Bundled angle steel top identification method and identification and grabbing device

Abstract

The application discloses a kind of whole bundle angle steel top identification method, comprising the following steps, trigger photographing step: PLC control truss moves to the preset high position photographing point above material frame, sends trigger signal.Data acquisition step: camera gathers the depth map and infrared map of stacking area, generates 3D point cloud data.Algorithm processing step: the point cloud is segmented, identified and pose estimation by IPC, and the best grabbing target coordinates are output.The kind of identification method has the following beneficial effects: a kind of "3D dimensionality reduction analysis-2D feature filtering-3D pose inverse calculation" hybrid algorithm is proposed, through to candidate point cloud is sliced with multiple sections and morphological classification, accurately identifies single angle steel, solves the problem of false grabbing under complex stacking environment.

Description

Technical Field

[0001] This invention relates to the field of angle steel equipment, and more particularly to a method for identifying the top of a bundle of angle steel and an identification and gripping device. Background Technology

[0002] Angle steel, as an important structural steel material, is widely used in the manufacture of power transmission towers, bridges, and building frames. In traditional production logistics, the loading, unloading, and unstacking of angle steel mainly rely on teach-pen robotic arms, which poses high safety risks and poor flexibility. Now, with the development of intelligent manufacturing, more and more robotic arms are beginning to use machine vision-based "bin-picking" to grasp angle steel. However, this grasping method faces the following problems in unstacking angle steel: First, angle steel has an L-shaped cross-section, which makes it easy for them to interlock and nest when stacked, making it impossible for traditional 2D image recognition to segment individual pieces; second, the metal surface has strong specular reflection, resulting in missing or noisy 3D point cloud data. Summary of the Invention

[0003] To address the above-mentioned problems, this invention proposes a method for identifying the top of a bundle of angle steel.

[0004] The technical solution adopted in this invention is as follows:

[0005] A method for identifying the top of a bundle of angle steel includes the following steps:

[0006] Triggering the photo-taking process: The PLC controls the gantry to move to the preset high-position photo-taking point above the material frame and sends a trigger signal.

[0007] Data acquisition steps: The camera acquires depth and infrared images of the stacked area to generate 3D point cloud data. .

[0008] Algorithm processing steps: IPC performs point cloud segmentation, recognition, and pose estimation, and outputs the optimal target coordinates for grasping. .

[0009] 2. Key Algorithms (Methodology) for Point Cloud Processing

[0010] To address the complexity of stacked angle steel, this method designs a layered processing algorithm.

[0011] 2.1 Point Cloud Preprocessing and ROI Extraction

[0012] The original point cloud contains background noise from the material frame, ground, etc. First, a pass-through filter is used to limit the noise based on the physical dimensions of the material frame. Range, extracting the region of interest (ROI):

[0013]

[0014] Subsequently, statistical outlier removal filter (SOR) was used to remove discrete noise caused by metallic reflection.

[0015] 2.2 Coarse Matching Based on Edge Features

[0016] The most prominent geometric feature of angle steel is its longitudinal ridge. This is determined by calculating the surface normal of the point cloud. and curvature Set curvature threshold Extract the set of high-curvature points, i.e., suspected edge points. Use Euclidean clustering to segment the discontinuous edge points. candidate clusters Each cluster represents a potential individual angle steel.

[0017] 2.3 Fine Matching Based on Multi-Cross-Section Average Filtering

[0018] This is the core algorithm of this method. Since coarse clustering may contain two overlapping angle steels, cross-sectional analysis is required to distinguish them.

[0019] 2.3.1 Multi-section slice

[0020] For each candidate cluster Along its principal component direction (PCA principal axis), at intervals of Cut a vertical section Assume a total of [number] segments are captured. One cross section.

[0021] 2.3.2 Cross-sectional projection and averaging filtering

[0022] cross section The 3D points within the point cloud are projected onto a local 2D plane. Due to noise in the point cloud, direct contour fitting is prone to distortion. This method applies a sliding window averaging filter to the cross-sectional contour points:

[0023]

[0024] in The window size is specified. The filtered outline is smoother, preserving the true "L" or "V" shaped features of the angle steel.

[0025] 2.3.3 Dual-channel deep learning cross-section classification based on attitude decoupling

[0026] To further improve the robustness of recognition under complex working conditions, this method proposes a dual-channel classification strategy based on pose prior. Considering the significant differences between the illumination reflection characteristics and point cloud missing patterns of positive V-shaped (opening downwards) and negative V-shaped (opening upwards) angle steel during imaging, this method constructs and trains two independent binary classification convolutional neural network models: a positive V-shaped discrimination model and a negative V-shaped (opening upwards) discriminant model. Compared with the inverse V discriminant model Attitude prediction and model routing:

[0027] For the candidate cluster of angle steel obtained from coarse matching First, calculate the average direction of its principal normal vector. .like Pointing outwards and upwards, it is initially determined to be an inverted V-shape. The cross-sectional slice is then input into the model. .like Pointing to the lower side, it is initially determined to be a positive V-shape. The cross-sectional slice is then input into the model. This routing mechanism effectively reduces the feature space dimension of a single model, allowing the model to focus on distinguishing between "single roots" and "overlapping / noise".

[0028] Dual-model architecture and classification:

[0029] Both models employ a lightweight CNN architecture, with input being a binarized image generated from a cross-sectional point cloud.

[0030] Positive V-model Output category is .

[0031] Inverse V-model Output category is .

[0032] Model output layer after Processing to obtain classification confidence scores .

[0033] 2.4 3D Pose Estimation and Coordinate Transformation

[0034] Loading CAD point cloud template for standard angle steel The Iterative Closest Point (ICP) algorithm is used to select the template. With target point cloud Perform fine registration to minimize the error function:

[0035]

[0036] Solve for the rotation matrix Translation vector By combining the hand-eye calibration matrix, the grasping posture in the truss coordinate system is calculated.

[0037] Among the above methods, a hybrid algorithm of "3D dimensionality reduction analysis - 2D feature filtering - 3D pose inverse calculation" is proposed. By performing multi-section slicing and morphological classification on the candidate point cloud, it can accurately identify single angle steel and solve the problem of misgrabbing in complex stacking environment.

[0038] A gripping device suitable for the top identification method of a bundle of angle steel as described above includes a three-axis truss robot, a 3D structured light camera, and an end effector. The end effector and the 3D structured light camera are both mounted on the three-axis truss robot and are electrically connected to a host controller.

[0039] After receiving the optimal target coordinates, the host controller (which can be a PLC) controls the gantry to perform the picking, lifting, and placement actions.

[0040] Optionally, the three-axis gantry robot is equipped with multiple end effector manipulators.

[0041] The beneficial effects of this invention are: it proposes a hybrid algorithm of "3D dimensionality reduction analysis - 2D feature filtering - 3D pose inverse calculation", which accurately identifies single angle steel by performing multi-section slicing and morphological classification on candidate point clouds, thus solving the problem of misgrabbing in complex stacking environments. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0043] Figure 1 This is a simplified schematic diagram of the gripping device.

[0044] The attached figures are labeled as follows: 1. Three-axis gantry robot; 2. 3D structured light camera; 3. End effector; 4. Angle steel. Detailed Implementation

[0045] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0046] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0047] In the description of this application, it should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. For ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items, and therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0048] Example 1

[0049] A method for identifying the top of a bundle of angle steel includes the following steps:

[0050] Triggering the photo-taking process: The PLC controls the gantry to move to the preset high-position photo-taking point above the material frame and sends a trigger signal.

[0051] Data acquisition steps: The camera acquires depth and infrared images of the stacked area to generate 3D point cloud data. .

[0052] Algorithm processing steps: IPC performs point cloud segmentation, recognition, and pose estimation, and outputs the optimal target coordinates for grasping. .

[0053] 3. Key Algorithms (Methodology) for Point Cloud Processing

[0054] To address the complexity of stacked angle steel, this method designs a layered processing algorithm.

[0055] 3.1 Point Cloud Preprocessing and ROI Extraction

[0056] The original point cloud contains background noise from the material frame, ground, etc. First, a pass-through filter is used to limit the noise based on the physical dimensions of the material frame. Range, extracting the region of interest (ROI):

[0057]

[0058] Subsequently, statistical outlier removal filter (SOR) was used to remove discrete noise caused by metallic reflection.

[0059] 3.2 Coarse Matching Based on Edge Features

[0060] The most prominent geometric feature of angle steel is its longitudinal ridge. This is determined by calculating the surface normal of the point cloud. and curvature Set curvature threshold Extract the set of high-curvature points, i.e., suspected edge points. Use Euclidean clustering to segment the discontinuous edge points. candidate clusters Each cluster represents a potential individual angle steel.

[0061] 3.3 Fine Matching Based on Multi-Cross-Section Average Filtering

[0062] This is the core algorithm of this method. Since coarse clustering may contain two overlapping angle steels, cross-sectional analysis is required to distinguish them.

[0063] 3.3.1 Multi-section slice

[0064] For each candidate cluster Along its principal component direction (PCA principal axis), at intervals of Cut a vertical section Assume a total of [number] segments are captured. One cross section.

[0065] 3.3.2 Cross-sectional projection and averaging filtering

[0066] cross section The 3D points within the point cloud are projected onto a local 2D plane. Due to noise in the point cloud, direct contour fitting is prone to distortion. This method applies a sliding window averaging filter to the cross-sectional contour points:

[0067]

[0068] in The window size is specified. The filtered outline is smoother, preserving the true "L" or "V" shaped features of the angle steel.

[0069] 3.3.3 Dual-channel deep learning cross-section classification based on attitude decoupling

[0070] To further improve the robustness of recognition under complex working conditions, this method proposes a dual-channel classification strategy based on pose prior. Considering the significant differences between the illumination reflection characteristics and point cloud missing patterns of positive V-shaped (opening downwards) and negative V-shaped (opening upwards) angle steel during imaging, this method constructs and trains two independent binary classification convolutional neural network models: a positive V-shaped discrimination model and a negative V-shaped (opening upwards) discriminant model. Compared with the inverse V discriminant model Attitude prediction and model routing:

[0071] For the candidate cluster of angle steel obtained from coarse matching First, calculate the average direction of its principal normal vector. .like Pointing outwards and upwards, it is initially determined to be an inverted V-shape. The cross-sectional slice is then input into the model. .like Pointing to the lower side, it is initially determined to be a positive V-shape. The cross-sectional slice is then input into the model. This routing mechanism effectively reduces the feature space dimension of a single model, allowing the model to focus on distinguishing between "single roots" and "overlapping / noise".

[0072] Dual-model architecture and classification:

[0073] Both models employ a lightweight CNN architecture, with input being a binarized image generated from a cross-sectional point cloud.

[0074] Positive V-model Output category is .

[0075] Inverse V-model Output category is .

[0076] Model output layer after Processing to obtain classification confidence scores .

[0077] 3.4 3D Pose Estimation and Coordinate Transformation

[0078] Loading CAD point cloud template for standard angle steel The Iterative Closest Point (ICP) algorithm is used to select the template. With target point cloud Perform fine registration to minimize the error function:

[0079]

[0080] Solve for the rotation matrix Translation vector By combining the hand-eye calibration matrix, the grasping posture in the truss coordinate system is calculated.

[0081] Example 2

[0082] A gripping device suitable for the top identification method of a bundle of angle steel as shown in Embodiment 1 includes a three-axis truss robot 1, a 3D structured light camera 2, and an end effector 3. The end effector 3 and the 3D structured light camera 2 are both mounted on the three-axis truss robot 1, and both the end effector 3 and the 3D structured light camera 2 are electrically connected to a host controller (not shown in the figure, which performs the calculation process as shown in Embodiment 1).

[0083] After receiving the optimal coordinates of the grasping target, the host controller (which can be a PLC) controls the gantry robot to perform the picking, lifting, and placing actions. Specifically, the end effector robot 3 performs the action of grasping the angle steel 4.

[0084] The three-axis gantry robot is equipped with multiple end effector manipulators.

[0085] The above-described embodiments only illustrate some aspects of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying the top of a bundle of angle steel, characterized in that, Includes the following steps The triggering process involves sending a trigger signal to the camera, which then begins operation upon receiving the trigger signal. Data acquisition involves using cameras to capture depth and infrared images of the angle steel stacking area and generating 3D point cloud data. ; The algorithm processes the point cloud, performs segmentation, recognition, and pose estimation, and outputs the optimal target coordinates for grasping. .

2. The method for identifying the top of a bundle of angle steel according to claim 1, characterized in that, In the algorithm processing steps, the original point cloud includes the material frame and ground background noise. A pass-through filter is used to limit the X, Y, and Z ranges based on the physical dimensions of the material frame, and the region of interest is extracted. Subsequently, a statistical outlier removal filter was used to remove discrete noise caused by metal reflection.

3. The method for identifying the top of a bundle of angle steel according to claim 1, characterized in that, The algorithm processing steps also include calculating the surface normal of the point cloud. and curvature Set curvature threshold Extract the high-curvature point set, i.e., the suspected edge points, and use Euclidean clustering to segment the discontinuous edge points into... candidate clusters Each cluster represents a potential individual angle steel.

4. The method for identifying the top of a bundle of angle steel according to claim 3, characterized in that, For each candidate cluster Along its principal component direction at intervals Cut a vertical section A total of One cross section; Project the 3D points within section S_j onto a local 2D plane; apply a sliding window averaging filter to the section contour points. .

5. The method for identifying the top of a bundle of angle steel according to claim 3, characterized in that, For the candidate cluster of angle steel obtained from coarse matching First, calculate the average direction of its principal normal vector. ,like Pointing outwards and upwards, it is initially determined to be an inverted V-shape. The cross-sectional slice is then input into the model. ,like Pointing to the lower side, it is initially determined to be a positive V-shape. The cross-sectional slice is then input into the model. The model output layer is processed to obtain the classification confidence score. .

6. The method for identifying the top of a bundle of angle steel according to claim 1, characterized in that, The algorithm processing steps also include loading the CAD point cloud template of the standard angle steel. The Iterative Closest Point (ICP) algorithm is used to select the template. With target point cloud Perform fine registration to minimize the error function: ； Solve for the rotation matrix Translation vector The calibration matrix is ​​used to calculate the grasping pose.

7. A gripping device applicable to the top identification method of a bundle of angle steel as described in any one of claims 1 to 6, characterized in that, It includes a three-axis gantry robot, a 3D structured light camera, and an end effector. The end effector and the 3D structured light camera are both mounted on the three-axis gantry robot and are electrically connected to a host controller.

8. The gripping device according to claim 7, characterized in that, The three-axis gantry robot is equipped with multiple end effector manipulators.

Images