A method for pose detection of scattered irregularly shaped small workpieces
By combining high-definition industrial cameras with edge detection and polar coordinate transformation, the problems of large computational load and poor interpretability in the pose estimation of irregular workpieces are solved, realizing efficient and accurate pose recognition and grasping, and improving the processing efficiency and accuracy of the flexible feeding system.
Patent Information
- Application Number
- CN202410634871.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-16
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-05-16
AI Technical Summary
Existing technologies for flexible feeding and processing of multi-variety, small-batch, irregularly shaped workpieces involve large computational loads, high costs, and are difficult to interpret. The application of deep learning models is limited, making it difficult to achieve efficient and reliable pose estimation.
By employing a high-definition industrial camera combined with edge detection, logarithmic polar coordinate transformation, and 3D coordinate transformation, and by analyzing image features through deterministic rules, a local polar coordinate system is established to achieve precise measurement and grasping of workpiece pose.
It achieves high-precision, high-speed pose recognition for irregularly shaped small workpieces, reduces the dimensionality of data processing, provides intuitive and interpretable processing decisions, and improves the efficiency and accuracy of flexible automated assembly.
Smart Images

Figure CN118485718B_ABST
Abstract
Description
Technical Field
[0001] This invention mainly relates to the fields of computer vision and robotics, and in particular to a detection method for identifying workpiece features and determining the pose of scattered workpieces using image data analysis technology. Background Technology
[0002] In the field of automated production, there is a large demand for flexible material handling of multi-variety, small-batch, irregularly shaped workpieces, and automated assembly needs to be able to adapt to processing scenarios. The deep integration of measurement, recognition, and grasping technologies is key to solving common problems in intelligent processing. One of the key technologies for achieving intelligent robotic grasping is pose estimation. Currently, this mainly relies on 3D vision systems to collect data and uses point cloud matching technology to estimate the pose of randomly placed objects. Meanwhile, deep learning technology has made significant progress in computer vision, providing new technical ideas and foundations for perception technology in manufacturing. However, both of these methods typically suffer from drawbacks such as high computational load and high cost, hindering practical applications. For example, the currently popular deep neural network (DNN) black-box model is limited by its difficulty in interpretation and dependence on large amounts of labeled data, while industrial automation requires high reliability and time cost control. Based on the camera calibration method of Zhang Zhengyou, this invention designs robust image feature vectors for the attitude recognition of irregular small workpieces to effectively reduce the dimensionality of data processing. Furthermore, by establishing a local polar coordinate system, it further analyzes the fine-grained features of the image and adopts a rule-based modeling method to provide an intuitive and interpretable processing decision and control logic for a flexible feeding system for multiple varieties and small batches of irregular workpieces. Summary of the Invention
[0003] Scattered, irregularly shaped small workpieces are small in size and light in weight. The workpieces have little distinguishability in different postures and working conditions, and it is difficult for the human eye to judge their posture based on the workpiece features. This invention uses a high-definition industrial camera to acquire images and, based on deterministic rules, combines mathematical methods such as edge detection, logarithmic polar coordinate transformation, feature analysis, and three-dimensional coordinate transformation to accurately capture the pose of irregularly shaped small workpieces in three-dimensional space, assisting the robot arm in realizing the flexible feeding and grasping process in the automated processing. First, a flexible vibrating device is used to separate the scattered workpieces, and the workpieces are placed on the vibrating platform with a high probability of a stable posture with a large convex area contact. An industrial camera captures images of the scattered workpieces, and edge detection technology is used to extract the precise contours of the workpieces. Next, the orientation axis is determined based on the common features of the workpieces under different posture conditions, a local polar coordinate system is established, and the image data is processed using a logarithmic polar coordinate transformation method to standardize the posture representation of the workpieces and eliminate the complexity caused by rotation and scale changes. Then, through feature selection and analysis, deterministic rules are used to determine the posture of the workpieces under different working conditions. Finally, combined with depth information, the two-dimensional image coordinates of the workpieces are converted into three-dimensional spatial coordinates using three-dimensional coordinate transformation technology to achieve accurate measurement of the workpiece position and posture.
[0004] In a first aspect, the present invention discloses a method for detecting the pose of a miniature irregular workpiece, comprising the following steps:
[0005] Step 1: Fix the shooting platform, select the flexible vibrating platform, and align the camera with the flexible vibrating platform to ensure that the camera is perpendicular to the image acquisition plane. This step also includes camera calibration, calculating the transformation relationship between the image coordinate system and the camera coordinate system, and calculating the relationship between the camera coordinate system and the world coordinate system.
[0006] Step 2: Add the workpiece to the flexible vibrating tray, start the tray to vibrate to separate the workpiece, and start the industrial camera to capture images when the vibration stops. Perform image preprocessing, and use Gaussian filtering and Sobel operator to perform image enhancement and edge detection to obtain the edge information of the workpiece to be detected.
[0007] Step 3: Screening workpieces. Based on edge information, calculate the area and shape characteristics enclosed by the outer contour edge of each workpiece and compare it with a preset threshold to exclude workpieces with abnormal area or shape, remove workpieces with abnormal postures such as stacking and non-large convex area contacting the vibrating disc, select workpieces with large convex area contacting the vibrating disc as workpieces to be gripped for further analysis, and determine the position coordinates of the centroid of the circumscribed rectangle of each workpiece based on the outer contour of each workpiece to be gripped, and take the area enclosed by the outer contour as the region of interest (ROI).
[0008] Step 4: Feature localization. Utilize the common features of the workpiece under stable posture conditions to construct feature invariants, determine the orientation axis, establish a local polar coordinate system, and map the image to logarithmic polar coordinates. By analyzing the unique features of a single workpiece, accurately determine the workpiece's posture, and simultaneously calculate the workpiece posture confidence and the suitable gripping position for the robot arm.
[0009] Step 5: 3D information conversion. Based on the coordinates of the circumscribed rectangle centroid obtained in Step 3 and the workpiece posture information determined in Step 4, and combined with the coordinate system transformation relationship established in Step 1, the 2D image coordinate data is converted into a specific position and posture in 3D space.
[0010] Step 6: Based on confidence level and attitude classification, the robot arm sequentially grabs all workpieces in normal attitude in the vibrating tray, repeating steps 2 to 5 until all workpiece pose estimation and grabbing are completed.
[0011] Secondly, the present invention relates to a system for detecting the pose of scattered, irregularly shaped small workpieces, specifically designed for implementing the aforementioned pose detection method. The system includes:
[0012] Camera and calibration equipment: A high-resolution industrial camera is used to capture images of the workpiece. The camera is used in conjunction with specialized calibration equipment to correct camera lens distortion, ensure image quality, and calculate accurate transformation relationships between the camera coordinate system, the two-dimensional image coordinate system, and the world coordinate system.
[0013] Flexible vibrating feeder equipment includes a storage bin, a flexible vibrating plate, a feed tray and a feed tray frame. The material is manually fed into the storage bin, and the storage bin conveys the workpiece to the feed tray. The feed tray separates the scattered workpieces with a specific vibration mode and places the workpieces in a stable posture on the feed tray.
[0014] Image preprocessing unit: Responsible for performing image preprocessing operations, including Gaussian filtering, binarization, and edge detection. This unit uses algorithms such as the Sobel operator to accurately extract workpiece edge information, providing necessary data for subsequent calculations and analysis.
[0015] Feature extraction and analysis unit: Receives the output of the image preprocessing unit, extracts the outer contour data of the workpiece, analyzes its shape, size, and other fine-grained features, performs abnormal posture workpiece discrimination and removal, and performs local logarithmic polar coordinate transformation, feature localization, and other processing. This unit is also responsible for analyzing workpiece images with normal postures, identifying and calculating key feature points of the workpiece to establish an appropriate coordinate system, and then accurately determining the workpiece's posture in the logarithmic polar coordinate system.
[0016] 3D Transformation and Interface Unit: Based on the obtained workpiece posture and position information, and utilizing previously established coordinate system transformation relationships, this unit converts the workpiece's 2D image information into 3D spatial coordinates. This unit is also responsible for outputting the transformation results to downstream systems (such as robotic arm control systems) to guide the robotic arm in precise grasping and manipulation.
[0017] Beneficial Effects: The method for detecting the pose of scattered, irregularly shaped small workpieces provided by this invention utilizes defined rules to analyze the common and unique features of workpieces under different posture conditions, achieving high-precision and high-efficiency identification of workpiece posture. Through a series of sophisticated image processing and analysis steps, including edge detection, polar coordinate transformation, feature analysis, and 3D coordinate transformation, this method can accurately capture the subtle differences and complex postures of irregularly shaped small workpieces, simultaneously providing front and back view identification results and calculating the gripping position of the robotic arm. It can accurately convert the two-dimensional image information of the workpiece into three-dimensional spatial coordinates. This invention can significantly improve the processing efficiency and accuracy of workpieces in flexible automated assembly and quality control processes, while reducing production costs. Attached Figure Description
[0018] Figure 1 A step diagram of a method for detecting the pose of scattered, irregularly shaped small workpieces is shown.
[0019] Figure 2 This diagram illustrates the pose detection platform for scattered irregularly shaped small workpieces according to an example of the present invention, as well as the relationship between the various coordinate systems.
[0020] Figure 3 This invention illustrates a flowchart of the process by which scattered workpieces undergo attitude determination and flow, as exemplified by the present invention.
[0021] Figure 4 This diagram illustrates an example of establishing a local polar coordinate system based on a single workpiece according to an example of the present invention.
[0022] Figure 5 This diagram illustrates the fine-grained feature analysis of workpieces in different postures under the local coordinate system of this invention.
[0023] Figure 6 The following diagram shows an example of the detection results of the present invention;
[0024] See Figure 2 201 is a robotic arm, 202 is an industrial camera, 203 is a material tray and tray frame, 204 is a flexible vibrating plate, and 205 is a storage bin.
[0025] See Figure 4 , Figure 5The workpiece is an example of an embodiment of the present invention. Image 401 is the front view of the workpiece, image 402 is the back view of the workpiece, image 403 is the center of the circular hole in the workpiece, image 404 is the centroid (polar point) of the circumscribed rectangle, image 405 is the polar coordinate axis of the local polar coordinate system, image 406 is the horizontal axis passing through the polar point of the local polar coordinate system, image 407 is the Cartesian coordinate system xoy of the image plane, images 501 and 502 are edge data extracted within the range of 0-π and π-2π in the local coordinate system of image 401, respectively, and images 503 and 504 are edge data extracted within the range of 0-π and π-2π in the local coordinate system of image 402, respectively. Common features on workpieces with different postures include the circular hole and the centroid coordinates of the circumscribed rectangle. Using the centroid coordinates of the circumscribed rectangle of a single workpiece as the polar point, a ray is drawn from the polar point to the center of the hole, serving as the polar axis, to construct a local polar coordinate system for the single workpiece. The workpiece area is divided into two parts by rotating 180 degrees clockwise around the polar axis. The features are aligned and the feature data of the two parts are compared and analyzed to determine the front and back of the workpiece.
[0026] See Figure 6 The image shows the workpiece detection result of an embodiment of the present invention. "front" indicates that the front of the workpiece is facing up, "back" indicates that the back of the workpiece is facing up, 601 is the centroid position x of the outer rectangle of the single workpiece, and 602 is the calculated appropriate position for workpiece gripping (in this example, the suction method is used). Detailed Implementation
[0027] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0028] See Figure 1 As shown in the step diagram, this invention proposes a method for pose detection of scattered, irregularly shaped small workpieces, including the following steps:
[0029] Step S101: Install an industrial camera directly above the center of the flexible vibrating disc, while a robotic arm waits to grab the material above and to the side of the flexible vibrating disc; calculate the transformation matrix between the planar two-dimensional image coordinate system and the world coordinate system.
[0030] The camera was calibrated using Zhang Zhengyou's calibration method, and the height d of the camera and the flexible vibrating disc, the rotation matrix R, and the translation vector T were determined.
[0031] The formula for transforming a 2D image to the camera coordinate system is:
[0032]
[0033] Where (u, v) are the image pixel coordinates, (x, y, d) are the coordinates in the camera coordinate system, and K is the camera intrinsic parameter matrix.
[0034] Secondly, the transformation formula from the camera coordinate system to the world coordinate system is:
[0035]
[0036] Where (X, Y, Z) are the coordinates in the world coordinate system, R is a 3×3 rotation matrix, and T is a 3×1 translation vector.
[0037] Step S102: Add workpieces to the flexible vibrating feed pan and start the pan to vibrate to separate the workpieces. After the vibration stops, start the industrial camera to capture images of the scattered workpieces and perform preprocessing on the images. The preprocessing process includes Gaussian filtering, Sobel operator, and binarization to enhance the image and detect the edge information of the workpieces to be detected.
[0038] Step S103: Screen workpieces with normal posture. Calculate the area enclosed by the outer contour edge of each workpiece in the image, and compare each area value with the set threshold Thresh. Further analyze the outer contour shape, remove stacked workpieces and workpieces that do not have large convex surface contact with the vibratory feeder, and select workpieces with large convex surface contact with the vibratory feeder as the workpieces to be grasped, and perform further posture analysis.
[0039] This step sets a single workpiece region pixel area threshold of Thresh to eliminate workpieces with abnormal postures. In this embodiment, when the workpiece region area is higher than high_thresh = 1.2 * Thresh, it is determined to be a workpiece stacking condition; if it is lower than the low_thresh = 0.8 * Thresh threshold, it may be a workpiece with a non-large convex area contacting the vibratory feeder and an unstable posture. Further analysis is performed by calculating the outer contour shape of the workpiece. For those that do not meet the conditions, they are determined to be workpieces with abnormal postures, and further analysis is performed after the next vibration of the vibratory feeder changes their posture.
[0040] For the screened workpieces, the center of the rectangle circumscribed by the outer contour is the center of each workpiece, and the region enclosed by the largest contour as the radius is set as the region of interest (ROI).
[0041] Step S104: Establish a local polar coordinate system based on the common features of workpieces in different postures, transform the workpiece coordinates to logarithmic polar coordinates, compare the features of workpieces in different postures, determine the front and back postures of the workpiece, and calculate the local coordinates for workpiece grasping, including:
[0042] The invention determines the centroid of the circumscribed rectangle of a single workpiece and identifies common features on the workpiece. In this example, a circular hole on the workpiece is used as a common feature. The center of the circular hole is located, and the coordinates of the centroid of the circumscribed rectangle of the single workpiece are taken as the pole. A ray is drawn from the pole to the center of the hole as the polar axis, and the angle is defined as positive in the clockwise direction. A local polar coordinate system for the single workpiece is constructed. Then, the natural logarithm of the polar radius coordinate is taken, denoted as (r, θ), to normalize the workpiece dimensions. Figure 4 As shown.
[0043] Furthermore, in logarithmic polar coordinates, the image data of a single workpiece is analyzed to identify its unique features; the outer contour edge data of the workpiece is extracted clockwise from the 0-π angle, and the area enclosed by its outer contour data and polar diameter is calculated as Area_half_one; the edge data of the workpiece is extracted clockwise from the π-2π angle, and the area enclosed by its outer contour data and polar diameter is calculated as Area_half_other. The ratio of the two parts is:
[0044] ratio=Area_half_one / Area_half_other
[0045] Determine the front and back orientation of the workpiece according to the following rules and calculate the confidence level:
[0046]
[0047] Among them, conf front conf back The confidence scores are for the face-up and face-down orientations, respectively, and the confidence score is conf. front conf back Calculate the condition that satisfies the condition with th respectively gt ,th lt Let Δ1 and Δ2 be the mean, and let Δ1 and Δ2 be the deviations, which are normally distributed.
[0048]
[0049] Where N = {conf front conf back}, u={th gt ,th lt}, σ={Δ1,Δ2}.
[0050] Through calculation, the reference values for this embodiment of the invention are: the front face ratio ∈ [1.25, 1.35], the back face ratio ∈ [0.67, 0.77], that is, Δ1 ∈ (0, 0.28), Δ2 ∈ (0, 0.28), th gt =1.3,th lt =0.72.
[0051] The workpiece orientation is determined as either face up or back up. Figure 6 As shown. In this example of the invention, the grab position is set to a local coordinate system. Place.
[0052] The image information of the workpiece is transformed into a logarithmic polar coordinate system. The logarithmic polar coordinate system and the Cartesian coordinate system have the following relationship:
[0053]
[0054] In the formula: (x o y o Let r be the geometric center of the workpiece, r be the radial coordinate in the logarithmic coordinate system, and θ be the polar coordinate in the logarithmic coordinate system. The angle between the polar coordinate axis and the horizontal line is calculated to obtain the workpiece rotation amount β, which can provide a reference rotation angle for the robot to grasp the workpiece.
[0055] Step S105: 3D information conversion. Based on the workpiece posture and position information obtained in step S104, combined with the coordinate system transformation relationship established in step S101 and the inverse of the camera intrinsic parameter matrix, the local_grab coordinates in the local polar coordinate system of the 2D image are converted into the specific position Globle_grab(x′, y′) in the world coordinate system using the coordinates (x, y) of the centroid point of the circumscribed rectangle of the workpiece obtained in step S103. Given the camera height and workpiece thickness, the workpiece height Highth can be calculated, or the workpiece height data can be obtained using laser ranging or other methods. Finally, the pose of the robot arm grasping the workpiece is calculated through the inverse kinematics of the robot.
[0056] Step S106: Based on confidence level and attitude classification, the robot sequentially grabs all workpieces in normal attitude in the vibrating tray, repeating steps S102-S105 until all workpiece pose estimation and grabbing are completed.
[0057] In practical applications, the rotation θ of a single workpiece in its local coordinate system can be fine-tuned based on the specific workpiece shape and material, as well as the characteristics of the gripping device, to determine a stable gripping position (Globle_grab) to adapt to the specific needs of different scenarios. For different types of workpieces, this invention can switch discrimination conditions and comprehensively apply features such as contour, area, texture, and grayscale as criteria for posture classification. Based on common features, a local polar coordinate system for a single workpiece is established, which can eliminate the influence of workpiece rotation. By utilizing the principle of workpiece rotation and mirroring, fine-grained local feature data of the image is obtained in different regions, significantly reducing the dimensionality of scattered workpiece image data and quickly providing workpiece pose detection results, thereby improving production efficiency.
[0058] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.
Claims
1. A method for pose detection of scattered, irregularly shaped small workpieces, characterized in that, include: Step 1: Fix the shooting platform, select the flexible vibrating platform, and align the camera with the flexible vibrating platform to ensure that the camera is perpendicular to the image acquisition plane. This step also includes camera calibration, calculating the transformation relationship between the image coordinate system and the camera coordinate system, and calculating the relationship between the camera coordinate system and the world coordinate system. Step 2: Add the workpiece to the flexible vibrating tray, start the tray to vibrate to separate the workpiece, and start the industrial camera to capture images when the vibration stops. Perform image preprocessing, and use Gaussian filtering and Sobel operator to perform image enhancement and edge detection to obtain the edge information of the workpiece to be detected. Step 3: Workpiece screening. Based on edge information, the area and shape characteristics enclosed by the outer contour edges of each workpiece are calculated and compared with a preset threshold to exclude workpieces with abnormal area or shape. Workpieces with abnormal postures, such as stacking or non-large convex areas contacting the vibrating disc, are removed. Workpieces with large convex areas contacting the vibrating disc are selected as the workpieces to be gripped for further analysis. The coordinates of the centroid of the circumscribed rectangle of each workpiece are determined based on its outer contour, and the area enclosed by the outer contour is designated as the Region of Interest (ROI). Step 4: Feature localization. Utilize the common features of the workpiece under stable posture conditions to construct feature invariants, determine the orientation axis, establish a local polar coordinate system, and map the image to logarithmic polar coordinates. By analyzing the unique features of a single workpiece, accurately determine the workpiece's posture, and simultaneously calculate the workpiece posture confidence and the suitable gripping position for the robot arm. Specifically, a local polar coordinate system is established based on the common features of the workpiece, and the workpiece information is transformed into logarithmic polar coordinates for feature alignment, including: A local polar coordinate system is established based on the common features of workpieces in different postures, and the workpiece coordinates are transformed to a logarithmic polar coordinate system. By comparing the features of workpieces in different postures under the local polar coordinate system, the front and back postures of the workpiece are determined, and the local coordinates for workpiece grasping are calculated, including: Determine the centroid of the circumscribed rectangle of a single workpiece and identify common features on the workpiece. Locate the center of the common features. Using the coordinates of the centroid of the circumscribed rectangle of the single workpiece as the pole, the ray from the pole to the center of the common features is taken as the polar axis. The angle is defined as positive in the clockwise direction. Construct a local polar coordinate system for the single workpiece and take logarithmic polar coordinates (r, θ) to normalize the workpiece dimensions. Furthermore, in logarithmic polar coordinates, the image data of a single workpiece is analyzed to identify its unique features; workpiece edge data is extracted clockwise from an angle of 0-π, and its outer contour data and the area enclosed by its polar diameter are calculated as Area_half_one; workpiece edge data is extracted clockwise from an angle of π-2π, and its outer contour data and the area enclosed by its polar diameter are calculated as Area_half_other. The ratio of the two parts is: ratio=Area_half_one / Area_half_other Determine the front and back orientation of the workpiece according to the following rules and calculate the confidence level: wherein, conf front , conf back are the confidence levels of the front-up and back-up postures, respectively, and the confidence levels conf front , conf back are calculated to satisfy a normal distribution with th gt , th lt as the mean values and Δ1, Δ2 as the deviations, respectively; where N={conf front , conf back}, u={th gt , th lt}, σ={Δ1, Δ2}; When the image information of the workpiece is transformed into logarithmic polar coordinates, the logarithmic polar coordinates and the Cartesian coordinate system have the following relationship: In the formula: (x o y o Let r be the geometric center of the workpiece, r be the radial coordinate in the logarithmic coordinate system, and θ be the polar coordinate in the logarithmic coordinate system. The workpiece rotation β is obtained by calculating the angle between the polar coordinate axis and the horizontal line, which can provide a reference rotation angle for the robot to grasp the workpiece. Step 5: 3D information conversion. Based on the coordinates of the circumscribed rectangle centroid obtained in Step 3 and the workpiece posture information determined in Step 4, and combined with the coordinate system transformation relationship established in Step 1, the 2D image coordinate data is converted into a specific position and posture in 3D space. Step 6: Based on confidence level and attitude classification, the robot arm sequentially grabs all workpieces in normal attitude in the vibrating tray, repeating steps 2-5 until all workpiece pose estimation and grabbing are completed.
2. The method for detecting the pose of scattered, irregularly shaped small workpieces according to claim 1, characterized in that... In step 1, the two-dimensional plane coordinate system is transformed to obtain the transformation matrix, including: An industrial camera is installed directly above the center of the flexible vibrating disc, and a robotic arm waits to grab the material above and to the side of the flexible vibrating disc. The camera is calibrated to determine the height d between the camera and the flexible vibrating disc, the rotation matrix R, and the translation vector T. The formula for transforming a 2D image to the camera coordinate system is: Where (u, v) are the image pixel coordinates, (x, y, d) are the coordinates in the camera coordinate system, and K is the camera intrinsic parameter matrix; Secondly, the transformation formula from the camera coordinate system to the world coordinate system is: Where (X, Y, Z) are the coordinates in the world coordinate system, R is a 3×3 rotation matrix, and T is a 3×1 translation vector.
3. The method for detecting the pose of scattered, irregularly shaped small workpieces according to claim 1, characterized in that... Step 2, preprocess the image to obtain workpiece edge information in the image, including: Workpieces are added to a flexible vibrating feed pan, and the pan is started to vibrate to separate the workpieces. After the vibration stops, an industrial camera is started to capture images of the scattered workpieces, and the images are preprocessed. The preprocessing process includes Gaussian filtering, Sobel operator, and binarization to enhance the images and detect the edge information of the workpieces to be detected.
4. The method according to claim 1, characterized in that... In step 3, the edge information of the workpieces in the image is analyzed to obtain the Region of Interest (ROI) for each workpiece, including: Select workpieces in normal posture, calculate the area enclosed by the outer contour edge of each workpiece in the image, compare each area value with the set threshold Thresh, analyze the outer contour shape, remove stacked workpieces and workpieces that do not have large convex surface contact with the vibratory feeder, select workpieces with large convex surface contact with the vibratory feeder as the workpieces to be grasped, and perform further posture analysis. This step sets two thresholds, high_thresh and low_thresh, for comparison with the pixel area of a single workpiece region to eliminate workpieces with abnormal postures. When the workpiece region area is higher than high_thresh, it is determined to be a workpiece stacking condition. When it is lower than low_thresh, it is determined to be a non-large convex area contacting the vibrating plate, i.e., an unstable posture. Further, the outer contour shape of the workpiece is calculated and analyzed. Workpieces that do not meet the conditions are determined to be abnormal and wait for the next vibration of the vibrating plate to change their posture before further analysis. For the screened workpieces, the center of the rectangle circumscribed by the outer contour is the center of each workpiece, and the area within the radius of the largest contour is set as the region of interest (ROI).
5. The method for detecting the pose of scattered, irregularly shaped small workpieces according to claim 1, characterized in that... Step 5: After obtaining the workpiece posture, convert the two-dimensional information into a specific position and posture in three-dimensional space, including: The 3D information conversion is performed based on the workpiece posture and position information obtained in step 4, combined with the coordinate system transformation relationship established in step 1 and the inverse of the camera intrinsic parameter matrix. Using the coordinates (x, y) of the centroid point of the circumscribed rectangle of the image workpiece obtained in step 3, the grab coordinates Local_grab in the local polar coordinate system of the 2D image are converted into the specific position Globle_grab(x′, y′) in the world coordinate system. The camera height and workpiece thickness are known, so the height of the workpiece to be grasped (Highth) can be calculated, or the workpiece height data can be obtained using laser ranging. Finally, the pose of the robot arm grasping the workpiece is calculated by inverse kinematics of the robot.
6. A pose detection system for scattered irregularly shaped small workpieces, characterized in that it implements the pose detection method for scattered irregularly shaped small workpieces as described in claim 1, comprising: Industrial cameras, after precise camera calibration, are used to capture images of scattered workpieces, providing crucial two-dimensional image data; The robotic arm, after determining the precise gripping position of the workpiece, performs a precise gripping operation on the workpiece; The tray and tray frame are used to place the workpiece and serve as a backing for image acquisition. Selecting a backing that is significantly different from the workpiece helps the camera clearly distinguish the edge features of the workpiece. The tray frame is used to prevent workpieces from shaking off and falling off the tray; Flexible vibratory feeders have a vibration function and are used to separate stacked workpieces and change their posture to ensure that more workpieces are in a normal gripping posture. The storage bin serves as the initial storage for workpieces. It is manually replenished, and the storage bin delivers the workpieces to the material tray.
