Material lumpiness statistical method based on multi-view three-dimensional laser point cloud

Abstract

The invention belongs to the technical field of material lumpiness statistics, and particularly discloses a material lumpiness statistics method based on multi-view three-dimensional laser point clouds, which comprises the following steps of: obtaining point cloud data of materials through a three-dimensional laser scanner, preprocessing the obtained point cloud data of the materials, and storing the preprocessed point cloud data in a tree structure; determining point cloud groups of all surfaces of the material blocks by using a point cloud region growth algorithm, dividing the relationship between adjacent point cloud groups by using spatial feature normal vectors of the point cloud groups, and obtaining all point cloud groups of each material block in combination with a nearest neighbor algorithm so as to judge the spatial relationship of adjacent materials; according to the method, the point cloud data are preprocessed to reduce illumination variation, shielding objects, noise interference and the like, and the material identification precision and accuracy are improved, so that the problems of low material identification precision and poor accuracy caused by factors such as illumination variation, shielding objects, noise interference and the like in a complex environment during statistics of different forms of materials are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of material block size statistics, and in particular to a material block size statistics method based on multi-view three-dimensional laser point cloud. Background Art

[0002] Block size refers to the size of the fragments formed after blasting or crushing of ore or rock. It is an important production indicator in the fields of industrial production, mining, logistics and transportation, and resource management. It is also an important factor that cannot be ignored in the production, processing and use process. With the rapid development of industrial automation and intelligence, the demand for accurate identification of material block size is becoming increasingly urgent. Accurate block size identification can not only improve production efficiency and reduce energy consumption, but also optimize resource allocation and reduce waste. At the same time, it is related to safety, stability, transportation and stacking, resource utilization, and environmental impact.

[0003] The material size recognition methods mainly include vision-based recognition methods, laser scanning-based recognition methods, and comprehensive recognition methods that combine 3D reconstruction and computing technology. Among them, the comprehensive recognition method of 3D reconstruction and computing technology is based on the principle of multi-view geometry. It uses a 3D laser scanner to obtain images or scan data from multiple perspectives, and combines computer graphics and image processing technology to infer the 3D structure of objects or scenes. Although the material size recognition technology has made significant progress, there are still some problems and challenges. First, the morphology of materials varies greatly, which puts high demands on the adaptability and generalization ability of the recognition algorithm. Secondly, factors such as lighting changes, obstructions and noise interference in complex environments lead to low material recognition precision and poor accuracy. Summary of the invention

[0004] The purpose of the present invention is to provide a material block size statistics method based on multi-view three-dimensional laser point cloud to solve the problem of low material recognition precision and poor accuracy caused by factors such as lighting changes, obstructions and noise interference in complex environments when counting materials of different shapes.

[0005] To achieve the above object, the basic solution provided by the present invention is: a material block size statistics method based on multi-view three-dimensional laser point cloud, comprising the following steps:

[0006] S1. Set up several 3D laser scanners in the area where the material block size is counted, and set the threshold of the angle between adjacent point cloud groups in the industrial computer according to actual needs;

[0007] S2. Scan the material block by a 3D laser scanner to obtain point cloud data;

[0008] S3. Preprocess the point cloud data to obtain material point cloud data;

[0009] S4. Construct a tree structure in the industrial control computer and save the material point cloud data in the tree structure;

[0010] S5. Using the point cloud region growth algorithm, the point clouds of all faces of the material block are transformed into point cloud groups of corresponding faces;

[0011] S6. Determine the spatial relationship between adjacent materials through the normal vector features of the point cloud group and the nearest neighbor distance method.

[0012] The principle and beneficial effect of the present invention are as follows: when in use, according to the complexity and diversity of the on-site environment, multiple three-dimensional laser scanners are set up, and the threshold of the normal vector angle of adjacent point cloud groups is set in the industrial control computer. The point cloud data of the material is obtained by the three-dimensional laser scanner, and then the point cloud data of the material is preprocessed and stored in a tree structure. Then, the point cloud region growth algorithm is used to determine the point cloud groups of all faces of the material block, and the spatial characteristic normal vectors of the point cloud group are used to divide the relationship between adjacent point cloud groups. Combined with setting the threshold of the normal vector angle of adjacent point cloud groups, the spatial relationship of adjacent materials can be judged. The point cloud data has high-precision and high-resolution geometric information, and can intuitively represent the shape, surface and texture of objects in space. The preprocessing of the point cloud data reduces the illumination changes, obstructions and noise interference, etc., and increases the precision and accuracy of material recognition.

[0013] Solution 2 is the preferred basic solution. In step S1, a threshold ∠θ1 of the normal vector angle of adjacent point cloud groups is set; and different thresholds are set for the normal vector angle of different materials according to the characteristics of different materials.

[0014] Solution three is the preferred basic solution. In step S3, the point cloud data is denoised by statistical filtering, and the denoised point cloud is spliced ​​by the ICP algorithm. Statistical filtering is mainly used to remove noise in the image. The principle is to sort the pixel values ​​around each pixel and select the middle value as the new value of the pixel. It can effectively remove noise points while maintaining the edge information of the image and reducing interference information caused by the environment. The ICP algorithm minimizes the distance between the two point clouds through iterative calculation, and can obtain a very accurate registration effect. There is no need to segment the point cloud and extract features, which simplifies the splicing of pre-processed point cloud data.

[0015] Solution 4 is the preferred basic solution. In step S4, the point cloud data is divided into different levels and regions through an octree. The octree can efficiently process three-dimensional data by dividing the space into eight identical subspaces, which can reduce memory usage and increase data query and update speed.

[0016] Scheme 5 is the preferred basic scheme. In step S5, the steps of the point cloud region growing algorithm are:

[0017] a. Select initial point cloud: select a set of initial point clouds from the point cloud data by farthest sampling;

[0018] b. Determine the growth criteria: set the growth criteria according to the geometric characteristics and statistical requirements of the material, and set the threshold of the distance between the points;

[0019] c. Perform region growing: Starting from the initial point cloud, search for adjacent points in the point cloud data according to the set growth criteria. When the distance between two points is less than the threshold, they belong to the same surface of the same material block; when the distance between two points is greater than the threshold, they do not belong to the same surface of the same material block.

[0020] d. Repeat step c.: perform the region growing process for each initial point cloud until all point clouds are processed and multiple point cloud groups are obtained;

[0021] The principle of the point cloud region growing algorithm is to start from one or more point clouds, and gradually add neighboring points with similar geometric features to the point cloud into the same region until there are no more points to add. It is mainly used to cluster the points in the point cloud data into different regions according to geometric features, and cluster the point clouds into point cloud groups through the point cloud region growing algorithm. The algorithm is highly flexible, simple to execute and robust, and can adapt to different needs by adjusting parameters according to the properties of different materials.

[0022] Solution 6, which is the preferred basic solution, in step S6, the step of determining the spatial relationship between adjacent materials using the nearest neighbor distance method:

[0023] a. Normal vector calculation: By calculating the local covariance matrix of a point cloud group and finding the eigenvector corresponding to its minimum eigenvalue

[0024] b. Calculation of normal vector angle: through the eigenvectors of two point cloud groups and Get the angle value ∠θ2; use the formula: Get the angle value ∠θ2;

[0025] c. Compare ∠θ1 and ∠θ2. When ∠θ2>∠θ1, the two point cloud groups do not belong to the same material block; when ∠θ2<∠θ1, the two point cloud groups belong to the same material block;

[0026] The nearest neighbor distance method calculates the distance between the unknown sample and the sample in the training set, selects multiple samples with the nearest distance, and then performs majority voting or average calculation based on the labels of these samples to determine the category of the unknown sample or predict its value. It calculates the normal vector angle ∠θ2 of the point cloud group, compares ∠θ2 with the threshold ∠θ1 for setting the normal vector angle of the adjacent point cloud group, and then classifies the adjacent materials according to the comparison results. It is convenient to process the denoised point cloud data, and the construction is simple, intuitive and easy to understand. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 The present invention is a flow chart of a material block size statistics method based on multi-view three-dimensional laser point cloud. DETAILED DESCRIPTION

[0028] The present invention is further described in detail below through specific embodiments:

[0029] Example

[0030] like Figure 1 As shown: A material block size statistics method based on multi-view three-dimensional laser point cloud, comprising the following steps:

[0031] S1. Set up several 3D laser scanners in the area where the material block size is counted, and set the threshold ∠θ1 of the angle between the normal vectors of adjacent point cloud groups in the industrial control computer according to actual needs;

[0032] S2. Scan the material block by a 3D laser scanner to obtain point cloud data;

[0033] S3. De-noise the point cloud data by statistical filtering, and splice the de-noised point cloud by ICP algorithm to obtain material point cloud data;

[0034] S4. construct an octree in the industrial control computer, and divide the point cloud data into different levels and areas through the octree;

[0035] S5. Use the point cloud region growth algorithm to obtain the point cloud group of all faces of the material block. The steps of the point cloud region growth algorithm are:

[0036] a. Select initial point cloud: select a set of initial point clouds from the point cloud data by farthest sampling;

[0037] b. Determine the growth criteria: set the threshold of the distance between points according to the geometric characteristics and statistical requirements of the material;

[0038] c. Perform region growing: Starting from the initial point cloud, search for adjacent points in the point cloud data according to the set growth criteria. When the distance between two points is less than the threshold, they belong to the same surface of the same material block; when the distance between two points is greater than the threshold, they do not belong to the same surface of the same material block.

[0039] d. Repeat step c.: perform the region growing process for each initial point cloud until all point clouds are processed, and multiple point cloud groups can be obtained;

[0040] S6. Steps for determining the spatial relationship of adjacent materials by using the nearest neighbor distance method through the normal vector features of the point cloud group:

[0041] a. Normal vector calculation: By calculating the local covariance matrix of a point cloud group and finding the eigenvector corresponding to its minimum eigenvalue

[0042] b. Calculation of normal vector angle: through the eigenvectors of two point cloud groups and Get the angle value ∠θ2, using the formula: Get the angle value ∠θ2;

[0043] c. Compare ∠θ1 and ∠θ2. When ∠θ2>∠θ1, the two point cloud groups do not belong to the same material block; when ∠θ2<∠θ1, the two point cloud groups belong to the same material block.

[0044] The implementation method of this embodiment is as follows: when in use, first set up multiple three-dimensional laser scanners according to the complexity and diversity of the on-site environment, set the threshold value ∠θ1=90° of the angle between the normal vectors of adjacent point cloud groups in the industrial control computer according to the geometric characteristics and statistical requirements of the material, and then obtain the point cloud data of the material through the three-dimensional laser scanner, and then reduce the noise of the point cloud data through statistical filtering. The filtering here can also use radius filtering, and its basic principle is to calculate the distance between the central pixel point and the neighboring pixel point, and perform weighted averaging processing on each pixel point according to the size of the distance, so as to achieve the filtering effect, and the edge information of the point cloud can be maintained after removing the noise points, and at the same time While reducing the interference information caused by the environment, the ICP algorithm is used to minimize the distance between the two point clouds through iterative calculation, and the noise-reduced point clouds are spliced ​​to obtain a very accurate registration effect, and there is no need to segment and extract features from the point clouds. An octree is constructed in the industrial computer, and the point cloud data is divided into different levels and regions through the octree, which can reduce memory usage and increase the query and update speed of data. KD tree can also be used here. KD tree can efficiently perform operations such as range query and nearest neighbor query in multi-dimensional space, which is convenient for subsequent search of adjacent points. Secondly, the point cloud region growth algorithm is used to aggregate the point cloud into a point cloud group. The specific steps are:

[0045] First, a group of initial point clouds are selected by farthest sampling. Random sampling without replacement can be used here. The principle is that after an individual is selected into the sample, he will not be put back into the population to continue to participate in the subsequent sampling process. Then, according to the geometric characteristics and statistical requirements of the material, the threshold of the distance between points is set. Then, starting from the initial point cloud, according to the set growth criterion, adjacent points are searched in the point cloud data. When the distance between two points is less than the threshold, they belong to the same surface of the same material block; when the distance between two points is greater than the threshold, they do not belong to the same surface of the same material block. A group of initial point clouds are clustered according to the growth criterion to obtain multiple point cloud groups.

[0046] Finally, the nearest neighbor distance method is used to determine the spatial relationship between adjacent materials. The specific steps are as follows:

[0047] First calculate the local covariance matrix of a point cloud group and find the eigenvector corresponding to its minimum eigenvalue Then, the eigenvectors of the two point cloud groups are and Using the formula: The angle value ∠θ2 is obtained, and ∠θ2 and ∠θ1 are compared. When ∠θ2>∠θ1, the two point cloud groups do not belong to the same material block; when ∠θ2<∠θ1, the two point cloud groups belong to the same material block.

[0048] The above is only an embodiment of the present invention, and the common knowledge such as the known specific structure and characteristics in the scheme is not described in detail here. It should be pointed out that for those skilled in the art, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A material block size statistics method based on multi-view three-dimensional laser point cloud, characterized in that: The following steps are involved: S1. Set up several 3D laser scanners in the area where the material block size is counted, and set the threshold of the angle between adjacent point cloud groups in the industrial computer according to actual needs; S2. Scan the material block by a 3D laser scanner to obtain point cloud data; S3. Preprocess the point cloud data to obtain material point cloud data; S4. construct a tree structure in the industrial control computer and save the material point cloud data in the tree structure; S5. Using the point cloud region growth algorithm, the point clouds of all faces of the material block are transformed into point cloud groups of corresponding faces; S6. Determine the spatial relationship between adjacent materials through the normal vector features of the point cloud group and the nearest neighbor distance method.

2. The material block size statistics method based on multi-view three-dimensional laser point cloud according to claim 1, characterized in that: In step S1, a threshold ∠θ1 of the normal vector angle between adjacent point cloud groups is set.

3. The material block size statistics method based on multi-view three-dimensional laser point cloud according to claim 1, characterized in that: In step S3, the point cloud data is denoised by statistical filtering, and the denoised point clouds are spliced ​​by the ICP splicing algorithm.

4. The material block size statistics method based on multi-view three-dimensional laser point cloud according to claim 1, characterized in that: In step S4, the point cloud data is divided into different levels and regions through an octree.

5. The material block size statistics method based on multi-view three-dimensional laser point cloud according to claim 1, characterized in that: In step S5, the steps of the point cloud region growing algorithm are: a. Select initial point cloud: select a set of initial point clouds from the point cloud data by farthest sampling; b. Determine the growth criteria: set the threshold of the distance between points according to the geometric characteristics and statistical requirements of the material; c. Perform region growing: Starting from the initial point cloud, according to the set growth criteria, search for adjacent points in the point cloud data. When the distance between two points is less than the threshold, they belong to the same surface of the same material block; when the distance between two points is greater than the threshold, they do not belong to the same surface of the same material block; d. Repeat step c: perform region growing on each initial point cloud until all point clouds are processed and multiple point cloud groups are obtained.

6. The material block size statistics method based on multi-view three-dimensional laser point cloud according to claim 1, characterized in that: In step S6, the step of determining the spatial relationship between adjacent materials using the nearest neighbor distance method is as follows: a. Normal vector calculation: By calculating the local covariance matrix of a point cloud group and finding the eigenvector corresponding to its minimum eigenvalue b. Calculation of normal vector angle: through the eigenvectors of two point cloud groups and Using the formula Get the angle value ∠θ2; c. Compare ∠θ1 and ∠θ2. When ∠θ2>∠θ1, the two point cloud groups do not belong to the same material block; when ∠θ2<∠θ1, the two point cloud groups belong to the same material block.

Images