Stock yard point cloud denoising method and system based on clustering and classification algorithm

Abstract

The invention provides a material field point cloud denoising method and system based on a clustering and classification algorithm, and the method comprises the steps: obtaining original point cloud data of a material field in the moving process of a bucket wheel machine through a laser scanner carried on the bucket wheel machine, and carrying out the clustering processing, and obtaining N cluster sets; performing feature extraction on each cluster to obtain a feature vector of each cluster; and carrying out classification processing on each clustering cluster by using a classification algorithm, retaining and screening classification categories of the clustering clusters according to a preset classification rule, integrating all retained class cluster point clouds, and outputting filtered point clouds. According to the method, original point clouds are divided into a plurality of clustering clusters through clustering processing, feature vectors of the clustering clusters are extracted, then the feature vectors are divided into a stock yard stock pile point cloud class, a stock yard non-stock pile point cloud class, a ground point cloud class and a noise point cloud class by using a classification algorithm, and finally the point clouds of the corresponding classes are reserved and screened out according to a preset rule. The accuracy and the automatic operation efficiency of point cloud modeling of the stock ground are improved.

Description

Technical Field

[0001] This invention relates to the field of point cloud data processing technology for digital material yards in bulk material scenarios, and more particularly to a method and system for denoising point clouds in material yards based on clustering and classification algorithms. Background Technology

[0002] In bulk material handling scenarios such as ports, mines, and power plants, a digital material yard is a system that manages the material yard through digital technology. It primarily involves installing laser scanners on bucket wheel excavators to acquire point cloud data of the material yard and converting it into point cloud coordinates within the material yard coordinate system. This point cloud data can not only be used for visualization, helping managers to more intuitively understand the distribution of materials in the on-site material yard, but also guide the automated operation of the bucket wheel excavator, providing data support for automation strategies. The modeling of digital material yards has significant practical implications for material yard management and is currently widely used in numerous bulk material handling sites.

[0003] Currently, in the field of point cloud denoising in bulk material handling scenarios, traditional methods mainly employ point cloud filtering techniques. These filtering methods are applied in multiple fields and are relatively mature. They can remove noisy point clouds generated by LiDAR to a certain extent, thereby improving the accuracy of point clouds in material handling scenarios. However, existing technologies primarily address random noise and outliers in point cloud data. For point cloud sets adjacent to the material pile, such as those generated by stacker-reclaimer booms, dropped materials, dust, parked vehicles, and other equipment, existing technologies often struggle to effectively separate and remove them. This means that in complex bulk material handling scenarios, traditional methods cannot meet the demand for accurate point cloud data in digital material handling, thus affecting the management efficiency and accuracy of automated operations in material handling.

[0004] While existing point cloud denoising techniques can handle random noise generated by LiDAR to some extent, these methods have significant limitations in bulk material scenarios. Specifically, existing technologies cannot effectively distinguish and remove point cloud sets that are not adjacent to the material pile, such as those generated by stacker-reclaimer booms, dropped materials, dust, parked vehicles, and other equipment. If these non-material pile point cloud sets appear in the final generated digital material yard point cloud, they will directly affect the accuracy of material pile management and material pile volume calculation, thus negatively impacting the overall management of the digital material yard. Furthermore, existing technologies often lack sufficient versatility and flexibility when dealing with complex scenarios, making it difficult to adapt to the specific needs of different bulk material sites. Therefore, the industry urgently needs a point cloud denoising method that meets the characteristics of bulk material sites to effectively remove non-material pile point clouds that traditional filtering algorithms cannot remove, thereby improving the accuracy of material yard point cloud modeling and the efficiency of automated operations. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and system for denoising point clouds in material yards based on clustering and classification algorithms. The invention divides the original point cloud into multiple clusters through clustering, extracts the feature vectors of each cluster, and then uses a classification algorithm to categorize them into point clouds representing material piles, non-material piles, ground surfaces, and noise. Finally, point clouds of the corresponding categories are retained or removed according to preset rules, thereby improving the accuracy of point cloud modeling in material yards and the efficiency of automated operations.

[0006] The technical means employed in this invention are as follows: A method for denoising point clouds in a material yard based on clustering and classification algorithms includes: S1. Acquire the original point cloud data of the material yard in the current scanning area by using the laser scanner mounted on the bucket wheel excavator during the movement of the bucket wheel excavator; S2. Perform clustering processing on the original point cloud data of the material yard to obtain... N A set of clusters; S3. Extract features from each cluster to obtain the feature vector of each cluster; S4. Use a classification algorithm to classify each cluster into point clouds of material piles in the material yard, point clouds of non-material piles in the material yard, ground point clouds, and noise point clouds. S5. Based on the preset classification rules, retain and filter the classification categories of the clusters, integrate all retained cluster point clouds, and output the filtered point cloud.

[0007] Further, in step S1, the original point cloud data of the material yard is represented as follows: Specifically, it is expressed as follows:

[0008] in, Indicating the first point cloud One point; Indicates the first The three-dimensional coordinates of the points Indicates the first Reflectance values ​​at each point; This represents the total number of point clouds.

[0009] Furthermore, in step S1, the current scanning area refers to the area that the laser scanner can cover during one movement of the bucket wheel machine. This area is determined jointly by the movement path of the bucket wheel machine and the scanning range of the laser scanner. Further, step S2 includes: S21. Select a clustering algorithm, wherein the clustering algorithm includes one or more of the following: DBSCAN algorithm, K-Means algorithm, hierarchical clustering algorithm, and spectral clustering algorithm; S22. Based on the clustering algorithm selected in step S1, process the original point cloud data of the material yard. Cluster each point in the data and output the results. N A set of clusters ,as follows:

[0010] Each cluster contains several points, and each point belongs to one and only one cluster.

[0011] Further, step S3 includes: S31. By calculating the minimum and maximum coordinate values ​​of all points in each cluster, the coverage area of ​​the point cloud is determined; S32. Determine the point density of the point cloud by calculating the ratio of the number of points in each cluster to the area it covers; S33, By calculating all points in each cluster z The average value of the coordinates determines the average height of the point cloud; S34. Determine the total number of point clouds in the point cloud cluster set by counting the number of points in each cluster; S35. Determine the average reflectance of the point cloud by calculating the average reflectance value of all points in each cluster; S36. Combine the features determined above into a feature vector, which serves as the feature representation of the cluster.

[0012] Further, step S4 includes: S41. Prepare a training dataset, which includes feature vectors of labeled point cloud classes for material piles, non-material piles in the material yard, ground point clouds, and noise point clouds. S42. Select a classification algorithm, wherein the classification algorithm includes, but is not limited to, one of support vector machine (SVM), decision tree, random forest and neural network; S43. Train the selected classification algorithm using the training dataset to obtain the classification model; S44. For the feature vector of each cluster, use the trained classification model to make a classification prediction; S45. The classification prediction results of each cluster are divided into material pile point cloud, non-material pile point cloud, ground point cloud and noise point cloud.

[0013] Further, step S5 includes: S51. Retain all point clouds of material stockpiles and ground point clouds; S52. Screen out point clouds of noise type and point clouds of non-material pile type in the material yard; S53. Integrate all retained cluster point clouds and output the filtered point cloud.

[0014] This invention also provides a point cloud denoising system for material yards based on clustering and classification algorithms, implemented using the aforementioned point cloud denoising method based on clustering and classification algorithms, comprising: The point cloud acquisition module is used to acquire the original point cloud data of the material yard in the current scanning area by using a laser scanner mounted on the bucket wheel excavator during the movement of the bucket wheel excavator; The clustering module is used to perform clustering processing on the original point cloud data of the material yard to obtain... N A set of clusters; The feature extraction module is used to extract features from each cluster to obtain the feature vector of each cluster; The classification module is used to classify each cluster using a classification algorithm, dividing the clusters into material pile point clouds, non-material pile point clouds, ground point clouds, and noise point clouds. The filtering and integration module is used to retain or remove the classification categories of clusters according to preset classification rules, integrate all retained cluster point clouds, and output the filtered point cloud.

[0015] Compared with the prior art, the present invention has the following advantages: 1. This invention processes raw point cloud data using clustering algorithms (such as DBSCAN, K-Means, hierarchical clustering, spectral clustering, etc.), dividing the point cloud into multiple clusters, thereby effectively separating the non-pile point cloud set adjacent to the material pile. This technique can accurately identify and separate the point clouds of different objects such as material piles, equipment, and vehicles, significantly improving the accuracy and clarity of point cloud data, and solving the problem that traditional denoising filtering methods can only remove noise but cannot remove reasonable targets.

[0016] 2. This invention utilizes feature extraction technology to extract features such as coverage area, point density, average height, total number of point clouds, and average reflectivity for each cluster, and combines these features into a feature vector. The extraction of these features provides rich information for subsequent classification processing, enabling the classification algorithm to more accurately identify different types of point clouds, thereby improving the accuracy and reliability of classification.

[0017] 3. This invention classifies clusters using classification algorithms (such as Support Vector Machine (SVM), Decision Tree, Random Forest, Neural Network, etc.), dividing them into point clouds belonging to material piles, non-material piles, ground, and noise. This technique accurately identifies which point clouds belong to material piles and which belong to noise or other non-material pile objects, effectively removing point cloud sets unrelated to material piles and improving the purity and usability of material pile point clouds.

[0018] 4. This invention filters and integrates classification results through preset classification rules, retaining point cloud types of material piles and ground point clouds, while filtering out noise point clouds and non-material pile point clouds, ultimately outputting filtered point clouds. This technique allows for flexible adjustment of retention and rejection rules according to specific business needs, making the output point cloud data more consistent with practical application requirements and improving the system's flexibility and practicality. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the method of the present invention.

[0021] Figure 2 This is a block diagram of the system structure of the present invention.

[0022] Figure 3 This is a flowchart of the clustering and classification algorithm for denoising in this invention.

[0023] Figure 4 This is an overall flowchart of the digital material yard point cloud acquisition and update part of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0026] like Figure 1 As shown, this invention provides a method for denoising point clouds in material yards based on clustering and classification algorithms, including: S1. Acquire the original point cloud data of the material yard in the current scanning area by using the laser scanner mounted on the bucket wheel excavator during the movement of the bucket wheel excavator; S2. Perform clustering processing on the original point cloud data of the material yard to obtain... N A set of clusters; S3. Extract features from each cluster to obtain the feature vector of each cluster; S4. Use a classification algorithm to classify each cluster into point clouds of material piles in the material yard, point clouds of non-material piles in the material yard, ground point clouds, and noise point clouds. S5. Based on the preset classification rules, retain and filter the classification categories of the clusters, integrate all retained cluster point clouds, and output the filtered point cloud.

[0027] In a specific implementation, as a preferred embodiment of the present invention, in step S1, the original point cloud data of the material yard is represented as follows: Specifically, it is expressed as follows:

[0028] in, Indicating the first point cloud One point; Indicates the first The three-dimensional coordinates of the points Indicates the first Reflectance values ​​at each point; This represents the total number of point clouds.

[0029] In a specific implementation, as a preferred embodiment of the present invention, the current scanning area refers to the area that the laser scanner can cover during one movement of the bucket wheel machine. This area is determined by the movement path of the bucket wheel machine and the scanning range of the laser scanner.

[0030] In a specific implementation, as a preferred embodiment of the present invention, step S2 includes: S21. Select a clustering algorithm, wherein the clustering algorithm includes one or more of the following: DBSCAN algorithm, K-Means algorithm, hierarchical clustering algorithm, and spectral clustering algorithm; S22. Based on the clustering algorithm selected in step S1, process the original point cloud data of the material yard. Cluster each point in the data and output the results. N A set of clusters ,as follows:

[0031] Each cluster contains several points, and each point belongs to one and only one cluster.

[0032] In a specific implementation, as a preferred embodiment of the present invention, step S3 includes: S31. By calculating the minimum and maximum coordinate values ​​of all points in each cluster, the coverage area of ​​the point cloud is determined; S32. Determine the point density of the point cloud by calculating the ratio of the number of points in each cluster to the area it covers; S33, By calculating all points in each cluster z The average value of the coordinates determines the average height of the point cloud; S34. Determine the total number of point clouds in the point cloud cluster set by counting the number of points in each cluster; S35. Determine the average reflectance of the point cloud by calculating the average reflectance value of all points in each cluster; S36. Combine the features determined above into a feature vector, which serves as the feature representation of the cluster.

[0033] In specific implementation, as a preferred embodiment of the present invention, such as Figure 2 As shown, step S4 includes: S41. Prepare a training dataset, which includes feature vectors of labeled point cloud classes for material piles, non-material piles in the material yard, ground point clouds, and noise point clouds. S42. Select a classification algorithm, wherein the classification algorithm includes, but is not limited to, one of support vector machine (SVM), decision tree, random forest and neural network; S43. Train the selected classification algorithm using the training dataset to obtain the classification model; S44. For the feature vector of each cluster, use the trained classification model to make a classification prediction; S45. The classification prediction results of each cluster are divided into material pile point cloud, non-material pile point cloud, ground point cloud and noise point cloud.

[0034] In a specific implementation, as a preferred embodiment of the present invention, step S5 includes: S51. Retain all point clouds of material stockpiles and ground point clouds; S52. Screen out point clouds of noise type and point clouds of non-material pile type in the material yard; S53. Integrate all retained cluster point clouds and output the filtered point cloud.

[0035] Corresponding to the point cloud denoising method for material yards based on clustering and classification algorithms in this application, such as Figure 3 As shown, the present invention also provides a point cloud denoising system for material yards based on clustering and classification algorithms, comprising: The point cloud acquisition module is used to acquire the original point cloud data of the material yard in the current scanning area by using a laser scanner mounted on the bucket wheel excavator during the movement of the bucket wheel excavator; The clustering module is used to perform clustering processing on the original point cloud data of the material yard to obtain... N A set of clusters; The feature extraction module is used to extract features from each cluster to obtain the feature vector of each cluster; The classification module is used to classify each cluster using a classification algorithm, dividing the clusters into material pile point clouds, non-material pile point clouds, ground point clouds, and noise point clouds. The filtering and integration module is used to retain or remove the classification categories of clusters according to preset classification rules, integrate all retained cluster point clouds, and output the filtered point cloud.

[0036] The embodiments of the present invention are described simply because they correspond to those in the embodiments above. For any similarities, please refer to the descriptions in the embodiments above, which will not be elaborated here.

[0037] Example Implementation location: Bulk material site, equipped with bucket wheel excavator with scanner, to establish a digital material yard.

[0038] Implementation plan: such as Figure 4 As shown, a laser scanner (model: Sagitar RS16P) is deployed on top of the bucket wheel excavator at the bulk material handling site, with a scanning frequency of 10Hz and a point cloud resolution of 0.1 meters. The scanner is activated when the bucket wheel excavator moves to receive point cloud data. Data reception stops when the excavator stops moving, thus obtaining raw point cloud data for a specific range within the current time period. The DBSCAN clustering algorithm was applied to the original point cloud. Clustering is performed. After the clustering algorithm, the original point cloud... Each point in Each point belongs to one and only one cluster. For each cluster, a classification algorithm is used for classification. The point cloud clusters are ultimately divided into four categories: material yard / stockpile point cloud, non-material yard / stockpile point cloud, ground point cloud, and noise point cloud. Each cluster belongs to one and only one of these four categories. In this embodiment, the classification algorithm uses the SVM algorithm. The input features are the cluster features obtained in the previous step, and the output is the category to which the cluster belongs. After classification calculation, for each cluster's classification category, retention and filtering are performed according to business needs. Generally, all point clouds from the material yard / stockpile and ground point cloud categories are retained, while the noise and non-material yard / stockpile point cloud categories are filtered out. All retained cluster point clouds are integrated, and the final filtered point cloud is output. The filtered point cloud is then rasterized and blind spot completion is performed to obtain the final material yard point cloud.

[0039] This invention is applicable to bulk material handling sites equipped with bucket wheel excavators and laser scanners, such as ports, mines, and power plants. After implementation, the accuracy of point cloud data at material yards will increase by approximately 30%, and the efficiency of automated operations will increase by 15%-20%, with an estimated annual economic benefit of 500,000 to 1,000,000 yuan (depending on the size of the material yard).

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A stockyard point cloud denoising method based on clustering and classification algorithm, characterized in that, include: S1. Acquire the original point cloud data of the material yard in the current scanning area by using the laser scanner mounted on the bucket wheel excavator during the movement of the bucket wheel excavator; S2, performing clustering processing on the stockyard original point cloud data to obtain N a cluster set; S3. Extract features from each cluster to obtain the feature vector of each cluster; S4. Use a classification algorithm to classify each cluster into point clouds of material piles in the material yard, point clouds of non-material piles in the material yard, ground point clouds, and noise point clouds. S5. Based on the preset classification rules, retain and filter the classification categories of the clusters, integrate all retained cluster point clouds, and output the filtered point cloud.

2. The method for denoising point clouds in a material yard based on clustering and classification algorithms according to claim 1, characterized in that, In step S1, the stockyard original point cloud data is represented as and is specifically represented as follows: in, Indicating the first point cloud One point; Indicates the first The three-dimensional coordinates of the points Indicates the first Reflectance values ​​at each point; This represents the total number of point clouds.

3. The method for denoising point clouds in a material yard based on clustering and classification algorithms according to claim 1, characterized in that, In step S1, the current scanning area refers to the area that the laser scanner can cover during one movement of the bucket wheel machine. This area is determined by the movement path of the bucket wheel machine and the scanning range of the laser scanner.

4. The method for denoising point clouds in a material yard based on clustering and classification algorithms according to claim 1, characterized in that, Step S2 includes: S21. Select a clustering algorithm, wherein the clustering algorithm includes one or more of the following: DBSCAN algorithm, K-Means algorithm, hierarchical clustering algorithm, and spectral clustering algorithm; S22. Based on the clustering algorithm selected in step S1, process the original point cloud data of the material yard. Cluster each point in the data and output the results. N A set of clusters ,as follows: Each cluster contains several points, and each point belongs to one and only one cluster.

5. A method for denoising point clouds in a material yard based on clustering and classification algorithms according to claim 1, characterized in that, Step S3 includes: S31. By calculating the minimum and maximum coordinate values ​​of all points in each cluster, the coverage area of ​​the point cloud is determined; S32. Determine the point density of the point cloud by calculating the ratio of the number of points in each cluster to the area it covers; S33, By calculating all points in each cluster z The average value of the coordinates determines the average height of the point cloud; S34. Determine the total number of point clouds in the point cloud cluster set by counting the number of points in each cluster; S35. Determine the average reflectance of the point cloud by calculating the average reflectance value of all points in each cluster; S36. Combine the features determined above into a feature vector, which serves as the feature representation of the cluster.

6. The method for denoising point clouds in a material yard based on clustering and classification algorithms according to claim 1, characterized in that, Step S4 includes: S41. Prepare a training dataset, which includes feature vectors of labeled point cloud classes for material piles, non-material piles in the material yard, ground point clouds, and noise point clouds. S42. Select a classification algorithm, wherein the classification algorithm includes, but is not limited to, one of support vector machine (SVM), decision tree, random forest and neural network; S43. Train the selected classification algorithm using the training dataset to obtain the classification model; S44. For the feature vector of each cluster, use the trained classification model to make a classification prediction; S45. The classification prediction results of each cluster are divided into material pile point cloud, non-material pile point cloud, ground point cloud and noise point cloud.

7. A method for denoising point clouds in a material yard based on clustering and classification algorithms according to claim 1, characterized in that, Step S5 includes: S51. Retain all point clouds of material stockpiles and ground point clouds; S52. Screen out point clouds of noise type and point clouds of non-material pile type in the material yard; S53. Integrate all retained cluster point clouds and output the filtered point cloud.

8. A point cloud denoising system for a material yard based on clustering and classification algorithms, implemented according to the point cloud denoising method for material yards based on clustering and classification algorithms as described in any one of claims 1-7, characterized in that, include: The point cloud acquisition module is used to acquire the original point cloud data of the material yard in the current scanning area by using a laser scanner mounted on the bucket wheel excavator during the movement of the bucket wheel excavator; The clustering module is used to perform clustering processing on the original point cloud data of the material yard to obtain... N A set of clusters; The feature extraction module is used to extract features from each cluster to obtain the feature vector of each cluster; The classification module is used to classify each cluster using a classification algorithm, dividing the clusters into material pile point clouds, non-material pile point clouds, ground point clouds, and noise point clouds. The filtering and integration module is used to retain or remove the classification categories of clusters according to preset classification rules, integrate all retained cluster point clouds, and output the filtered point cloud.

Images