical solution: the method of removing redundant points in overlapping areas using weighted fusion and fusing into a continuous point cloud comprises:

[0037] Spatially matching the registered point cloud data, and for point pairs with a spatial distance less than a set threshold, judging them as overlapping point pairs in the same area;

[0038] For each overlapping point pair, obtaining the distance weight and normal consistency of the two adjacent overlapping points, and performing weighted summation to obtain a confidence coefficient, and based on the confidence coefficient, removing and fusing the overlapping points.

[0039] As a further description of the above technical solution: the method of removing and fusing the overlapping points based on the confidence coefficient comprises: presetting a confidence coefficient threshold, and when the obtained confidence coefficient is less than the preset confidence coefficient threshold, the overlapping point pair is removed, otherwise the overlapping point pair is retained;

[0040] For the retained overlapping point pair, using the average coordinates of the two overlapping points to synthesize a representative point.

[0041] As a further description of the above technical solution: the method of identifying the size of the target rock block comprises:

[0042] Extracting feature parameters of the target rock block, the feature parameters including minimum circumscribed cuboid size, occlusion rate, average curvature, and rock type;

[0043] Inputting the collected feature parameters into the trained rock block size prediction model to output the rock block size parameter.

[0044] As a further description of the above technical solution: the training method of the rock block size prediction model comprises:

[0045] The broken rock blocks are taken and photographed by using image recognition technology, and different sizes of rock block contours are recognized by a special image analysis software to obtain rock block size parameters, and the rock block size parameters accurately measured are taken as output labels of the model, and a mapping relationship between the minimum circumscribed cuboid size, the occlusion rate, the average curvature, the rock type and the rock block size is established as training data.

[0046] Check whether there are missing values and abnormal values in the collected training data, for missing values, mean, median or interpolation method can be used for filling, and for abnormal values, identification and elimination are performed, and the minimum circumscribed cuboid size, the occlusion rate and the average curvature feature parameters collected are standardized.

[0047] A random forest regression model in machine learning is selected.

[0048] The preprocessed data is divided into a training set, a validation set and a test set according to a preset ratio, the training set is used for parameter learning of the model, and the validation set is used for adjusting the hyperparameters of the model during the training process.

[0049] The training set data is used to train the random forest regression model, the hyperparameters of the model are adjusted during the training process by using the validation set, and the optimized model is finally tested by using the test set.

[0050] As a further description of the above technical solution: the method for obtaining the minimum circumscribed cuboid size comprises:

[0051] For each rock block spatial coordinate point set, a covariance matrix of the point cloud is constructed.

[0052] The eigenvalues and corresponding eigenvectors of the covariance matrix are calculated.

[0053] The three eigenvectors are taken as three orthogonal directions of the rock point cloud in space, i.e. the principal axis, the secondary axis and the minimum axis direction, and the spatial orientation of the subsequent minimum circumscribed cuboid is determined through the three orthogonal directions.

[0054] A local coordinate system is constructed with the three orthogonal directions as the reference, and the rock point cloud is projected into the local coordinate system, so that the principal axis direction is aligned with the X axis, the secondary axis direction is aligned with the Y axis, and the minimum axis direction is aligned with the Z axis.

[0055] In the transformed coordinate system, the projection extreme values of the rock block spatial coordinate point set in the three axial directions are extracted, and the length, width and height dimensions of the minimum circumscribed cuboid are calculated.

[0056] Advantages:

[0057] In the above technical solution, by synchronously projecting laser grid light in multiple directions and collecting grid deformation size data from different viewing angles, the complementarity of different direction viewing angles can be realized, thereby significantly reducing the "dead angle" and information missing problem caused by mutual stacking and shielding between rocks, improving the coverage rate and reconstruction integrity of the overall point cloud, and through multi-view point cloud splicing and fusion, the point cloud model has higher completeness and richer edge information, which is beneficial to subsequent processing steps such as block size recognition, shielding rate analysis and visible point screening based on machine learning model, and compared with single ranging or image recognition, the scheme has stronger robustness and recognition stability under complex lighting, dust interference, irregular rock shape and other actual working conditions of the open-pit mine.

[0058] Further, the weighted fusion (based on distance weight, normal consistency) is used to remove redundant points in the overlapping area. In the point cloud fusion stage after multi-view registration, the point cloud data in the overlapping area is de-redundant, purified, and weighted fused to construct a more real and clean rock pile geometry contour model.

[0059] Finally, by calculating the shielding rate of each rock block and eliminating rock blocks with high shielding rate, the block size judgment error caused by shielding is avoided. Especially in the breaking process, the rock accumulation is complex, and part of the rock blocks may be shielded. Through this method, the judgment result can be effectively avoided by the shielded rock block. Through automatic screening of target rock blocks with less shielding, the processing efficiency and judgment accuracy are improved, human intervention is reduced, and more accurate reference is provided for subsequent breaking operation. By presetting the shielding rate threshold, the system can quickly judge and eliminate invalid point cloud data, reduce the calculation burden, and improve the overall efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0060] The present application will be further explained in conjunction with the accompanying drawings and embodiments:

[0061] Figure 1 A structural schematic diagram of a rock block size laser indicating device provided for an embodiment of the present application;

[0062] Figure 2 A flowchart of a method for calculating the shielding rate provided for an embodiment of the present application;

[0063] Figure 3 A flowchart of a method for obtaining a rock pile geometry contour model provided for an embodiment of the present application;

[0064] Figure 4 A working timing diagram of a rock block size laser indicating device provided for an embodiment of the present application;

[0065] Figure 5 A working flowchart of a rock block size laser indicating device provided for an embodiment of the present application. DETAILED DESCRIPTION

[0066] In order to make the technical means, creative features, purposes and effects of the present application easy to understand, the present application will be further described below in conjunction with specific drawings. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0067] Embodiment 1

[0068] Please refer to Figures 1-5 The embodiment of the present application provides a technical solution: a rock lumpness laser indication device;

[0069] The laser projector comprises at least L groups, L≥3, and is respectively installed at different positions of the mechanical arm of the breaking hammer; for example, such as the front, the left front side, the right front side and the bottom;

[0070] The acquisition module is configured to cooperate with the laser projector to obtain grid deformation size data;

[0071] In the embodiment, the acquisition module is an industrial camera, which is arranged on one side of each laser projector and is configured to record the grid deformation size data projected on the rock surface; the industrial camera captures the distorted grid deformation size data of the grid on the rock surface, and forms multiple perspective grid deformation size data at different angles; wherein the grid deformation size data is a grid deformation image;

[0072] The central processing unit is configured to obtain the image collected by the industrial camera and analyze and process the image to obtain a rock lumpness index;

[0073] The central processing unit comprises:

[0074] The geometric measurement module uses an internal and external parameter calibration method to perform geometric calibration and registration on the images at different angles, uses the deformation information of the projected grid to extract a set of spatial coordinate points of the surface profile of the rock pile, and fuses and splices the point cloud data at multiple perspectives in space to obtain a complete geometric profile model of the rock pile.

[0075] In the embodiment, the laser grid light is synchronously projected in multiple directions, and the deformed images are collected from different perspectives, which can realize the complementation of different direction perspectives, thereby significantly reducing the "dead angle" and information missing problem caused by the mutual stacking and shielding of rocks, improving the coverage rate and reconstruction integrity of the overall point cloud, and obtaining a point cloud model with higher completeness and richer edge information, which is beneficial to subsequent processing steps such as AI-based lumpness identification, shielding rate analysis and visible point screening, and compared with the single ranging or image recognition mode, the scheme has stronger robustness and recognition stability under the actual working conditions of the open-pit mine such as complex lighting, dust interference and irregular rock shape.

[0076] Embodiment 2

[0077] Please refer to Figure 1 and Figure 3 , the embodiment of the present application provides a technical scheme: the method for obtaining the geometric contour model of the rock pile comprises:

[0078] Step S01: using a standard checkerboard pattern calibration board, pictures are collected at multiple known angles, the corner points of the checkerboard in the image are extracted, the focal length, principal point position and distortion coefficient of each camera are calculated using Zhang Zhengyou calibration algorithm, and the intrinsic matrix of each camera is obtained;

[0079] Step S02: using the same feature points appearing in different camera fields of view of the standard checkerboard pattern calibration board, the rotation matrix R and translation vector T of each camera relative to the world coordinate system or the main camera are solved, and the extrinsic calibration of all cameras is completed;

[0080] Step S03: the laser projector emits a regular orthogonal grid light, which is projected on the surface of the rock pile. Due to the unevenness of the rock pile surface, the grid produces distortion in the image, and multiple industrial cameras synchronously collect the grid deformation image at the current angle;

[0081] Step S04: filtering, edge detection and sub-pixel fitting processing are performed on the grid deformation image, the position coordinates of each deformed grid line are extracted, and the grid intersection points in the grid deformation image are extracted as feature points;

[0082] Step S05: using the triangulation geometric relationship of each industrial camera-laser projector, the pixel points corresponding to each grid intersection point are back projected into three-dimensional coordinate points, the intrinsic matrix and extrinsic calibration are obtained, the three-dimensional reconstruction formula is used, and the intersection points with the laser plane are solved to obtain the three-dimensional coordinates of the corresponding points in the actual space, which are recorded as target three-dimensional coordinate points. Repeat steps S03-S05 to obtain the dense point cloud data at the current viewing angle;

[0083] Wherein, the three-dimensional reconstruction formula is, , wherein, is the target three-dimensional coordinate point, is the intrinsic matrix of the camera, is the scaling factor or depth scale, , is the two-dimensional coordinate of the pixel point corresponding to the grid intersection point;

[0084] Step S06: aligning the point cloud data from multiple viewing angles (the viewing angles collected by multiple industrial cameras) to the same coordinate system to obtain a complete and continuous geometric contour model of the rock pile.

[0085] The method for aligning the point cloud data from multiple perspectives (perspectives collected by multiple industrial cameras) into the same coordinate system to obtain a complete and continuous rock pile geometry profile model comprises the following steps:

[0086] Step S061: using the rotation matrix and the translation vector obtained in the camera calibration process , performing a rigid transformation on the local point cloud u in each perspective to convert it to a unified global coordinate system;

[0087] The calculation formula of the rigid transformation is: ; in the formula, is the point cloud after the rigid transformation, is the rotation matrix, is the translation vector;

[0088] Step S062: selecting the point cloud data seen by the industrial camera in the front direction as the main point cloud, calculating a nearest neighbor point pair set for each other sub-point cloud data, calculating a rigid transformation matrix by minimizing the Euclidean distance error, updating the source point cloud position by applying the rigid transformation matrix, and repeating the iteration until the maximum number of iterations is reached;

[0089] Step S063: after the registration of all sub-point cloud data and the main point cloud is completed, merging all point cloud sets, removing redundant points in the overlapping area by weighted fusion (such as distance weight, normal consistency, etc.), and fusing into a continuous point cloud, and performing triangular mesh reconstruction (such as Poisson reconstruction, Ball-Pivoting algorithm) on the fused point cloud to obtain a rock pile geometry profile model.

[0090] The method for removing redundant points in the overlapping area by weighted fusion comprises the following steps:

[0091] Obtaining a distance weight and a normal consistency of adjacent two overlapping points, performing weighted summation to obtain a confidence coefficient, presetting a confidence coefficient threshold, when the obtained confidence coefficient is less than the preset confidence coefficient threshold, removing the overlapping point pair, and otherwise, retaining the overlapping point pair.

[0092] In this embodiment, the weighted fusion (based on distance weight and normal consistency) is used to remove the redundant points in the overlapping area, and in the point cloud fusion stage after the multi-perspective registration, the point cloud data in the overlapping area is de-redundant, purified, and weighted fused to construct a more real and clean rock pile geometry profile model.

[0093] The method for removing redundant points in the overlapping area by weighted fusion and fusing into a continuous point cloud comprises the following steps:

[0094] Performing spatial adjacent point matching on the registered point cloud data, and judging that a point pair with a spatial distance less than a set threshold (for example, <3 cm) is an overlapping point pair in the same area; ​

[0095] For each overlapping point pair, the distance weight and the normal consistency of the two adjacent overlapping points are weighted and summed to obtain a confidence coefficient, a preset confidence coefficient threshold is set, when the obtained confidence coefficient is less than the preset confidence coefficient threshold, the overlapping point pair is removed, otherwise the overlapping point pair is retained.

[0096] Preferably, the weighted fusion calculation formula is: ;

[0097] wherein, is the confidence coefficient, , is the distance of the corresponding point from the reference view camera, the closer point is more reliable, represents the distance-based weight; is the cosine of the angle between the normal vectors of the two overlapping points, used to represent the normal consistency.

[0098] =cos(θ), the cosine of the angle between the normal vectors of the two overlapping points is calculated, the closer the cosine value is to 1, the more consistent the surface of the overlapping points is, and the higher the reliability is; wherein , represents the normal vector of the two overlapping points, and θ is the angle between the normal vectors of the two overlapping points.

[0099] When cos(θ)=0 degrees (the directions are completely consistent), =1;

[0100] When cos(θ)=90 degrees (the directions are orthogonal), =0;

[0101] When cos(θ)=180 degrees (the directions are completely opposite), =-1;

[0102] , is a weight coefficient, satisfying + =1, preferably, =0.6, =0.4.

[0103] It should be noted that the size of the weight coefficient is to obtain a specific numerical value by quantifying each data, which is convenient for subsequent comparison. The size of the weight coefficient depends on the number of comprehensive parameters and the corresponding weight coefficient preliminarily set by the person skilled in the art for each group of comprehensive parameters.

[0104] Specifically, the higher the normal consistency (i.e. The weight is close to 1, which means that the point repeatedly appears in multiple views and has a consistent direction, and the reliability is high, and the point should be preferentially reserved; on the contrary, it indicates that the point may be at the edge of the occlusion, a discontinuous surface or a mismatch region, and should be removed or given a lower weight; using the weight for point cloud fusion helps to construct a smoother, more accurate and real surface model.

[0105] For the reserved overlapping point pairs, an average coordinate of the two overlapping points is used to synthesize a representative point.

[0106] The segmentation module performs spatial clustering processing on the rock pile geometric contour model, and divides the whole rock pile geometric contour model into i rock blocks.

[0107] Examples:

[0108] For example, the spatial clustering processing on the geometric contour model generates spatial coordinate point set 1, spatial coordinate point set 2, …, and spatial coordinate point set s; wherein the spatial coordinate point set 1 is the point of rock block 1; the spatial coordinate point set 2 is the point of rock block 2; and the spatial coordinate point set s is the point of rock block s.

[0109] The spatial clustering processing method includes DBSCAN algorithm, Region Growing algorithm and Euclidean clustering algorithm, wherein the DBSCAN algorithm is based on the spatial density clustering of points, automatically removes noise, the Region Growing algorithm is based on curvature and normal continuity from seed points, and the Euclidean clustering algorithm is based on the Euclidean distance between points.

[0110] The rock marking module performs occlusion rate analysis on the i rock blocks, predefines an occlusion rate threshold, removes the rock blocks with an occlusion rate greater than the threshold, and marks the remaining rock blocks as target rock blocks.

[0111] The method for performing occlusion rate analysis on the i rock blocks, predefining an occlusion rate threshold, and removing the rock blocks with an occlusion rate greater than the threshold includes:

[0112] The occlusion analysis is performed on each rock block spatial coordinate point set, and the occlusion rate is calculated by using the ratio of visible surface points to theoretical envelope area;

[0113] When the occlusion rate is greater than the preset occlusion rate threshold, the rock block is automatically removed;

[0114] When the occlusion rate is less than or equal to the preset occlusion rate threshold, the rock block is retained and marked as a target rock block.

[0115] Example: The occlusion rate threshold is set to 0.8 according to engineering experience or data training results;

[0116] If the occlusion rate is greater than 0.8 (i.e., the visible area is less than 20%), the rock block is removed;

[0117] If the shielding rate is less than or equal to 0.8, the rock block is retained.

[0118] In this embodiment, by calculating the shielding rate of each rock block and removing the rock block with high shielding rate, the error in block size judgment caused by shielding is avoided. Especially in the crushing process, the rock pile is complex, and part of the rock block may be shielded. Through this method, the judgment result can be effectively avoided by the shielded rock block. Through automatic screening of target rock blocks with less shielding, the processing efficiency and judgment accuracy are improved, human intervention is reduced, and more accurate reference is provided for subsequent crushing operations. By presetting the shielding rate threshold, the system can quickly judge and remove invalid point cloud data, reduce the calculation burden, and improve the overall efficiency.

[0119] The rock block space coordinate point set is the point cloud data corresponding to an independent rock block.

[0120] As shown in Figure 2 The shielding analysis is performed on each rock block space coordinate point set, and the method for calculating the shielding rate by using the ratio of visible surface points to theoretical envelope area includes:

[0121] The rock block space coordinate point set is filtered to remove noise and abnormal isolated points.

[0122] The target rock block is wrapped using a geometric envelope algorithm, and the model constructed is a multi-face triangular patch, which is used to represent the complete theoretical shell of the rock.

[0123] It should be noted that the geometric envelope algorithm includes a Convex Hull algorithm (convex hull algorithm) and an Alpha Shape algorithm (alpha shape reconstruction). If the rock surface is regular, the Convex Hull algorithm is used, and if the rock surface is uneven or complex in shape, the Alpha Shape algorithm is used.

[0124] The areas of all triangular patches in the complete theoretical shell of the rock are summed to obtain the theoretical complete surface area Af of the rock block.

[0125] The calculation formula of the theoretical complete surface area Af is: ; in the formula is the area of the kth triangular patch, and n represents the number of triangular patches.

[0126] The normal vector of all points in the rock block space coordinate point set is estimated, and based on a preset screening condition, the visible surface points are screened to obtain the visible point set of the rock block.

[0127] The screening condition includes: through line-of-sight projection detection, the normal vector of the point is outward, and there is no other shielding point nearby or through curvature detection, the area point located at the edge of the point cloud is obtained.

[0128] Input the visible point set into the surface reconstruction algorithm (the surface reconstruction algorithm includes Greedy Projection Triangulation, Poisson Surface Reconstruction), and construct the visible surface as a triangular mesh model. Add up the areas of all triangular facets to obtain the visible surface area Av;

[0129] Calculate the occlusion rate by the theoretical complete surface area Af of the rock block and the visible surface area Av;

[0130] The occlusion rate calculation formula is: occlusion rate = 1-Av / Af.

[0131] The block prediction module is configured to identify the block size of the target rock block, preset a block size threshold, and obtain the number and size of rock blocks exceeding the block size threshold.

[0132] The method for identifying the block size of the target rock block comprises:

[0133] Extract the feature parameters of the target rock block, including the minimum circumscribed cuboid size, the occlusion rate, the average curvature, and the rock type.

[0134] Input the collected feature parameters into the trained rock block size prediction model, and output the rock block size parameter.

[0135] The training method of the rock block size prediction model comprises:

[0136] Using image recognition technology, take the broken rock blocks and take photos, identify the outlines of rock blocks of different sizes through special image analysis software, and then obtain the rock block size parameter. Accurately measure the rock block size parameter as the output label of the model, so that the model can learn the input features and establish an accurate mapping relationship between the minimum circumscribed cuboid size, the occlusion rate, the average curvature, the rock type and the rock block size, as training data.

[0137] Check whether there are missing values and abnormal values in the collected training data. For missing values, mean, median or interpolation method can be used for filling. For abnormal values, statistical analysis methods (such as 3σ principle) can be used for identification and elimination. Standardize the feature parameters such as the minimum circumscribed cuboid size, the occlusion rate and the average curvature, such as using Z-score standardization to convert the data to a distribution with a mean of 0 and a standard deviation of 1.

[0138] Data cleaning can improve data quality and avoid adverse effects of missing values and abnormal values on model training. Data standardization can eliminate the dimensional differences between different feature parameters, so that the model can pay equal attention to each feature during training, speed up the model convergence speed, and improve the training stability.

[0139] A random forest regression model in machine learning is selected. There is a complex nonlinear relationship between rock block size and input features. The random forest regression model is composed of multiple decision trees. By averaging the prediction results of multiple decision trees, the nonlinear problem can be effectively handled, and the generalization ability and noise resistance are good. Compared with a single decision tree, the random forest can avoid overfitting, and is very suitable for rock block size prediction which needs to consider multiple features. In the construction process of the decision tree, the sample is divided step by step according to the input features. The random forest can more comprehensively learn the complex relationship between the features and the rock block size by integrating a large number of decision trees.

[0140] The preprocessed data is divided into training set, validation set and test set according to a certain proportion (such as 7:2:1). The training set is used for parameter learning of the model, and the validation set is used for adjusting the hyperparameters of the model during training.

[0141] The random forest regression model is trained using the training set data. During training, the model constructs multiple decision trees according to the input features such as the minimum bounding box size, the occlusion rate, the average curvature and the rock type, and the corresponding rock block size parameter label. Each decision tree is trained by randomly selecting part of the features and samples during construction. By calculating indicators such as information gain (or Gini impurity), the best split feature and split point are selected to divide the samples into different child nodes step by step until the specified stop condition (such as reaching the maximum depth or the number of samples in the child node is less than the minimum sample number) is met. With the training, the model continuously adjusts the structure and parameters of the decision tree, so that the prediction results of multiple decision trees gradually approach the true rock block size parameter, improving the prediction accuracy on the training data. The hyperparameters of the model are adjusted during training using the validation set. The optimized model is finally tested using the test set. The MSE and MAE indicators of the model on the test set are calculated. If the test result meets the preset indicators, the model has good generalization ability and can be applied to actual rock engineering to accurately predict the rock block size parameter according to the characteristics of the rock block. If the test result is not ideal, the data preprocessing, model selection, hyperparameter adjustment and other steps need to be rechecked to find out the problem and improve it.

[0142] Embodiment 3

[0143] Please refer to Figures 1-3 The technical scheme provided by the embodiment of the application comprises:

[0144] For each rock block spatial coordinate point set, the three-dimensional distribution direction information is extracted by principal component analysis (PCA). The specific process is as follows:

[0145] The covariance matrix of the point cloud is constructed.

[0146] obtaining eigenvalues and corresponding eigenvectors of the covariance matrix;

[0147] The three eigenvectors are respectively taken as three orthogonal directions of the rock point cloud in space, i.e., a principal axis direction, a secondary axis direction and a minimum axis direction, and the three orthogonal directions are used to determine a spatial orientation of a subsequent minimum circumscribed cuboid;

[0148] A local coordinate system is constructed with the three orthogonal directions as a reference, and the rock point cloud is projected into the local coordinate system so that the principal axis direction is aligned with the X axis, the secondary axis direction is aligned with the Y axis, and the minimum axis direction is aligned with the Z axis. This processing can simplify the boundary calculation of the circumscribed cuboid.

[0149] In the transformed coordinate system, projection extreme values of the rock block spatial coordinate point set in the three axis directions are extracted, and the length, width and height dimensions of the minimum circumscribed cuboid are calculated, which are respectively represented as:

[0150] Length C = Xmax-Xmin;

[0151] Width W = Ymax-Ymin;

[0152] Height H = Zmax-Zmin;

[0153] Wherein, Xmax, Xmin, Ymax, Ymin, Zmax and Zmin are respectively the maximum and minimum coordinate values of the point cloud in the X axis, Y axis and Z axis directions of the transformed coordinate system.

[0154] It should be noted that different types of rocks have large differences in physical properties, for example, granite and limestone differ in texture, hardness, etc. Different origin rocks have different internal structures, and different weathering degrees also cause different fragmentation and blockiness of rocks. Collecting a wide range of samples can enable the model to learn the relationship between rock characteristics and blockiness under various conditions, thereby improving the generalization ability of the model.

[0155] The average curvature calculation method is based on point cloud data, using a surface fitting algorithm such as least squares method to fit a surface, then calculating the curvature of the point cloud data, and statistically averaging the curvature of the entire rock block surface to obtain the average curvature.

[0156] The average curvature reflects the degree of fluctuation and bending of the rock block surface. The rock block with a large surface curvature has an uneven internal stress distribution, and is more likely to produce cracks and fragmentation during the fragmentation process. The block size distribution is also different from that of the rock block with a flat surface. This parameter helps the model understand the internal structure and fragmentation characteristics of the rock block.

[0157] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and the description in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A rock lump laser indicating device, characterized in that, The application relates to a rock blockage prediction method and device. The application comprises: a laser projector configured with L groups, L >= 3, which are respectively installed at different positions of a mechanical arm of a rock breaking hammer; a collection module used for acquiring grid deformation size data in cooperation with the laser projector; a central processing unit used for acquiring the grid deformation size data and analyzing and processing the grid deformation size data to obtain a rock blockage index, wherein the central processing unit comprises: a geometric measurement module which uses an internal and external parameter calibration method to perform geometric calibration and registration on the grid deformation size data at different angles and obtains a rock pile geometric contour model; a segmentation module which performs spatial clustering processing on the rock pile geometric contour model and divides the rock pile geometric contour model into i rock blocks; a rock marking module which performs occlusion rate analysis on the i rock blocks, predefines an occlusion rate threshold, removes the rock blocks with an occlusion rate greater than the threshold, and marks the remaining rock blocks as target rock blocks; a blockage prediction module which performs blockage identification on the target rock blocks, predefines a blockage threshold, and obtains the number and size of the rock blocks exceeding the blockage threshold; the method for removing the rock blocks with an occlusion rate greater than the threshold comprises: performing occlusion analysis on each rock block spatial coordinate point set, and calculating the occlusion rate by using a ratio of visible surface points to theoretical envelope area; when the occlusion rate is greater than the preset occlusion rate threshold, the rock block is removed; when the occlusion rate is less than or equal to the preset occlusion rate threshold, the rock block is retained and marked as a target rock block; the method for calculating the occlusion rate by using a ratio of visible surface points to theoretical envelope area comprises: filtering the rock block spatial coordinate point set to remove noise and abnormal isolated points; performing geometric envelope algorithm on the target rock block to construct a model of a multi-face triangular facet which is used for representing a complete theoretical shell of the rock block; summing up the areas of all triangular facets in the complete theoretical shell of the rock block to obtain a theoretical complete surface area of the rock block; performing normal vector estimation on all points in the rock block spatial coordinate point set, screening out visible surface points based on a preset screening condition to obtain a visible point set of the rock block, wherein the screening condition is that the normal vector of the point is outward and there is no other occlusion point in the vicinity or the point is located in an edge region of the point cloud through curvature detection; inputting the visible point set into a surface reconstruction algorithm to construct a triangular mesh model of a visible surface, and summing up the areas of all triangular facets to obtain a visible surface area; 2. A rock lump size laser indicating device according to claim 1, wherein, calculating the occlusion rate by using the theoretical complete surface area and the visible surface area of the rock block. The method for obtaining the rock pile geometric contour model comprises: step S01: using a standard checkerboard pattern calibration board to collect grid deformation size data at multiple known angles, extracting corner points of the checkerboard in the grid deformation size data, calculating the focal length, principal point position and distortion coefficient of each camera by using Zhang Zhengyou calibration algorithm to obtain the internal parameter matrix of each camera; step S02: using the same feature points of the standard checkerboard pattern calibration board appearing in different camera fields of view to solve the rotation matrix R and the translation vector T of each camera relative to a world coordinate system or a main camera, and completing external parameter calibration of all cameras. Step S03: The laser projector emits a regular orthogonal grid light, which is projected on the surface of the rock pile, and multiple industrial cameras synchronously collect the grid deformation size data under the current angle; Step S04: The grid deformation size data is filtered, edge detected and sub-pixel fitted to extract the position coordinates of each deformed grid line, and the grid intersection points in the grid deformation size data are extracted as feature points; Step S05: Using the triangulation geometric relationship of each industrial camera-laser projector, the pixel points corresponding to each grid intersection point are back-projected into three-dimensional coordinate points, the intrinsic matrix and extrinsic calibration are obtained, the three-dimensional reconstruction formula is used, and the intersection points with the laser plane are obtained to obtain the three-dimensional coordinates of the corresponding points in the actual space, denoted as target three-dimensional coordinate points, and the steps S03-S05 are repeated to obtain the dense point cloud data under the current viewing angle; Step S06: Aligning the point cloud data from multiple viewing angles to the same coordinate system to obtain a geometric contour model of the rock pile.

3. A rock lump size laser indicating device according to claim 2, wherein, The method of aligning the point cloud data from multiple viewing angles to the same coordinate system to obtain a geometric contour model of the rock pile comprises: Step S061: Using the rotation matrix R and translation vector T obtained in the camera calibration process, performing a rigid transformation on the local point cloud under each viewing angle to convert it to a unified global coordinate system; Step S062: Selecting the point cloud data seen by the industrial camera in the front as the main point cloud, calculating the nearest neighbor point pair set for each other sub-point cloud data, calculating the rigid transformation matrix by minimizing the Euclidean distance error, updating the source point cloud position by applying the rigid transformation matrix, and repeating the iteration until the maximum iteration number is reached; Step S063: After the registration of all sub-point cloud data with the main point cloud is completed, merging all point cloud sets, removing redundant points in the overlapping area by weighted fusion, and fusing into a continuous point cloud to obtain a geometric contour model of the rock pile.

4. A rock lump size laser indicating device according to claim 3, wherein, The method of removing redundant points in the overlapping area by weighted fusion and fusing into a continuous point cloud comprises: Performing spatial adjacent point matching on the registered point cloud data, and judging that the point pairs with a spatial distance less than a set threshold are overlapping point pairs in the same area; For each overlapping point pair, the distance weight and normal consistency of the two adjacent overlapping points are weighted and summed to obtain a confidence coefficient, and the overlapping points are removed and fused based on the confidence coefficient.

5. A rock lump size laser indicating device according to claim 4, wherein, The method of removing and fusing the overlapping points based on the confidence coefficient comprises: presetting a confidence coefficient threshold, and when the obtained confidence coefficient is less than the preset confidence coefficient threshold, the overlapping point pair is removed, otherwise the overlapping point pair is retained; For the retained overlapping point pair, an average coordinate of the two overlapping points is used to synthesize a representative point.

6. A rock lump size laser indicating device according to claim 1, wherein, The method of identifying the block degree of the target rock block comprises: Extracting feature parameters of the target rock block, the feature parameters comprising minimum circumscribed cuboid size, occlusion rate, average curvature and rock type; Inputting the collected feature parameters into a trained rock block degree prediction model to output rock block degree parameters.

7. A rock lump size laser indicating device according to claim 6, wherein, The training method of the rock block degree prediction model comprises: The broken rock block is taken and photographed by using image recognition technology, the rock block contour is recognized by using image analysis software, the rock block size parameter is obtained as the output label of the model, and the mapping relationship between the minimum circumscribed cuboid size, the occlusion rate, the average curvature, the rock type and the rock block size is established as the training data; Check if there are missing values and abnormal values in the collected training data, fill in the missing values by using the mean, median or interpolation method, identify and eliminate the abnormal values, and standardize the minimum circumscribed cuboid size, the occlusion rate and the average curvature collected; Select a random forest regression model in machine learning; The preprocessed data is divided into a training set, a validation set and a test set according to a preset proportion, the training set is used for parameter learning of the model, and the validation set is used for adjusting the hyperparameters of the model during the training process; The training set data is used to train the random forest regression model, the hyperparameters of the model are adjusted during the training process by using the validation set, and the optimized model is finally tested by using the test set.

8. A rock lump size laser indicating device according to claim 7, characterised in that, The method for obtaining the minimum circumscribed cuboid size comprises: For each rock block spatial coordinate point set, a covariance matrix of the point cloud is constructed; Eigenvalues and corresponding eigenvectors of the covariance matrix are calculated; Three eigenvectors are respectively taken as three orthogonal directions of the rock point cloud in space, i.e., the principal axis, the secondary axis and the minimum axis direction, and the spatial orientation of the minimum circumscribed cuboid is determined by the three orthogonal directions; A local coordinate system is constructed with the three orthogonal directions as the reference, and the rock point cloud is projected into the local coordinate system, so that the principal axis direction is aligned with the X axis, the secondary axis direction is aligned with the Y axis, and the minimum axis direction is aligned with the Z axis; In the transformed coordinate system, the projection extreme values of the rock block spatial coordinate point set in the three axial directions are extracted, and the length, width and height dimensions of the minimum circumscribed cuboid are calculated.

Images