A non-destructive and rapid segmentation method, device and medium for particle point cloud of construction waste.

Abstract

This invention relates to the field of intelligent detection of building materials and resource utilization of waste, and discloses a non-destructive and rapid segmentation method, device, and medium for point clouds of construction waste particles. The invention reads the particle point cloud file and calculates the axis-aligned bounding box of the particle point cloud; calculates the projection image of the point cloud along the maximum face direction of the AABB of the particle point cloud, and saves the pixel-point cloud correspondence; segments the particle projection image using a watershed algorithm; converts the segmented projection image into point cloud data using the pixel-point cloud correspondence; and outputs the segmented particle point cloud. On a computer with conventional hardware configuration, this invention improves algorithm efficiency by more than 95% compared to DBSCAN, ensuring real-time requirements and meeting the need for rapid segmentation of large-scale construction waste particle point clouds. This invention employs the mapping relationship between pixels and point cloud data to convert the segmented two-dimensional image result back into three-dimensional point cloud data, achieving accurate mapping from the projection image to the actual particle point cloud.

Description

Technical Field

[0001] This invention belongs to the field of intelligent detection of building materials and resource utilization of waste, specifically relating to a non-destructive and rapid segmentation method, device and medium for point cloud of slag particles. Background Technology

[0002] In the field of engineering waste resource utilization, point cloud data obtained from 3D camera scanning provides rich information for the analysis of the physical properties of engineering waste. Rapid segmentation of the waste particle point cloud provides the data foundation for subsequent particle feature extraction and is one of the most important steps in the analysis process. Currently, the segmentation of waste particle point clouds mainly relies on clustering algorithms for 3D point cloud data, such as DBSCAN (density-based spatial clustering algorithm) and its improved version (publication number CN119360363A, patent name: a particle clustering segmentation algorithm (PPC) in a real-time detection method for waste particle gradation characteristics). However, these algorithms have the following shortcomings:

[0003] Slow processing speed: When processing large-scale point cloud data, DBSCAN has a slow processing speed due to its high computational complexity (usually O(N log N)), which cannot meet the real-time requirements.

[0004] Information loss: In order to improve speed, the PPC algorithm needs to downsample the point cloud data, which will lead to the loss of point cloud information, affecting the segmentation accuracy and the accuracy of subsequent analysis.

[0005] Parameter sensitivity: DBSCAN depends on parameters such as neighborhood radius ε and minimum number of points minPts. Improper parameter selection will lead to poor segmentation results, especially when the point cloud density is uneven.

[0006] In summary, existing technologies for segmenting aggregate particle point clouds suffer from problems such as slow processing speed, information loss, and parameter sensitivity, making it difficult to meet the needs for rapid and flexible segmentation of aggregate particle point clouds. Therefore, it is urgent to study a new method. Summary of the Invention

[0007] To address the problems of slow processing speed, information loss, and parameter sensitivity in the segmentation of slag particle point clouds, this invention proposes a lossless and rapid segmentation method for slag particle point clouds, which includes the following steps:

[0008] S1: Read the grain point cloud file and calculate the axis-aligned bounding box (AABB) of the grain point cloud.

[0009] S2: Based on the axis-aligned bounding box, calculate the particle projection image of the point cloud in the direction of the largest face of the particle point cloud AABB, and save the pixel-point cloud correspondence.

[0010] S3: Segment the particle projection image using the watershed algorithm;

[0011] S4: Convert the segmented projection image into point cloud data and output it using the pixel-point cloud correspondence formula.

[0012] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0013] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above method.

[0014] A computer program product includes a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0015] The beneficial effects of the technical solution provided by this invention are as follows: By avoiding downsampling of point cloud data, this invention effectively solves the shortcomings of existing technologies in terms of poor segmentation results under conditions of uneven point cloud density, ensuring stability and accuracy in complex scenes. By employing a strategy combining projection dimensionality reduction and the watershed algorithm, the complete information of the point cloud data is preserved, avoiding information loss and ensuring segmentation accuracy and the accuracy of subsequent analysis, achieving lossless segmentation. This invention transforms high-dimensional point cloud data into a two-dimensional projected image using projection technology. Through morphological processing, distance transformation, and background and foreground region generation, the information of the particle point cloud in the two-dimensional projected image is effectively extracted, and then the watershed algorithm is used for accurate segmentation. Compared with the simple clustering algorithm used in traditional methods, the watershed algorithm provides more refined segmentation results, especially suitable for scenes with complex particle morphologies and irregular distributions. Compared with DBSCAN, the algorithm of this invention has higher computational efficiency on high-dimensional point cloud data. Under the same hardware conditions, the processing speed of traditional methods is relatively slow, while the algorithm of this invention can significantly improve the segmentation speed and enhance real-time processing capabilities. By combining projection dimensionality reduction and the watershed algorithm, the processing speed is significantly improved. Efficient projection and segmentation algorithms enhance processing speed and achieve high-efficiency segmentation, meeting real-time requirements. On computers with conventional hardware configurations, compared to the DBSCAN algorithm, this invention improves efficiency by over 95% when processing point cloud data of the same scale, ensuring real-time performance and meeting the need for rapid segmentation of large-scale slag particle point clouds. This invention employs a mapping relationship between pixels and point cloud data, converting the segmented two-dimensional image results back into three-dimensional point cloud data, achieving accurate mapping from the projected image to the actual particle point cloud. This technique overcomes the limitations of traditional algorithms, ensuring consistency between segmentation accuracy and subsequent analysis. This method is particularly suitable for segmenting slag particle point clouds and can also be extended to point cloud segmentation scenarios for other particulate materials, such as construction sand and gravel, and mineral particles. It is applicable to the gradation feature detection of slag particles and can also be applied to real-time detection needs of particulate materials in other fields, such as mineral particle screening, construction sand and gravel detection, and agricultural granular fertilizer grading. This multifunctional detection method enables rapid and accurate detection of particle size distribution characteristics in various industries, meeting the needs of diverse industrial production and engineering applications. Attached Figure Description

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0017] Figure 1 This is a flowchart of the non-destructive and rapid segmentation method for point clouds of slag particles in this embodiment;

[0018] Figure 2 This is a schematic diagram of the particle point cloud data in this embodiment;

[0019] Figure 3 This is a schematic diagram of the projection image of the particles on the XY plane of the AABB in this embodiment;

[0020] Figure 4 This is a schematic diagram of the results of the watershed algorithm on the particle projection image in this embodiment;

[0021] Figure 5 According to this embodiment Figure 4 The segmented point cloud data is obtained by converting the segmentation result into a pixel-point cloud correspondence. Detailed Implementation

[0022] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0023] Example 1

[0024] Please refer to Figure 1 , Figure 1 This is a flowchart of a non-destructive and rapid segmentation method for point clouds of slag particles according to an embodiment of the present invention, which specifically includes the following steps:

[0025] S1. Calculate the axis-aligned bounding box (AABB) of the particle point cloud.

[0026] (I) Point cloud data reading: The reading format is as follows .ply Point cloud data, where each point cloud contains a series of three-dimensional coordinate points (x, y, z). For example... Figure 2 As shown, in this embodiment, the original point cloud data of particles obtained by scanning with a 3D camera is treated as a whole because the computer cannot distinguish the point cloud data of different particles at this stage.

[0027] (II) Finding the minimum and maximum coordinates: For each coordinate axis (x, y, and z), find the minimum and maximum values ​​of all points on that axis. These two points represent the minimum and maximum ranges of the point cloud data in each coordinate axis direction, i.e., the two diagonal points of the AABB.

[0028] (III) AABB Bounding Box Generation: An AABB is an axis-aligned cuboid with 8 vertices, which can be determined by the minimum and maximum bounding boxes. Specifically, given the minimum bounding box of an AABB... and maximum boundary Their eight vertices can be obtained by combining all their coordinates. These eight vertices are respectively... , , , , , , , These points are the eight vertices of the AABB cuboid, representing all the corners of the bounding box in three-dimensional space.

[0029] S2. Calculate the projected image of the point cloud along the maximum face direction of the AABB of the particle point cloud, and save the pixel-point cloud data correspondence function.

[0030] Point cloud data can be represented as The corresponding pixels of the projected image are two-dimensional data, which can be represented as... Now assume that the particle point cloud data obtained by the 3D camera scan is as follows: A A point cloud containing n points can be represented by formula (1):

[0031] (1)

[0032] Point cloud data A Each point is read and stored by the computer, and each point is assigned a fixed location index. The projected image obtained by projection is: I It can be expressed as formula (2):

[0033] (2)

[0034] Point cloud data A With projected image I They all store the same mapping function: ,in Is it point cloud or pixels in point cloud data? Projected images The position in the middle, It refers to the amount of point cloud data. This is the first point cloud data. This is the second point cloud data. This is the third point cloud data. It is the first Point cloud data, It is the pixel corresponding to the first point cloud data. It is the pixel corresponding to the first point cloud data. It is the first Each point cloud data corresponds to a pixel.

[0035] The pixel corresponding to the point cloud data can be found based on the location information, and the point cloud data corresponding to the pixel can also be found through the location information. That is, the pixel-point cloud data correspondence function can be used to realize the conversion of the projected image to point cloud data.

[0036] (I) AABB is a cuboid with three distinct faces, corresponding to the XY, XZ, and YZ planes respectively. The area of ​​each face can be obtained by calculating its corresponding side length:

[0037] XY plane area:

[0038] XZ plane area:

[0039] Area of ​​YZ plane:

[0040] in, These are the maximum and minimum values ​​on the X-axis. These are the maximum and minimum values ​​on the Y-axis. These are the maximum and minimum values ​​on the Z-axis.

[0041] The largest surface is the largest surface in AABB. Generally speaking, in granular point cloud data, the largest surface is the XY surface.

[0042] (II) After determining the maximum surface, project the particle point cloud onto the XY plane along the Z-axis. Figure 3 This is a projected image obtained by projecting point cloud data. Similarly, the computer cannot distinguish particles within the image; the data is treated as a single entity. The projected point coordinates are mapped to a fixed-size two-dimensional grid (image coordinate system) using integer pixel indices. np.digitize The coordinates of each projection point are converted into a pixel index on the image. By mapping each projection point to a pixel position on the image, a binary image is finally generated, in which the pixel value of the projection point is 255, and the value of other pixels is 0.

[0043] S3. Segment the particle projection image using the watershed algorithm. The specific steps are as follows:

[0044] (I) Morphological Opening Operation: To remove noise from the image, the watershed algorithm applies a morphological opening operation (cv2.morphologyEx). This step operates on the image using a small convolutional kernel to remove small objects and noise, extracting larger, connected regions. The opening operation is essentially a process of first performing erosion and then dilation to remove small regions.

[0045] (II) Background and Foreground Generation: A distinct background region is generated using a dilation operation (cv2.dilate). This background region is used to mark the background portion in the watershed algorithm. Next, a distance transform (cv2.distanceTransform) is applied to calculate the distance from each pixel in the foreground region to the nearest background pixel. This helps determine the boundaries of the foreground region. Then, thresholding is used to transform these regions into foreground regions.

[0046] (III) Generating the labeled image: By subtracting the foreground region from the background, the uncertain parts of the image are obtained, which often become the boundaries of the watershed algorithm. The OpenCV function `cv2.connectedComponents` is used to generate labels for the foreground region. Then, each region of the labeled image is given a unique identifier for labeling each region in the watershed algorithm.

[0047] (IV) Applying the Watershed Algorithm: The watershed algorithm is applied to the image using OpenCV's `cv2.watershed` function. Using the previously generated labeled image, the watershed algorithm segments the image into multiple regions. The watershed algorithm works by finding watershed lines in the image based on the labeling information and the gradient in the image, thus segmenting the image into different regions. These regions are then labeled with different colors. In the projected image processed by the watershed algorithm, different particles are distinguished by color, thus completing the image segmentation task; each color represents one particle.

[0048] S4. Convert the segmented projection image into point cloud data using the pixel-point cloud correspondence.

[0049] After the watershed algorithm is executed, the projected image is distinguished into different colors, such as Figure 4 The red, yellow, and green colors in the image represent the same particle, thus completing the segmentation task of the projected image, meaning the computer can distinguish different particles. Now, the computer can determine the point cloud data for each segmented region based on the different colors used to label pixels. The specific steps are as follows:

[0050] (I) Obtain the label value of each pixel from the watershed segmentation results.

[0051] (II) Find the corresponding point cloud index through the pixel position, map each pixel position back to the point cloud index, and at the same time pass the color information of the pixel assigned by the watershed algorithm to the point cloud data.

[0052] (III) such as Figure 5 As shown, point cloud data for each segmented region is extracted based on the label values ​​of the watershed segmentation results and the pixel-point cloud correspondence.

[0053] Example 2

[0054] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0055] Example 3

[0056] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above method.

[0057] Example 4

[0058] A computer program product includes a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A non-destructive and rapid segmentation method for point clouds of slag particles, characterized in that, The method includes the following steps: S1: Read the particle point cloud file and calculate the axis-aligned bounding box of the particle point cloud; S2: Based on the axis-aligned bounding box, calculate the particle projection image of the point cloud in the direction of the largest face of the particle point cloud AABB, and save the pixel-point cloud correspondence. In S2, point cloud data A for The corresponding pixels of the projected image are two-dimensional data. The projected image is I ,in: （1） （2） The pixel-to-point cloud data correspondence function is: ; in, Is it point cloud or pixels in point cloud data? Projected images The position in the middle, It refers to the amount of point cloud data. This is the first point cloud data. This is the second point cloud data. This is the third point cloud data. It is the first Point cloud data, It is the pixel corresponding to the first point cloud data. It is the pixel corresponding to the first point cloud data. It is the first Each point cloud data corresponds to a pixel; S3: Segment the particle projection image using the watershed algorithm; S4: Convert the segmented projection image into point cloud data and output it using the pixel-point cloud correspondence formula.

2. The non-destructive and rapid segmentation method for point clouds of slag particles according to claim 1, characterized in that, S1 specifically refers to: S1.1: Point cloud data reading Read point cloud data, where each point cloud contains a series of three-dimensional coordinate points (x, y, z); S1.2: Finding the minimum and maximum coordinates For each coordinate axis, find the minimum and maximum values ​​of all points on that coordinate axis. The point corresponding to the minimum value and the point corresponding to the maximum value represent the minimum and maximum ranges of the point cloud data in each coordinate axis direction, that is, the two diagonal points of AABB. S1.3: AABB bounding box generation An AABB is an axis-aligned cuboid with 8 vertices, defined by its minimum and maximum boundaries. Specifically, given the minimum boundary of an AABB... and maximum boundary The eight vertices of the AABB cuboid are obtained by combining all the coordinates. , , , , , , , , representing all the angles of the bounding box in three-dimensional space.

3. The non-destructive and rapid segmentation method for point clouds of slag particles according to claim 1, characterized in that, S2 specifically refers to: S2.1: AABB is a cuboid with three faces, corresponding to the XY plane, XZ plane, and YZ plane respectively. Calculate the area of ​​each face: XY plane area: XZ plane area: YZ plane area: in, These are the maximum and minimum values ​​on the X-axis. These are the maximum and minimum values ​​on the Y-axis. The maximum and minimum values ​​on the Z-axis represent the largest and largest areas, and the face with the largest area is the largest face of the AABB. S2.2: Project the particle point cloud onto the largest surface along the Z-axis. The coordinates of the projected points are mapped to a fixed-size two-dimensional grid using integer pixel indices. np.digitize The coordinates of each projection point are converted into pixel indices on the image. By mapping each projection point to a pixel position on the image, a binary image is finally generated.

4. The non-destructive and rapid segmentation method for point clouds of slag particles according to claim 1, characterized in that, S3 specifically refers to: S3.1: Perform morphological opening operations on the image to remove small objects and noise, and extract larger, connected regions; S3.2: A background region is generated through dilation to mark the background part in the watershed algorithm. Then, a distance transformation is applied to calculate the distance from each pixel in the foreground region to the nearest background pixel. Finally, thresholding is used to convert these regions into foreground regions. S3.3: By subtracting the foreground region from the background, the uncertain parts of the image are obtained, which will become the boundaries of the watershed algorithm; the cv2.connectedComponents function of OpenCV is used to generate labels for the foreground region; each region of the labeled image will have a unique identifier, which is used to label each region in the watershed algorithm; S3.4: Use OpenCV's cv2.watershed function to apply the watershed algorithm to the image, dividing the image into multiple regions based on the previously generated labeled image; S3.5: First, initialize all markers to black. Then, for different markers, randomly generate colors and map them to the corresponding areas, and draw red boundaries between the marked areas.

5. The non-destructive and rapid segmentation method for point clouds of slag particles according to claim 4, characterized in that, S4 specifically refers to: S4.1: Obtain the label value of each pixel from the watershed segmentation results; S4.2: Find the corresponding point cloud index by pixel position and map each pixel position back to the point cloud index; S4.3: Extract point cloud data for each segmented region based on the label values ​​of the watershed segmentation results and the pixel-point cloud correspondence.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes a computer program to implement the steps of the non-destructive and rapid segmentation method for point clouds of slag particles as described in any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the steps of the non-destructive and rapid segmentation method for point clouds of slag particles as described in any one of claims 1-5.

8. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a processor, implement the steps of the non-destructive and rapid segmentation method for point clouds of slag particles as described in any one of claims 1-5.

Images