A method and system for intelligent identification of ore fragmentation based on 3D cloud images

Abstract

The present application relates to the technical field of ore identification, and provides an intelligent identification method and system for ore block size based on 3D cloud images, wherein the system includes a controller, a scanning laser transmitter and a three-dimensional camera. The line scanning laser transmitter and the three-dimensional camera are arranged directly above the conveyor belt. In the actual application process, the controller is used to execute the intelligent identification method, including: obtaining a three-dimensional point cloud image of the ore particles and converting it into three-dimensional point cloud coordinates; performing point cloud filtering processing on the three-dimensional point cloud coordinates to obtain filtered point cloud coordinates; performing threshold segmentation on the filtered point cloud coordinates to obtain the target area point cloud coordinates; performing particle segmentation processing on the target area point cloud coordinates to obtain the single particle point cloud coordinates; performing particle size calculation on all single particle point cloud coordinates to obtain particle size distribution information, and the particle size distribution information includes particle size and particle number.

Description

Technical Field

[0001] The present application relates to the technical field of ore identification, and in particular to a method and system for intelligently identifying ore block size based on 3D cloud images. Background Art

[0002] Intelligent perception and identification of ore properties is the basis for realizing the intelligentization of mineral processing production process in the field of intelligent mining. The core development directions of intelligent mineral processing: ore property identification, intelligent flotation equipment and genetic mineral processing technology. The grade of valuable elements, particle size and crushing characteristics of ore are the three most important basic properties among the various physical and chemical properties of ore. These three basic properties determine the theoretical basis and process measures for the separation and sorting of useful components and the rational or comprehensive utilization of products. In actual production, the fluctuations of the three directly affect the stability of mineral processing production and the economic benefits, which is of great significance to the efficient and intelligent control of mineral processing production.

[0003] Among them, for the particle size identification of ore, with the rapid development of image processing and intelligent control methods, machine vision technology has begun to be studied and applied to the online monitoring of ore particle size information. However, ore images have ore adhesion and cannot be effectively segmented and positioned, resulting in low accuracy of particle size analysis. With the development of technology, commercial systems for online ore particle size measurement have gradually emerged. For example, the patent application number is: CN201810202367.X, and the name is "A method for detecting ore strength based on deep learning". Although it can predict the ore particle size through machine learning, it uses two-dimensional image learning, and the prediction accuracy is low; for example, the name is "Explosion pile block size analysis based on three-dimensional laser point cloud". Although it can identify the size and distribution of ore particles, it focuses more on the identification of large pieces of ore. In the field of sintering technology, the ore particle size is generally below 10mm, and the analysis method of this paper cannot meet the accuracy requirements.

[0004] In summary, in order to accurately identify the particle size and distribution of ore in the field of sintering technology, providing a high-precision ore particle size identification method or system is a technical problem that needs to be urgently solved by technical personnel in this field. Summary of the invention

[0005] In order to accurately identify the particle size and distribution of ores in the field of sintering technology, the present application provides a method and system for intelligently identifying ore block size based on 3D cloud maps.

[0006] On the one hand, an embodiment of the present application provides a method for intelligently identifying ore fragmentation based on a 3D cloud map, comprising:

[0007] A three-dimensional point cloud image of ore particles is obtained by a line scanning laser transmitter and a three-dimensional camera, and the three-dimensional point cloud image is converted into three-dimensional point cloud coordinates;

[0008] Performing point cloud filtering processing on the three-dimensional point cloud coordinates to obtain filtered point cloud coordinates, wherein the filtering processing includes downsampling, removing outliers and repairing missing points;

[0009] Extract the Z coordinate value of the filtered point cloud coordinates, perform K-means clustering segmentation on the Z coordinate value, and obtain K 1 clusters, the K 1 is a positive integer greater than 1;

[0010] Calculate the center point value of each cluster and set K 1 The maximum value of the center point of the cluster is used as the segmentation threshold;

[0011] Remove the point clouds whose Z coordinate values ​​in the filtered point cloud coordinates are greater than the segmentation threshold to obtain the point cloud coordinates of the target area;

[0012] Performing particle segmentation processing on the point cloud coordinates of the target area to obtain the point cloud coordinates of a single particle;

[0013] The particle size is calculated for all single particle point cloud coordinates to obtain particle size distribution information, which includes particle size and particle quantity.

[0014] Optionally, the downsampling includes: creating a plurality of voxel grids in the three-dimensional point cloud coordinate region, wherein the size of the voxel grids is smaller than the resolution of the three-dimensional camera;

[0015] Calculate the center coordinates of all point cloud coordinates within each voxel grid;

[0016] The center coordinates of the voxel grid are used to replace all the point cloud coordinates in the voxel grid to generate downsampled point cloud coordinates.

[0017] Optionally, the outlier point removal includes: traversing the downsampled point cloud coordinates, selecting any point cloud coordinate as the current point, and determining the distance between the current point and the nearest K 2 The average distance of each field point is calculated, the mean and standard deviation of the average distance are calculated, and a proportional coefficient is obtained, where the proportional coefficient is a constant;

[0018] Determine the distance threshold according to the mean and standard deviation of the average distance and the proportionality coefficient;

[0019] If the average distance is greater than the distance threshold, the current point is deleted to obtain the point cloud coordinates without outliers.

[0020] Optionally, the repairing of missing points includes: obtaining a resolution and a spacing variation coefficient of a three-dimensional camera, and calculating a missing point threshold according to the resolution and the spacing variation coefficient of the three-dimensional camera;

[0021] Traverse the point cloud coordinates to remove outliers and calculate the distance between any adjacent points;

[0022] The size of the distance between any adjacent points and the missing point threshold is determined. If the distance between any adjacent points is greater than the missing point threshold, local range surface fitting is performed between the adjacent points, the missing points are added, and the filtered point cloud coordinates are obtained.

[0023] Optionally, after the step of adding the missing point, the method further includes: calculating the distance between the missing point and the adjacent point of the missing point;

[0024] Determine the size of the distance between the adjacent points of the missing point and the missing point and the missing point threshold. If the distance between the adjacent points of the missing point and the missing point is greater than the missing point threshold, perform local range surface fitting between the adjacent points of the missing point and the missing point and add a new missing point.

[0025] Optionally, the size of the voxel grid is 0.2 mm.

[0026] Optionally, the spacing variation coefficient is 2.

[0027] Optionally, the step of performing particle segmentation processing on the target area point cloud coordinates to obtain single particle point cloud coordinates includes:

[0028] Traverse the point cloud coordinates of the target area and use the Kd tree to search for the K of each point cloud 3 field points, through each point cloud K 3 The normal and curvature of each point cloud are calculated for each domain point;

[0029] According to the normal line and the curvature, the point cloud coordinates of the target region are segmented into particles by using the region growing method to obtain the point cloud coordinates of a single particle.

[0030] On the other hand, the embodiment of the present application provides an intelligent identification system for ore block size based on 3D cloud images, including a controller, and a line scanning laser transmitter and a three-dimensional camera connected to the controller;

[0031] The line scanning laser transmitter and the three-dimensional camera are arranged directly above the conveyor belt, the conveyor belt is used to transport block ore, the line scanning laser transmitter is used to emit a line laser perpendicular to the upper surface of the conveyor belt to the conveyor belt; the 3D camera is used to obtain a line laser image irradiated on the conveyor belt and the block ore, and the line laser image is a three-dimensional point cloud image;

[0032] The controller is further configured to perform the following steps:

[0033] A three-dimensional point cloud image of ore particles is obtained by a line scanning laser transmitter and a three-dimensional camera, and the three-dimensional point cloud image is converted into three-dimensional point cloud coordinates;

[0034] Performing point cloud filtering processing on the three-dimensional point cloud coordinates to obtain filtered point cloud coordinates, wherein the filtering processing includes downsampling, removing outliers and repairing missing points;

[0035] Extract the Z coordinate value of the filtered point cloud coordinates, perform K-means clustering segmentation on the Z coordinate value, and obtain K 1 clusters, the K 1 is a positive integer greater than 2;

[0036] Calculate the center point value of each cluster and set K 1 The maximum value of the center point of the cluster is used as the segmentation threshold;

[0037] Remove the point clouds whose Z coordinate values ​​in the filtered point cloud coordinates are greater than the segmentation threshold to obtain the point cloud coordinates of the target area;

[0038] Performing particle segmentation processing on the target area point cloud coordinates to obtain single particle point cloud coordinates, wherein the particle segmentation processing includes supervoxel area clustering, area concavity marking and area particle clustering;

[0039] The particle size is calculated for all single particle point cloud coordinates to obtain particle size distribution information, which includes particle size and particle quantity.

[0040] Optionally, it is characterized in that the three-dimensional camera has a pixel of 5 million and a resolution of 0.3 mm.

[0041] It can be seen from the above scheme that the present application provides a method and system for intelligent identification of ore block size based on 3D cloud map, and the intelligent identification system for ore block size based on 3D cloud map includes a controller, and a line scanning laser transmitter and a three-dimensional camera connected to the controller. The line scanning laser transmitter and the three-dimensional camera are arranged directly above the conveyor belt, and the conveyor belt is used to convey block ore. The line scanning laser transmitter is used to emit a line laser perpendicular to the upper surface of the conveyor belt to the conveyor belt; the 3D camera is a 3D industrial camera with 5 million pixels and a resolution of 0.3mm, which is used to obtain the line laser image irradiated on the conveyor belt and the block ore, and the line laser image is a three-dimensional point cloud image.

[0042] In actual application, the controller is used to execute an intelligent identification method for ore block size based on a 3D cloud map, obtain a 3D point cloud image of ore particles through a line scanning laser transmitter and a 3D camera, and convert the 3D point cloud image into a 3D point cloud coordinate; perform point cloud filtering on the 3D point cloud coordinates to obtain filtered point cloud coordinates, wherein the filtering process includes downsampling, removing outliers and repairing missing points; extract the Z coordinate value of the filtered point cloud coordinates, perform K-means clustering segmentation on the Z coordinate value, and obtain K 1 clusters, the K 1is a positive integer greater than 1; calculate the center point value of each cluster, and K 1 The maximum value of the center point of each cluster is used as the segmentation threshold; the point cloud with a Z coordinate value greater than the segmentation threshold in the filtered point cloud coordinates is removed to obtain the point cloud coordinates of the target area; the point cloud coordinates of the target area are subjected to particle segmentation processing to obtain the point cloud coordinates of a single particle; the particle size is calculated for all the single particle point cloud coordinates to obtain the particle size distribution information, which includes the particle size and the number of particles. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0044] Figure 1 A schematic diagram of the overall structure of an intelligent identification system for ore block size based on a 3D cloud map provided in an embodiment of the present application;

[0045] Figure 2 A schematic diagram of a process flow of an intelligent identification method for ore block size based on a 3D cloud map provided in an embodiment of the present application;

[0046] Figure 3 A schematic diagram of a downsampling process of three-dimensional point cloud coordinates provided in an embodiment of the present application;

[0047] Figure 4 A schematic diagram of a process for removing outliers in downsampling point cloud coordinates provided in an embodiment of the present application;

[0048] Figure 5 A schematic diagram of a process for repairing missing points by removing outliers from point cloud coordinates provided in an embodiment of the present application;

[0049] Figure 6 This is a statistical chart of ore particle size provided in the embodiments of the present application. DETAILED DESCRIPTION

[0050] In order to accurately identify the particle size and distribution of ores in the field of sintering technology, the present application provides a method and system for intelligently identifying ore block size based on 3D cloud maps.

[0051] like Figure 1 As shown, it is a schematic diagram of the overall structure of an intelligent identification system for ore block size based on 3D cloud map provided in an embodiment of the application. The first aspect of an embodiment of the application provides an intelligent identification system for ore block size based on 3D cloud map, including a controller 1, and a line scanning laser transmitter 2 and a three-dimensional camera 3 connected to the controller 1.

[0052] The line scanning laser emitter 2 and the three-dimensional camera 3 are arranged directly above the conveyor belt 4. The conveyor belt 4 is used to transport block ore. The line scanning laser emitter 2 is used to emit a line laser perpendicular to the upper surface of the conveyor belt to the conveyor belt; the 3D camera 3 is a 3D industrial camera with 5 million pixels and a resolution of 0.3mm. It is used to obtain the line laser image irradiated on the conveyor belt and the block ore. The line laser image is a three-dimensional point cloud image.

[0053] The controller 1 is also configured to execute a method for intelligently identifying ore block size based on a 3D cloud map provided in the second aspect of the present application, such as Figure 2 As shown, it is a schematic flow chart of a method for intelligently identifying ore fragmentation based on 3D cloud images provided in an embodiment of the present application, and the method for intelligently identifying ore fragmentation includes steps S201 to S211.

[0054] The controller 1 obtains a three-dimensional point cloud image of the ore particles through a line scanning laser transmitter and a three-dimensional camera, and converts the three-dimensional point cloud image into three-dimensional point cloud coordinates. The specific process is step S201 to step S205.

[0055] Step S201, emitting a line laser perpendicular to the upper surface of the conveyor belt toward the conveyor belt.

[0056] Step S202, obtaining a line laser image irradiated on the conveyor belt and the block ore, wherein the line laser image is a three-dimensional point cloud image.

[0057] The line laser transmitter 2 emits a line laser perpendicular to the upper surface of the conveyor belt 4, and the three-dimensional camera 3 captures the line laser image irradiated on the conveyor belt 4 and the block ore. As the conveyor belt 4 moves, the three-dimensional camera 3 can obtain a complete three-dimensional point cloud image of the conveyor belt 4 and the block ore.

[0058] Step S203, sending the three-dimensional point cloud image.

[0059] Step S204, receiving a three-dimensional point cloud image.

[0060] Step S205, converting the three-dimensional point cloud image into three-dimensional point cloud coordinates.

[0061] The controller 1 receives the 3D point cloud image sent by the 3D camera 3, and converts the 3D point cloud image into 3D point cloud coordinates, that is, a point cloud in (x, y, z) format that can be directly processed by the controller.

[0062] Step S206: performing point cloud filtering processing on the three-dimensional point cloud coordinates to obtain filtered point cloud coordinates.

[0063] The filtering process includes downsampling, removing outliers and repairing missing points.

[0064] like Figure 3 As shown, a schematic diagram of the downsampling process of three-dimensional point cloud coordinates provided in an embodiment of the present application is shown.

[0065] Step S301 : creating a plurality of voxel grids in the three-dimensional point cloud coordinate region, wherein the size of the voxel grids is smaller than the resolution of the three-dimensional camera.

[0066] That is, the three-dimensional point cloud coordinate area is divided into a plurality of small squares, and the side length of each small square needs to be smaller than the resolution of the three-dimensional camera 3. In the embodiment of the present application, the side length of the voxel grid is set to 0.2 mm.

[0067] Step S302, calculating the center coordinates of all point cloud coordinates in each voxel grid.

[0068] Step S303: All point cloud coordinates in the voxel grid are replaced by the center coordinates of the voxel grid to generate downsampled point cloud coordinates.

[0069] All point cloud coordinates in each voxel grid are counted to calculate the center coordinates of a voxel grid area. The obtained center coordinates are used to replace all point cloud coordinates in the entire voxel grid area. By downsampling, the point volume in the three-dimensional point cloud coordinate area can be reduced without changing the ore characteristics, thereby reducing the amount of calculation for subsequent processing.

[0070] like Figure 4 As shown, a schematic diagram of the process of removing outliers in downsampling point cloud coordinates provided in an embodiment of the present application is provided.

[0071] Step S401, traverse the downsampled point cloud coordinates, select any point cloud coordinate as the current point, and determine the distance between the current point and the nearest K 2 The average distance of the field points is calculated, the mean and standard deviation of the average distance are calculated, and the proportionality coefficient is obtained.

[0072] The proportional coefficient is a constant, such as 0.5 or 1, which is used to control the degree of point cloud removal. The smaller the proportional coefficient is, the greater the amount of point cloud removed is. It can be set according to actual engineering requirements.

[0073] Step S402: determining a distance threshold according to the mean and standard deviation of the average distance and the proportionality coefficient.

[0074] The distance threshold is determined according to the following model:

[0075] D1=U+S×a1

[0076] Among them, D1 is the distance threshold, U is the mean of the average distance, S is the standard deviation of the average distance, and a1 is the proportional coefficient.

[0077] Step S403: determine whether the average distance is greater than a distance threshold.

[0078] Step S404: if the average distance is greater than the distance threshold, the current point is deleted to obtain the point cloud coordinates after removing the outlier points.

[0079] If the average distance is greater than the distance threshold and less than or equal to the distance threshold, then return to step S401, continue to select any point cloud coordinates as the current point, until the point clouds in all downsampled point cloud coordinate areas are traversed, and the loop process ends. The downsampled point cloud coordinates are statistically filtered by removing outliers, and the noise point cloud and outlier point cloud are filtered by querying the distance between the point cloud and the adjacent points to obtain the point cloud coordinates after removing outliers.

[0080] like Figure 5 As shown, a schematic diagram of the process of repairing missing points of point cloud coordinates by removing outliers provided in an embodiment of the present application.

[0081] Step S501: Obtain the resolution and spacing variation coefficient of the three-dimensional camera, and calculate the missing point threshold according to the resolution and spacing variation coefficient of the three-dimensional camera.

[0082] The missing point threshold is a2×C, where a2 is the spacing variation coefficient and C is the resolution of the three-dimensional camera. In the embodiment of the present application, the spacing variation coefficient a2=2.

[0083] Step S502, traverse the point cloud coordinates after removing outliers, and calculate the distance between any adjacent points.

[0084] Step S503, determine the size of the distance between any adjacent points and the missing point threshold. If the distance between any adjacent points is greater than the missing point threshold, perform local range surface fitting between the adjacent points, add the missing points, and obtain the filtered point cloud coordinates.

[0085] That is, when the distance between any adjacent points is greater than twice the resolution of the 3D camera, it means that there is a missing point between the two adjacent points, and the missing point is added through local range surface fitting. It should be noted that when there is a missing point between two adjacent points, there may be more than one missing point, so after adding the missing point, the added missing point will be repaired twice. The secondary repair process includes: calculating the distance D2 between the adjacent point of the missing point and the missing point; judging the size of the distance D2 between the adjacent point of the missing point and the missing point and the missing point threshold a2×C. If the distance between the adjacent point of the missing point and the missing point is greater than the missing point threshold D2>a2×C, then performing local range surface fitting between the adjacent point of the missing point and the missing point, and adding a new missing point.

[0086] When obtaining the filtered point cloud coordinates, since the upper surface of the conveyor belt 4 is inevitably captured in the image captured by the three-dimensional camera 3, in order to remove the image interference of the conveyor belt 4, it is necessary to remove the point cloud data of the conveyor belt 4 from the filtered point cloud coordinates, and the removal process is step S207 to step S209.

[0087] Since the data acquired by the line scanning laser transmitter 2 and the three-dimensional camera 3 include the point cloud of the conveyor belt surface, the ore area and the conveyor belt 4 area need to be segmented before particle segmentation, and the segmentation method is steps S207 to S209.

[0088] Step S207: extract the Z coordinate value of the filtered point cloud coordinates, perform K-means clustering segmentation on the Z coordinate value, and obtain K 1 Clusters, K 1 is a positive integer greater than 1.

[0089] Since the block ore is on the upper surface of the conveyor belt 4, it is only necessary to ensure that the number of clusters is greater than 1, that is, at least 2 clusters, to remove the point cloud of the conveyor belt 4, but K 1 The smaller the value, the more point clouds containing blocky ores are removed, so the value of K1 is usually set larger. 1 A larger value will increase the amount of calculation. In actual projects, a relatively reasonable value will be set according to project needs.

[0090] Step S208, calculate the center point value of each cluster, and set K 1 The maximum value of the center point of the clusters is used as the segmentation threshold.

[0091] Among them, the cluster with the largest center point value is the cluster that includes the image point cloud of conveyor belt 4, so the value with the largest center point value is used as the segmentation threshold.

[0092] Step S209, removing the point clouds whose Z coordinate values ​​are greater than the segmentation threshold in the filtered point cloud coordinates to obtain the point cloud coordinates of the target area.

[0093] By removing the point cloud whose Z coordinate value in the filtered point cloud coordinates is greater than the segmentation threshold, it is ensured that the image point cloud of the conveyor belt 4 can be removed. However, this is not limited to the removal of the point cloud whose Z coordinate value is greater than the segmentation threshold. When the accuracy requirement is not high, it is also possible to consider directly removing the cluster with the largest center point value. By removing the entire cluster, it is also possible to remove the image point cloud of the conveyor belt 4. This method can ensure a smaller amount of data calculation, which is beneficial to improving the recognition speed of the ore block size intelligent recognition method.

[0094] Step S210, performing particle segmentation processing on the point cloud coordinates of the target area to obtain the point cloud coordinates of a single particle.

[0095] The step of performing particle segmentation processing on the point cloud coordinates of the target area to obtain the point cloud coordinates of a single particle comprises:

[0096] Traverse the point cloud coordinates of the target area and use the Kd tree to search for the K of each point cloud 3 field points, through each point cloud K 3 The normal and curvature of each point cloud are calculated for each domain point.

[0097] According to the normal line and the curvature, the point cloud coordinates of the target region are segmented into particles by using the region growing method to obtain the point cloud coordinates of a single particle.

[0098] According to the normal and curvature of each point cloud, the region growing algorithm is used to cluster and segment individual particles. The specific process is as follows:

[0099] Step 1: Determine the initial point cloud, total clustering clusters, angle threshold, and curvature threshold.

[0100] Among them, the initial point cloud is the point cloud coordinates of the target area, the initial state of the total clustering cluster is an empty cluster, the angle threshold and the curvature threshold are determined by the concavity and convexity of the block ore produced under actual working conditions, and the angle threshold provided in the embodiment of the present application is 11 degrees and the curvature threshold is 0.1.

[0101] Step 2: Generate an empty current seed group, select the point with the smallest curvature from the initial point cloud as the initial growth point, add it to the current seed group, and remove the point from the initial point cloud.

[0102] Step 3: Traverse the current seed group and remove a seed point as the current seed point.

[0103] Step 4: Use KD tree to search for the K of the current seed point 3 Neighborhood points.

[0104] Step 5: Traverse K 3 Neighborhood points are found, and the angle between the normal of the current neighbor point and the current seed point is calculated. If the angle between the normal of the current neighbor point and the current seed point is less than the set angle threshold, the current neighbor point is added to the current cluster group and the current neighbor point is removed from the initial point cloud. Then, the curvature of the current neighborhood point is determined. If the curvature is less than the set curvature threshold, the neighborhood point is added to the current seed group as a new seed point. After traversing K3 neighborhood points, return to step 3 until the current seed group is empty, at which time the current cluster group is added to the total cluster group.

[0105] Step 6: Re-execute steps 2 to 5 until the initial point cloud is empty, terminate the region growing, and return the clustering result.

[0106] Step S211, calculating the particle size of all single particle point cloud coordinates to obtain particle size distribution information, wherein the particle size distribution information includes particle size and particle quantity.

[0107] Among them, the number of particles can be directly counted by the coordinates of the segmented single particle points. In order to obtain the particle size of the single particle point cloud, the embodiment of the present application uses the following model and adopts the minimum circumscribed sphere fitting method to calculate the particle size of the particles.

[0108] R 2 =∑((x i -x c ) 2 +(y i -y c ) 2 +(z i -z c ) 2 ) / N

[0109] Among them, (x i ,y i ,z i ) is the edge point of the particle point cloud, (x c ,y c ,z c ) is the center of the fitting sphere, N is the number of point clouds, and R is the radius of the fitting sphere.

[0110] In order to intuitively understand the particle size and particle size distribution of the bulk ore, the particle size distribution information can be made into a chart, for example Figure 6 As shown, it is a statistical diagram of ore particle size provided in an embodiment of the present application, and a display is provided for display, so that the particle size distribution of the ore on the conveyor belt 4 can be intuitively and clearly understood.

[0111] It can be seen from the above scheme that the embodiment of the present application provides a method and system for intelligent identification of ore block size based on 3D cloud map, and the intelligent identification system for ore block size based on 3D cloud map includes a controller 1, and a line scanning laser transmitter 2 and a three-dimensional camera 3 connected to the controller 1. The line scanning laser transmitter 2 and the three-dimensional camera 3 are arranged directly above the conveyor belt 4, and the conveyor belt 4 is used to convey block ore. The line scanning laser transmitter 2 is used to emit a line laser perpendicular to the upper surface of the conveyor belt to the conveyor belt; the 3D camera 3 is a 3D industrial camera with 5 million pixels and a resolution of 0.3mm, which is used to obtain a line laser image irradiated on the conveyor belt and the block ore, and the line laser image is a three-dimensional point cloud image.

[0112] In the actual application process, the controller 1 is used to execute an intelligent identification method for ore block size based on a 3D cloud map, obtain a 3D point cloud image of ore particles through a line scanning laser transmitter 2 and a 3D camera 3, and convert the 3D point cloud image into a 3D point cloud coordinate; perform point cloud filtering processing on the 3D point cloud coordinates to obtain filtered point cloud coordinates, and the filtering processing includes downsampling, removing outliers and repairing missing points; extract the Z coordinate value of the filtered point cloud coordinates, perform K-means clustering segmentation on the Z coordinate value, and obtain K 1 clusters, the K 1 is a positive integer greater than 1; calculate the center point value of each cluster, and K 1 The maximum value of the center point of each cluster is used as the segmentation threshold; the point cloud with a Z coordinate value greater than the segmentation threshold in the filtered point cloud coordinates is removed to obtain the point cloud coordinates of the target area; the point cloud coordinates of the target area are subjected to particle segmentation processing to obtain the point cloud coordinates of a single particle; the particle size is calculated for all the single particle point cloud coordinates to obtain the particle size distribution information, which includes the particle size and the number of particles.

[0113] The present application is described in detail above in conjunction with specific implementation methods and exemplary examples, but these descriptions cannot be understood as limiting the present application. Those skilled in the art understand that, without departing from the spirit and scope of the present application, a variety of equivalent replacements, modifications or improvements can be made to the technical solution of the present application and its implementation methods, all of which fall within the scope of the present application. The scope of protection of the present application shall be subject to the attached claims.

Claims

1. A method for intelligent identification of ore fragmentation based on 3D cloud images, characterized in that: include: A three-dimensional point cloud image of ore particles is obtained by a line scanning laser transmitter and a three-dimensional camera, and the three-dimensional point cloud image is converted into three-dimensional point cloud coordinates; Performing point cloud filtering processing on the three-dimensional point cloud coordinates to obtain filtered point cloud coordinates, wherein the filtering processing includes downsampling, removing outliers and repairing missing points; Extract the Z coordinate value of the filtered point cloud coordinates, perform K-means clustering segmentation on the Z coordinate value, and obtain K1 clusters, where K1 is a positive integer greater than 1; Calculate the center point value of each cluster, and take the maximum center point value of K1 clusters as the segmentation threshold; Remove the point clouds whose Z coordinate values ​​in the filtered point cloud coordinates are greater than the segmentation threshold to obtain the point cloud coordinates of the target area; Performing particle segmentation processing on the point cloud coordinates of the target area to obtain the point cloud coordinates of a single particle; Calculate the particle size of all single particle point cloud coordinates to obtain particle size distribution information, which includes particle size and particle quantity; The repairing of missing points includes: Obtain the resolution and spacing variation coefficient of the 3D camera, and calculate the missing point threshold according to the resolution and spacing variation coefficient of the 3D camera; Get the point cloud coordinates after removing outliers, and traverse the point cloud coordinates after removing outliers to calculate the distance between any adjacent points; The size of the distance between any adjacent points and the missing point threshold is determined. If the distance between any adjacent points is greater than the missing point threshold, local range surface fitting is performed between the adjacent points, the missing points are added, and the filtered point cloud coordinates are obtained.

2. According to claim 1, a method for intelligently identifying ore block size based on 3D cloud images, characterized in that: The downsampling comprises: In the three-dimensional point cloud coordinate area, a plurality of voxel grids are created, wherein the size of the voxel grids is smaller than the resolution of the three-dimensional camera; Calculate the center coordinates of all point cloud coordinates within each voxel grid; The center coordinates of the voxel grid are used to replace all the point cloud coordinates in the voxel grid to generate downsampled point cloud coordinates.

3. According to claim 2, a method for intelligently identifying ore block size based on 3D cloud images, characterized in that: The removal of outliers includes: Traverse the downsampled point cloud coordinates, select any point cloud coordinate as the current point, and determine the average distance between the current point and the nearest K2 field points, calculate the mean and standard deviation of the average distance, and obtain the proportional coefficient, which is a constant; Determine the distance threshold according to the mean and standard deviation of the average distance and the proportionality coefficient; If the average distance is greater than the distance threshold, the current point is deleted to obtain the point cloud coordinates without outliers.

4. According to the method for intelligent identification of ore fragmentation based on 3D cloud map in claim 1, it is characterized in that: After the step of adding the missing points, also include: Calculate the distance between the adjacent points of the missing point and the missing point; Determine the size of the distance between the adjacent points of the missing point and the missing point and the missing point threshold. If the distance between the adjacent points of the missing point and the missing point is greater than the missing point threshold, perform local range surface fitting between the adjacent points of the missing point and the missing point and add a new missing point.

5. According to claim 2, a method for intelligently identifying ore fragmentation based on 3D cloud images, characterized in that: The size of the voxel grid is 0.2 mm.

6. The method for intelligently identifying ore fragmentation based on 3D cloud images according to claim 1 is characterized in that: The spacing variation coefficient is 2.

7. The method for intelligently identifying ore block size based on 3D cloud images according to claim 1 is characterized in that: The step of performing particle segmentation processing on the point cloud coordinates of the target area to obtain the point cloud coordinates of a single particle comprises: Traverse the point cloud coordinates of the target area, use the Kd tree to search for the K3 domain points of each point cloud, and calculate the normal and curvature of each point cloud through the K3 domain points of each point cloud; According to the normal line and the curvature, the point cloud coordinates of the target region are segmented into particles by using the region growing method to obtain the point cloud coordinates of a single particle.

8. An intelligent identification system for ore fragmentation based on 3D cloud images, characterized in that: It includes a controller, and a line scanning laser transmitter and a three-dimensional camera connected to the controller; The line scanning laser transmitter and the three-dimensional camera are arranged directly above the conveyor belt, the conveyor belt is used to convey block ore, the line scanning laser transmitter is used to emit a line laser perpendicular to the upper surface of the conveyor belt to the conveyor belt; the three-dimensional camera is used to obtain a line laser image irradiated on the conveyor belt and the block ore, and the line laser image is a three-dimensional point cloud image; The controller is further configured to perform the following steps: A three-dimensional point cloud image of ore particles is obtained by a line scanning laser transmitter and a three-dimensional camera, and the three-dimensional point cloud image is converted into three-dimensional point cloud coordinates; Performing point cloud filtering processing on the three-dimensional point cloud coordinates to obtain filtered point cloud coordinates, wherein the filtering processing includes downsampling, removing outliers and repairing missing points; Extract the Z coordinate value of the filtered point cloud coordinates, perform K-means clustering segmentation on the Z coordinate value, and obtain K1 clusters, where K1 is a positive integer greater than 2; Calculate the center point value of each cluster, and take the maximum center point value of K1 clusters as the segmentation threshold; Remove the point clouds whose Z coordinate values ​​in the filtered point cloud coordinates are greater than the segmentation threshold to obtain the point cloud coordinates of the target area; Performing particle segmentation processing on the target area point cloud coordinates to obtain single particle point cloud coordinates, wherein the particle segmentation processing includes supervoxel area clustering, area concavity marking and area particle clustering; Calculate the particle size of all single particle point cloud coordinates to obtain particle size distribution information, which includes particle size and particle quantity; Wherein, the repairing missing points includes: Obtain the resolution and spacing variation coefficient of the 3D camera, and calculate the missing point threshold according to the resolution and spacing variation coefficient of the 3D camera; Get the point cloud coordinates after removing outliers, and traverse the point cloud coordinates after removing outliers to calculate the distance between any adjacent points; The size of the distance between any adjacent points and the missing point threshold is determined. If the distance between any adjacent points is greater than the missing point threshold, local range surface fitting is performed between the adjacent points, the missing points are added, and the filtered point cloud coordinates are obtained.

9. The ore block size intelligent identification system based on 3D cloud map according to claim 8 is characterized in that: The three-dimensional camera has 5 million pixels and a resolution of 0.3 mm.

Images